Hazell, Hobler, Do Deficits Cause Inflation? A High Frequency Narrative Approach (LSE CFM Discussion Paper 2025-05, 2025)
Summary
A paper dated February 7, 2025, titled Do Deficits Cause Inflation? A High Frequency Narrative Approach, by Jonathon Hazell and Stephan Hobler. The abstract states that the paper measures the causal effect of deficits on inflation by identifying an event that released news about the 2021 deficits in the United States, the Georgia Senate election runoff, and by calculating the size of the shock from narrative data from investment banks. It reports an estimated "inflation multiplier" of 0.19% price level growth over two years for a 1% deficit-to-GDP shock. The abstract states that this estimate implies the 2021 deficits caused around a third of the 2021-22 inflation.
Summary drafted by a model from the document's text below and checked by script against that text before publication. It is a navigation aid, not a reading of what the document proves. Where AI is used
Cite this document
“Hazell, Hobler, Do Deficits Cause Inflation? A High Frequency Narrative Approach (LSE CFM Discussion Paper 2025-05, 2025).”
Original at www.lse.ac.uk. Archived at Pandemic Darlings, https://pandemicdarlings.com/source-documents/hazell-hobler-do-deficits-cause-inflation-a-high-frequency-narrative-a-102da54914/
, SHA-256 e2b874cdda389e3485c76eb14eb2728c1eba8d36c8840df0719844ca600aa0c0.Full text
Do Deficits Cause Inflation?
A High Frequency Narrative Approach*
Jonathon Hazell Stephan Hobler
February 7, 2025
Abstract
This paper measures the causal effect of deficits on inflation using a “high frequency
narrative approach”. We identify an event that released news about the 2021 deficits in the
United States—the Georgia Senate election runoff. We calculate the size of the shock using
new narrative data from investment banks. We then study the high frequency response of
inflation forecasts from asset prices, in order to separate deficits from other factors affecting
inflation. We estimate an “inflation multiplier” of 0.19% price level growth over two years,
for a 1% deficit-to-GDP shock. Our estimate implies that the 2021 deficits caused around a
third of the 2021-22 inflation—meaning deficits were important but not the only cause. A
heterogeneous agent New Keynesian model quantitatively matches the size and dynamics
of the inflation multiplier.
* First draft: February 2024. Hazell: London School of Economics (email: j.hazell@lse.ac.uk); Hobler: Lon-
don School of Economics (s.j.hobler@lse.ac.uk). We thank Mark Aguiar for early conversations that inspired this
project, as well as Silvia Ardagna, Saleem Bahaj, Gabriel Chodorow-Reich, Thomas Drechsel, Salomé Fofana, Niels
Gormsen, Yuriy Gorodnichenko, Steve Hou, Ethan Ilzetzki, Xavier Jaravel, Gwen Jing, Danial Lashkari, Chen Lian,
Hanno Lustig, Matthew Luzzetti, Atif Mian, Michael McMahon, Robbie Minton, Benjamin Moll, Dmitry Mukhin,
Emi Nakamura, Ivan Petrella, Carolin Pflueger, Noemie Pinardon-Touati, Franck Portier, Morten Ravn, Ricardo
Reis, Julian Richers, Jari Stehn, Silvana Tenreyro, Brian Wheaton, Gianluca Violante, Mark Zandi and numerous
seminar participants. Seyed Hosseini and Nicholas Tokay provided outstanding research assistance. We gratefully
acknowledge data from PredictIt and ElectionBettingOdds.com.
1 Introduction
One classic question in macroeconomics is whether fiscal deficits raise inflation. The question
has become even more important in recent years. Around the world, governments ran large
budget deficits. Inflation followed. For instance, in December 2020 and March 2021, the United
States carried out deficit financed fiscal stimulus worth 13% of GDP. Inflation rose soon after-
wards, reaching a peak of 8% during the following summer.
How much did deficits contribute to the post-Pandemic inflation? Some research finds that
deficits were a primary cause (e.g. Barro & Bianchi 2023). Others argue that supply side factors
such as bottlenecks and commodity prices mattered more (e.g. Bernanke & Blanchard 2023).
These other factors act as omitted variables, making the effect of deficits on inflation hard to
single out. More generally, it is hard to estimate the causal effect of a single, episode-specific
shock. Several shocks typically hit the economy at once, and isolating one shock is difficult.
Certain episodes are a crucible for macroeconomic models, meaning evidence about their
cause is valuable. The Great Depression and the 1960s-80s inflation both led economists to
question the prevailing model (Keynes 1936, Friedman 1968). The post-Pandemic inflation may
prove just as influential. The setting—a large, deficit financed transfer—is a powerful test of
state-of-the-art macroeconomic models, which often study precisely this shock. As a result,
the post-Pandemic inflation has already motivated new theoretical work (e.g. Angeletos, Lian &
Wolf 2024). Clearly, estimating the cause of the post-Pandemic inflation is important.
This paper measures the causal effect of the 2021 deficits on inflation—using a “high fre-
quency narrative approach” designed to study single, episode-specific shocks. The first, narra-
tive step is in the tradition of Friedman & Schwartz (1963). We find an event that released news
about deficits, and then calculate the size of the shock using new narrative data.
The event is the Georgia Senate election runoffs of early 2021. In November 2020, Democrats
won the presidency and held 48 seats out of 100 in the Senate. Both Georgia Senate seats were
to be decided by runoff elections on January 5th, 2021. The main implication of the runoff
was for fiscal policy. If Democrats won both seats they would have a majority in the Senate
for fiscal stimulus. Under Senate procedure, only fiscal legislation can be passed with a simple
majority. Non-fiscal legislation requires a supermajority of 60 votes, unattainable regardless of
the runoffs. By January 7th, Democrats won both seats. Afterwards, in March 2021 Democrats
passed $1.9 trillion of deficit financed fiscal stimulus (8.8% of GDP). This stimulus added to
the $900 billion (4.2% of GDP) passed in December 2020, for a total of 13% of GDP in stimulus
1
during late 2020 and early 2021. Shortly after, inflation started to rise.1
We then measure the size of news about deficits due to the Democrat victory. The challenge
is determining how much deficit spending was expected on the eve of the election—not only if
Democrats, but also Republicans, were to win. We introduce a new dataset: hand collected re-
ports from 20 investment banks and other macroeconomic research groups. Investment banks
distribute time stamped reports widely, around market moving events, with quantitative infor-
mation about various scenarios.
Using the reports, we size the deficit news from the Georgia runoff. The median investment
bank expected Democrats to win both seats with 50% probability, and to spend $900 billion of
stimulus if they won. If Republicans had won at least one seat, banks expected no stimulus.
Therefore the Democrat victory in both seats was a shock to expected deficits, due to going
from a 50% chance of $900 billion stimulus to a 100% chance—i.e. a shock of $450 billion,
or 2.1% of GDP. Banks expected that the stimulus would be deficit financed in the short run.
70% of stimulus was expected to be transfers, such as “stimulus checks”; with the remainder
government spending.
We also use the reports to assess the main consequences of Democrat victory. In principle,
markets could have expected the election to have many consequences, each with effects on
inflation. For instance, Democrats could have been expected to pass regulations that would
lead to inflation. However we find that according to investment banks, the main consequence of
Democrat victory was fiscal stimulus—consistent with the overall approach of the paper. There
is inevitably some uncertainty about how to interpret the reports. Throughout, we buttress our
interpretation with extensive quotations from the text.
The second, high frequency step studies the response of inflation forecasts from asset prices,
as in Gürkaynak, Sack & Swanson (2004) and others. Our main identification strategy is a single
event study examining inflation swaps in a window around the runoff. High frequency variation
separates deficits from other factors that could have mattered during the post-Pandemic infla-
tion. For instance, oil shocks will not confound our estimate, unless news about these shocks
occurred in the window.
We estimate that the Democrat victory led to an increase in expected prices of 0.39% (stan-
dard error of 0.05%) over 2 years. The shock is expected to have persistent effects until 2026.
The shock seems to have increased demand, since dividend futures suggest strong expected
real GDP growth, and investment banks revise their growth forecasts upward because of the
1
Levy (2021) linked the Georgia runoff to inflation expectations in a Twitter thread. Mian, Straub & Sufi (2024)
used this event to study the response of convenience yields on government debt, instead of inflation.
2
Democrat victory. The identification concern is that other inflationary shocks happened dur-
ing the window around the runoff. The main candidate is the January 6th Capitol Hill riots;
however, a range of evidence suggests the riots do not confound our estimates.
The drawback of the event study is that it relies on a single, high-powered observation. We
therefore add a supplementary, regression based strategy. Our motivation is that between the
November presidential election and the January runoff, there were large changes in the chance
of Democrats winning, and markets paid attention. In effect, during November to January there
were many events during which news about deficits were released. We use the daily probability
of Democrat victory, measured from betting markets, as a measure of news about deficits. The
regression strategy finds similar estimates of how deficits affect inflation to the single event
study, albeit larger.
Our evidence shows that deficits caused forecasts of inflation to increase. What are the im-
plications for actual inflation, as opposed to forecasts? Studying the effect of deficits on actual
inflation requires additional assumptions. Using the framework of Coibion & Gorodnichenko
(2015), we show that forecasts underreact to shocks, relative to the effect of these shocks on ac-
tual inflation. Then according to the framework, the response of forecasts to deficits is a lower
bound for the response of actual inflation to deficits.
To complete the approach, we combine narrative and high frequency information to calcu-
late the causal effect of the 2021 deficits on inflation. We summarize our estimates with an “in-
flation multiplier”, which divides the high frequency response by the narrative measure of the
shock. The inflation multiplier is a price level increase of 0.19% over 2 years after a 1% deficit-
to-GDP shock. In a “back of the envelope” exercise, we calculate the effect of the stimulus as
the product of the inflation multiplier and the 13% of GDP from the 2021 deficits. The result
is an increase in prices of 2.5% over 2021-22, which is around a third of total 2021-22 headline
inflation. We conclude that the 2021 deficits were important for the post-Pandemic Inflation,
though not the only cause.2
Our final empirical result suggests that loose monetary policy was one reason why deficits
caused inflation. We find that short term nominal interest rates did not respond to the Georgia
shock—meaning that in real terms, rates fell. Our result matches several papers finding mone-
tary policy was loose after the Pandemic.3
In the past, influential episodes led macroeconomists to re-evaluate their models. We ask
2
Our calculation does not include the effect of the April 2020 CARES Act, which was $2.2 trillion (10% of GDP).
If the same inflation multiplier applies, the CARES Act raised prices by another 1.8% over 2 years.
3
See, for instance, Gagliardone & Gertler (2023), Cieslak, McMahon & Pang (2024), Bocola, Dovis, Jørgensen &
Kirpalani (2024) and Bauer, Pflueger & Sunderam (2024).
3
how the state of the art model addressing deficits and inflation, the Heterogeneous Agent New
Keynesian (HANK) model, fares after the Pandemic. It is well known that in HANK models,
stimulus raises inflation—qualitatively matching our findings. What is less clear is whether the
model can quantitatively match the size and dynamics of the inflation response. We calibrate
a version of the HANK model to pre-2020 statistics, including a flat but upward sloping Phillips
Curve (Hazell, Herreno, Nakamura & Steinsson 2022), and estimates of the marginal propensity
to consume (Fagereng, Holm & Natvik 2021).4
We show that the standard model can match the size and dynamics of the inflation mul-
tiplier after the Georgia shock. The Georgia runoff is a reduced form combination of various
structural shocks, including shocks to government spending, transfers, interest rates and the
path of debt. We measure the structural shocks using our high frequency and narrative ev-
idence, and feed them into the model. The model quantitatively matches both the size and
the persistence of inflation dynamics, even though these dynamics are not directly targeted.
The dynamics of output in the model are “plausible” in the sense of Orchard, Ramey & Wieland
(2023)—being consistent with contemporaneous forecasts of real GDP around the Georgia runoff.
One part of the mechanism is loose monetary policy.
According to narrative reports, a secondary consequence of the Democrat victory—alongside
the primary impact of the stimulus—was a delayed and deficit-neutral infrastructure bill. We
consider an extension of the model with infrastructure. Consistent with past work, the effect of
infrastructure spending is small relative to the stimulus (Ramey 2021), and does not affect our
calculations about the inflation multiplier much.
Our results have two lessons for macroeconomic models. First, in HANK models deficits
generate significant inflation in empirically relevant cases—supporting recent work advanc-
ing this view (Angeletos et al. 2024).5 Second, a stable and flat but upward sloping Phillips
Curve is consistent with the post-Pandemic inflation, as some evidence suggests (Beaudry, Hou
& Portier 2024; Barnichon & Shapiro 2024).6 A major caveat is that the HANK model cannot ex-
plain the response of inflation or output when calibrated to lower and more transitory marginal
propensities to consume (e.g. Orchard et al. 2023; Boehm, Fize & Jaravel 2023).
Related literature. A defining challenge in empirical macroeconomics is how to estimate
4
We study several HANK models, including the tractable overlapping-generations version of Wolf (2021) and
Angeletos et al. (2024); the bond-in-utility and one-asset models of Auclert, Rognlie & Straub (2023); and versions
with “sticky expectations” (Auclert et al. 2020). Appropriately calibrated, the models deliver similar results.
5
We show that a simple version of the fiscal theory of the price level can also fit the inflation response (Bianchi,
Faccini & Melosi 2023, Cochrane 2023).
6
See Cerrato & Gitti (2022), Benigno & Eggertsson (2023) or Gitti (2024) for the opposite view that the Phillips
Curve became steeper after the Pandemic.
4
the causal effect of shocks to the economy. One method is the narrative approach pioneered
by Friedman & Schwartz (1963): searching the historical record for moments when an impor-
tant shock happened, and studying the response of the economy.7 A second method is the
high frequency approach: studying movements in asset prices around a series of events such as
monetary policy announcements or macroeconomic data releases.8 Both the high frequency
and the narrative approach have limitations for understanding single, episode specific shocks.
The narrative approach typically studies the economy at a monthly or quarterly frequency. At
this lower frequency, other confounding shocks may matter. The high frequency approach typ-
ically studies a time series of shocks, spanning a range of episodes. However the behavior of the
economy during a single episode is often of particular interest. Combining high frequency and
narrative information, as in this paper, is a way to estimate the causal effect of single, episode
specific shocks. We apply the approach to the 2021 deficits, but one can apply the same method
to other important episodes and shocks.
There is previous academic work that also combines high frequency and narrative identifi-
cation, such as Velde (2009) and Bahaj (2020). Closely related to our paper, Gomez Cram, Kung
& Lustig (2023) ask how inflation forecasts respond at high frequency to a series of announce-
ments about deficits from the Congressional Budget Office. One distinguishing feature of our
approach is to measure the size of the shock associated with the event, using new narrative data
from investment banks. Combining the narrative shock and the high frequency response, one
can calculate useful targets for quantitative models such as the inflation multiplier.
There are many papers on the causes of the post-Pandemic inflation. Some argue that
deficits were important (e.g. Reis 2022, Cochrane 2022, Bianchi, Faccini & Melosi 2023, Barro &
Bianchi 2023, Giannone & Primiceri 2024, Bigio, Caramp & Silva 2024). Other papers empha-
size different causes of inflation, and occasionally suggest that deficits were not important (e.g.
Mondragon & Wieland 2022, Bernanke & Blanchard 2023, Gagliardone & Gertler 2023, Guerri-
eri, Marcussen, Reichlin & Tenreyro 2023, Ferrante, Graves & Iacoviello 2023, Crump, Eusepi,
Giannoni & Şahin 2024). Disentangling the effect of deficits from the other shocks is challeng-
ing with monthly or quarterly data—we suggest that higher frequency information combined
with narrative methods can advance the debate.
Paper outline. Section 2 introduces the data. Section 3 identifies an event that released
7
Important papers within this vast literature include Romer & Romer (1989), Ramey & Shapiro (1998), Romer
& Romer (2010), Ramey (2011), Cloyne (2013), Antolín-Díaz & Rubio-Ramírez (2018), Barnichon & Mesters (2020),
Coglianese, Olsson & Patterson (2023), Drechsel (2024), Cloyne, Dimsdale & Postel-Vinay (2024) and Cloyne, Dims-
dale & Hürtgen (2025).
8
See, for instance, Gürkaynak et al. (2004), Gertler & Karadi (2015), Nakamura & Steinsson (2018), Känzig (2021,
2023), Bauer & Swanson (2023) and Swanson & Jayawickrema (2023).
5
news about deficits, the Georgia Senate runoff, and measures the size of the shock with new
narrative data. Section 4 studies the high frequency response of asset prices. Section 5 com-
bines narrative and high frequency information to form an inflation multiplier. Section 6 shows
that versions of the HANK model can quantitatively match the inflation multiplier.
2 Data
This paper uses three main datasets. The first dataset is new hand collected narrative data, from
which we will measure the shock: markets’ expectations about stimulus during the Georgia Sen-
ate runoffs. Investment banks such as Goldman Sachs or Barclays Capital provide regular in-
formation about market news, as do other macroeconomic research outfits such as Bloomberg
Economics or Moody’s Analytics. Banks tend to issue reports directly before and after market
moving events, such as major data releases, monetary policy announcements, and electoral
events. Banks also provide regular summaries and discussion of market behavior. We hand col-
lected these data by contacting the chief economist of each bank. We assembled reports from
20 organizations in total. Typically a bank gave us access to a research portal, containing the
universe of reports written by the bank.
There are three qualities of these reports which will let them proxy markets’ expectations
about stimulus. First, the reports are widely distributed. They are available to be sent by email,
for free, to anyone who trades with an investment bank, which includes most inflation swaps
traders. Second, the reports are time stamped. Therefore one can use these reports to gauge
when information has been revealed to markets. For instance, emailed reports discussing ma-
jor data releases are normally released within an hour of the release. Finally, banks discuss
quantitative statements about various scenarios as well as their likelihood, including around
market moving events such as fiscal and monetary policy announcements. Appendix Figure
C.1 gives the example of a report from Goldman Sachs that illustrates these qualities.
The second dataset contains asset prices measured at the daily and intra-daily level, to mea-
sure high frequency responses. One asset price is inflation swaps, with which we measure infla-
tion expectations. An inflation swap is a financial derivative used to exchange a fixed cash flow,
for a cash flow linked to the US Consumer Price Index (CPI). Inflation swaps allow parties to
hedge against or speculate about future inflation levels, and as such provide a measure of mar-
kets’ inflation expectations, albeit including risk premia. Over this period, inflation risk premia
seem to be stable (Cieslak & Pflueger 2023), and we will interpret movements in inflation swaps
6
as changes in expected inflation at various horizons.9 We obtain zero coupon inflation swaps
for inflation over the following 1, 2, 5 and 10 years. The data are reported at ten minute intervals
during market hours, as the median price quoted by broker-dealers in Bloomberg.10
We measure expectations about dividends following Gormsen & Koijen (2020), by using div-
idend futures on the S&P 500 stock market index. S&P dividend futures allow investors to spec-
ulate on or hedge against the future dividends paid by the companies in the S&P 500 index.
The n year contract settles based on the actual dividends distributed by the index’s companies
during over the course of that year. We will interpret movements in dividend futures as chang-
ing expectations about dividends.11 We obtain dividend futures for 1 and 2 years ahead (longer
horizon futures are not traded at intraday frequency). The data are reported at ten minute inter-
vals during market hours, based on transactions from the Chicago Mercantile Exchange (CME).
We measure intraday interest rates on government bonds. We obtain transaction prices on
2, 3, 5, 7 and 10 year positive coupon bonds from CME. We aggregate to ten minute windows
and then infer the zero coupon yield curve using the procedure of Cieslak, Morse & Vissing-
Jorgensen (2019), which pins down the short end of the yield curve using 3 month treasury bills.
We will also use daily end-of-day data on swaps, rates and futures.
The third dataset is election probabilities from online betting exchanges. Our main source
is PredictIt. PredictIt is an online exchange that allows traders to buy and sell securities whose
value is indexed to political events. The market-clearing price represents the market’s probabil-
ity of the political event. PredictIt provides intradaily and daily information on the likelihood
that Democrats would win both Senate seats in Georgia and hence overall control of the Senate.
We supplement PredictIt with election probabilities from BetFair. BetFair provides daily proba-
bilities that each individual Senate election would be won in Georgia, though not a probability
that Democrats would win both Senate seats or intraday information.
3 A Narrative Shock: the 2021 Georgia Senate Runoffs
The first step of our approach is to find an event that released information about the 2021
deficits, and then use investment bank reports to size the shock.
9
See also the measure of inflation risk premia from the Cleveland Federal Reserve, which was unchanged be-
tween December 2020 and February 2021.
10
Pflueger & Viceira (2016) and Cieslak & Pflueger (2023) discuss how an alternative measure of expected
inflation—the gap between inflation protected (TIPS) and nominal government bonds—is less reliable.
11
This interpretation requires that dividend risk premia are stable at short horizons, which is consistent with
options data (Gormsen, Koijen & Martin 2021).
7
3.1 The Georgia Senate Election Runoffs and the 2021 Deficits
The key event that released information about the 2021 deficits was the Georgia Senate elec-
tion runoffs of early 2021. In November 2020, Democrat Joe Biden won the presidency, while
Democrats held 48 seats out of 100 in the Senate. In November, there were elections for both
Senate seats in Georgia, but neither produced a majority for one candidate. By Georgia law,
the top two candidates in each election—Democrat Jon Ossoff and Republican David Perdue in
one, and Democrat Raphael Warnock and Republican Kelly Loeffler in the other—would con-
test runoff elections on January 5th.
The Georgia Senate runoff would determine fiscal policy over the next two years, but matter
less for other policy. If Democrats were to win, they would have 50 seats in the Senate. Given
Democratic Vice President Harris as a tie-breaking vote, they would have a majority. Under Sen-
ate procedure, legislation relating to fiscal policy can be passed by a simple majority, through
a procedure known as budget reconciliation. For this procedure, fiscal policy is defined as leg-
islation related to spending, taxes and the federal debt limit. Other legislation that does not
relate to fiscal policy requires a supermajority of 60 votes to pass in the Senate—although what
qualifies as fiscal policy is often contested. Therefore other legislation would be impossible for
Democrats, regardless of the runoffs. A Senate majority also allows the President to confirm
appointments without bipartisan support.
In December, between the presidential election and the Georgia runoffs, bipartisan stimulus
was passed. This bill, the Consolidated Appropriations Act, involved $900 billion of stimulus, or
4.2% of 2020Q4 annualized GDP. 70% of the stimulus was transfers, principally unemployment
insurance, stimulus checks of $600 and transfers to businesses; while the remainder was gov-
ernment spending, principally education and pandemic-related funding for tests and vaccines.
Before the Senate runoffs, Democrats campaigned for additional support for the economy
through stimulus checks (see article). However Senate Republicans did not support the stim-
ulus. An attempt by Democrats at the end of December 2020 to pass “stimulus checks” was
blocked by the then Republican Senate majority leader, Mitch McConnell (see article).
Immediately after the November presidential election, a Democrat victory in both races
seemed unlikely. However, the probability of a Democrat victory increased, particularly in the
days just before the election. Appendix Figure C.2 plots the daily probability of Democrat vic-
tory from betting markets.
In the event, Democrats won both seats. On January 5th, the election took place. Networks
confirmed that Warnock had won by the early hours of January 6th, and determined that Ossoff
8
had also won by the late afternoon. As such, the Democrat victory released news about deficits.
Beforehand, there was some chance that Democrats would win and pass stimulus. Afterwards,
given that Democrats had won, some kind of stimulus would almost definitely pass.
Another major event happened at 2 PM on January 6th. Protestors against Democrat vic-
tory in the presidential election hurdled barricades and invaded buildings on Capitol Hill, in
Washington DC. This moment initiated the Capitol Hill Riots of January 6th.
The Democrat majority led to two significant pieces of legislation. In March 2021 Democrats
passed $1.9 trillion of deficit financed fiscal stimulus through the American Rescue Plan (8.8%
of 2020Q4 annualized GDP). The Plan was 60% transfers, primarily “stimulus checks” of $1400
dollars and an extension of the generous unemployent insurance benefits of the Pandemic. The
remaining 40% was government spending, primarily state and local aid (Edelberg & Sheiner
2021). This stimulus added to the $900 billion (4.2% of GDP) passed in December 2020 in a bi-
partisan bill, for a total of 13% of GDP in stimulus during late 2020 and early 2021 (collectively
the “2021 deficits”).12 The second consequential piece of legislation passed by Democrats was
the $891 billion Inflation Reduction Act (IRA). Passed in August 2022, the IRA was an approxi-
mately deficit neutral bill that increased infrastructure spending financed by prescription drug
price reform and corporation tax, supported only by Democrats.13
After fiscal stimulus, in the spring and summer of 2021, inflation started to rise. Inflation had
been around 2% prior to fiscal stimulus. During and after the stimulus, inflation rose towards
its peak of 8% in the summer of 2022.
3.2 Sizing the Shock to Deficits after the Democrat Victory
Clearly, the Democrat victory in Georgia led to some news about stimulus—but how much? One
needs a range of information to measure the deficit news shock. One must measure not only
i) expectations about how much Democrats would spend if they were to win, but also ii) what
would happen in the counterfactual scenario in which Republicans were to win, and iii) the
likelihood of each scenario. Our narrative information from investment bank reports contains
this information.
In brief, we find that Democrat victory represented a shock to expectations of fiscal stim-
ulus worth 2.1% of GDP. There is inevitably some uncertainty in how to interpret the reports.
12
There was also an earlier large stimulus during the start of the Pandemic in March 2020—this stimulus, the
CARES Act, was 10.2% of GDP.
13
Two bipartisan infrastructure bills were also passed around the same time, namely the $550 billion Infrastruc-
ture Investment and Jobs Act (November 2021); and the $280 billion CHIPS and Science Act (July 2022).
9
Therefore this section and Appendix Tables C.1-C.10 contains extensive quotes from the re-
ports, which we use to support our interpretation.
To arrive at the narrative information, we search each investment bank’s reports for infor-
mation about the Georgia runoff and deficit spending, in a window from a week before to a
week after Democrats’ victory.14 We extract by hand from the reports each bank’s view about
the relevant aspects of fiscal policy. As we discuss, investment banks often but not always dis-
cuss key information both before and after the election. In many cases banks provided only
qualitative information about certain variables, which we discard.
The main objective is to measure news about stimulus. One requires three pieces of infor-
mation: (i) the expected stimulus if Democrats were to win both seats, (ii) the expected stimulus
if Republicans were to win at least one seat, and (iii) the probability that Democrats would win.
First, we measure expected stimulus if Democrats were to win. In total, 11 investment banks
forecasted the size of the Democrat stimulus in the week after the election. The median stimu-
lus size is $900 billion, or 4.2% of 2020Q4 annualized GDP. Table 1 and Figure 1 summarize this
information. A typical quote, from JPMorgan, reads “our best guess ... is a spending package
of around $900 billion passed in the next few months.” All reports expect that the stimulus will
be passed early in 2021. We will interpret the $900 billion number as an expectation. Table 1
shows that banks use language consistent with this interpretation, with phrases like “expect”
and “anticipate”.15 The stimulus expected in January was smaller than the $1.9 trillion even-
tually passed in the American Rescue Plan. The stimulus ended up being unexpectedly large,
relative to beliefs in early January, due to successful efforts by party leadership to sway moder-
ate Democrats in February and March (see article).
Next, we measure the stimulus that before the election, banks expected would pass with di-
vided government. 5 investment banks forecasted stimulus in the case of divided government,
prior to the election. The median forecast was no further stimulus, as Appendix Table C.2 sum-
marizes. A typical quote, from Rabobank on 5th January, reads “in this case Republicans are
likely to shoot down the ambitious spending plans of the Democrats”. This forecast is consis-
tent with a casual reading of political events, since as we discussed, Senate Republicans were
unwilling to pass further stimulus.
We then measure the probability that Democrats would win both seats. Five investment
14
We do not consider reports from more than a week after the Georgia election, since the Biden administration
released the first details of the American Rescue Plan on January 15th.
15
Only 3 investment banks provided a numerical forecast of the size of the Democrat stimulus before the election,
with a median of 700 billion (Appendix Table C.1). Pooling pre- and post-election forecasts, the median remains
$900 billion.
10
Figure 1: Expected Stimulus after Democrat Victory
Barclays
Morgan Stanley
Jefferies
BNP Paribas
Bank of America Corp
JP Morgan
Bank
Deutsche Bank
Moody's Analytics Median = 900
JP Morgan Wealth Management
Goldman Sachs
UBS
Capital Economics
0 500 1000
Number, $ bn
Notes: this graph contains expected stimulus after Democrat victory, taken from reports of investment banks after the election.
banks provided a probability of Democrat victory before the election (Appendix Table C.3). No
bank commits to a specific numerical probability. However, four of the banks use language
such as “toss-up”, “very close” and “a very slim advantage” which we interpret as a 50% chance
of Democrat victory in both seats. Consistent with this interpretation, several reports cite pre-
diction markets, which had a 50% probability of Democrat victory in both seats. A typical quote,
from Goldman Sachs on 5th January, states that the “race remains a toss-up with a slight Repub-
lican lean”.
With this information, we can calculate the news to deficits from the Democrat victory in
Georgia. The reports imply a shock to expected deficits worth $450 billion, or 2.1% of 2020Q4
annualized GDP.
The reports provide additional information that is useful for interpreting the stimulus. First,
five banks stated what they expect the composition of stimulus will be (Appendix Table C.4).
The median bank expects that the stimulus will be 70% transfers, principally unemployment
insurance and stimulus checks; and the remainder government spending, principally state and
local aid. One bank provided this information before the election and the rest afterwards, pro-
viding similar numbers in each case. The expected breakdown between spending and transfers
would prove to be roughly the same as the final American Rescue Plan. 7 investment banks
discuss financing, and agree that the stimulus was going to be partly or perhaps entirely deficit
financed (see Appendix Table C.5).
11
Figure 2: Outcomes after the Democrat Victory
Notes: this figure contains a word cloud of outcomes of the Democrat victory discussed by investment banks in the week after the election. In
the cloud, an outcome is larger if more banks discuss it, and darker if banks on average assess that it is more likely.
3.3 Other Outcomes of Democrat Victory
According to banks, the main outcome of the election, other than deficit financed stimulus, is
a delayed and tax financed environment and infrastructure bill. The infrastructure bill is quan-
titatively less important than the stimulus. Before the election five banks discuss expected in-
frastructure spending (Appendix Table C.6). On the whole, the banks suggest that infrastructure
spending would happen with either a Democrat majority or divided government, albeit more
with the Democrat majority. No bank commits to a numerical forecast. After the election, 8
banks forecast a specific size of the infrastructure package (Appendix Table C.7). The median
bank expected a $1 trillion dollar stimulus, and banks expected that the infrastructure would
take place in late 2021 or 2022. Banks agree that the infrastructure package would be partly or
wholly financed by tax increases, especially corporation and capital gains taxes (Appendix Ta-
bles C.8 and C.9). In Section 6, we will use a version of our model to quantitatively study the
effects of the infrastructure bill. The effects are small, consistent with past work (Ramey 2021).
As a result, infrastructure is not a primary consequence of the Democrat victory.
We claim that the main effect of the Democrat victory was to raise expected deficits. Al-
ternatively, markets could have expected the election to raise inflation for other reasons. For
instance, Democrats could have been expected to pass regulations that would lead to inflation.
If the election were to affect inflation through other channels, then isolating the effect of deficits
would be difficult.
We now support our view that the main effect of Democrat victory was to raise deficits. For
each bank we read their discussion of what would happen after the Democrats won. We then
manually collect the various outcomes and their perceived likelihood. We do the exercise for all
banks, but for illustration Appendix Table C.10 contains a summary for a single bank, Barclays.
12
We find that the main outcome discussed by banks was the stimulus, followed by tax rises and
infrastructure spending. Banks hardly discuss other outcomes. To visualize this information,
Figure 2 presents a word cloud. In the word cloud, a word is larger if more banks discuss this
outcome. The word is shaded darker if banks typically believe the outcome is more likely. Evi-
dently banks believe that stimulus is important and likely. The second most important policies
are tax changes and infrastructure (as well as the related green stimulus), though they are less
likely. Other issues are less important. Appendix Figure C.3 carries out the same exercise, but
uses ChatGPT to read the reports, with similar results.16
4 High Frequency Response of Inflation Forecasts from Swaps
We have the first ingredient of our approach: a shock to news about the 2021 deficits from the
Georgia Senate election runoffs, identified and sized from the narrative. We now turn to the
second ingredient: the high frequency response of inflation forecasts from swaps, in a narrow
window around the election.
High frequency changes in inflation forecasts allow us to eliminate omitted variables. At
monthly or quarterly horizons, other inflationary shocks hit the economy after the Pandemic,
such as oil shocks or bottlenecks. Disentangling these omitted variables from deficits is difficult.
However changes in inflation forecasts, in a narrow window around the election, will not be
affected by the omitted variables—unless news about the other variables coincides with the
election. Our method requires that movements in swaps prices represent changes in inflation
forecasts, and not changes in inflation risk. Consistent this interpretation, we have discussed
that inflation risk premia were stable around the election.17
4.1 Main Identification Strategy: Single Event Study
We now introduce our main identification strategy: a single event study in a narrow window
around the election. Single event studies of this kind are common in corporate finance, mean-
ing we can use standard methods (e.g. MacKinlay 1997). We assume that around the Georgia
shock, an asset price y t follows the process
16
Some banks do believe that confirming Biden administration nominees is important. The main appointment
that matter for the macroeconomy is the Federal Reserve Board; we study monetary policy in Section 5.
17
Inflation risk premia are only available at monthly frequency. Daily proxies for other risk premia, such as VIX,
are stable or declining during the election (Appendix Figure C.4).
13
ε t
if t < T
yt =
ε t + α t
if t ≥ T.
In this equation, the event happens at time T . Before time T , the “typical” movement in the
asset price is some process εt , due to factors such as the liquidity shocks that are common in
inflation swaps markets. Then, αT + j is the causal effect of the election on asset prices, j periods
after the event occurs.
The estimate of the causal effect is α̂T + j = y T + j − ET y T + j |αT + j = 0 . That is, the estimate
£ ¤
of the causal effect is the actual asset price y T + j , minus the expectation of what the asset price
would have been, using information from just prior to the event, and supposing that the event
had not come to pass. In practice, one estimates ET y T + j |αT + j = 0 using a simple ARIMA pro-
£ ¤
cess estimated in a relatively short window before the event date T .
The identification assumption is that the distribution of typical shocks to asset prices, εt ,
did not change from just before versus just after the event date T . That is, there where no
other “atypical” shocks to asset prices just after the Senate runoff. However typical shocks to
asset prices are allowed after the Senate election. Information on the distribution of the typical
shocks—measured from asset prices before the election—will let us construct standard errors
for the estimate of the causal effect. The ARIMA model easily provides these standard errors,
and flexibly deals with missing data due to weekends and public holidays.
With this identification strategy, the key decision is the width of the event window. The event
window should be wide enough to capture the full effect of the shock on asset prices. However
it should be narrow enough to exclude other atypical shocks that otherwise would confound
estimates of the causal effect.
In our baseline analysis, the event window begins on the morning of election day, January
5th, and ends at the end of January 7th. This window is our baseline because state of the art
estimates, from Bahaj, Czech, Ding & Reis (2023), suggest that inflation swaps markets take 2-
3 days to incorporate news about inflation. Therefore the window should be 2-3 days long to
account for the full effect of the shock. The election outcome was known in the early morn-
ing of January 6th. For instance, Goldman Sachs wrote a report at 2:01 AM on January 6th
stating “[d]emocratic Senate control looks likely”. Similarly, high frequency data from betting
markets suggests that the Democrat victory was known in the small hours of January 6th (see
Appendix Figure C.5). Therefore ending the window two full days after the early morning of
January 6th seems reasonable. We start on the morning of January 5th in order to account for
14
“pre-announcement drift”.18
Consistent with the identification assumption, narrative evidence suggests that the Georgia
election was the main shock to asset prices within this window. Consider for instance Goldman
Sachs’ Global Rates Trader report, which summarizes the main movements in macro-related
markets over the last week. For the week around the election, Goldman Sachs wrote: “[t]he
Georgia senate runoff results remain the key event of the week for rates, notwithstanding the
pandemic-driven drop in December payrolls.” The latter event, an unexpectedly low value for
the December employment data release, happened on January 8th i.e. directly after the event
window.19 Bloomberg News carries out daily summaries of major market moving events (Ap-
pendix Table C.11). The only potential confounder mentioned by Bloomberg is the January 6th
Capitol Hill Riots. To avoid this confounder, we consider a specification that ends the event
window at 2 PM on January 6th, i.e. before the riots began, as well as other robustness exercises
to be discussed.
Effect on inflation. Figure 3 presents the single event study. The figure shows a jump in
inflation expectations around the election, representing the causal effect of the Georgia shock.
The outcome variable is the expected increase in the price level over two years, which is calcu-
lated from the two year inflation swap, deducting the first value on January 5th. In the two weeks
prior to January 5th (vertical dashed orange line), expected price level growth is fairly constant,
meaning a small pre-trend if any. From January 5th to January 7th—as news about the Georgia
election is released—the inflation swap jumps upwards. Afterwards, the inflation swap price is
stable, suggesting that transitory factors such as market liquidity were not responsible for the
jump.
The graph also contains the estimate of the causal effect. The green line is the estimate of
what would have happened to inflation expectations absent the Georgia election, with the shad-
ing representing the 95% confidence interval. The estimate is generated from an ARIMA(1,1,2)
estimated on the prior two weeks of data; the order is selected using the Bayesian Information
Criterion, allowing for trends. The difference between the blue and green lines is the causal
effect of the election on inflation expectations. Therefore in our baseline estimate the election
18
Two factors suggest drift. First, Democrats’ best poll was released after markets closed on January 4th, and
after investment banks had released their pre-election reports. Given this news, betting markets moved towards
Democrats on January 5th (Appendix Figure C.5), as did financial markets. Second, hedge funds often have ad-
vance information around political events, bought from exit polls. Consistent with advance information, we will
see pre-announcement drift is present only in shorter durations where hedge funds are active (Bahaj et al. 2023).
19
Consistent with this view, Bloomberg releases an intraday “data surprise” index, which releases information
about whether an important macroeconomic news announcement surprised consensus forecasts. This index does
not register surprises until January 8th (Appendix Figure C.6).
15
Figure 3: Expected Percentage Point Increase in the Price Level Over 2 Years
Percentage Point Increase in the Price Level over 2 Years
0.4
0.2
Estimate at 2 PM on
Jan. 6: 0.19 (0.04)
Estimate at close of
0.0
Jan. 7: 0.39 (0.05)
First 2pm Last ARIMA(1,1,2)
Jan 5 Jan 6 Jan 7
−0.2
Dec 21 Dec 28 Jan 04 Jan 11 Jan 18
Date
Notes: this plot is the intraday percentage point increase in price level over the next two years implied by the 2 year inflation swap, subtracting
the first value on January 5th. Dashed lines are missing data from holidays and weekends. The green line is the forecast if the policy had not
taken place, the gray shade is the 95% confidence interval.
causes an increase in expected price level growth over the next 2 years of 39 basis points. Ex-
cluding data after 2PM on January 6th leads to a qualitatively similar but smaller estimate of 19
basis points. The standard error of the baseline estimate is 0.05, meaning that the jump in ex-
pectations around the election is far outside the typical range. Appendix Figures C.7-C.9 show
similar event plots for the 1 year, 5 year and 10 year expected increase in the price level. These
figures have the same distinctive pattern, i.e. stable swaps prices prior to the election shock,
followed by a jump around the election that continues to affect prices afterwards.20
Table 2 reports the estimates associated with the single event study. In the first column, we
report the baseline estimate for the percentage point increase in prices over 1 year (first row)
through 10 years (last row). The effect of deficits on prices is expected to grow over time: prices
grow by 28 basis points in the first year and 77 basis points after 10 years. The other columns re-
port robustness tests, with similar results to the baseline. In column 2 we end the event window
at 2 PM on January 6th, before the Capitol Hill Riots. The result is always directionally similar,
albeit smaller. In column 3 we drop missing data (the baseline ARIMA imputes missing data).
Column 4 estimates the ARIMA over a longer, 6 week pre-period. Column 5 estimates the coun-
terfactual for forecasts using a linear trend. Column 6 aggregates to hourly frequency. Column
7 calculates the causal effect more simply, as the difference in forecasts over the event window.
20
Inflation swaps index to inflation with a lag. For instance, the 1 year inflation swap price measures expected
inflation between months t − 3 and t + 9. We adjust swaps prices by a factor that converts them into annual units
and accounts for the fact that inflation prior to time t is pre-determined with respect to the shock. For instance we
adjust the 1 year swap price by a factor of (4/3), the 2 year swap price by a factor of (8/7) and so on. We apply this
adjustment to swaps prices in all of the analysis that follows.
16
Figure 4: Single Event Study—Effect on Inflation by Horizon
Percentage Point Increase in Expected Inflation
0.4
0.2
0.0
−0.2
'21 '22 '23 '24 '25 '26 '27 '28 '29
Years
Notes: the figure plots the causal effect of the Georgia shock on forward annual inflation rates, for 2021-29.
We can also measure dynamics—that is, how markets expect inflation will change in each
year after the shock. To construct expected inflation in each year for swaps, we impute the
forward yield curve for swaps following Cieslak, Morse & Vissing-Jorgensen (2019). Figure 4
reports the result. The figure shows that inflation is expected to increase significantly in 2021,
but also is expected to persistently increase until 2026.21
Real outcomes. One question is whether the deficit shock behaved as a typical “demand
shock”, by increasing GDP. We study two pieces of information on real outcomes, from narrative
reports and dividend futures. Both suggest that real growth increased after the Georgia shock.
First, we study dividend futures. As we have discussed, the S&P500 n year ahead dividend
future is a contract whose value is indexed to the value of nominal dividends paid by S&P 500
companies in year n. Existing evidence suggests that dividend futures are a good proxy for ex-
pected dividends (Gormsen et al. 2021). As such, movements in the 1 and 2 year ahead dividend
measure how expected dividends change around the shock.
Figure 5 presents the single event study for 2022 dividends and shows that expected nominal
dividends increased. The outcome variable is the percent increase in S&P500 dividends. The
event study suggests that nominal dividends in 2022 grew by 3.32% due to the shock. Again,
there is no pre-trend before the shock, and the price remains persistently high after the shock.
Appendix Figure C.13 shows a similar graph for the 2021 dividend future.
Dividend futures are nominal and not real. However, their behavior suggests that expected
real dividends increased over 2021-23 as well. To approximate this behavior, we subtract from
the estimate of the effect of the shock on nominal dividends, the effect of the shock on the price
21
Consistent with the inflationary shock, oil and food futures jump around the Georgia Shock (Appendix Figures
C.10 and C.11). Consistent with markets’ expectations, Appendix Figure C.12 shows that at quarterly frequency,
professional forecasters’ expectations of inflation jump after the passage of stimulus in 2021Q1.
17
Figure 5: Single Event Study—2022 Dividend Futures
6
100 x Log Expected Dividend − 2 Years
4
2
Estimate at 2 PM on
Jan. 6: 1.23 (0.51)
Estimate at close of
0
Jan. 7: 3.32 (0.51)
First 2pm Last
Jan 5 Jan 6 Jan 7 ARIMA(2,0,0)
−2
Dec 21 Dec 28 Jan 04 Jan 11 Jan 18
Date
Notes: this graph plots the intraday percent increase in dividends 2 years ahead, implied by the 2 year S&P 500 dividend future, subtracting the
last value on January 4th. Dashed lines are missing data from holidays and weekends. The green line is the forecast if the policy had not taken
place, the gray shade is the 95% confidence interval.
level at the same horizon from Table 2. Expected real dividends rose by 2.4% in 2021 and by
2.9% in 2022 due to the Georgia shock.
Under more speculative assumptions, one can use dividend futures to estimate how the
shock affected expected real GDP growth. In particular, at quarterly frequency and using a long
time series, Gormsen & Koijen (2020) show that a percent increase in dividends associates with
a 0.67% increase in real GDP. If the same relationship holds at high frequency, then one can use
the 0.67 factor to convert changes in dividend into changes in real GDP. This method suggests
that markets expected real GDP to be 1.6% higher in 2021 and 1.9% higher in 2022 due to the
Georgia shock. This method to infer real GDP from dividend futures should be treated with
caution, since the relationship between dividends and GDP may change over time. Moreover,
at intraday frequency dividend futures are traded less frequently than swaps.
A second way to gauge the effect of the Georgia shock on real outcomes is to use information
on how investment banks revised their real GDP forecasts in response to the Georgia shock. 7
investment banks provided quantitative information on how they changed their growth fore-
casts after the Democrat victory in Georgia. The median bank states that over 2 years, real GDP
is expected to grow by 1.8% more due to the Georgia shock (Appendix Table C.12). This estimate
is similar to the estimates from dividend futures. Many other banks provide similar, qualitative
information, but do not immediately update their quantitative growth forecast; banks do not
update their quantitative forecasts of inflation at a high enough frequency to be useful.
Robustness and Capitol Hill Riots. The identification assumption of the single event study
18
is that no other atypical events occurred during a narrow window around the election. As we
have discussed, the main potential confounder is the Capitol Hill Riots. We have already seen
three pieces of evidence suggesting that this potential confounder does not affect our estimates.
First, results are qualitatively similar excluding the January 6th riots from the window. Second,
the event study plots show that asset prices remain persistently different after the event, even
when the Capitol Hill riots have subsided. Third, information on real variables suggests that
expected real GDP growth increased around the Georgia election. If the riots were expected to
be important, they presumably would have contracted real GDP.
We now present two further pieces of evidence suggesting that the Capitol Hill Riots do not
confound our estimate. First, narrative evidence from news suggests that the Capitol Hill Riots
did not affect asset prices. Appendix Figure C.14 is a collage of news articles, which summarize
the prevailing view that the Capitol Hill Riots were not important. One quote, from Bloomberg
Economics on January 19th, reads: “[t]he markets appear to be putting zero probability on the
U.S. becoming a banana republic ... [o]n Jan. 6, as a mob stormed the Capitol, the S&P 500
merely trimmed its gains.”
Second, Appendix Figure C.15 plots how credit default swaps on 5 year US government debt
evolved around the Capitol Hill Riots. Credit default swaps measure the likelihood of default
on US government debt. Presumably, if the Riots were perceived to be important, then extreme
outcomes such as default on government debt would become more likely. In the event, credit
default swaps fell slightly.
4.2 Additional Regression-Based Identification Strategy
Our first identification strategy is a simple event study. The drawback is that it relies on a single,
high-powered observation. We therefore add a supplementary, regression-based identification
strategy.
Our motivation is that between the presidential election of November, and the Georgia elec-
tion in January, markets paid a great deal of attention to the likelihood of Democrat victory.
As we have discussed, markets knew that if Democrats were to win, fiscal stimulus was likely.
Moreover, the perceived probability that Democrats would win varied between the presidential
election and the runoff election. Appendix Table C.13 illustrates these points, using informa-
tion from a single investment bank, Barclays. Barclays discussed the likelihood of Democrats
winning in Georgia 5 times between the November election and the runoff. Barclays’ perceived
probability varies significantly, from 0.2 after the presidential election, to 0.5 just before the
19
runoff.
With this motivation, we use the daily probability of Democrat victory from betting markets
as measure of news about deficits (see Appendix Figure C.5 for a plot of the daily probability).
We run the regression
y t = α + βprobabilityt + εt . (1)
Here, y t is an asset price such as an inflation swap, and probabilityt is the end of day proba-
bility that Democrats would control the Senate after the Georgia elections, from PredictIt. β
is the coefficient of interest—how changing news about Democrat victory affects inflation ex-
pectations. We expect β to be positive: a higher chance that Democrats would win in Georgia
means fiscal stimulus is more likely, presumably leading to higher inflation. The sample is daily
data, from one week after the November presidential election to one week after the Georgia
election. The identification assumption is that changes in the probability of Democrat victory
were not caused by macroeconomic factors, and that there were no other correlated shocks to
both macroeconomic factors and to the probability of Democrat victory. The identification as-
sumption is at least plausible: as we shall see, changes in the probability that Democrats would
win was driven in part by factors such as better polls.
Figure 6 presents the results in a scatter plot, and finds that a higher probability of Democrat
victory associates with more expected inflation. In the graph, the y axis variable is the expected
percentage point increase in the price level over the subsequent two years. The x-axis is the end
of day probability that Democrats would win the Senate after the Georgia election, from Pre-
dictIt. Each dot is the observation from a single day. One can see a strong positive relationship
between the win probability and price level growth. The blue line, the regression line for the
full sample, has a slope of 1.48 (standard error 0.28). Restricting to data before January 5th, i.e.
before the election, leads to similar although larger results. Appendix Figures C.16-C.18 show
similar figures for expected price level growth over 1, 5 and 10 years.22
Table 3 collects the regression results and provides robustness. In Panels A through D of
the table we study the percentage point increase in prices over 1, 2, 5 and 10 years. In the first
column we study the baseline specification. In the second column we restrict to before January
5th. In the third we drop outlier observations—i.e. the election days of the 6th and 7th, as well as
days when news about December’s Consolidated Appropriations Act was released. In the final
22
One concern is data quality from PredictIt. In Appendix Figure C.19 we report similar plots for 2 year price
level growth using data from an alternative online prediction market, BetFair, for Jon Ossoff’s Senate election; and
in Appendix Figures C.20 we report our estimate for PredictIt data, using the sample for which BetFair is available.
20
Figure 6: Regression Estimates—Effect on Price Level Growth over 2 Years
Percentage Point Increase in the Price Level over 2 Years
5.0
4.5
Full Sample Regression: 1.48 (0.28)
Pre Jan 5 Regression: 4.69 (0.38)
4.0
3.5
0.25 0.50 0.75 1.00
Prob. Democrats Win Senate
After Election Before Election
Notes: this graph plots the end-of-day expected percentage point increase in prices over two years, against end of day probability of Democrat
victory (PredictIt). Standard errors, in brackets, are Newey-West with automatic lag length (Lazarus et al. 2018).
column we estimate the regression in weekly differences. For all specifications, the response of
the price level grows with the horizon, i.e. prices respond by more over 10 years than 1 year.
The estimates are larger restricting to before January 5th, and smaller in differences—though
still significant except at short horizons.
We now consider additional robustness tests. The main identification concern is that other
macro factors affected asset prices at the same time as news about the Democrat victory was
released. To deal with this concern we adopt a strategy in two parts. First, we instrument
for the likelihood of Democrat victory with daily polling data. Specifically, we use FiveThir-
tyEight.com’s daily measure of polls for Jon Ossoff’s campaign. Given that this measure is avail-
able only before the election, we restrict the sample to before January 5th. The idea behind this
instrument is to isolate movements in Democrat victory likelihood that are only due to polling
information. Secondly, we control for various other determinants of inflation expectations. We
control for weekly lags of the 10 year US government bond, the Oil price, the S&P500 index
value, and daily data from the Cleveland Fed of households expectations of the effect of Covid-
19 on the economy. Since the regression only has 30 observations, we add each these controls
in separate specifications. Appendix Table C.14 contains the results. The estimates are large
and more imprecise, but qualitatively consistent with the baseline specification, and the robust
first stage F statistic is always above 10. The main caveats are that after controlling for oil prices
or S&P 500 prices, the effects are no longer significant at short horizons. The effect of deficit
news on 5 and 10 year expectations is always large, positive and significant.
How do the magnitudes of the two identification strategies compare? The single event study
21
measures the response of asset prices to a change in the probability of Democrat victory of 0.5.
After halving the regression coefficients, the regression-based strategy also measures the re-
sponse of asset prices to a change in the probability of 0.5. Comparing Tables 2 and 3, and
halving the latter estimates, we can see that the regression-based effects are roughly 90% larger
than the single event study. One reason why the regression estimates are bigger could be that
the expected size of deficits, if Democrats won, may also have changed between November and
January. Our estimates cannot account for such changes, because daily measures of the ex-
pected size of deficits are not available. In our quantitative estimates, we will target the smaller
and easier-to-interpret numbers from the single event study.23
4.3 Mapping from Inflation Forecasts to Actual Inflation
The main evidence in this paper is about how the Georgia shock affected inflation forecasts
from swaps. What are the implications for actual inflation, as opposed to forecasts? Studying
actual inflation requires additional assumptions. We argue that the effect of the Georgia shock
on inflation forecasts is likely to be a lower bound for the effect on actual inflation. The reason
is that inflation expectations tend to underreact to shocks.
We formalize this argument in two steps. First, we document underreaction. We estimate a
regression following Coibion & Gorodnichenko (2015),
πt ,t +1 − F t πt ,t +1 = α + β F t πt ,t +1 − F t −1 πt ,t +1 + εt
¡ ¢
(2)
where πt ,t +1 is consumer price index inflation over the next 12 months, F t πt ,t +1 is the forecast
of inflation over the next 12 months, and F t −1 πt ,t +1 is the forecast from a year ago. In the regres-
sion, if β > 0, then new information F t πt ,t +1 −F t −1 πt ,t +1 predicts forecast errors πt ,t +1 −F t πt ,t +1 .
If so, then inflation expectations under-react to new information.24
We estimate the regression over three periods: from 1983 to 2023, from 2004 to 2023 and
from 2019 to 2023. Before 2004, we impute inflation swaps using the Cleveland Federel Re-
serve’s measure. Restricting to 2019 onwards allows us to measure underreaction specifically
during the post-Pandemic inflation. Appendix Table C.15 shows the results. Across all three
sample periods there is underreaction, consistent with Coibion & Gorodnichenko (2015). Un-
derreaction is particularly strong during the post-Pandemic Inflation, consistent with the fact
that expectations rose and fell more slowly than inflation at the time (e.g. Hazell 2024).
23
Appendix Figures C.21-C.22 show that the response of dividends is similar with the regression-based strategy.
24
Unlike Coibion & Gorodnichenko (2015), we implement the using forecasts of inflation over the next 12 months
instead of the next quarter, because swaps only provide 12 month forecasts of inflation.
22
In the second step, we formalize the sense in which the response of inflation forecasts to
swaps is a lower bound for the response of actual inflation to swaps, if there is underreaction. To
do so, we study the noisy information model of Coibion & Gorodnichenko (2015) (see Appendix
Section A). Combining the two steps, the response of inflation expectations to the Georgia shock
is likely a lower bound for the effect of the Georgia shock on actual inflation.
5 The Inflation Multiplier and the Effect of the 2021 Deficits
At this stage, we have both ingredients of our high frequency narrative approach: a shock, iden-
tified and sized from the narrative; and a response, measured at high frequency. We now com-
bine the shock and response into an “inflation multiplier”, which summarizes the causal effect
on inflation of a 1% deficit-to-GDP shock. We then use this multiplier to calculate the total
effect of the 2021 deficits on inflation, via a back of the envelope exercise.
We calculate the inflation multiplier by combining high frequency and narrative informa-
tion as follows. The narrative evidence of Section 3 implies a shock to expected deficits worth
2.1% of GDP. The high frequency response of Section 4 implies a response of expected prices of
0.39% over 2 years. The inflation multiplier divides the response by the shock. Therefore for a
1% of GDP shock to deficits, expected prices grow by 0.19% over 2 years. As we have discussed,
this inflation multiplier is likely a lower bound for the effect of deficits on actual inflation, given
the underreaction of expectations.
The inflation multiplier summarizes the causal effect of a marginal 1% of GDP deficit shock
on inflation during 2021. To calculate the causal effect of the whole 2021 deficits on inflation,
one needs the product of the inflation multiplier and the total size of deficits. The 2021 deficits
were 13% of 2020Q4 annualized GDP. Therefore with our estimate of the multiplier, total deficits
caused at least 13% × 0.19 = 2.5% inflation, cumulatively over 2021-22.
This calculation implies that deficits were an important contributor to the subsequent in-
flation. Cumulatively, between the start of 2021 and the end of 2022, the price level increased
by about 7.5% in excess of the normal rate.25 Our calculations suggest that deficits accounted
for around a third of the excess increase in headline prices over this period.
This calculation is only suggestive, because it involves an extrapolation. The calculation as-
sumes that the marginal effect of deficits on inflation was the same for the Georgia runoff, as
25
Typical annual inflation is 2% for PCE inflation and 2.4% for CPI inflation. Headline CPI prices grew by 14%
over 2021-22, or 9.2% more than typical. Core CPI grew by 11.5%, or 6.7% more than typical. Headline PCE grew
by 11.5%, or 7.5% more than typical. Core PCE grew by 10%, or 6% more than typical.
23
for the 2021 deficits as a whole. While this assumption is plausible, it is hard to test directly
and may not be correct. For instance, the exercise assumes monetary policy should have re-
sponded similarly to the Georgia runoff as to the other deficit shocks, which may or may not
have been true. Likewise, if the economy responds nonlinearly to larger shocks, then our back
of the envelope estimate will be incorrect.26
Nevertheless, our calculation suggests that deficits were probably important for inflation
after the Pandemic. However deficits were not the only cause—other factors must explain the
remaining two thirds of the increase in inflation. For instance, energy shocks from the Russia-
Ukraine war probably also mattered (Gagliardone & Gertler 2023).
The total contribution of deficits to the post-Pandemic Inflation could have been even higher.
The March 2020 CARES Act was an additional stimulus worth 10.2% of GDP. Applying the same
inflation multiplier to the CARES Act implies a further 1.9% increase in prices over 2 years. How-
ever, we caution that the inflation multiplier could have been different for the CARES Act.
This calculation highlights that deficits were a significant contributor to the post-Pandemic
inflation in part because they were so large. The inflation multiplier, of 0.39 over 2 years, does
not seem particularly big. But the size of the 2021 stimulus, at 13% of GDP, was big by recent
standards. For instance the sum of the Economic Stimulus Act of 2008 and the American Re-
covery and Reinvestment Act of 2009, the stimulus in response to the Great Recession, was only
6.7% of 2008 GDP.
5.1 Inflation Multiplier: The Role of Loose Monetary Policy
In the coming section, we will ask whether standard models can quantitatively explain the in-
flation multiplier that we have estimated. However in order to interpret the inflation multiplier
with a standard model, one additional piece of evidence is useful. In standard models, the infla-
tion multiplier is partly determined by monetary policy. To gauge monetary policy, we estimate
how interest rates responded to the Georgia shock.
Short term nominal interest rates did not change after the Georgia Shock, whereas long term
interest rates rose. Figure 7 presents the single event study using intraday 1 year nominal inter-
est rates on government bonds as the outcome variable. Clearly, one year nominal interest rates
do not respond to the Georgia shock. Appendix Figure C.23 is the single event study for the 5
year forward 5 year nominal interest rate, which rises significantly—as originally discovered by
26
One plausible source of nonlinearity arises because large output changes may have greater marginal effects on
inflation (e.g. Blanco, Boar, Jones & Midrigan 2024). If true, then our back of the envelope exercise understates the
total effect of the 2021 deficits on inflation.
24
Figure 7: Single Event Study—1 Year Nominal Interest
Percentage Point Change in Nominal Interest Rates, 1Y
0.02
0.00
Estimate at 2 PM on
Jan. 6: 0.014 (0.006)
Estimate at close of
Jan. 7: 0.005 (0.009)
−0.02
ARIMA(0,1,1)
First 2pm Last
Jan 5 Jan 6 Jan 7
−0.04
Dec 21 Dec 28 Jan 04 Jan 11 Jan 18
Date
Notes: This plot shows the intraday increase in 1 year nominal interest rates, subtracting the first value on January 5th. Dashed lines indicate
missing data from holidays and weekends. The green line is the forecast if the policy had not taken place, and the grey shade is the 95%
confidence interval. The dashed orange lines mark the first observation on January 5th, 2 PM on January 6th, and the final observation on
January 7th.
Mian et al. (2024).27
These moments will be crucial for the quantitative model of the next section. Since nominal
interest rates did not change in the short term, and inflation expectations rose, real interest
rates must have fallen—suggesting loose monetary policy in the short run. On the other hand,
long term rates rose, consistent with an increase in the stock of government debt, as Mian et al.
(2024) previously argued.
6 HANK Models and the Inflation Multiplier
To recap: in Section 5 we found that the 2021 caused a significant share of the post-Pandemic
inflation. This episode is a test for the Heterogeneous Agent New Keynesian (HANK) model, the
workhorse model for studying how deficits affect inflation. It is well known that in HANK mod-
els, deficit financed fiscal stimulus raises inflation—clearly the model can qualitatively match
our finding. This section asks whether HANK models can pass a more powerful test: quanti-
tatively matching the size and dynamics of the inflation multiplier. There is a range of HANK
models in the literature. In the main text, we study a tractable version of the HANK model pro-
posed by Wolf (2021) and Angeletos et al. (2024). In the Appendix we study additional versions
27
Appendix Figures C.24-C.25 show similar results with the regression specification using daily data and the
probability of Democrat victory. Appendix Figures C.26-C.29 show that Federal Funds rate futures, up to a year
ahead, also did not change after the Georgia shock.
25
of the HANK model.28
6.1 A Tractable HANK Model
Time is discrete, with t ∈ {0, 1, ...}. We study linearized dynamics in response to a shock that is
realized at date 0. The economy is at steady state prior to the shock, with X ss denotes the steady
state value of variable X t before the shock.
Households. Following Wolf (2021) and Angeletos et al. (2023), there is a unit mass of house-
holds, comprising of two types: hand-to-mouth and overlapping generations (OLG) agents.
There is a mass 1 − µ of perpetual-youth, overlapping-generations households as in Blan-
chard (1985). Each period, households die with probability 1 − φ, with φ ∈ (0, 1]. New house-
holds replace those that die, and deceased households do not value the utility of new born
households. Given mortality risk, OLG household i in period t has expected utility
∞ ¡ ¢s £ ¡
Et βφ u C i ,t +s − v Ni ,t +s
X ¢ ¡ ¢¤
s=0
where C i ,t +s and Ni ,t +s denote the consumption and labor supply of OLG household i , given
1 ¡
that they survive. Utility over consumption and labor supply take standard forms u (C ) = C 1− σ / 1 − σ1
¢
1+ ϕ1
³ ´
1
and v (N ) = N / 1+ ϕ .
Households use a risk-free and actuarially fair nominal annuity in order to save and bor-
row. If households survive, they have a nominal rate of return (1 + I t −1 ) /φ, where I t −1 is the net
nominal interest rate on government bonds. Additionally, households receive real labor income
Wi t Ni t net of labor income tax τ y and given real wage Wi t . OLG households also pay lump
sum taxes T t . A negative lump sum tax is equivalent to a transfer from the government such
as a “stimulus check”. Last, newborn households receive contributions from a “social fund” to
which older households contribute.
As such, the budget constraint of household i at date t is
1 1 + I t −1
A i ,t −1 + 1 − τ y Wi t Ni t − T t + Zi t ,
¡ ¢
Ci t + Ai t =
φ Πt
where A i t is the end of period real saving of agent i at date t ; and Πt is the gross inflation rate
between periods t and t − 1. Here, Zi t is the contribution towards, or transfer away from the
28
Specifically, we study the one-asset HANK model and the bond-in-utility model of Auclert, Rognlie & Straub
(2023), as well as versions with “sticky expectations” as in Auclert et al. (2020); appropriately calibrated, all models
deliver similar results.
26
social fund. We have Zi t = Z new > 0 for newborns and Zi t = Z old < 0 for older households, with
1 − φ Z new +φZ old = 0. We also set Z new = (1+r ss )A ss , where A ss denotes the steady state level
¡ ¢
of assets held by savers and r ss is the steady state real interest rate. As Angeletos et al. (2023)
explain, the role of the social fund is to ensure that all generations have the same wealth and
consumption in steady state.
The remaining mass µ of households are hand to mouth. These households do not par-
ticipate in asset markets. The budget constraint of a hand to mouth household i is C iHt =
(1 − τ y )Wi t NiHt − T tH , where C iHt is the consumption of hand to mouth households, NiHt is the
labor supply, and T tH is their lump sum tax. The steady state lump sum taxes on OLG and hand
H
to mouth households, T ss and T ss , ensure that steady consumption is the same for both types
of households.
As in Wolf (2021) and Angeletos et al. (2023), a mix of OLG and hand to mouth households
provides a tractable form of the HANK model. Mortality risk can be interpreted as the probabil-
ity that a borrowing constraint might bind in the future. As a result, deficit financed transfers af-
fect aggregate demand, and Ricardian Equivalence breaks. Relative to the canonical permanent
income model, households have a larger marginal propensity to consume (MPC) in the short
run, and discount future income more heavily. By adding a share µ of hand to mouth house-
holds, our model will be able to match “intertemporal marginal propensities to consume” and
approximate the behavior of a richer HANK model (Auclert, Rognlie & Straub 2023).29
Firms. Firms operate in perfectly competitive goods markets, selling output at a flexible
price P t . Each firm has a production function that is linear in a single input, which is produced
by a “labor packer” and sold to firms at a real price Wt . Aggregate output Y t satisfies Y t = N t ,
where N t is the aggregate output of the labor packer. Output is sold to either household con-
sumption or the government. That is, we have Y t = C t + G t , where G t is government purchases
and C t is aggregate consumption across OLG and hand to mouth households.
Nominal Wage Rigidity. There is nominal wage rigidity, as in Erceg, Henderson & Levin
(2000), and Auclert, Bardóczy & Rognlie (2023). Since the ingredients are standard, we discuss
them only briefly. Appendix Section B.1.2 presents a detailed derivation of the Phillips Curve.
There is a continuum of unions k, and each worker i belongs to a union. Within the union, all
workers are of equal productivity, receive equal after-tax wages, and work the same number of
hours. Each union is representative of the entire population. Unions have quadratic costs of
adjusting wages as in Rotemberg (1982), and set the nominal wage in order to maximize the
29
In Appendix B.4.3 we study additional models of consumption, including a bonds-in-utility model, a one-asset
HANK model, and versions with sticky expectations.
27
equally weighted utility of their members, discounted by β. The labor packer then combines
the labor from each union into aggregate employment, using a standard CES aggregator.
The Phillips Curve linking price inflation to real variables takes a familiar form. Since prices
are flexible, wage and price inflation coincide absent shocks to total factor productivity. To first
order around the zero inflation steady state, inflation satisfies
( )
ϕ
πt = κ y t − g t + βπt +1 , (3)
ϕ + σ CYssss
where κ is the “slope” of the Phillips Curve defined in Appendix Section B.1.2, and y t = YtY−Y
ss
ss
and g t = G tY−G
ss
ss
are deviations of output and government spending from their steady state val-
ues, normalized by steady state output. Equation (3) is the Phillips Curve, with the standard
adjustment for how government spending alters wealth effects on labor supply.
Policy and Equilibrium. The government carries out two kinds of policy: fiscal policy, in-
volving government spending, debt and taxes; and a monetary authority that sets nominal in-
terest rates. Regarding monetary policy, the central bank sets the nominal interest rate I t .
Regarding fiscal policy, the government issues nominal one period bonds B t , spends G t on
final goods output, levies total lump sum taxes µT tH + 1 − µ T t on hand to mouth and OLG
¡ ¢
households, and collects labor income taxes τ y Y t . As a result, the government budget constraint
is
1 + I t −1
Bt = B t −1 − S t , (4)
Πt
where S t ≡ µT tH + 1 − µ T t + τ y Y t −G t is the primary budget surplus.
¡ ¡ ¢ ¢
Given these elements, an equilibrium is (i) a sequence of consumption, employment, wages
and savings for OLG and hand to mouth households; as well as prices, aggregate output, and
government debt; which (ii) satisfies household optimality, the Phillips Curve, asset and goods
market clearing and the government’s budget constraint (as well as a no-Ponzi condition).
ª∞
The equilibrium is conditional on the “policy block”—a sequence I t , T t , T tH ,G t t =0 of nom-
©
inal interest rates, lump sum taxes on OLG and hand to mouth households, and government
spending. We will discuss how we treat the policy block momentarily. In Appendix Section B.2,
we present the full set of linearized equations associated with the equilibrium of the model.
Calibration. We calibrate the model to parameters from before 2020. Most important are
the parameters governing household spending and the slope of the Phillips Curve. As Auclert,
Rognlie & Straub (2023) point out, intertemporal marginal propensities to consume should
match in the model and data, in order to generate the correct consumption response to fis-
28
cal shocks. We calibrate the share of hand to mouth consumers µ, and the OLG mortality rate
φ, in order to match the first and second year MPCs from Fagereng et al. (2021). We calibrate
the slope of the Phillips Curve to the value of Hazell et al. (2022), with κ = 0.055.30
Additionally, we calibrate the discount factor, the intertemporal elasticity of substitution
and the Frisch elasticity of labor supply to standard values: β = 0.99, σ = 1 and ϕ = 1; and
also choose r ss so that β (1 + r ss ) = 1. We calibrate the steady state debt-to-GDP ratio, the steady
state government spending to GDP ratio, and the labor income tax rate to 2019 values from CBO
(2019). We report our calibration in Table 5, which also contains parameters about fiscal policy
to be discussed directly.
6.2 Modeling the 2021 Deficits
The main exercise will be to feed the shock to deficits from the Georgia election into the model.
We will then ask whether the model, with its calibration to pre-2020 data, can replicate the infla-
tion multiplier that we have estimated. The Georgia runoff is a reduced form shock, combining
structural shocks to fiscal and monetary policy. We therefore discuss how we discipline shocks
ª∞
to fiscal and monetary policy, i.e. I t , T t , T tH ,G t t =0 , using our narrative and high frequency
©
evidence as well as additional information sources.
Size of deficit shock. We take the size of the deficit shock, 2.1% of steady state output, from
our reading of narrative information in Section 3.
Composition of deficit shock: spending vs. transfers. We now allocate the total deficit
shock between transfers and government spending. The narrative reports provide information
about the expected composition of deficits. From Appendix Table C.4, the median bank expects
that the stimulus will be 70% transfers, with the remainder government spending, principally
state and local aid. We opt for similar, but more precise information, based on the realized com-
position of spending and transfers from the American Rescue Plan (Edelberg & Sheiner 2021).
The American Rescue Plan had four major components: (1) federal spending, state aid, and
COVID-19 containment (40%) (2) direct aid to families (30%) (3) aid to financially vulnerable
households (21%) and (4) aid to businesses (8%). We allocate (1) as government spending and
(2)-(4) as transfer payments. Therefore, we assume that 60% of the shock to deficits around
Georgia is transfers—similar to the expectations from the narrative reports. We assume that per
capita transfers are the same for OLG and hand to mouth households, consistent with the lump
sum behavior of “stimulus checks”, and consider other forms of distribution in robustness.
30
Hazell et al. (2022) report the slope of the Phillips Curve, at annual frequency and including housing, in Foot-
note 24 of the paper.
29
Figure 8: Fiscal and Monetary Shocks
Notes: the left panel shows innovations to government spending and total transfers, normalized by steady state output. The innovations are for
2021 onwards, expected after the Georgia shock, and measured from narrative information. The right panel shows the path of innovations to
one-year interest rates, expected from 2021 onwards, in the aftermath of the Georgia shock. The shocks are inferred from single event studies
applied to the one-year-ahead forward yield curve.
Timing of deficit shock. We now allocate when the components of the deficit shock are ex-
pected to be spent. The narrative reports do not provide information about how quickly they
expect the shock to deficits around the Georgia election to be spent. Instead, we study the time
path of how the overall American Rescue Plan was spent, and assume that the expected time
path of deficits after the Georgia election was the same. CBO (2021a) (detailed Tables 1-11)
projects how quickly various detailed components of the American Rescue Plan would be spent
over the subsequent 10 years. We aggregate the detailed components into the broad categories
of Edelberg & Sheiner (2021), which results in a projected spending path for government spend-
ing and transfers. The CBO provides the spending path of appropriations—that is, when spend-
ing would be allocated from the American Rescue Plan towards its intended purpose. However
there is typically a lag between appropriations and realized government spending. Following
Ramey (2021), we assume a “time to spend” delay between appropriations and realized spend-
ing of 1.5 years. We assume that spending on COVID-19 containment does not have this delay.
Together, our information on the size, composition and timing of the deficit shock defines
the fiscal stimulus. We will denote fiscal stimulus—the sequence of innovations to taxes and
ª∞
government spending—by T̃ t , T̃ tH , G̃ t t =0 . Fiscal stimulus is associated with an innovation that
©
lowers the primary surplus εt ≡ G̃ t − µT̃ tH + 1 − µ T̃ t . Figure 8, left panel, plots the sequence
¡ ¡ ¢ ¢
of innovations to government spending and total transfers, normalized by steady state output.
The figure contains innovations for 2021 onwards, expected after the Georgia shock.
Fiscal Rule and Path of Debt. We now specify the fiscal rule associated with our model,
which describes how debt will be paid back as well as the initial stimulus. Our fiscal rule speci-
30
ª∞
fies the entire sequence of fiscal policy, T t , T tH ,G t t =0 , as follows:
©
• When t ≤ H: government spending, and lump sum taxes on OLG households and hand
to mouth households are determined entirely by fiscal stimulus. That is, we have that
ªH ªH
T t , T tH ,G t t =0 = T ss + T̃ t , T ss
H
+ T̃ tH ,G ss + G̃ t t =0 .
© ©
• When t > H: government spending and lump sum taxes on hand to mouth households
continue to be determined by fiscal stimulus. Lump sum taxes on OLG households con-
repay
tain an additional component T t , set in order to pay back debt. That is, we have
∞ repay H
ª∞ repay
T t , T tH ,G t t =H +1 = T ss + T̃ t + T t + T̃ tH ,G ss + G̃ t t =H +1 , where T t
© ª ©
, T ss is set so that
primary surpluses satisfy
S t = τB B t −1 − B̄ + r t B t −1 − εt .
¡ ¢
(5)
This fiscal rule has various features that suit our setting. First, the fiscal rule has two phases. Be-
fore period H , there is a stimulus phase: lump sum taxes are set to provide stimulus transfers.
After period H , there is a repayment phase: lump sum taxes are associated with an additional
component in order to pay back debt. This feature, which is shared by the fiscal rule of Angele-
tos et al. (2023), matches the data. In Figure 8, left panel, stimulus transfers take place almost
entirely within years 0, 1 and 2 (i.e. 2021-23). Meanwhile the CBO (2021a) (Summary Table 1)
forecasts that increases in taxes to partially pay for the American Rescue Plan would start in
year 3 (i.e. 2024). Given these features, it is natural to calibrate H = 3.
Second, during the repayment phase, the fiscal rule of equation (5) takes a sensible form.
The fiscal rule is written in terms of primary surpluses (as in for instance Blanchard 2023).
Taxes change, so that primary surpluses gradually move towards a level that is consistent with
the steady state level of debt, B̄ . The speed of adjustment is parameterized by τB . We will al-
low the steady state level of debt after the shock, B̄ , to potentially be higher than the level of
debt before the shock. This aspect of the fiscal rule is motivated by our finding that long term
interest rates rise, which in HANK models is consistent with steady state debt increases (e.g.
Campos, Fernández-Villaverde, Nuño & Paz 2024). With higher steady state debt, and higher
accompanying interest rates, primary surpluses must also be higher in the new steady state.
The penultimate term in equation (5) allows the fiscal rule to accommodate these extra interest
costs. The final term is the continual impact of the fiscal stimulus on primary deficits.
Finally, we assume that the long term fiscal adjustment takes place entirely through lump
sum taxes on OLG households. This assumption is appealing because consumption and output
31
will be unchanged in the new steady state with higher debt, after the shock. According to the
text of the legislation itself, the eventual tax rises associated with the American Rescue Plan
would be taxes on high income individuals and corporations. We will consider fiscal adjustment
via distortionary taxes and cuts in government spending in robustness exercises.
Overall, we calibrate the fiscal rule as follows (see the last three rows of Table 5). As dis-
cussed, we set H = 3. We calibrate B̄ , the increase in steady state debt after the shock, in order
to match the increase in the 9 year ahead 1 year interest rate from the data. We calibrate τB
to match the long-horizon persistence of debt forecasted by the CBO (2021b). We elaborate on
this procedure in Appendix Section B.3, where we show how to identify τB from the CBO debt
projections.
Monetary policy. Estimates from Section 4 pin down expectation of future short-term nom-
inal interest rates up to 10 years. In Figure 8, right panel, we convert these estimates into 1 year
forward rates, between 0 and 9 years after the shock. After 10 years we assume a determinate
Taylor rule with a time-varying intercept that ensures convergence to steady state output and
inflation.
Overall shock. The previous information is enough to describe the shock to the 2021 deficits,
around the Georgia elections, summarized in the top panels of Figure 9. Specifically, the fiscal
ª∞
rule and the path of interest rates define the shock to the policy block I t , T t , T tH ,G t t =0 , which
©
we will feed into the model.
6.3 Matching the Inflation Multiplier in the Model
Figure 9 plots the impulse response of the economy to the shock after the Georgia election. The
model fits the size and dynamics of the response of inflation well. This finding is in the middle
panel. The orange circles are the response of inflation expectations after the Georgia runoff, as
we estimated in Section 4. The blue line is predicted inflation from the model. The two series
match closely, both in the initial and later stages of the stimulus, and the model prediction is al-
ways well within the confidence intervals of the estimates from data (shaded orange area).31 As
a result, the HANK model seems to quantitatively match the size and dynamics of the inflation
multiplier.
In the standard model, inflation increases via the Phillips Curve: higher output raises marginal
costs and therefore inflation. Consistent with this logic, output increases significantly and per-
sistently in our calibration, as the left panel of Figure 9 shows. The right panel shows the cumu-
31
Appendix Figure C.30 plots the same graph with price levels, and with similar results.
32
Figure 9: Impulse Response to Georgia Shock
Notes: this graph plots impulse responses of output y t in percentage deviation from steady state (left panel), and inflation πt in basis point
deviations from steady state (middle panel); as well as the cumulative output multiplier (right panel). All responses are to the shocks shown in
Figure 8. Inflation forecast estimates from the data are shown with their 95% confidence intervals.
Pt j Pt j
lative output multiplier. The cumulative output multiplier is defined as j =0 β y t + j / j =0 β ε̃t + j ,
where ε̃t ≡ εt /Y ss is the sum of government spending and transfers normalized by steady state
output, plotted in Figure 8. The cumulative output multiplier starts at a little below 1.5 and
gradually rises with the horizon.
One question is whether the changes in output predicted by the model are reasonable. We
draw on three pieces of evidence to argue that the output response is indeed plausible. First, in
the spirit of Orchard et al. (2023), we consider how private sector forecasts of real GDP changed.
As we discussed in Section 4, forecasters revised their forecasts of real GDP upwards after the
Georgia election. The median forecaster predicted that 2022 output would be 1.8% higher due
to the Georgia election. Second, consider dividend futures. We found in Section 4 that markets
predicted real GDP in 2022 would be roughly 1.9% higher due to the Georgia shock. The model
predicts that output would be 1.25% higher in 2022 due to the shock, which is similar albeit
smaller than the narrative and dividend evidence. Third, contemporary accounts predicted a
similar multiplier to what we have found. In particular Blanchard (2021) predicted a short run
multiplier associated with the American Rescue Plan. His central tendency was a multiplier of
1.2, with high and low values of 2 and 0.4. The short run output multiplier predicted by the
model is well within this range.
Our estimates are at least somewhat consistent with historical estimates of the multiplier.
In her survey of empirical work, Ramey (2019) suggests that cumulative output multipliers with
respect to government spending during periods of accommodative monetary policy could be
33
Figure 10: Impulse Response with Alternative Monetary Policy
Notes: the figure plots the impulse responses of output y t and inflation πt in the model with the interest rate path {I t }t ≥0 estimated from US
government bonds (blue), or under the assumption that the monetary authority follows a robust policy rule as in Orphanides & Williams (2002)
(orange).
1.5 or higher. Our estimates are within this range within the first three years. However the 2021
stimulus comprises of various kinds of transfers as well as government spending, meaning the
multiplier from our exercise is not directly comparable to pure government spending multipli-
ers. Aggregate evidence on transfer multipliers is relatively scarce (Ramey 2019).
6.4 The Role of Monetary Policy
We now show that loose monetary policy partly explains why the model can match the inflation
multiplier from the data. As we have discussed, monetary policy was expected to be loose in
response to the 2021 deficits. Consider the behavior of interest rates in the right panel of Figure
8. 1 year nominal interest rates were unchanged, and rose by little in the short term, meaning
real interest rates fell in the short term after the Georgia shock.
To quantitatively evaluate the role of relatively loose monetary policy, one must specify “nor-
mal” monetary policy. Our model of normal monetary policy is the Orphanides & Williams
(2002) rule, recently popularized by Campos et al. (2024). By this rule, nominal interest rates
are set according to i t = i t −1 + φπ πt , with φπ = 1.5. As Campos et al. (2024) explain, this rule
is appealing because it allows nominal interest rates to respond to increases in inflation away
from the central bank’s target. However, the rule does not require information about the steady
state level of real interest rates—which in our model changes with steady state increases in gov-
ernment debt.
34
We show that with the historical monetary policy rule, the response of inflation to the deficit
shock would have fallen by roughly 40%. Figure 10 reports this result. In the blue line we plot
the impulse response from the baseline model. In the orange line we plot the impulse response
under the alternative, historical monetary policy rule. The right panel shows that the response
of inflation would have been lower by 40% or more in each year of the stimulus. In the left panel
the reason is evident. Tighter monetary policy under the historical rule leads to a smaller output
boom, which dampens the rise in inflation.
As such, we conclude that one reason for the response of inflation to deficits observed in the
data is relatively loose monetary policy. This message is shared by other work about the post-
Pandemic inflation. For instance, using a vector autoregression and a structural model based
approach, Gagliardone & Gertler (2023) finds that monetary policy was an important contribu-
tor to the post-Pandemic inflation. Narrative accounts also find that monetary policy was loose
over this period (e.g. Cieslak et al. 2024); and estimates of the monetary policy reaction function
are notably looser around this time (Bocola et al. 2024; Bauer et al. 2024).
6.5 Robustness and Discussion
The main takeaway is that a tractable version of the HANK model can match the size and dy-
namics of the inflation multiplier around the Georgia shock. We now consider various robust-
ness exercises that support this message. We also emphasize an important caveat. If the HANK
model is calibrated to match lower and more transitory intertemporal MPCs, in line with many
estimates, then it can no longer match the inflation multiplier. Finally, we also discuss an alter-
native theory linking deficits and inflation—the fiscal theory of the price level.
Infrastructure investment. As we discussed in Section 3, a second component of the Demo-
crat victory was the likelihood of greater infrastructure spending. So far, we have omitted infras-
tructure spending from our analysis. The reason is that, as we show in Appendix Section B.4.1,
the effect of the infrastructure package on the economy is small according to the model. In the
Appendix, we extend the model to include infrastructure, following a simplified version of the
model of Ramey (2021), which in turn builds on Leeper et al. (2010). In this extended model, we
look at the impulse response to the combined shock of the policy shocks from Figure 8 and the
infrastructure spending shock discussed in Section 3. We pin down the path for infrastructure
spending from detailed projections from Moody’s Analytics about the spending path of the In-
flation Reduction Act. According to the model, infrastructure increases the first year response
of inflation by 6 basis points or less, depending on our calibration, and the first year response
35
of output by 5 basis points or less.
As we discuss in the Appendix, the small effect of infrastructure is for three reasons. First, the
infrastructure program was balanced budget, and the tax rises dampen the output multiplier.
Second, the wealth effects induced by greater infrastructure spending dampen the effects on
inflation and output. Third, following Ramey (2021), we incorporate a realistic “time to build”
lag, meaning little of the infrastructure is spent in the first three years.
Alternative HANK models. Our baseline exercise studies a tractable version of the HANK
model, with a mix OLG and hand to mouth households. In robustness, we explore four other
HANK models. The first is a mix of hand to mouth and bond-in-utility households, following
Auclert, Rognlie & Straub (2023). The second is three types of OLG households who differ in
their mortality risk, following Wolf (2021) and Angeletos et al. (2024). The third is a one asset
HANK economy with borrowing constrained households and idiosyncratic income risk. The
fourth is sticky information as in Auclert et al. (2020), which has been shown to be important to
produce hump shaped aggregate responses while being consistent with responses at the micro-
level. Appendix Section B.4.3 describes these alternative possibilities and Figure B.4 shows that
when these alternative models match the same intertemporal marginal propensities to con-
sume, the implications for output and inflation are similar.
Taxes to pay back debt. In the baseline model, we assume that the taxes to pay back the
debt are entirely levied as lump sum transfers on OLG households. Some plausible alternative
assumptions are that the primary surplus is raised (i) by cutting government spending or (ii)
via distortionary taxation on labor. Appendix Figure B.3 studies the impulse response of the
economy under these alternative assumptions, and finds that the inflation response changes
little.
Alternative allocations of transfers in the American Rescue Plan. In the baseline, we as-
sume that transfers from the American Rescue Plan are allocated equally to OLG and hand to
mouth households. Other assumptions are possible. For instance, unemployment insurance
could have been received only by hand to mouth households. Similarly, aid to businesses po-
tentially went to lower MPC households. With these alternative assumptions, the impulse re-
sponse of inflation changes little, as Appendix Figure B.5 shows. The output response is more
front loaded but less persistent. As the inflation response is based on the net present value of
the output response these two effects roughly cancel.
Alternative monetary policy. Instead of assuming that monetary policy switches to a deter-
minate Taylor Rule after 10 years, we also consider an alternative rule which implements steady
state output and inflation after 10 years. Appendix Figure B.7 plots the impulse response under
36
this alternative terminal condition, which changes little.
Calibrating to alternative intertemporal marginal propensities to consume. Our base-
line model calibrates to the intertemporal marginal propensities to consume of Fagereng et al.
(2021), following for instance Angeletos et al. (2023) and Auclert, Rognlie & Straub (2023). Fagereng
et al. (2021) find a marginal propensity to consume of 0.51 after 1 year and 0.16 after two years.
However, other work estimates a smaller and more transitory consumption response to trans-
fers (e.g. Orchard et al. 2023; Boehm et al. 2023). Matching the alternative consumption re-
sponse means that the HANK model can no longer match the inflation multiplier. For instance,
Boehm et al. (2023) estimate a marginal propensity to consume of 0.23 in the first quarter, and
zero afterwards. We calibrate the model to this alternative moment, and present the results in
Appendix Section B.4.5. Figure B.6 shows that with the alternative calibration, the inflation re-
sponse in the model is significantly lower than the data. This result is an important qualification
to our finding that HANK models perform well during the post-Pandemic Inflation. One rea-
son to prefer our baseline calibration is that it matches the fact that long horizon interest rates
rise after the fiscal shock, given the increase in the long-run stock of government debt. As we
discuss in the Appendix, the alternative calibration requires an implausibly low intertemporal
elasticity of substitution in order to account for the long term behavior of interest rates.
Fiscal Theory of the Price Level. We now discuss how our results relate to the fiscal theory
of the price level (FTPL). We have shown that a HANK model can match our estimates of the
causal effects of deficits on inflation. One question is how a leading alternative model linking
deficits to inflation fares, namely the FTPL. In Appendix Section B.4.7 we explore a simple FTPL
model similar to Bianchi et al. (2023), with partly unfunded deficits, long term debt, and a Taylor
Rule. The alternative model is also capable of matching our estimate of the size and response
of inflation dynamics to the deficit shock. Our result echoes a message from Angeletos et al.
(2024), that both HANK and FTPL models are capable of matching the response of inflation to
deficits.
7 Conclusion
An important question in macroeconomics is whether deficits raise inflation, especially in the
context of the recent, post-Pandemic inflation. This paper proposes a “high frequency narrative
approach”, to measure the causal effect of the 2021 deficits on inflation. We identify an event
that released news about the 2021 deficits—the Georgia Senate election runoff. We calculate
the shock to expected deficits from the runoff, using new narrative data from investment banks.
37
We next measure the high frequency response of inflation forecasts, using inflation swaps. We
combine the high frequency and narrative information to estimate an inflation multiplier of
0.18% over two years, implying the 2021 deficits caused around 30% of the excess inflation in
2021 and 2022. Last, we confront standard HANK models with our estimate of the inflation
multiplier. With a calibration to pre-2020 data, HANK models successfully match the size and
dynamics of the inflation response.
We believe our high frequency narrative approach could be fruitfully applied to estimate the
causal effect of other single, episode specific shocks on the economy. This method is useful be-
cause certain episodes, such as the 1980s Disinflation or the Great Depression, are particularly
influential to macroeconomists.
38
8 Tables
Table 1: Expected Stimulus after Democrat Victory
Date Bank Number, Exact Phrasing
$(billion)
06.01.2021 Goldman Sachs 750 “With control of the Senate by a narrow margin, Democrats are likely to pass further fiscal
stimulus in Q1 that we expect to total about $750bn.”
06.01.2021 BNP Paribas 1000 “We expect the unified Democratic government to enact significantly more near-term
spending – upwards of $1trn, split between Covid-19 and non-Covid related fiscal sup-
port – than under our previous assumption of a GOP-led Senate and divided govern-
ment.”
06.01.2021 Jefferies 1000 “Jefferies LLC economists ... see Democratic victories in both seats spurring an addi-
tional $1 trillion of stimulus in the next few months.”
06.01.2021 Capital Economics 0 “We are not going to be factoring in any further fiscal stimulus into our forecasts yet.”
07.01.2021 JP Morgan Wealth Man- 750 “We are assuming another support bill of around $750 billion will be passed sometime
agement between February and early April.”
07.01.2021 JP Morgan 900 “Our best guess ... is a spending package of around $900 billion passed in the next few
months.”
07.01.2021 Deutsche Bank 900 “In the first quarter, we anticipate passage of a bill of approximately $900bn.”
08.01.2021 UBS 500 “We would expect a fiscal package of roughly $500bn following the inauguration.”
08.01.2021 Barclays 1400 “We assume over $1.4trn in additional aid following the outcome of this week’s Senate
runoffs in Georgia.”
10.01.2021 Moody’s Analytics 750 “Fiscal support from the new Biden administration and Congress is expected to include
an additional $750 billion to help the economy through to the end of the pandemic.”
11.01.2021 Bank of America Corp 1000 “A Blue Wave increases the likelihood of an immediate $1 trillion Covid stimulus.”
11.01.2021 Morgan Stanley 1000 “We expect an additional US$1 trillion for Covid-19 aid in the near term.”
Median of Expected Stimulus after Democrat Victory: $900 bn
Notes: The number is taken from the reports of investment banks after elections. The sample is restricted to be from 6th of January until 13th of
January. For cases where the range is given, the median of the range is taken.
Table 2: Single Event Study—Effect on Inflation Forecasts from Swaps
Panel A: Percentage point increase in the price level from inflation swaps over 1 year
Baseline Ends Jan 6 Drop 6 weeks Linear Hourly Difference
missing Trend
(1) (2) (3) (4) (5) (6) (7)
Change in Forecasts 0.28 0.18 0.28 0.28 0.27 0.25 0.24
(0.07) (0.05) (0.12) (0.12) (0.008) (0.03) (0.06)
Observations 231 231 231 715 232 80 80
Panel B: Percentage point increase in the price level from inflation swaps over 2 years
Change in Forecasts 0.39 0.19 0.38 0.32 0.39 0.39 0.41
(0.05) (0.04) (0.08) (0.09) (0.01) (0.05) (0.13)
Observations 656 656 656 2032 657 139 139
Panel C: Percentage point increase in the price level from inflation swaps over 5 years
Change in Forecasts 0.58 0.29 0.58 0.58 0.56 0.58 0.58
(0.22) (0.16) (0.23) (0.17) (0.03) (0.16) (0.19)
Observations 1048 1048 1048 3064 1049 194 194
Panel D: Percentage point increase in the price level from inflation swaps over 10 years
Change in Forecasts 0.77 0.44 0.75 0.75 0.72 0.75 0.74
(0.18) (0.14) (0.49) (0.28) (0.06) (0.25) (0.38)
Observations 646 646 646 2060 647 145 145
Note: Each panel corresponds to the expected percentage point increase in the price level over a specific maturity. The
data for forecasts comes from inflation swaps at 10-minute frequency. In all panels, we calculate the increase in inflation
expectations compared to a counterfactual estimated on data from before January 5th. In Column (1), we fit an ARIMA
model to the data from the start of December 18th, 2020 to the start of January 5th, and calculate the effect at the end
of January 6th. Column (2) calculates the effect at 2 PM on January 6th. Column (3) drops missing values. Column (4)
estimates the ARIMA over a 6 week period before January 5th. Column (5) estimates the counterfactual as a linear trend,
and uses Newey-West standard errors with lag length from Lazarus et al. (2018). Column (6) aggregates to hourly frequency
and estimates the ARIMA. Column (7) aggregates to hourly frequency and calculates the effect simply as the change over
the event window. Here, the standard error is the standard error of price changes over the pre-period, scaled to the length
of the event window and assuming that hourly price changes in the pre-period are uncorrelated.
Table 3: Regression Estimates—Effect on Inflation
Panel A: Percentage point increase in the price level from inflation swaps over 1 year
Full Sample Before Jan 5 Outliers Dropped Diff
(1) (2) (3) (4)
Democrat Win Probability 0.96 3.28 1.03 0.1
(0.20) (0.26) (0.25) (0.06)
Observations 46 40 41 44
Panel B: Percentage point increase in the price level from inflation swaps over 2 years
Democrat Win Probability 1.48 4.69 1.57 0.16
(0.28) (0.38) (0.34) (0.11)
Observations 46 40 41 44
Panel C: Percentage point increase in the price level from inflation swaps over 5 years
Democrat Win Probability 2.2 6.87 2.35 0.46
(0.41) (0.61) (0.48) (0.15)
Observations 46 40 41 44
Panel D: Percentage point increase in the price level from inflation swaps over 10 years
Democrat Win Probability 3.05 10.9 3.27 0.49
(0.69) (1.14) (0.81) (0.2)
Observations 46 40 41 44
Note: Each panel in the table presents a different horizon for changes in the price level. In all panels, we adjust the price of
the inflation swap to take into account the 3-month lag of the inflation index used in the contracts. For all panels, we regress
the expected increase in the price level on the lagged probability of a Democratic win in the 2021 Georgia Senate election. We
use Newey-West standard errors with automatic lag length selection from Lazarus et al. (2018). Our dataset is daily, sourcing
expected increases in the price level from zero-coupon inflation swaps from Bloomberg, and probabilities of a Democratic
victory from PredictIt’s 2020 Senate election betting prices. The data spans November 9, 2020, to January 12, 2021. Column
(1) analyzes the entire dataset. Column (2) considers only data gathered before January 5, 2021. Column (3) omits data from
outliers, namely the 6th and 7th of January and the 2nd-4th December. Lastly, in Column (4), the analysis uses the differenced
values of both dependent and independent variables with lag 5. Counts refer to the number of daily observations.
41
Table 4: Single Event Study—Effect on Nominal Interest Rates
Panel A: Percentage point increase in nominal interest rates over 1 year
Baseline Ends Jan 6 Drop Missing 4 weeks Linear Trend Hourly Difference
(1) (2) (3) (4) (5) (6) (7)
Jump in Interest Rate 0.005 0.014 0.005 0.005 0.004 0.007 0.007
(0.009) (0.006) (0.015) (0.009) (0.003) (0.006) (0.01)
Observations 570 570 570 1311 571 100 100
Panel B: Percentage point increase in nominal interest rates over 5 years, after 5 years
Jump in Interest Rate 0.202 0.161 0.202 0.202 0.202 0.218 0.199
(0.08) (0.059) (0.106) (0.098) (0.006) (0.019) (0.069)
Observations 570 570 570 1311 571 101 100
Note: Each panel corresponds to the percentage point increase in the interest rate over a specific maturity. The data for
interest rates come from the intraday prices of US government treasuries at 10-minute frequency, sourced from CME
group. We calculate the zero-coupon yield of the treasuries using bootstrapping and interpolate using a cubic smooth-
ing spline. In all panels, we calculate the increase in interest rates compared to the counterfactual scenario where the
series would have continued to behave as before the beginning of January 5th, 2021, just before the announcement of
the Georgia election results. In column (4) we force the algorithm to choose a stationary ARIMA model, in all other
columns we let the algorithm to choose either a stationary or non-stationary ARIMA model, using the Bayesian Infor-
mation Criterion. In Column (1), we fit a non-stationary ARIMA model to the data from the start of December 18th,
2020 to the start of January 5th. Column (2) sets the counterfactual at 2:00 PM on January 6th, 2021 and fits a non-
stationary ARIMA model as well. drops all missing values and then fits the non-stationary ARIMA model. Column (3)
drops all missing values and then fits the non-stationary ARIMA model. Column (4) estimates the ARIMA over a 4 week
period before January 5th. Column (5) estimates the counterfactual as a linear trend, and uses Newey-West standard
errors with lag length from Lazarus et al. (2018). Column (6) aggregates to hourly frequency and estimates the ARIMA.
Column (7) aggregates to hourly frequency and calculates the effect simply as the change over the event window. Here,
the standard error is the standard error of price changes over the pre-period, scaled to the length of the event window
and assuming that hourly price changes in the pre-period are uncorrelated.
42
Table 5: Calibration of Model
Parameter Description Value Target
Households
µ Share of hand-to-mouth 0.275
1 & 2 year intertemporal MPC
φ OLG survival rate 0.68
σ Intertemporal elasticity of substitution 1 Standard
ϕ Frisch elasticity 1 Standard
β Discount factor 0.99 Standard
Nominal rigidities
κ Phillips Curve slope 0.055 Hazell et al. (2022)
Steady State Fiscal
B ss /Y ss Steady state Debt-to-GDP 0.8 OMB (2024)
τy Marginal tax rate 0.27 CBO (2019)
G ss /Y ss Gov’t spending-to-GDP 0.2 BEA (2024)
Fiscal Rule
τB Response of surpluses to debt 0.189 Persistence of debt, CBO (2021b)
H Period where debt repayment starts 3 CBO (2021a)
B̄ /Y ss Steady state Debt-to-GDP after shocks 80.6% 9 year ahead 1 year interest rate
Notes: this table reports each parameter and its source for the calibration. The intertemporal MPCs are from Fagereng et al. (2021). We discuss
in Appendix Section B.3 how we calibrate τB to match the long horizon persistence of debt, after the American Rescue Plan, from CBO (2021b).
43
References
Angeletos, G.-M., Lian, C. & Wolf, C. K. (2023), Can deficits finance themselves?, Technical report, Na-
tional Bureau of Economic Research.
Angeletos, G.-M., Lian, C. & Wolf, C. K. (2024), ‘Deficits and inflation: Hank meets FTPL’.
Antolín-Díaz, J. & Rubio-Ramírez, J. F. (2018), ‘Narrative sign restrictions for svars’, American Economic
Review 108(10), 2802–2829.
Auclert, A., Bardóczy, B. & Rognlie, M. (2023), ‘MPCs, MPEs, and multipliers: A trilemma for New Keyne-
sian models’, Review of Economics and Statistics 105(3), 700–712.
Auclert, A., Rognlie, M. & Straub, L. (2020), Micro Jumps, Macro Humps: Monetary Policy and Business
Cycles in an Estimated HANK Model, NBER Working Papers 26647, National Bureau of Economic Re-
search, Inc.
Auclert, A., Rognlie, M. & Straub, L. (2023), The intertemporal keynesian cross, Technical report, National
Bureau of Economic Research.
Bahaj, S. (2020), ‘Sovereign spreads in the euro area: Cross border transmission and macroeconomic
implications’, Journal of Monetary Economics 110, 116–135.
Bahaj, S., Czech, R., Ding, S. & Reis, R. (2023), ‘The market for inflation risk’, Available at SSRN 4488881 .
Barnichon, R. & Mesters, G. (2020), ‘Identifying modern macro equations with old shocks’, The Quarterly
Journal of Economics 135(4), 2255–2298.
Barnichon, R. & Shapiro, A. H. (2024), ‘Phillips meets beveridge’, Journal of Monetary Economics
p. 103660.
Barro, R. J. & Bianchi, F. (2023), Fiscal influences on inflation in OECD countries, 2020-2022, Technical
report, National Bureau of Economic Research.
Bauer, M. D., Pflueger, C. E. & Sunderam, A. (2024), Changing perceptions and post-pandemic monetary
policy, in ‘Jackson Hole Symposium, Federal Reserve Bank of Kansas City, available at https://www.
kansascityfed. org/Jackson% 20Hole/documents/10337/pflueger_jh. pdf’.
Bauer, M. D. & Swanson, E. T. (2023), ‘A reassessment of monetary policy surprises and high-frequency
identification’, NBER Macroeconomics Annual 37(1), 87–155.
Beaudry, P., Hou, C. & Portier, F. (2024), The dominant role of expectations and broad-based supply
shocks in driving inflation, Technical report, National Bureau of Economic Research.
Benigno, P. & Eggertsson, G. B. (2023), Its baaack: The surge in inflation in the 2020s and the return of the
non-linear Phillips curve, Technical report, National Bureau of Economic Research.
Bernanke, B. & Blanchard, O. (2023), ‘What caused the US pandemic-era inflation?’, Hutchins Center
Working Papers .
Bianchi, F., Faccini, R. & Melosi, L. (2023), ‘A fiscal theory of persistent inflation’, The Quarterly Journal of
Economics p. qjad027.
Bigio, S., Caramp, N. & Silva, D. (2024), ‘Sticky inflation: Some unpleasant new keynesian arithmetic’.
Blanchard, O. (2021), In defense of concerns over the $1.9 trillion relief plan, Technical report, Peterson
Institute for International Economics.
Blanchard, O. (2023), Macroeconomics Global Edition, Pearson Deutschland.
URL: https://elibrary.pearson.de/book/99.150005/9781292351520
Blanchard, O. J. (1985), ‘Debt, deficits, and finite horizons’, Journal of political economy 93(2), 223–247.
44
Blanco, A., Boar, C., Jones, C. J. & Midrigan, V. (2024), Nonlinear dynamics in menu cost economies?
evidence from us data, Technical report, National Bureau of Economic Research.
Bocola, L., Dovis, A., Jørgensen, K. & Kirpalani, R. (2024), Bond market views of the fed, Technical report,
National Bureau of Economic Research.
Boehm, J., Fize, E. & Jaravel, X. (2023), ‘Five facts about MPCs: Evidence from a randomized experiment’.
Campos, R. G., Fernández-Villaverde, J., Nuño, G. & Paz, P. (2024), Navigating by falling stars: monetary
policy with fiscally driven natural rates, Technical report, National Bureau of Economic Research.
CBO (2019), Marginal Federal Tax Rates on Labor Income: 1962 to 2028 , Technical report, Congressional
Budget Office.
CBO (2021a), Estimated Budgetary Effects of H.R. 1319, American Rescue Plan Act of 2021 , Technical
report, Congressional Budget Office.
CBO (2021b), Three Scenarios for the Budget as Specified by Senator Graham , Technical report, Con-
gressional Budget Office.
Cerrato, A. & Gitti, G. (2022), ‘Inflation since covid: Demand or supply’, Available at SSRN 4193594 .
Cieslak, A., McMahon, M. & Pang, H. (2024), ‘Did I make myself clear? the fed and the market in the
post-2020 framework period’.
Cieslak, A., Morse, A. & Vissing-Jorgensen, A. (2019), ‘Stock returns over the FOMC cycle’, The Journal of
Finance 74(5), 2201–2248.
Cieslak, A. & Pflueger, C. (2023), ‘Inflation and asset returns’, Annual Review of Financial Economics
15(1), 433–448.
Cloyne, J. (2013), ‘Discretionary tax changes and the macroeconomy: new narrative evidence from the
united kingdom’, American Economic Review 103(4), 1507–1528.
Cloyne, J., Dimsdale, N. & Hürtgen, P. (2025), ‘Are tax cuts contractionary at the zero lower bound? evi-
dence from a century of data’, Journal of Political Economy 133(2), 000–000.
Cloyne, J., Dimsdale, N. & Postel-Vinay, N. (2024), ‘Taxes and growth: new narrative evidence from inter-
war britain’, Review of Economic Studies 91(4), 2168–2200.
Cochrane, J. (2023), The Fiscal Theory of the Price Level, Princeton University Press.
Cochrane, J. H. (2022), Inflation past, present and future: Fiscal shocks, Fed response, and fiscal limits,
Technical report, National Bureau of Economic Research.
Coglianese, J. M., Olsson, M. & Patterson, C. (2023), ‘Monetary policy and the labor market: a quasi-
experiment in Sweden’, University of Chicago, Becker Friedman Institute for Economics Working Paper
(2023-123).
Coibion, O. & Gorodnichenko, Y. (2015), ‘Information rigidity and the expectations formation process: A
simple framework and new facts’, American Economic Review 105(8), 2644–2678.
Crump, R. K., Eusepi, S., Giannoni, M. & Şahin, A. (2024), ‘The unemployment–inflation trade-off revis-
ited: The phillips curve in covid times’, Journal of Monetary Economics p. 103580.
Diercks, A. M., Campbell, C., Sharpe, S. A. & Soques, D. (2023), ‘The swaps strike back: Evaluating expec-
tations of one-year inflation’.
Drechsel, T. (2024), Estimating the effects of political pressure on the Fed: a narrative approach with new
data, Technical report, National Bureau of Economic Research.
Edelberg, W. & Sheiner, L. (2021), ‘The macroeconomic implications of bidens $1.9 trillion fiscal package’.
45
Erceg, C. J., Henderson, D. W. & Levin, A. T. (2000), ‘Optimal monetary policy with staggered wage and
price contracts’, Journal of monetary Economics 46(2), 281–313.
Fagereng, A., Holm, M. B. & Natvik, G. J. (2021), ‘MPC heterogeneity and household balance sheets’,
American Economic Journal: Macroeconomics 13(4), 1–54.
Federal Reserve Bank of Dallas (2024), ‘Market value of gross federal debt [mvgfd027mnfrbdal]’, Retrieved
from FRED, Federal Reserve Bank of St. Louis. Accessed September 2, 2024.
Ferrante, F., Graves, S. & Iacoviello, M. (2023), ‘The inflationary effects of sectoral reallocation’, Journal of
Monetary Economics 140, S64–S81.
Friedman, M. (1968), ‘The role of monetary policy’, The American Economic Review 58(1).
Friedman, M. & Schwartz, A. J. (1963), A monetary history of the United States, 1867-1960, Vol. 9, Princeton
University Press.
Gagliardone, L. & Gertler, M. (2023), Oil prices, monetary policy and inflation surges, Technical report,
National Bureau of Economic Research.
Gertler, M. & Karadi, P. (2015), ‘Monetary policy surprises, credit costs, and economic activity’, American
Economic Journal: Macroeconomics 7(1), 44–76.
Giannone, D. & Primiceri, G. E. (2024), The drivers of post-pandemic inflation, in ‘Conference Proceed-
ings: Monetary Policy in an era of transformation, ECB Forum on Central Banking’.
Gitti, G. (2024), ‘Nonlinearities in the regional phillips curve with labor market tightness’.
Gomez Cram, R., Kung, H. & Lustig, H. N. (2023), ‘Can US treasury markets add and subtract?’, Available
at SSRN .
Gormsen, N. J. & Koijen, R. S. (2020), ‘Coronavirus: Impact on stock prices and growth expectations’, The
Review of Asset Pricing Studies 10(4), 574–597.
Gormsen, N. J., Koijen, R. S. & Martin, I. W. (2021), Implied dividend volatility and expected growth, in
‘AEA Papers and Proceedings’, Vol. 111, American Economic Association 2014 Broadway, Suite 305,
Nashville, TN 37203, pp. 361–365.
Guerrieri, V., Marcussen, M., Reichlin, L. & Tenreyro, S. (2023), ‘Geneva report: The art and science of
patience: relative prices and inflation’.
Gürkaynak, R. S., Sack, B. P. & Swanson, E. T. (2004), ‘Do actions speak louder than words? the response
of asset prices to monetary policy actions and statements’, The Response of Asset Prices to Monetary
Policy Actions and Statements (November 2004) .
Hazell, J. (2024), ‘Comment on âthe dominant role of expectations and broad-based supply shocks in
driving inflationâ(beaudry, hou and portier, nber macro annual 2024)’.
Hazell, J., Herreno, J., Nakamura, E. & Steinsson, J. (2022), ‘The slope of the Phillips curve: evidence from
US states’, The Quarterly Journal of Economics 137(3), 1299–1344.
Känzig, D. R. (2021), ‘The macroeconomic effects of oil supply news: Evidence from OPEC announce-
ments’, American Economic Review 111(4), 1092–1125.
Känzig, D. R. (2023), The unequal economic consequences of carbon pricing, Technical report, National
Bureau of Economic Research.
Kaplan, G., Moll, B. & Violante, G. L. (2018), ‘Monetary policy according to hank’, American Economic
Review .
Keynes, J. M. (1936), The General Theory of Employment, Interest and Money, Macmillan. 14th edition,
1973.
46
Lazarus, E., Lewis, D. J., Stock, J. H. & Watson, M. W. (2018), ‘Har inference: Recommendations for prac-
tice’, Journal of Business & Economic Statistics 36(4), 541–559.
Leeper, E. M., Walker, T. B. & Yang, S.-C. S. (2010), ‘Government investment and fiscal stimulus’, Journal
of monetary Economics 57(8), 1000–1012.
Levy, A. (2021), ‘âthe slope of the phillips curve: having fun with a back of the envelope computation
based on the georgia runoff election ...â’. Tweet.
URL: https://twitter.com/LevyAntoine/status/1358367460555165699
MacKinlay, A. C. (1997), ‘Event studies in economics and finance’, Journal of economic literature
35(1), 13–39.
Mian, A., Straub, L. & Sufi, A. (2024), ‘A goldilocks theory of fiscal policy’, NBER Working Paper
29351(6), 18.
Mondragon, J. A. & Wieland, J. (2022), Housing demand and remote work, Technical report, National
Bureau of Economic Research.
Nakamura, E. & Steinsson, J. (2018), ‘High-frequency identification of monetary non-neutrality: the in-
formation effect’, The Quarterly Journal of Economics 133(3), 1283–1330.
Orchard, J., Ramey, V. A. & Wieland, J. F. (2023), Micro mpcs and macro counterfactuals: the case of the
2008 rebates, Technical report, National Bureau of Economic Research.
Orphanides, A. & Williams, J. C. (2002), ‘Robust monetary policy rules with unknown natural rates’,
Brookings Papers on Economic Activity 2002(2), 63–145.
Pflueger, C. E. & Viceira, L. M. (2016), ‘Return predictability in the treasury market: real rates, inflation,
and liquidity’, Handbook of Fixed-Income Securities pp. 191–209.
Ramey, V. A. (2011), ‘Identifying government spending shocks: It’s all in the timing’, The Quarterly Journal
of Economics 126(1), 1–50.
Ramey, V. A. (2019), ‘Ten years after the financial crisis: What have we learned from the renaissance in
fiscal research?’, Journal of Economic Perspectives 33(2), 89–114.
Ramey, V. A. (2021), ‘The macroeconomic consequences of infrastructure investment’, Economic Analysis
and Infrastructure Investment p. 219.
Ramey, V. A. & Shapiro, M. D. (1998), Costly capital reallocation and the effects of government spending,
in ‘Carnegie-Rochester conference series on public policy’, Vol. 48, Elsevier, pp. 145–194.
Reis, R. (2022), The burst of high inflation in 2021-22: How and why did we get here?, Bank for Interna-
tional Settlements, Monetary and Economic Department.
Romer, C. D. & Romer, D. H. (1989), ‘Does monetary policy matter? a new test in the spirit of friedman
and schwartz’, NBER macroeconomics annual 4, 121–170.
Romer, C. D. & Romer, D. H. (2010), ‘The macroeconomic effects of tax changes: estimates based on a
new measure of fiscal shocks’, American Economic Review 100(3), 763–801.
Rotemberg, J. J. (1982), ‘Sticky prices in the United States’, Journal of political economy 90(6), 1187–1211.
Swanson, E. & Jayawickrema, V. (2023), ‘Speeches by the Fed chair are more important than FOMC an-
nouncements: An improved high-frequency measure of US monetary policy shocks’, Unpublished
Manuscript .
U.S. Bureau of Economic Analysis (2024), ‘Shares of gross domestic product: Government consumption
expenditures and gross investment [a822re1q156nbea]’, Retrieved from FRED, Federal Reserve Bank
of St. Louis. Accessed September 2, 2024.
47
U.S. Office of Management and Budget and Federal Reserve Bank of St. Louis (2024), ‘Gross federal debt
held by the public as percent of gross domestic product [fypugda188s]’. Retrieved September 2, 2024.
Velde, F. R. (2009), ‘Chronicle of a deflation unforetold’, Journal of Political Economy 117(4), 591–634.
Wolf, C. K. (2021), Interest rate cuts vs. stimulus payments: An equivalence result, Technical report, Na-
tional Bureau of Economic Research.
48
Appendix (For Online Publication)
Jonathon Hazell Stephan Hobler
A Response of Forecasts vs. Actual Inflation
This section shows that if inflation forecasts underreact to shocks, so that β > 0 in regression
equation (2), then the response of inflation forecasts is a lower bound for the response of ac-
tual inflation. To make this point, we study the noisy information model of Coibion & Gorod-
nichenko (2015).
Suppose that inflation follows an AR(1) process
πt = ρπt −1 + νt (A.1)
where νt is an i.i.d. normally distributed innovation to inflation. A deficit shock could represent
one such innovation. Agents cannot observe inflation directly but instead receive a signal
y i t = πt + ωi t
where ωi t is normally distributed mean-zero noise that is identically distributed across time
and agents.
Coibion & Gorodnichenko (2015) show that agents forecast inflation h periods ahead, F i t πt +h ,
according to
F i t πt = G y i t + (1 −G) F i ,t −1 πt (A.2)
F i t πt +h = ρ h F i t πt (A.3)
where G is the Kalman gain representing the weight on new information relative to previous
forecasts.
Averaging across agents and rearranging implies a relationship between ex post mean fore-
cast errors and ex ante mean forecast revisions given by
1 −G
πt +h − F t πt +h = (F t πt +h − F t −1 πt +h ) + νt +h,t , (A.4)
G
Ph h− j
where νt +h,t = j =1 ρ νt + j . Comparing this equation to regression equation (2) from the
main text, we can see that β = (1 − G)/G. That is, the regression coefficient is positive only if
G < 1 so that there is underreaction.
49
We now show that if there is underreaction, then the reponse of the forecast is a lower bound
for the response of the actual variable. It follows from the AR(1) process (A.1) that the actual re-
sponse of inflation at t +h to a unit impulse of the shock is ρ h . From equation (A.2), the response
of the current belief about inflation, F i t πt , to the shock is G, since: F i ,t −1 πt is predetermined
with respect to the shock, ωi t is unaffected by the shock , and πt responds by 1 unit. Therefore,
using equation (A.3) the response of F i t πt +h to the shock is Gρ h . Therefore the response of ex-
pectations to the shock, Gρ h , is a lower bound for the response of the actual variable ρ h , as long
as G < 1 i.e. there is underreaction.
B Model Appendix
B.1 Additional Derivations
This subsection contains additional derivations for the consumption and wage setting blocks
of the model. We will need these derivations in order to present the full set of log-linearized
equations characterizing the equilibrium of the model, which we present in the next section.
B.1.1 Consumption Block
It will be convenient to index household i by the cohort j that they belong to. As age is the only
source of heterogeneity amongst savers this can be done without loss of generality. A household
in cohort j chooses the sequence of consumption and savings to maximize
1− σ1
+∞ C j +s,t +s − 1
(βφ)s
X
max (B.1)
1 − σ1
© ª
C j +s,t +s ,A j +1+s,t +s s=0
subject to
p
1 + Rt
C j ,t + A j +1,t = A j ,t −1 + Y td + Z j t , (B.2)
φ
p
where 1 + R t = 1+I
Πt
t −1
denotes the ex-post real return on the nominal asset, Y td = (1 − τ y )Y t − T t
denotes disposable income of OLG households, and Z j ,t the cohort-specific social fund pay-
ments.
Optimality conditions. The first order condition of the household problem gives the standard
Euler equation as annuity markets compensate households for the mortality risk
−1 p ¢ −σ
1
C j ,tσ = β 1 + R t +1 C j +1,t
¡
+1
. (B.3)
50
Combining the Euler equation with the net present value budget constraint lets us characterize
the consumption function for a household in cohort j as
·+∞ ¸−1 Ã p !
1 + R
(φβσ )s R σ−1 t
A j t −1 + Ωhj,t + Ωzj,t ,
X
C j ,t = t ,t +s (B.4)
s=0 φ
Qs p
where R t ,t +s = k=1
(1 + R t +k ) with the normalization R t ,t = 1, and human wealth is defined
recursively
φ
Ωhj,t = Y td + h
p Ω j +1,t +1 , (B.5)
1 + R t +1
and the net present value of social fund payments is given by
φ
Ωzj,t = Z j ,t + p Ωzj +1,t +1 . (B.6)
1 + R t +1
The budget constraint (B.2) completes the characterization of the individual cohort problem.
Aggregation. Next we aggregate individual policies across cohorts. As is well known, the con-
stant survival probability gives rise to a geometric distribution across cohorts. Summing over
households, let us define aggregates as
+∞
(1 − φ)φ j C j ,t
X
Ct =
j =0
+∞
1 X
At = (1 − φ)φ j A j ,t
φ j =1
+∞
Ωht = (1 − φ)φ j Ωhj,t
X
j =0
+∞
Ωzt = (1 − φ)φ j Ωzj,t
X
j =0
First, since assets are defined in terms of end of period and cohorts are born with zero assets,
aggregation of assets starts at j = 1 and is pre-multiplied by 1/φ. Noting that the social fund
payments net out on aggregate, the budget constraint aggregates to
p
C t + A t = (1 + R t )A t −1 + Y td . (B.7)
On aggregate current generations need to fund the social fund payments for future newborn
generations. Specifically, the social fund must raise (1−φ)Z new in the next period to fund trans-
fers to the mass of 1 − φ newborn households. This must be financed by households that are
51
currently alive. As households need to pay this cost only in the next period it is discounted by
1
p . It follows that the dynamics of the net present value of the aggregate social fund pay-
1+R t +1
ments of households currently alive is given by
1−φ φ z
Ωzt = − p Z
new
+ p Ωt +1 . (B.8)
1 + R t +1 1 + R t +1
Given these elements and for a given sequence of disposable aggregate income and ex-post real
p
rates {Y td , R t }, we can fully characterize the aggregate consumption function as
·+∞ ¸−1 ³ ´
σ s σ−1 p
(φβ ) R t ,t +s (1 + R t )A t −1 + Ωht + Ωzt ,
X
Ct = (B.9)
s=0
where the net present value of human wealth is given by
φ
Ωht = Y td + h
p Ωt +1 , (B.10)
1 + R t +1
and the net present value of social fund payments Ωzt is given by (B.8). Finally, the aggregate
budget constraint is given by (B.7).
Savings. Finally, we derive an expression for the steady state level of savings of households in
the baseline overlapping generations model. From equations (B.10) and (B.8) we have Ωhss =
1 1 1−φ
φ Y ssd and Ωzss = − φ Z new . Plugging into the aggregate consumption function and
1− 1+r 1− 1+r 1+r ss
ss ss
using the fact that Y ssd = C ss − r ss A ss and Z new = (1 + r ss )A ss yields
φ ¡
(β(1 + r ss ))σ − 1
¢
1+r ss 1
A ss = iC ss + Z new . (B.11)
φ
h
(1 − φ) 1 − 1+r ss (β(1 + r ss ))σ 1 + r ss
∂ log A φβ
The elasticity of steady state savings with respect to the real rate is ∂ log(1+r ) = σ (1−φ)(1−φβ) CAssss −1,
which we use to pin down the new steady state value of government debt B .
B.1.2 Wage Setting
We follow standard practice and assume that wages are sticky as in Erceg et al. (2000) and Au-
clert, Rognlie & Straub (2023). The exposition in this Appendix follows Auclert, Rognlie & Straub
(2023). There is a continuum of unions k set nominal wages and uniform working hours for
their members. Each worker i is part of a union k and there is no sorting of workers into unions.
Workers are homogeneous and do not differ in their productivity. A competitive labor packer
52
combines labor from each union into an aggregate input using the standard CES aggregator
µZ ε−1
¶ ε
ε−1
ε
Nt = Nkt d k . (B.12)
These services are sold to firms at a nominal wage Wtn .
n
Union maximizes the utility of their members by setting the nominal wage Wkt subject to a
Rotemberg adjustment cost
(Z Ã n !2 )
s ψ Wk,t +s
β
X £ ¤
max u(C i ,t +s ) − v(Nk,t +s ) d i − n −1 , (B.13)
n
{Wkt }
+s s≥0 s≥0
2 Wk,t +s−1
subject to
n ¶−ε
Wkt
µ
Nkt = Nt (B.14)
Wtn
µZ ¶ 1
1−ε
n n 1−ε
Wt = (Wkt ) d k (B.15)
and n
(1 − τ y,t ) Wkt Nkt − T H
if i = H ,
Pt t
C i ,t = n (B.16)
1 1+I t −1 A i ,t −1 + (1 − τ y ) Wkt Nkt − Ti ,t + Zi ,t − A i ,t +1
h i
if i ⊂ S .
φ Πt Pt
where S denotes the indices of savers. For generality we include time-varying distortionary
labor taxation.
By an application of the Envelope Theorem unions only consider the direct effect of wages
on household utility and set a wage that satisfies the first order condition
n
1 − τ y,t +s ∂[Wkt N ] ∂Nkt +s
µZ ¶
′ +s kt +s
u (c i ,t +s )d i − v ′ (Nkt s )
P t +s ∂Wkt
n
+s
∂Wkt
n
s
à n + (B.17)
n Wkn
à ! !
Wkt +s 1 Wkt +s+1
−ψ n −1 n + βψ n −1 n
t +s+1
2
=0
Wkt +s−1
W kt +s−1
W kt +s
(W kt +s
)
n
∂[Wkt N
+s kt +s
]
Next, we know from the labor demand that ∂Wkt n = (1 − ε)Nkt +s . Plugging back in,
+s
n
n W N
multiplying both sides by Wkt +s
, using the fact that tP+st +st +s = Y t +s , and focusing on a symmetric
equilibrium we obtain the Wage New Keynesian Phillips Curve
ε ε−1
½ µZ ¶¾
πw w
t (1 + πt ) =
′
N t v (N t ) − (1 − τ y,t )Y t u(c i ,t )d i + βπw w
t +1 (1 − πt +1 ). (B.18)
ψ ε
53
Linearizing around the zero inflation steady state πw = 0, using the fact that all households
have the consumption level in steady state and plugging in the functional forms for u and v
1 d N t 1 dC t d Yt d Nt d τ y,t
½ µ ¶ ¾
πw
t = κW + − − + + βπw
t +1 , (B.19)
ϕ Nt σ C Y N 1 − τy
where κw = ψε v ′ (N )N and d X t = X t − X ss denote deviations from steady state. Since prices are
flexible, firms target a constant markup of one and πt = πw
t . In the baseline model, we abstract
from distortionary taxation, changes in TFP, and assume constant returns to scale in aggregate
labor
1 d N t 1 dC t
½ ¾
πw
t = κw + + βπw
t . (B.20)
ϕ N σ C
Calibration. Let y t = d Y t /Y ss and g t = dG t /Y ss . From the aggregate resource constraint c t =
y t − g t . From the production function y t = n t . Then, we can rearrange equation (B.19) as
¶Ã !
ϕ
µ
1 1 1
πt = κw + yt − g t + βπt +1 . (B.21)
ϕ σ C ss /Y ss ϕ + σ CYssss
³ ´
1 1 1
We calibrate κ = κw ϕ + σ C ss /Y ss to match the empirical estimate from Hazell et al. (2022) of
κ = 0.055 and determine κw = ψε v ′ (N )N residually given our parametrization of ϕ, σ, and the
steady state values for the consumption to output share. The disutility of labor is set to be
consistent with a zero inflation steady state.
B.2 Model Summary
This subsection reports the linearized equations that characterize the equilibrium of the model.
We linearize the model around the initial steady state. Quantity variables are expressed as de-
viations from the initial steady state normalized by steady state output and denoted by lower
case letters, e.g. x t = X tY−X
ss
ss
.
OLG household block. Given a sequence of disposable after-tax income y td = (1 − τ y )y t − t t
p
and ex-post real interest rates r t aggregate consumption and asset dynamics are characterized
by:
(i) Aggregate consumption function.
p
A ss p C ss +∞ r
µ ¶ µ ¶
h z s t +s+1
c t = (1 − φβ) (1 + r ss )a t −1 + r + ωt + ωt − (σ − 1)φβ
X
(φβ) , (B.22)
Y ss t Y ss s=0 1 + r ss
54
where the net present value of the aggregate human capital of households is given by
à ! p
d
βφ Y ss r t +1
ωht = y td − + βφωht+1 , (B.23)
1 − βφ Y ss 1 + r ss
and the net present value of social fund payments is given by
¶ p
(1 − φ)A ss r t +1
µ
1
ωzt = + βφωzt+1 . (B.24)
1 − φβ Y ss 1 + r ss
The aggregate consumption function is standard with the modification of the social fund.
Households have a marginal propensity to consume 1 − φβ out of the total of their finan-
p
cial wealth: (1+r ss )a t −1 + YAssss r t , their human wealth ωht and the net present value of social
fund payments ωzt . Larger interest rates increase the net present value of social fund pay-
ments as future contributions to the fund are discounted more strongly. The last term in
(B.22) reflects the standard income and substitution effects of changes in the interest rate.
(ii) Aggregate budget constraint. As social fund payments net out on aggregate, the aggregate
budget constraint is given by
A ss p
µ ¶
c t + a t = (1 + r ss )a t −1 + r + y td . (B.25)
Y ss t
Further details of the consumption and saving policies of cohorts are shown in Appendix B.1.1.
New Keynesian Phillips Curve. As shown in Appendix B.1.2 the New Keynesian Phillips Curve
is à !
ϕ
πt = κ y t − g t + βπt +1 . (B.26)
ϕ + σ CYssss
Fiscal Policy. The linearized dynamics of government debt are
B ss p
b t = (1 + r ss )b t −1 + r − st . (B.27)
Y ss t
As described in the main text, surpluses are set according to:
(i) Stimulus phase. No additional taxes are levied and the government provides stimulus
{t̃ t , t̃ tH , g̃ t }, with t̃ t , t̃ tH < 0 and g̃ t >. The primary surplus therefore is
s t = µt̃ tH + (1 − µ)t̃ t − g̃ t . (B.28)
55
(ii) Repayment phase. The government levies additional taxes t̂ t on OLG households to sta-
bilize debt dynamics. Specifically, t̂ t is set such that
B ss p
r t + τB (b t −1 − ∆b) − µt̃ tH + (1 − µ)t̃ t − g̃ t ,
£ ¤
s t = r ss b t −1 + (B.29)
Y ss
where ∆b := B −B
Y ss
ss
denotes the change in the target debt-to-output ratio.
Fisher equation. The ex-post real rate is given by the Fisher equation
p
rt it πt
= − . (B.30)
1 + r ss 1 + I ss Πss
Market clearing. The supply of nominal government bonds is absorbed by the OLG house-
holds. That is,
b t = (1 − µ)a t . (B.31)
By Walras’ Law, the good’s market clearing condition µ(1−τ y )y t +(1−µ)c t +g t = y t is redundant.
B.3 Mapping CBO Projections to Fiscal Rule
We elaborate on how we use the CBO projections on the time path of debt to calibrate τB in
the fiscal rule (5). We present the derivations for the case in which real rates are known ex-ante
since in our analysis we consider a time zero MIT shock and H > 0. For t > H , the fiscal rule is
S t = r t B t −1 + τB (B t −1 − B ) − εt . (B.32)
Combining the fiscal rule with the government budget constraint (5) gives
h i
B t = (1 + r t )B t −1 − τB (B t −1 − B ) + r t B t −1 − εt
(B.33)
= τB B + (1 − τB )B t −1 + εt .
Subtracting the initial steady state value from both sides and dividing by the initial steady state
level of output Y ss implies
B t − B ss B − B ss B t −1 − B ss εt
= τB + (1 − τB ) − . (B.34)
Y ss Y ss Y ss Y ss
n o
B t −B ss
CBO (2021b) (Table 1) implies a time series for the increase in the debt per GDP ratio Y ss
under the assumption of constant output, due to the American Rescue Plan, between 2021 and
56
2031. CBO (2021a) provides an estimate of εt , the change in primary deficits due the American
Rescue Plan. We estimate the persistence of debt 1 − τB via a time series regression of B tY−B
ss
ss
on
its first lag using projections from 2024 onward, consistent with our selection of H .
The CBO forecast holds fixed output and therefore ignores effects from an expansion from
the tax base. We restrict estimation to 2024 onwards, at which point according to our model
changes in output due to the American Rescue Plan are relatively small.
B.4 Robustness and Extensions
B.4.1 Effects of Infrastructure Spending
In this Section, we extend the baseline model to introduce infrastructure spending. As dis-
cussed in Section 3, Democrat victory increased expectations of infrastructure spending by an
upper bound of $1 trillion. However this bound is loose, because markets expected Republi-
cans to also enact some infrastructure spending if they were to win. To tighten the bound, we
size the portion of infrastructure spending that would have taken place even under a Republi-
can victory using the realized bipartisan share of infrastructure spending during 2021-2022 of
38%.32 Combined with the 50% chance of a Democrat victory we take the shock to the expecta-
tion of infrastructure spending as 0.5×(1−0.38)×$1trillion = $310 billion. We take the time path
of infrastructure spending from from detailed projections of Moody’s Analytics, which is part of
our narrative evidence. A distinctive feature is that infrastructure spending was expected to be
delayed with most of the spending not expected until after 4 years. The total fiscal stimulus
shock is shown in Figure B.1.
As is standard (e.g. Leeper et al. 2010, Ramey 2021), we model infrastructure spending as
productive government investment that increases public capital which, due to its non-rivalrous
nature, increases total factor productivity (TFP) of firms. We also include time to build de-
lays which have been found to be both empirically relevant and important for model dynamics
(Ramey 2021).
Denote government investment by G tI , public capital K tG , and let public capital be accumu-
lated via
K tG+1 = Φ(L)G tI + (1 − δ)K tG ,
where the operator Φ(L) = ωk L k captures "time to build" and L denotes the standard
P+∞
k=0
32
The two major realized infrastructure bills were a bipartisan bill of $550 billion dollars in November 2021 and
the Inflation Reduction Act of $900 billion dollars, which was opposed by Republicans. Therefore the bipartisan
share of spending is 38%. To be conservative, we not include the $280 billion CHIPS Act, which was also bipartisan,
but arguably a national security bill rather than an infrastructure bill.
57
Figure B.1: Total fiscal stimulus including infrastructure spending.
Notes: Total combined fiscal news from American Rescue Plan Act (ARPA) and the infrastructure spending. Section 3 details how we size the
shocks. Section 6 details how we pin down the time path for the fiscal news from ARPA while we take the time path for the infrastructure
spending from detailed projections by Moody’s Analytics.
lag operator. In our calibration, we set ω0 = ω1 = 1/2 with ωk = 0 for k > 1 such that Φ(L) =
(1/2)(I + L) to target a time to build of 1.5 years (Ramey 2021). Total factor productivity (TFP),
denoted by Θt , has two components: (i) a constant scaling factor Θ and (ii) a term capturing
the contributions of public capital K tG . That is, Θt = Θ(K tG )γ , where γ denotes the elasticity of
output with respect to public capital and is found to be small but positive (Ramey 2021). Then,
to first order, government investment and TFP are related via
γ +∞
Θ̂t = (1 − δ)s Φ(L)g tI−1−s ,
X
G
K ss /Y ss s=0
G tI −G ss
I
where Θ̂t = ΘtΘ−Θ
ss
ss
and g tI = Y ss .
G
Since in reasonable calibrations γ is small and K ss /Y ss
around 3.5 the total effect of infrastructure spending on TFP is dampened but persistent. Fi-
nal output is linear in labor but now includes a TFP term—Y t = Θt N t = Θ(K tG )γ N t .
The Phillips curve for final goods prices changes in two ways: (i) as firms target a constant
markup the relationship between final goods inflation, πt , and wage inflation, πw w
t , is πt = πt −
Θ̂t − Θ̂t −1 and (ii) the wage New Keynesian Phillips curve includes additional terms capturing
¡ ¢
the fact that employment and output no longer perfectly comove. Specifically, the extended
58
Table B.1: Calibration of additional parameters with infrastructure
Parameter Description Value Target
γ Elasticity of output to public capital 0.05 Ramey (2021)
δ Depreciation of public capital 0.04 Ramey (2021)
I
G ss /Y ss Ratio of gov’t investment of GDP 0.035 Ramey (2021)
wage Phillips curve is
(1 + ϕ)σ CYssss
( )
ϕ
πw
t = κ yt − (g t + g tI ) − Θ̂t + βπw
t +1 . (B.35)
ϕ + σ CYssss ϕ + σ CYssss
This expression shows that there are three channels through which infrastructure spend-
ing affects inflation dynamics: (i) through the output multiplier via changes in y t (ii) standard
wealth effects mediated by g tI and (iii) direct deflationary effects from productivity increases
Θ̂t .
The remaining model is unchanged from the baseline version in the main text.
Calibration. As discussed, we size the infrastructure spending shock as $310 billion and take
the time path of infrastructure spending from the narrative reports. We assume that 83% of the
infrastructure spending was tax financed and the remaining 17% deficit financed. This share
equals the realized financing of the Inflation Reduction Act. We use the realized financing share
because our narrative information on the financing of infrastructure, contained in Appendix
Table C.8, does not contain precise information on financing. We then calibrate the additional
model parameters to standard values from Ramey (2021). We report our calibration in Table
B.1.
Discussion. Figure B.2 reports the impulse response of output and inflation to the fiscal shock,
including infrastructure. We find that despite the fact that the expected infrastructure was large,
the effect on inflation is considerably smaller. The negligible effects of infrastructure spend-
ing on output in the short-run and large effects in the long-run are in line with Ramey (2021).
Overall, the small effect of infrastructure spending on inflation is for three reasons. First, the
infrastructure program was close to balanced budget (i.e. only 17% deficit financed), and the
tax rises dampen the output multiplier. Second, the wealth effects induced by greater infras-
tructure spending dampen the effects on inflation and output. Third, following Ramey (2021),
we incorporate a realistic “time to build” lag, meaning little of the infrastructure is spent in the
59
Figure B.2: Impulse response to Georgia shock including infrastructure
Notes: responses to output y t and inflation πt in the baseline two-agent OLG model. The blue line shows the impulse response in the baseline
model from the main text. The orange line adds the news about expected infrastructure spending.
first three years. This lag further lowers the output multiplier.
B.4.2 Alternative Financing of Stimulus
In the baseline model we assume that the fiscal authority raises lump sum taxes on OLG house-
holds to stabilize debt dynamics during the refinancing phase. In this Section, we consider
alternative financing mechanisms. Figure B.3 shows that inflation and output dynamics are
largely unaffected if instead the required primary surplus is raised by either (i) distortionary
taxes on labor or (ii) lowering government consumption. To isolate the role of the source of
financing the new debt target B is kept the same across specifications.
B.4.3 Alternative Consumption Models
We now consider alternative models of consumption, instead of baseline model of the main
text. We consider the following models: (i) two agent overlapping generations (OLG), the base-
line model of the main text; (ii) two agent bond-in-utility; (iii) three agent OLG and (iv) the
canonical one asset incomplete markets model. For each model we also consider a modifica-
tion of sticky information as in Auclert et al. (2020). We sketch these alternative model variants
and the calibration below.
We report the impulse responses of these models in Appendix Figure B.4. Sticky information
responses are generally larger and feature more persistent output responses. On the one hand,
sticky information lowers the general equilibrium effect through future income increases. On
60
Figure B.3: Impulse responses to Georgia shock under different financing assumptions.
Notes: impulse responses to output y t and inflation πt to the Goergia deficit news under three assumptions of paying back the debt: (i) ad-
justing lump-sum taxes on savers (blue) (ii) adjusting government consumption (orange) and raising primary surpluses through distortionary
taxes (gray).
Figure B.4: Impulse responses to Georgia shock under different consumption models.
Notes: impulse responses of output y t and inflation πt to Georgia shock for a variety of consumption models. All models are calibrated to the
same iMPCs (Fagereng et al. 2021).
the other hand, households are less attentive to future tax increases and rising real interest rates.
In our calibration the latter effect dominates.
Two agent Bond in Utility. As in the baseline model there is a mass µ of hand to mouth house-
1−1/η
holds and a mass 1 − µ of savers. Savers receive a direct utility benefit χ(A) = ζ A1−1/η−1 from
holding government bonds, which acts as a reduced form way to capture liquidity benefits or
61
convenience yields (Auclert, Rognlie & Straub 2023, Mian et al. 2024). Household i in period t
has expected utility
+∞ £
Et u(C i ,t +s ) − ν(Ni ,t +s ) + χ(A i ,t +1+s )
X ¤
s=0
subject to
1 + I t −1
Ci t + Ai t = A i t −1 + (1 − τ y )Wi t Ni t − T t .
Πt
We calibrate the curvature of the utility over wealth η to match the same intertemporal MPC
target from Fagereng et al. (2021) as in the baseline model. The scaling parameter ζ is calibrated
to match our steady state target for government debt. Due to the presence of additional utility
benefits of savings we choose a lower discount factor β = 0.96 such that β(1 + r ) < 1.33 The
remaining modeling blocks and calibration is the same as in the baseline model. The remaining
model blocks are as in the baseline model.
Three Agent OLG model. There is a mass µ1 of OLG households facing death risk φ1 and a
mass µ2 of OLG households facing death risk φ2 . The mass of hand to mouth households is
residually 1 − µ1 − µ2 . Annuity markets and social funds are perfectly segmented between OLG
types. This allows aggregation of each OLG block separately. We calibrate OLG parameters
and the distribution over fixed types to match the intertemporal MPCs from Fagereng et al.
(2021). We calibrate the wealth share of the near permanent income households to be 60% as
in Angeletos et al. (2024). We also set the intertemporal elasticity of substitution to σ = 0.5.
One-asset HA model. There is a continuum of households i ∈ [0, 1]. Each household decides
how much to consume and to invest in nominal public government bonds. Households are
subject to idiosyncratic income risk stemming from stochastic labor productivity. Markets are
incomplete and households face a borrowing constraint A ′ ≥ 0. households receive labor in-
come Wi t Ni t subject to a marginal tax rate τ y and pay lump-sum taxes proportional to their
income state T t . Stimulus checks will be captured by the unconditional transfer Tt . Household
i in period t maximizes their expected utility
+∞ £
Et u(C i ,t +s ) − ν(Ni ,t +s )
X ¤
max
{C i t +s ,A i t +s }s≥0 s=0
33
Additional saving incentives imply that under β(1 + r ) = 1 households savings would grow unboundedly.
62
Figure B.5: Impulse responses to targeted stimulus.
Notes: impulse responses of output y t and inflation πt to Georgia shock under different distributions of transfer payments: (i) equal distribu-
tion as in the baseline version (blue) and (ii) targeted spending towards financially vulnerable households as well as targeted spending towards
low MPC business owners (orange).
subject to
1 + I t −1
A i t −1 + e i t (1 − τ y )Wi t Ni t − T t + Tt
¡ ¢
Ci t + Ai t =
Πt
A i t +1 ≥ 0.
The idiosyncratic income process is taken from Kaplan et al. (2018). The discount factor β
is chosen to clear the asset markets while the total supply of government bonds is calibrated
to match the intertemporal MPCs from the baseline model. We set steady state unconditional
transfers Tss to zero and determine steady state lump-sum taxes residually, T ss = r A ss + G ss −
τ y Y ss .
B.4.4 Targeted Stimulus
In the baseline model of the main text, we assume that transfers are distributed equally to all
types of agents. In this Section, we consider an extension of the 3-type OLG model in which
fiscal stimulus is targeted as follows. Using the classification of Edelberg & Sheiner (2021), we
allocate (i) direct aid to families (e.g. stimulus checks) to all households equally (ii) direct aid
to financially vulnerable households (e.g. UI) to hand-to-mouth households only and (iii) aid
to businesses to the low MPC OLG households only. Figure B.5 shows that the effects of the
stimulus are largely unchanged.
63
B.4.5 Calibrating to Alternative Intertemporal MPCs
In this Section, we consider a version of the baseline OLG model with lower contemporaneous
MPCs, which decay rapidly with the horizon. These estimates of the MPC are in line with several
empirical studies (e.g. Orchard et al. 2023; Boehm et al. 2023). We find that calibrated to these
MPCs, the model can no longer match the inflation multiplier from the data. However for the
model to be consistent with the empirical evidence on interest rates, the intertemporal elasticity
of substitution (IES) must be very small.
Why does the model with transitory MPCs need a low IES? In order to fit transitory MPCs, the
calibration of the model requires a spender and a permanent income household. This model
has an infinite elasticity of savings with respect to the real rate. This is at odds with a change in
the long-run nominal rate after the fiscal stimulus, as we documented in Section 5.
For our calibration to transitory MPCs, we consider the case in which OLG households are
essentially permanent income consumers (i.e. φ = 0.99). The share of hand to mouth con-
sumers is then chosen to target the contemporaneous MPC. We use the expression for the elas-
ticity of household savings, equation (B.11), to calibrate the IES to match the new steady state
level of debt in the baseline model. The calibrated value for the IES is σ = 0.0012—much smaller
than typical macro estimates.
Figure B.6 plots the effect on output and inflation of the fiscal stimulus, in the models that
are calibrated to transitory MPCs. The specifications in the figure all share fast decaying iMPCs
in line with the estimates of Boehm et al. (2023). Each targets a different intertemporal MPC,
of between M 00 = 0.51 and M 00 = 0.23. These specifications can no longer fit the response of
inflation, as we see in the right panel; because the output response is now smaller (left panel).
B.4.6 Alternative Monetary Policy at End of Horizon
In the baseline model we assume that after T = 10 the monetary authority switches to a deter-
minate Taylor rule. In this Section, we consider two alternatives; one in which the central bank
switches to a constant real rate policy after T = 10 and a joint monetary and fiscal policy that
implements steady state output and inflation. Figure B.7 shows that the effects of the stimulus
on inflation and output are similar—though the inflation response is minimally smaller in the
model with the alternative terminal condition. Output is not fully back to steady state after 10
years and since the Phillips curve is a net present value formulation, these future output gaps
beyond the 10 year horizon contribute to higher inflation at shorter horizons.
64
Figure B.6: Impulse responses to Georgia shock under different iMPC assumptions.
Notes: impulse responses of output y t and inflation πt to Georgia shock under different calibrations of intertemporal marginal propensities to
consume. When changing iMPCs the intertemporal elasticity of substitution is re-calibrated – making sense of the empirical observation that
long-run interest rates rose after the Georgia shock. M00 is the impact MPC, and each line other than the baseline model refers to a model
calibrated to a different impact MPC.
Figure B.7: Impulse Responses with Different Assumptions About Terminal Condition
Notes: impulse responses of output y t and inflation πt to the Georgia shock under two different assumptions about the terminal condition: (i)
assume monetary authority switches to an active monetary policy after ten years and setting T = 500 (orange) (ii) impose steady state after ten
years (T = 10, orange) and (iii) assume monetary policy switches to a constant real rate policy after ten years (gray).
B.4.7 Fiscal Theory of the Price Level
In this Section, we explore whether a simple FTPL model calibrated to the 2021 stimulus can
match our estimate for the inflation multiplier. We consider two economies and follow Bianchi
65
et al. (2023). The first features flexible prices, long-term debt with an average maturity of 6
years, and a monetary authority that follows a Taylor rule with coefficient φ = 0.8. This admits
an analytical characterization of the inflation response. The second features nominal rigidities,
a representative household, and assumes a fully passive monetary policy, i.e. φ = 0.
The model exposition is standard. Outstanding nominal government debt is linked to real
primary surpluses S t via
+∞ +∞
X (t + j ) (t + j ) X (t + j ) (t + j )
Qt Bt = Qt B t −1 − P t S t , (B.36)
j =1 j =0
(t + j ) (t + j )
where Q t denotes the price of a nominal bond with outstanding maturity j and B t nom-
inal bond holdings. The real market value of debt relative to trend output Y t = (1 + g )t Y0 is
denoted as à !
+∞
X (t + j ) (t + j )
Vt = Qt Bt /(P t Y t ).
j =1
Assuming a geometric maturity structure ω and denoting primary surpluses relative to trend
output by S̃ t simplifies the expression to
1 + ωQ t 1 1
Vt = Vt −1 − S̃ t . (B.37)
Q t −1 1 + πt 1 + g
In a flexible price economy, equation (B.37) together with a passive monetary policy rule,
i t = φπt and |φ| < 1, is enough to characterize the inflation response. In the nominal rigidity
economy we supplement the equation with the standard Euler equation for a representative
household and the standard New-Keynesian Phillips curve.
Quantification. We size the fraction of unfunded stimulus as 50%–taken from Bianchi et al.
(2023)—which amounts to a total unfunded stimulus surprise of 1.05% of GDP due to the Geor-
ω
gia shock. We calibrate to an average maturity of 6 years and therefore set 1+i ss
= 5/6. Next, we
consider the case in which r ss ≈ g and take the gross market value of total outstanding debt to
GDP from the Federal Reserve Bank of Dallas (FRBD (2024)) to be Vss = 1.49.34
The inflation responses are shown in Figure B.8. The key point is that either version of the
FTPL model does a good job of fitting the inflation response,.
Flexible price model derivations. The response of inflation in the flexible price model can be
characterized analytically. We now present these results. Denote by 1 + r tn := (1 + ωQ t )/Q t −1
34
There are several notions of government debt. Arguably, for the FTPL the most natural is the gross market value
of debt rather than simply the value of privately held debt.
66
Figure B.8: Inflation Response In A Simple FTPL Model
the ex-post nominal return on the portfolio of government debt. To first order, d log(1 + r tn ) =
ω n
1+i ss d logQ t −d logQ t −1 where we have used the fact that 1+r ss = 1+i ss = ω+1/Q ss . Iterating on
the equation and using the fact that by no arbitrage Et d log(1+r tn+ j ) = Et d log(1 + r t + j ) + d log(1 + πt + j )
£ ¤
links the revaluation effect to future inflation and real rates (e.g. Cochrane 2023) as
¶j
+∞ ω
µ
∆Et d log(1 + r tn ) = − ∆Et d log(1 + r t + j ) + d log(1 + πt + j ) ,
X £ ¤
(B.38)
j =1 1 + i ss
where the operator ∆Et = Et − Et −1 denotes the expectation surprise in period t .
Linearizing and iterating forward the budget constraint (B.37) we obtain
¶j ¶j ·
1+g 1+g 1+g 1
µ µ ¸
∆Et d log(1+πt )−∆Et d log(1+r tn ) = ∆Et
X X
d log(1+r t + j )−∆Et d S̃ t + j .
j ≥1 1 + r ss j ≥0 1 + r ss 1 + r ss Vss
(B.39)
In the flexible price economy the real rate does not change. Next, as Bianchi et al. (2023)
explain—with constant real rates—the Taylor coefficient pins down the persistence of inflation
via the Fisher equation. That is, ∆Et d log(1 + πt + j ) = φ j ∆Et d log(1 + πt ). Combining all these
elements together with the expression for the ex-post nominal rate, we can characterize the
response of inflation in the simple FTPL model:
1. The response of inflation on impact is given by
ωφ X 1+g j 1+g 1
µ ¶ +∞ µ ¶ · ¸
∆Et d log(1 + πt ) = − 1 − ∆Et d S̃ t + j . (B.40)
1 + i ss j =0 1 + r ss 1 + r ss Vss
67
2. The persistence of the inflation response is given by ∆Et d log(1+πt + j ) = φ j ∆Et d log(1+πt )
for j ≥ 1.
Using the calibration discussed above, the initial inflation surprise evaluates to ∆Et d log(1+
πt ) = − 1 − 0.8 × 65 × 1.49
1
¡ ¢
× (−1.05)% = 0.234% which aligns well with our empirical estimate of
0.28%. The geometric persistence at the rate φ also qualitatively mimics the persistence of the
empirical estimates – though the data features less persistence.
68
C Additional Empirics
C.1 Tables
Table C.1: Expected Stimulus with Democratic Senate Majority in Week Before Elections
Date Bank Election Re- Number, Expectation Exact Phrasing
sults $(billion) Phrase
04.01.2021 Goldman before 600 “we would “If Democrats manage to win both of the Senate seats in
Sachs expect” play in Georgia, they would win 50 seats, which would al-
low Vice President-elect Harris to cast the tie-breaking vote.
This would lead to greater fiscal stimulus—we would expect
around $600bn more on top of the recently enacted $900bn—
but would also likely mean tax increases to finance additional
spending."
05.01.2021 Barclays before 2000 “the size of “If the Democrats control the Senate, a larger stimulus package
the package could be more likely, with a sizable portion dedicated to state
could possi- and local governments. With the focus in Q1 likely to be on
bly be” the virus, the size of the package could possibly be $2trn, and
Democrats might expand it to include significant spending for
infrastructure, clean energy initiatives, etc. if the political cli-
mate is advantageous."
05.01.2021 Bloomberg before 700 “we think” “In the event of a Democratic sweep in Georgia, we think addi-
tional near-term pandemic relief and accompanying stimulus
could stretch into the $600 billion to $800 billion range."
Median of Expected Stimulus with Democratic Senate Majority: $700 bn
Notes: The number is taken from the reports of investment banks. The window is restricted to 1st-5th of January 2021. The number on 6th of
January 2021 is already considered to be the number after the elections because the information about the results already started to appear. For
cases where a range is given, the median of the range is taken.
Table C.2: Expected Stimulus with Republican Senate Majority in Week Before Elections
Date Bank Number, Expectation Exact Phrasing
$(billion) Phrase
31.12.2020 Deutsche 0 “do not see” “As such, unless the Senate switches to Democratic control on the results of the Georgia elec-
Bank tion, we do not see much scope for further stimulus."
04.01.2021 Goldman 0 “we would “If Senate Republicans hold one or both of these Georgia seats, this will leave them with a nar-
Sachs not expect” row majority and probably will not have substantially different implications for legislation than
in the last Congress when they held 53 seats ... In that environment, we would not expect much
further fiscal stimulus. President Trump recently proposed $2000/person stimulus payments,
but these are unlikely to move forward under a Republican controlled Senate, we believe, as it
would cost around $450bn, Republican leaders and many Republican senators don’t support it,
and there is likely to be less momentum behind it once individuals start receiving the smaller
payments that Congress recently passed."
05.01.2021 Moody’s An- 0 “not pencil- “Our baseline forecast does not assume that Democrats will pick up both Georgia seats, which
alytics ing” would be necessary for that party to retake the Senate from Republicans. As a result, Moody’s
Analytics is not penciling in a sixth piece of federal pandemic legislation following the
$900 billion economic relief package that was enacted over the holidays, nor do we expect
President-elect Biden to get his tax and spending policy proposals from the campaign through
Congress."
05.01.2021 Rabobank 0 “we should “In contrast, if the Republicans manage to hold on to at least one of these two Georgia seats,
not expect” they will keep their majority in the Senate (either 51-49 or 52-48). In this case, the Senate Re-
publicans are likely to shoot down the ambitious spending plans of the Democrats. This means
that we should not expect major fiscal policy measures, at least until the 2022 midterms."
05.01.2021 Barclays 1000 “remains vi- “If Republicans keep control of the Senate, moderate virus-related relief and possible infras-
able... we tructure spending may be the only areas of bipartisan agreement in Congress, in our view. If the
think” GOP retains control of the Senate and the Biden administration faces a divided Congress, we
still a Q1 virus-related stimulus package—potentially around $1trn—remains viable ... [w]e
think moderate virus-related relief and possible infrastructure spending may be the only areas
of bipartisan agreement in Congress."
05.01.2021 Bloomberg 225 “we expect” “If Republicans hold the chamber by retaining at least one of the two seats, we expect only
must-have Covid relief in the vicinity of $150 billion to $300 billion by sometime in 2Q, at
most."
Median of Expected Stimulus with Republican Majority: $0 bn
Notes: The number is taken from the reports of investment banks. The window is restricted to 1st-5th of January 2021. For cases where a range
is given, the median of the range is taken.
Table C.3: Probability of Democratic Senate Majority in Week Before Elections
Date Bank Prob Democratic Gov- Exact Phrasing
ernment
04.01.2021 Deutsche Bank 0.5 “The web now has sites suggesting odds are only 52% in favour of the Re-
publicans maintaining control of the Senate - so a bit of a toss-up. Same
story on the individual races with the Ossoff-Perdue now essentially 50/50
while Warnock-Loeffler is 60/40 in the Democratic candidates favour. All
this well within the poll margin of errors, to say the least."
05.01.2021 Barclays 0.5 “Polling in both Georgia Senate run-off elections is well within the margin
of error, and we consider them both toss-ups."
05.01.2021 Goldman Sachs 0.5 “Polls show Democratic candidates with a very slim advantage and early
voting appears to have moved slightly in the Democratic direction (vs early
voting in November) ... race remains a toss-up with a slight Republican
lean ... Prediction markets appear to take the same view and imply nearly
even odds that Democrats win both seats"
05.01.2021 Moody’s Analyt- <0.5 “Our baseline forecast does not assume that Democrats will pick up both
ics Georgia seats, which would be necessary for that party to retake the Senate
from Republicans."
05.01.2021 Rabobank 0.5 “[A]fter the Georgia bifurcation point we enter one of two regimes that will
be very different in political dynamics, fiscal policy outcomes and pressure
on the various Fed policies. If we look at recent polls the probabilities of
the two regimes are close to fifty-fifty, although there appears to be a slight
advantage for both Democratic candidates."
Median of Expected Probability of Democratic Senate Majority: 0.5
Notes: the probability is taken from the investment bank reports before the election date. We take the closest probability to election
date, for each investment bank, from the window of 1st of January - 5th of January. The window is chosen before 6th of January 2021
because on 6th of January visible information about Democrat win started to appear.
Table C.4: Composition of Stimulus Package
Date Bank Initial Transfers Government Spending Other Spending
Number
05.01.2021 Bloomberg $850bn UI: state and local fiscal aid:
(before, - $250bn - $250bn
case of - 29.4% - 29.4%
Dem. win)
stimulus checks: Total: $250bn
- $350bn Total Share: 29.4%
- 41.2%
Total: $600bn
Total Share: 70.6%
06.01.2021 Goldman $750bn UI: state and local fiscal aid: other:
(after) Sachs - $150bn - $200bn - $100bn
- 20% - 26.7% - 13.3%
stimulus checks: Total: $200bn Total: $100bn
- $300bn Total Share: 26.7% Total Share: 13.3%
- 40%
Total: $450bn
Total Share: 60%
06.01.2021 BNP Paribas $1000bn stimulus checks: state and local fiscal aid: other non-COVID re-
(after) - $350bn - $300bn lated fiscal support:
- 35% - 30% - $350bn
- 35%
Total: $350bn Total: $300bn
Total Share: 35% Total Share: 30% Total: $350bn
Total Share: 35%
Date Bank Initial Transfers Government Spending Other Spending
Number
07.01.2021 JP Mor- $750bn stimulus checks: state and local fiscal aid:
(after) gan Wealth - $250bn - $150bn
Manage- - 33.3% - 20%
ment
UI: health/COVID related:
- $150bn - $50bn
- 20% - 6.7%
paycheck protection program (PPP): Total: $200bn
- $150bn Total Share: 26.7%
- 20%
Total: $550bn
Total Share: 73.3%
08.01.2021 Barclays $1425bn UI: state and local fiscal aid: other:
(after) - $125bn - $500bn - $50bn
- 8.77% - 35.09% - 3.51%
economic impact payments: Federal Medicaid funding: Total: $50bn
- $300bn - $50bn Total Share: 3.51%
- 21.05% - 3.51%
hazard pay for essential workers: Testing, tracing, vaccine distribu-
- $190bn tion:
- 13.33% - $100bn
- 7.02%
cover 100% COBRA costs:
- $100bn Total: $650bn
- 7.02% Total Share: 45.62%
expand emergency medical leave:
- $10bn
- 0.70%
Total: $725bn
Total Share: 50.87%
Median Share of Transfers: 0.69
Notes: The numbers for composition are taken from the reports of investment banks both before and after elections. When a range is given, the
median is taken. The share of transfers is calculated as transfers/(transfers + government spending), which assumes that “other spending” has
the same composition of transfers vs. government spending as the rest of the stimulus. We classify certain items (e.g. unemployment insurance
and stimulus checks) as transfers and other components (e.g. stage and local fiscal aid or vaccine distribution) as government spending, as in
the table.
Table C.5: Stimulus Package Financing
Date Bank Fiscal Package Exact Phrasing
30.12.2020 Financial Times Deficit Financed “The Treasury department plans to sharply shift its bond sales to-
wards debt maturing well into the future as the government seeks to
fund vast spending programmes."
06.01.2021 DWS North Deficit Financed “More fiscal support will likely require huge Treasury issuance to
America fund it, which is already pushing yields higher, and could increase
borrowing costs for companies."
06.01.2021 Bloomberg At least some “While stimulus will be the primary focus, high-earners and corpo-
deficit financing rations could be tasked with helping to pay for it ... tax hikes may be
implied limited and possibly delayed until the economy is on stronger foot-
ing."
06.01.2021 BNP Paribas Deficit Financed “In order to finance our increased 2021 fiscal deficit projection of
USD2.5trn+, we expect US Treasury issuance to remain at elevated
levels (averaging USD370bn/month) throughout 2021."
08.01.2021 HSBC Deficit Financed “The benchmark 10-year Treasury yield has moved above 1.0 per cent
for the first time since March 2020. This has been driven by expecta-
tions that the Senate elections in Georgia will pave the way for even
greater fiscal stimulus, which will ultimately have to be financed by
more bond issuance."
10.01.2021 Moody’s Analyt- Deficit Financed “Fiscal support from the new Biden administration and Congress is
ics expected to include an additional $750 billion to help the economy
through to the end of the pandemic. This will be entirely deficit-
financed, passed into law in February, and largely take effect in
March."
14.01.2021 Goldman Sachs Deficit Financed “The new stimulus programs should also translate into higher deficits
and larger net issuance."
14.01.2021 Barclays At least some “Taken together, we estimate that the FY21 fiscal deficit increases by
deficit financing about $1trn relative to our prior forecast, to $3.1trn (14.0% of GDP),
implied and the FY22 fiscal deficit increases to $1.9trn (8.0% of GDP) vs. 6.0%
of GDP previously."
Notes: discussion about financing of stimulus is taken from the reports of investment banks both before and after elections.
Table C.6: Infrastructure Discussion Before Elections
Date Bank Infrastructure Exact Phrasing
Number
30.12.2020 Moody’s Analytics Rep. win: “A divided government will prevent additional fiscal stimulus from being passed next
sizeable infras- year. However, there are reasonable odds that once the pandemic winds down, Biden
tructure is pos- will be able to get Congress to agree to a sizable infrastructure package, though likely
sible once the not in 2021."
pandemic winds
down
31.12.2020 Deutsche Bank Dem win: “However, if Democrats take both seats, another large fiscal stimulus package would be
possible infras- likely, possibly including some of the more structural priorities of the new Administra-
tructure package tion such as infrastructure."
04.01.2021 Goldman Sachs Dem win: “Infrastructure, for example, continues to be an area where some bipartisan support ap-
meaningful infras- pears possible...Democratic control of the Senate would increase the odds of a meaning-
tructure package; ful infrastructure package becoming law, though this is more of an indirect effect as such
legislation would still require bipartisan support to pass.”
Rep win:
some infrastruc-
ture package
05.01.2021 Rabobank Dem win: “Biden’s ambitious plans to boost the economy through expansive fiscal policy will
more expansive be shot down in the Senate if the Republicans keep a majority. ... So we can forget
fiscal policy; about all those plans to spend on education, public R&D, green infrastructure, health
care, unemployment benefits and social programs. The same is true for tax hikes for
Rep win: corporations and high income and high wealth individuals.
0 If the Democrats win both run-off elections in Georgia this would open the door to
a large fiscal stimulus package and more expansive fiscal policy in the coming years.
Part of this will likely be financed by higher taxes somewhere down the road."
05.01.2021 Barclays Dem win: “If the Democrats control the Senate, their first priority would likely be a stimulus pack-
possible signifi- age, with a sizable portion dedicated to state and local governments, and it might even
cant spending on get expanded to include significant spending for infrastructure and clean energy initia-
infrastructure; tives ...
If Republicans keep control of the Senate, moderate virus-related relief and possible
Rep win: infrastructure spending may be the only areas of bipartisan agreement in Congress, in
possible moder- our view."
ate infrastructure
spending
Notes: The discussion of infrastructure is taken from the reports of investment banks before the Georgia Senate election.
Table C.7: Expected Infrastructure Package After Elections
Date Bank Infrastructure, Type Exact Phrasing
$(billion)
06.01.2021 Cornerstone 1000 infrastructure “Infrastruct. Larger deal ($1 trillion) via budget recon; surface infrastr +
Research schools/housing"
06.01.2021 BNP Paribas 600 infrastructure “We also see a strengthened likelihood of a bipartisan passage of President-
and industrial elect Biden’s infrastructure and industrial policy plans (≈USD600bn)
policy roughly evenly spread across 2021 and 2022."
06.01.2021 Capital Eco- 0 infrastructure “Biden’s major legislative priorities, including a large Green New Deal-style
nomics infrastructure package partly funded by higher taxes on high-income indi-
viduals and corporations are still unlikely to become a reality, so we are not
minded to change our (above-consensus) forecasts for 2021 or 2022."
07.01.2021 Deutsche 1000 infrastructure “While at this point the size and scope of these policies are highly uncer-
Bank tain, we have in mind an infrastructure package of about $1tn and tax
reform raising revenues of about half that much."
10.01.2021 Moody’s Ana- 1150 net fiscal sup- “We also expect an additional $1.15 trillion in net fiscal support to be signed
lytics port into law later this year with government spending and tax increases in the
spirit of the "Build Back Better" policy agenda that Biden proposed during
the campaign."
11.01.2021 Saxo 3500 green infras- “With Harris to break the 50/50 potential Tie in the Senate, about $7 trillion
tructure in Green Infrastructure that Biden and Harris campaigned on has risen sev-
eral magnitudes in not just probability but scope ... We are not saying the
full $7 trillion will come into fruition, it could actually be more – but even if
it’s "only" $3.5 trillion the ripples are huge."
11.01.2021 Goldman 550 infrastructure “Our US economists see ... an ongoing 0.25% of GDP in new annual
Sachs and green spending financed by tax increases, which helps fund infrastructure and
stimulus green initiatives."
11.01.2021 Bank of Amer- 3000 infrastructure “A Blue Wave increases the likelihood of an immediate $1 trillion Covid
ica Corp stimulus and $2 trillion to $4 trillion infrastructure spending package
later in 2021"
Median of Expected Infrastructure Package: $1000 bn
Notes: The number is taken from the reports of investment banks after elections. For cases where the range is given, the median of
the range is taken. In Goldman Sachs report 0.25% of GDP for 10 years would equal approximately $550bn.
Table C.8: Infrastructure Package Financing
Date Bank Infrastructure Exact Phrasing
06.01.2021 Morgan Stanley Partially by taxes “US public policy strategist Michael Zezas ... sees ... a lighter touch
on taxes, used as a partial offset to infrastructure and/or healthcare
spending initiatives later in 2021."
06.01.2021 Capital Eco- Partly funded “But Biden’s major legislative priorities, including a large Green
nomics by higher taxes New Deal-style infrastructure package partly funded by higher taxes
on high-income on high-income individuals and corporations are still unlikely to
individuals and become a reality, so we are not minded to change our (above-
corporations consensus) forecasts for 2021 or 2022.”
(but unlikely)
07.01.2021 Deutsche Bank Half by tax “While at this point the size and scope of these policies are highly
uncertain, we have in mind an infrastructure package of about $1tn
and tax reform raising revenues of about half that much."
08.01.2021 UBS Partially fi- “Our Dem sweep scenario also assumed that there would be a multi-
nanced by taxes year fiscal package that included infrastructure spending along with
other measures. We had penciled in an annual flow rate of about
$275bn, but not starting until the second half of 2021. In addition,
we had assumed that there would be a set of tax increases, including
higher business taxes, that would be used to partially pay for the extra
spending."
10.01.2021 Moody’s Analyt- Financed by “We also expect an additional $1.15 trillion in net fiscal support to
ics taxes be signed into law later this year with government spending and tax
increases in the spirit of the "Build Back Better" policy agenda that
Biden proposed during the campaign."
11.01.2021 Goldman Sachs Fully tax fi- “Congress is likely to spend whatever tax revenue it raises on infras-
nanced tructure and social benefit spending. At the moment, infrastructure
appears to be the top priority."
14.01.2021 Barclays Financed by “[I]nfrastructure spending advanced under budget reconciliation
taxes would likely include revenue increases since it must score deficit neu-
tral outside of the 10-year budget window. This would put the focus
on Democrats agreeing on pay-fors, such as an increase in the corpo-
rate tax rate and/or changes to the taxation of capital gains.
Notes: discussion about financing of infrastructure is taken from the reports of investment banks both before and after elections.
Table C.9: Types of Tax Change for Infrastructure Financing
Date Bank Taxes
22.10.2020 UBS Personal tax:
- Dem win: increase to 0.396
- Rep win: taxes remain unchanged
Capital gains taxes:
- Dem win: capital gains taxed at higher rates at higher income levels
- Rep win: taxes remain unchanged
Corporate tax:
- Dem win: increase to 0.28
- Rep win: taxes remain unchanged
Alternative min tax on book income:
- Dem win: increase to 0.15
- Rep win: taxes remain unchanged
06.01.2021 Cornerstone Re- Personal tax:
search - Increase to 0.396
Capital gains taxes:
- Increase to 0.265
Corporate tax:
- Increase to 0.25
Social Security Tax and Payroll Tax:
- no change
Dividend rates:
- Increase to 0.265
Deductions and restorations:
- no TCJA extensions
- possible partial SALT deduction restoration
Date Bank Taxes
10.01.2021 Moody’s Analytics Capital gains taxes:
- Increase to 0.28
11.01.2021 Goldman Sachs Personal tax:
- no net increase in personal taxes
- increase in marginal rate on top earners: 0.396
Capital gains taxes:
- increase to 0.28
- $160bn
Corporate tax:
- increase to 0.25
- $400bn
Social Security Tax and Payroll Tax:
- no change
Deductions and restorations:
- increase to 0.28
- $225bn - itemized deductions
Notes: This table shows the changes in various tax categories as predicted by different banks and research institutions.
Table C.10: Policy Outcomes After Democratic Victory—Example from Barclays
Date Bank Outcome Probability Exact Phrasing
06.01.2021 Barclays aggressive progressive unlikely We believe ... the probability of an ‘aggressive progressive policy agenda’ is
policy agenda unlikely even if the Democrats win both seats in Georgia ... [w]e generally
agree with Maneesh that near-term corporate tax hikes are unlikely given
policy priorities during the pandemic.
06.01.2021 Barclays stimulus likely the outcome of the two Georgia elections, which are likely to give control
of Congress to Democrats, will raise expectations for further COVID-related
fiscal support and, potentially, spending on infrastructure
06.01.2021 Barclays infrastructure moderately the outcome of the two Georgia elections, which are likely to give control
likely of Congress to Democrats, will raise expectations for further COVID-related
fiscal support and, potentially, spending on infrastructure
07.01.2021 Barclays lower trade risks likely With a Democratic Congress, we expect the Biden administration likely will
pursue additional stimulus, revert to a more active regulatory agenda, and
lower trade risks.
07.01.2021 Barclays tax change moderately While infrastructure remains a distinct possibility, we assign a lower proba-
unlikely bility to significant tax changes or a public option.
07.01.2021 Barclays public option moderately While infrastructure remains a distinct possibility, we assign a lower proba-
unlikely bility to significant tax changes or a public option.
08.01.2021 Barclays confirm Biden adminis- likely With full control of Congress, we expect Democrats are more likely to con-
tration nominees firm all of the Biden administration’s nominees
08.01.2021 Barclays broader agenda setting likely With full control of Congress, we expect Democrats are more likely to con-
powers firm all of the Biden administration’s nominees, control the Congressional
policy agenda with the power to call hearings
08.01.2021 Barclays overturn some of the likely With full control of Congress, we expect Democrats are more likely to con-
Trump administration’s firm all of the Biden administration’s nominees, control the Congressional
de-regulatory efforts policy agenda with the power to call hearings, and overturn some of the
Trump administration’s de-regulatory efforts
08.01.2021 Barclays filibuster elimination unlikely Issues such as eliminating the legislative filibuster or expanding the
Supreme Court are very unlikely to gain traction
08.01.2021 Barclays Supreme Court expan- unlikely Issues such as eliminating the legislative filibuster or expanding the
sion Supreme Court are very unlikely to gain traction
14.01.2021 Barclays severe gas and oil regula- unlikely Regulatory risk. Even with the Georgia Senate results, our view is that near-
tory policy changes term policy changes are likely to be less punitive to oil & gas than initially
feared
Notes: this table an illustrative example for one of the banks, Barclays. Here, we show how we use the text of Barclays reports to discuss what
policy outcomes are associated with the Democratic victory, and what is their likelihood. In the main text, Figure 2 uses information of this kind
for all banks, not just Barclays, in order to create the word cloud.
Table C.11: Major Events News Summary from Bloomberg
Date News Summary US
5th Jan European stocks fluctuated with U.S. European stocks fluctuated with U.S. equity futures as traders weighed concerns about ✓
equity futures as traders weighed con- the impact of rising coronavirus cases and braced for key U.S. runoff elections. Energy
cerns about the impact of rising coro- firms and retailers led the Stoxx 600 Index as the U.K. went back into lockdown in an
navirus cases and braced for key U.S. attempt to prevent hospitals being overwhelmed.
runoff elections.
5th Jan The New York Stock Exchange said The New York Stock Exchange said it will no longer delist China’s three biggest state- ✓
it will no longer delist China’s three owned telecommunications companies, backtracking on a plan that had threatened to
biggest state-owned telecommunica- escalate tensions between the world’s largest economies. NYSE’s U-turn came with scant
tions companies explanation just four days after the exchange said it would remove the shares to comply
with a U.S. executive order barring investments in businesses owned or controlled by the
Chinese military.
5th Jan The U.S. Food and Drug Adminis- The U.S. Food and Drug Administration delivered a clear rebuke to health officials at- ✓
tration delivered a clear rebuke to tempting to alter the timing and dosage of Covid-19 vaccines: Don’t mess with our guide-
health officials:... [D]on’t mess with lines. The agency, in a statement late yesterday, urged that vaccines be given according
our guidelines. to how the FDA has authorized them after a key U.S. official proposed cutting dosage
levels for Moderna Inc.’s shot as a way to immunize more people.
5th Jan Qatar’s ruler landed in Saudi Arabia Qatar’s ruler landed in Saudi Arabia today to a warm embrace from host Crown Prince
today Mohammed bin Salman, hours after their nations re-established travel ties and eased a
regional dispute. Qatari Emir Sheikh Tamim bin Hamad Al Thani is attending the Gulf
Cooperation Council summit for the first time since a 2017 row that cut trade, travel, and
diplomatic ties with Saudi Arabia, the U.A.E., Bahrain, and Egypt. Saudi Arabia reopened
its air, land, and sea borders with Qatar on Monday, a leap toward easing the crisis that
had complicated U.S. efforts to isolate Iran amid heightened tensions.
5th Jan German joblessness unexpectedly de- German joblessness unexpectedly declined in December, though companies were heav-
clined in December ily dependent on government subsidies to keep workers employed through the latest
lockdowns. A decrease of 37,000 left the total number of unemployed people at 2.776
million and the rate unchanged at 6.1%, according to the Federal Labor Agency.
5th Jan The cost of moving freight from The cost of moving freight from France to the U.K. surged to more than four times the
France to the U.K. surged to more usual level last week after Brexit and a virulent new strain of the coronavirus compli-
than four times the usual level last cated supply chains. The spot rate for last-minute shipments across the English Channel
week reached more than 6 euros per kilometer ($4.56 a mile) for a full truckload in the final
week of 2020.
5th Jan Chancellor of the Exchequer Rishi Chancellor of the Exchequer Rishi Sunak rolled out 4.6 billion pounds ($6.2 billion) of
Sunak rolled out 4.6 billion pounds new support to help U.K. businesses forced to close during the pandemic, as the country
($6.2 billion) of new support to help enters its third lockdown. Retail, hospitality, and leisure businesses will be entitled to
U.K. businesses forced to close during one-off top-up grants of as much as 9,000 pounds to tide them over until the spring, the
the pandemic Treasury said in a statement.
6th Jan Benchmark Treasury yields touched Benchmark Treasury yields touched 1% for the first time since March and technology ✓
1% for the first time since March stock futures tumbled as investors speculated over the outcomes of the U.S. Senate races
and technology stock futures tumbled in Georgia. Markets started pricing in the possibility of a Democrat-controlled Congress
as investors speculated over the out- after Raphael Warnock ousted Republican Kelly Loeffler.
comes of the U.S. Senate races in
Georgia.
6th Jan The New York Stock Exchange is con- The New York Stock Exchange is considering reversing course a second time to delist ✓
sidering reversing course a second three major Chinese telecommunications firms after conferring further with senior au-
time to delist three major Chinese thorities on how to interpret an executive order Trump issued Nov. 12, according to peo-
telecommunications firms. ple familiar with the matter. Lawyers said the drama, whipsawing markets in recent days,
is exposing the ambiguities of the government’s instructions.
6th Jan President Donald Trump signed an President Donald Trump signed an order banning U.S. transactions with eight Chinese ✓
order banning U.S. transactions with apps including Ant Group’s Alipay and Tencent Holdings’ digital wallets, riling Beijing
eight Chinese apps. days before he’s slated to leave office. The order is the outgoing administration’s latest
bid to use national security powers against China’s largest companies, but it will be up to
President-elect Joe Biden to decide whether to enforce the policy.
6th Jan Hong Kong arrested dozens of oppo- Hong Kong arrested dozens of opposition figures under a controversial national security
sition figures under a controversial law, an unprecedented crackdown that included an American lawyer, as authorities work
national security law. to quash any dissent that remains in the former British colony. Police said they swept up
53 people in the operation and that around 1,000 officers had been dispatched to carry
out the detentions.
6th Jan The euro-area economy showed signs The euro-area economy showed signs of improvement at the end of last year that now
of improvement at the end of last year risk being derailed by longer pandemic lockdowns and discontent over the pace of vac-
that now risk being derailed by longer cinations against the coronavirus.
pandemic lockdowns
6th Jan Chancellor Angela Merkel called an Chancellor Angela Merkel called an emergency meeting to speed up Germany’s Covid-
emergency meeting to speed up Ger- 19 vaccine program in an effort to quell tensions within her government over claims the
many’s Covid-19 vaccine program rollout has been too slow.
6th Jan Indonesia sold a record amount of Indonesia sold a record amount of local-currency government bonds and issued over-
local-currency government bonds seas debt as it rushes to fund a fight against the worst Covid-19 pandemic in Southeast
and issued overseas debt Asia. In the first auction of the year, the finance ministry took advantage of the lowest
credit costs in a decade to sell 37.55 trillion rupiah ($2.7 billion) of debt excluding bills,
its biggest offering since at least 2002.
7th Jan Global markets are showing resilience Global markets are showing resilience after a day of violence rocked the U.S. Capitol, with ✓
after a day of violence rocked the U.S. investors firmly focused on the prospect for more economic stimulus and the likelihood
Capitol. that calm will prevail as Joe Biden takes the presidency. S&P 500 futures were up 0.5%
and stock benchmarks across Asia and Europe were in the green. Treasury yields stayed
above 1% and the yen weakened.
7th Jan Joe Biden was formally recognized by Joe Biden was formally recognized by Congress as the next U.S. president, ending two ✓
Congress as the next U.S. president. months of failed challenges by his predecessor, Donald Trump, that exploded into vio-
lence at the U.S. Capitol as lawmakers met to ratify the election result. The Democratic
president-elect’s victory was sealed after House and Senate members fended off a final
round of objections to the Nov. 3 election outcome raised by a handful of Republicans
on Trump’s behalf.
7th Jan President Donald Trump... [p]ledged President Donald Trump, minutes after Congress certified President-elect Joe Biden’s ✓
"an orderly transition." Electoral College Victory and hours after Trump supporters broke into the Capitol,
pledged "an orderly transition." "Even though I totally disagree with the outcome of the
election, and the facts bear me out, nevertheless there will be an orderly transition on
January 20th," Trump said in a statement posted on Twitter by his aide Dan Scavino.
7th Jan Euro-area headline inflation re- Euro-area headline inflation remained unchanged at -0.3% in December, with the core
mained unchanged at -0.3% in inflation rate steady at a record low 0.2%. Downward pressure from energy prices eased
December slightly on the month, but the impact on headline prices was offset by a decline in food
prices.
7th Jan Japanese PM Yoshihide Suga declared Japanese PM Yoshihide Suga declared a state of emergency for Tokyo and adjacent areas,
a state of emergency for Tokyo and trying to stem Covid-19 infections that hit a daily record in the capital. The declaration
adjacent areas covers the capital and the surrounding prefectures of Kanagawa, Saitama, and Chiba,
Suga said.
7th Jan Senior officials from Japan’s finance Senior officials from Japan’s finance ministry, central bank, and financial regulator met
ministry, central bank, and financial today in a show of vigilance over a strong yen and as the government announced a state
regulator met today of emergency in Tokyo over the coronavirus.
7th Jan Japanese wages fell in November for Japanese wages fell in November for an eighth straight month, dropping at more than
an eighth straight month double the pace expected by economists, as employers continued to be cautious about
the profit outlook amid a global resurgence of the coronavirus. Labor cash earnings
slid 2.2% from a year earlier, as year-end bonus payments plunged, labor ministry data
showed. Economists had predicted an overall 0.9% decline.
7th Jan Europe’s top auto markets posted Europe’s top auto markets posted their biggest annual sales declines in decades, with
their biggest annual sales declines in ongoing coronavirus restrictions expected to crimp a recovery early this year.
decades.
Notes: This table provides news which are listed in daily reports from Bloomberg as a list of important events for the day.
Table C.12: Conditional Forecast Table
Source Date Real GDP Increase Real GDP Phrasing
Goldman 06.01.2021 0.8% increase over 2 "We have revised our forecasts to reflect the results of the Georgia elections. With control of the
Sachs years Senate by a narrow margin, Democrats are likely to pass further fiscal stimulus.
We now forecast ... 2021 GDP growth of +6.4% on a full-year basis (vs. +5.9% previously and
+3.9% consensus) and +6.6% on a Q4/Q4 basis (vs. +5.6% previously and +3.3% consensus)...
Our 2022 GDP growth forecast is now +4% on a full-year basis (vs. +3.7% previously) and +2.4%
on a Q4/Q4 basis (vs. +2.7% previously)."
BNP Paribas 06.01.2021 1.4% increase over 2 "Both Democratic candidates are projected to win their Georgia run-off races...[w]e revise our
years annual average 2021 and 2022 GDP forecasts up by 0.5pp and 0.9pp, respectively, with growth
expected to register 4.2% and 4.1%."
Moody’s An- 10.01.2021 1.5% increase over 2 "The additional fiscal support will quickly boost the economy, pushing real GDP growth to ...
alytics years more than 5% for all of 2021. This is a percentage point more growth than we expected in last
month’s forecast, which was based on the incorrect assumption the Senate would remain in
Republican control. Real GDP should post another 5% gain in 2022, about 0.5 percentage point
more than previously forecast."
Deutsche 07.01.2021 1.8% increase over 2 "The first priority of the Biden administration and Democratic Congress is likely to be another
Bank years tranche of Covid-related fiscal support. ... In response, we have lifted our growth forecast for
2021 by about 2 percentage points to 6.3% (Q4/Q4) ... Beyond this year, we have modestly
downgraded 2022 growth expectations given a pull forward of activity into the next few quar-
ters."
JP Morgan 07.01.2021 1.9% increase over 2 "Democrats are now set to control the White House and to hold slim majorities in both cham-
years bers of Congress. This could set the stage for a dramatic increase in federal spending and fiscal
transfers to households ... If realized this would boost GDP growth this year by about 1.5%-
points to 5.3% (Q4/Q4), and 0.5%-point next year to 2.6%."
Barclays 14.01.2021 2.3% increase after 2 "With Democratic control of Congress, we expect another virus- related relief package of about
years $1.4trn ...[w]e now expect Q4/Q4 real GDP growth of 7.0% in 2021 (up 3.2pp) and 1.5% in 2022
(down 0.9pp). On a calendar-year basis, these revisions boost real GDP growth to 6.3% y/y in
2021 and 3.9% y/y in 2022."
Bloomberg 06.01.2021 2.3% increase after 2 "In the event of a Democratic sweep in Georgia, we think additional near-term pandemic relief
years and accompanying stimulus could stretch into the $600 billion to $800 billion range. The high
end could be sufficient to lift growth by roughly 1.7 percentage points in 2021, to 5.2% year-
over-year, with a faster pace continuing into 2022 (above 3%), compared to our current baseline
of 2.4%."
Median of the real GDP increase is 1.8%.
Notes: This table shows the change in forecasts of real GDP growth in the week after the Georgia election, by various investment banks.
Table C.13: Changing of Probability of Democratic Senate Majority over Time (Barclays)
Date Source of Probability of Demo- Exact Phrasing
Probabil- cratic Majority
ity
06.11.2020 Barclays < 0.5 “With a split Congress highly likely, prospects for another large fiscal package
seem remote, putting pressure on the Fed to boost monetary policy support. Al-
though many votes remain to be counted, the likelihood of a divided govern-
ment outcome is high."
04.12.2020 Prediction 0.2 “On November 3 (or shortly thereafter), we thought that we would have all the
Markets answers, but with the Senate’s fate still in limbo, the muni market faces a lot of
uncertainty. Prediction markets assign a nearly 80% probability of Republicans
winning at least one of the Georgia Senate seats in the January run-off."
11.12.2020 Barclays unlikely (< 0.5) “Looking ahead, as discussed in our 2021 municipal outlook, although it appears
somewhat unlikely, if Democrats win both Senate seats in Georgia, Treasuries
and tax-exempt yields might sell off sooner and to a larger degree."
18.12.2020 Prediction 0.35 “Regardless, the main focus of muni investors going into 2021 will be on the
Markets Georgia Senate elections, with a possibility of a large stimulus bill, with a siz-
able portion dedicated to municipalities, implemented if Democrats win both
races (although prediction markets assign less than a 35% probability to this
outcome)."
05.01.2021 Barclays 0.5 “Polling in both Georgia Senate run-off elections is well within the margin of er-
ror, and we consider them both toss-ups. ’
Notes: The probabilities are taken for Barclays as an illustrative example of change of assumed probabilities over time. They are
taken from after the presidential election up to the date of the Georgia senate runoff.
Table C.14: Polling Instrument Specification, With and Without Controls
Controls 1 Year 2 Years 5 Years 10 Years
1. No Control 3.32 5.27 9.32 15.05
(1.21) (1.74) (2.38) (4.00)
Observations 30 30 30 30
2. 10-year Bonds 3.16 5.08 9.22 15.07
(1.34) (1.98) (2.63) (4.25)
Observations 30 30 30 30
3. Oil Price -0.13 0.15 3.54 7.82
(0.61) (1.18) (1.03) (3.00)
Observations 30 30 30 30
4. S&P 500 0.73 1.58 5.6 10.7
(0.83) (1.41) (2.17) (4.65)
Observations 30 30 30 30
5. COVID Effect 2.74 4.43 8.01 13.13
(0.96) (1.45) (1.77) (3.19)
Observations 30 30 30 30
Note: Each panel in the table represents a different control variable added to the baseline IV specification. In
all panels, we adjust the price of the inflation swap to take into account the 3-month lag of the inflation index
used in the contracts. For all panels, we regress the expected increase in the price level on the probability of a
Democratic win in the 2021 Georgia Senate election, instrumented by polling data for the Georgia Senate elec-
tion from FiveThirtyEight.com. We use Newey-West standard errors. Our dataset is daily, sourcing expected in-
creases in the price level from zero-coupon inflation swaps from Bloomberg, and probabilities of a Democratic
victory from Predictit’s 2020 Senate election betting prices. The data spans from Nov 16, 2020, to Jan 4, 2021 (the
post-Presidential election dates with polling information). The first panel does not have any controls. All con-
trols are lagged 8 days. The 2nd panel controls for the zero-coupon yield of 10-year US government bonds from
Bloomberg. Panel 3 controls for the price of Brent crude oil from FRED. Panel 4 controls for the S&P 500 index
from Bloomberg. The last panel uses data from the Cleveland Fed. The robust first stage F-statistics are 10.7,
14.79, 20.6, 17.9, and 10.76 respectively.
86
Table C.15: Coibion-Gorodnichenko Regression for Different Sample Periods
Sample: Full July 2004 Jan 2019
Sample Onwards Onwards
Response of forecast errors 0.41 0.50 1.26
to news about inflation (0.53) (0.61) (0.45)
Observations 458 201 42
Notes: The table reports coefficient estimates with Newey-West standard errors
with automatic lag length (Lazarus et al. 2018) in parentheses, implementing the
regression of Coibion & Gorodnichenko (2015) outlined in equation (2) from the
main text, which regresses inflation forecast errors on news about inflation. Fol-
lowing Diercks et al. (2023), we drop observations during periods of financial dis-
tress. The first column uses the full sample from 1983 to 2023, the second uses data
from July 2004 onwards, and the last from January 2019 onwards. Data from before
July 2004 are imputed using the series for inflation expectations in financial mar-
kets calculated by the Cleveland Federal Reserve.
87
C.2 Figures
Figure C.1: Example of Report from Goldman Sachs
Email distribution
Time stamp
Narrative information
Notes: this figure contains a snapshot of a representative report from Goldman Sachs.
88
Figure C.2: Daily Probability of Democrat Majority in Senate
1.0
Prob. Democrats Win
0.8
0.6
0.4
0.2
Nov 15 Dec 01 Dec 15 Jan 01
Date
Notes: this graph plots the daily probability that Democrats would win both Georgia Senate seats and hence take a majority in the Senate, using
end-of-day probabilities from PredictIt.
89
Figure C.3: Outcomes after the Democrat Victory—GPT Reading
Notes: this figure contains a ChatGPT-read word cloud of outcomes of the Democrat victory discussed by investment banks in the week after
the election. In the cloud, an outcome is larger if more banks discuss it, and darker if banks on average assess that it is more likely.
Figure C.4: CBOE Volatility Index (VIX)
Jan 4 Jan 7
35
VIX Index
30
25
20
Nov Dec Jan Feb
Date
Notes: this figure contains a plot of CBOE Volatility Index (VIX) Index at daily frequency around Georgia runoff elections.
90
Figure C.5: High Frequency Betting Data around Georgia Runoff
Normalised Win Probability for Democrats with LOESS Smoothing and Sum of Volumes 50000
1.0 LOESS Smoothed Probability Jan 5 First Observation 2 PM on Jan 6 End of Jan 7
Democratic Win Probability
0.9
40000
Win Probability for Democrats
0.8
Sum of Volumes
30000
0.7
0.6
20000
0.5
0.4 10000
0.3 Total Volume
0
2021-01-01 2021-01-02 2021-01-03 2021-01-04 2021-01-05 2021-01-06 2021-01-07 2021-01-08 2021-01-09
Date
Notes: this figure contains tick by tick data on the probability that Democrats would control the 2020 Senate, based on trades from PredictIt.
We add a LOESS smoothed line to the figure. In shaded gray is the total trade volume at a given point in time.
91
Figure C.6: Bloomberg Surprise Index Around Georgia Runoff
Jan 4 Jan 7
0.3
Price
0.2
0.1
Dec 01 Dec 15 Jan 01 Jan 15
Date
Notes: this figure contains the square of Bloomberg’s intradaily “Surprise Index”, which measures the surprise from data releases.
92
Figure C.7: Expected Percentage Point Increase in the Price Level Over 1 Year
Percentage Point Increase in the Price Level over 1 Year
0.3
0.2
0.1
Estimate at 2 PM on
Jan. 6: 0.18 (0.05)
Estimate at close of
0.0
Jan. 7: 0.28 (0.07)
ARIMA(0,1,2)
−0.1 First 2pm Last
Jan 5 Jan 6 Jan 7
Dec 21 Dec 28 Jan 04 Jan 11 Jan 18
Date
Notes: this graph plots the intraday percentage point increase in the price level 1 year ahead, implied by the 1 year inflation swap, subtracting
the first value on January 5th. Dashed lines are missing data from holidays and weekends. The green line is the forecast if the policy had not
taken place, the gray shade is the 95% confidence interval.
93
Figure C.8: Expected Percentage Point Increase in the Price Level Over 5 Years
Percentage Point Increase in the Price Level over 5 Years
1.0
0.5
Estimate at 2 PM on
Jan. 6: 0.29 (0.16)
Estimate at close of
0.0
Jan. 7: 0.58 (0.22)
ARIMA(1,1,0)
First 2pm Last
Jan 5 Jan 6 Jan 7
−0.5
Dec 21 Dec 28 Jan 04 Jan 11 Jan 18
Date
Notes: this graph plots the intraday percentage point increase in the price level 5 years ahead, implied by the 5 year inflation swap, subtracting
the first value on January 5th. Dashed lines are missing data from holidays and weekends. The green line is the forecast if the policy had not
taken place, the gray shade is the 95% confidence interval.
94
Figure C.9: Expected Percentage Point Increase in the Price Level Over 10 Years
Percentage Point Increase in the Price Level over 10 Years
1.0
0.5
Estimate at 2 PM on
Jan. 6: 0.44 (0.14)
Estimate at close of
0.0
Jan. 7: 0.77 (0.18)
ARIMA(2,1,1)
First 2pm Last
−0.5 Jan 5 Jan 6 Jan 7
Dec 21 Dec 28 Jan 04 Jan 11 Jan 18
Date
Notes: this graph plots the intraday percentage point increase in the price level 10 years ahead, implied by the 10 year inflation swap, subtracting
the first value on January 5th. Dashed lines are missing data from holidays and weekends. The green line is the forecast if the policy had not
taken place, the gray shade is the 95% confidence interval.
95
Figure C.10: Oil Futures
55
50
Price Brent Crude Oil
WTI Crude Oil
45
40 Jan 4 Jan 7
Nov Dec Jan Feb
Date
Notes: this figure contains a plot of price of Brent crude oil and WTI crude oil futures.
96
Figure C.11: Food Futures
100 *Log Price(cents/bushel), Indexed to 1st Observ.
Jan 4 Jan 7
30
20
Corn
Wheat
10 Soybean
0
Nov Dec Jan Feb
Date
Notes: this figure contains a plot of 100*Log(Price) for food futures. Price is cents per bushel. Futures plotted are corn, wheat and soybean. The
observations are indexed to the first available value for comparison.
97
Figure C.12: Professional Forecasts for Quarterly Inflation in 2021Q3
2.5
Median CPI Forecast
2.4
2.3
2.2
2.1
2020Q3 2020Q4 2021Q1 2021Q2
Quarter
Notes: this figure contains a plot of forecasts of quarterly inflation (CPI) for 2021 Q3 from the Survey of Professional Forecasters (SPF) for both
financial and non-financial firms (those unclassified are excluded).
Figure C.13: Single Event Study—2021 Dividend Futures
4
100 x Log Expected Dividend − 1 Year
2
Estimate at 2 PM on
Jan. 6: 1.09 (1.27)
Estimate at close of
0
Jan. 7: 2.74 (1.63)
ARIMA(0,1,1)
−2
First 2pm Last
Jan 5 Jan 6 Jan 7
Dec 21 Dec 28 Jan 04 Jan 11 Jan 18
Date
Notes: this graph plots the intraday percent increase in dividends 1 year ahead, implied by the 1 year S&P 500 dividend future, subtracting the
last value on January 4th. Dashed lines are missing data from holidays and weekends. The green line is the forecast if the policy had not taken
place, the gray shade is the 95% confidence interval.
98
Figure C.14: Capitol Hill Riots—Mentions in Media
Figure C.15: Credit Default Swaps around Georgia Runoff
Jan 4 Jan 7
18
17
Price
16
15
14
Dec 01 Dec 15 Jan 01 Jan 15 Feb 01
Date
Notes: this figure plots end-of-day Credit Default Swaps Prices on 5 year US government debt in a month interval around the Georgia Senate
Runoff.
99
Figure C.16: Regression Specification—1 Year Swaps
Percentage Point Increase in the Price Level over 1 Year
2.7
Full Sample Regression: 0.96 (0.2)
2.4
Pre Jan 5 Regression: 3.28 (0.26)
2.1
0.25 0.50 0.75 1.00
Prob. Democrats Win Senate
After Election Before Election
Notes: this graph plots the end-of-day expected percentage point increase in prices over 1 years, implied by the 1 year inflation swap, against
end of day probability of Democrat victory from PredictIt. Standard errors, in brackets, are Newey-West with automatic lag length (Lazarus
et al. 2018).
100
Figure C.17: Regression Specification—5 Year Swaps
Percentage Point Increase in the Price Level over 5 Years
11.5
11.0
10.5 Full Sample Regression: 2.21 (0.41)
Pre Jan 5 Regression: 6.87 (0.61)
10.0
9.5
0.25 0.50 0.75 1.00
Prob. Democrats Win Senate
After Election Before Election
Notes: this graph plots the end-of-day expected percentage point increase in prices over 5 years, implied by the 5 year inflation swap, against
end of day probability of Democrat victory from PredictIt. Standard errors, in brackets, are Newey-West with automatic lag length (Lazarus
et al. 2018).
101
Figure C.18: Regression Specification—10 Year Swaps
Percentage Point Increase in the Price Level over 10 Years
23
22
Full Sample Regression: 3.05 (0.69)
Pre Jan 5 Regression: 10.91 (1.14)
21
20
0.25 0.50 0.75 1.00
Prob. Democrats Win Senate
After Election Before Election
Notes: this graph plots the end-of-day expected percentage point increase in prices over 10 years, implied by the 10 year inflation swap, against
end of day probability of Democrat victory from PredictIt.Standard errors, in brackets, are Newey-West with automatic lag length (Lazarus et al.
2018).
102
Figure C.19: Regression Specification—BetFair, Ossoff Election, 2 Year Swaps
Percentage Point Increase in the Price Level over 2 Years
4.8
4.4
Full Sample Regression: 0.91 (0.22)
Pre Jan 5 Regression: 1.75 (1.44)
4.0
0.4 0.6 0.8 1.0
Prob. Democrats Win Senate BetFair
After Election Before Election
Notes: this graph plots the end-of-day expected percentage point increase in prices over 2 years, implied by the 2 year inflation swap, against
end of day probability of Jon Ossof’s victory in the regular Senate Election, from Betfair. Standard errors, in brackets, are Newey-West with
automatic lag length (Lazarus et al. 2018).
103
Figure C.20: Regression Specification—PredictIt, 2 Year Swaps (same sample as BetFair)
Percentage Point Increase in the Price Level over 2 Years
5.0
4.5
Full Sample Regression: 0.96 (0.25)
Pre Jan 5 Regression: 3.05 (1.66)
4.0
0.4 0.6 0.8 1.0
Prob. Democrats Win Senate Predictit
After Election Before Election
Notes: this graph plots the end-of-day expected percentage point increase in prices over 2 years, implied by the 2 year inflation swap, against
end of day probability of Democrat victory from PredictIt, on the sample for which BetFair data are available. Standard errors, in brackets, are
Newey-West with automatic lag length (Lazarus et al. 2018).
104
Figure C.21: Regression Specification—2021 Dividends
100 x Log Expected 1 Year Dividend
406
404
Full Sample Regression: 6.14 (1.27)
Pre Jan 5 Regression: 18.25 (3.32)
402
400
0.25 0.50 0.75 1.00
Prob. Democrats Win Senate
After Election Before Election
Notes: this graph plots the end-of-day log dividend future for the S&P500 1 year ahead, against end of day probability of Democrat victory from
PredictIt. Standard errors, in brackets, are Newey-West with automatic lag length (Lazarus et al. 2018).
105
Figure C.22: Regression Specification—2022 Dividends
100 x Log Expected 2 Years Dividend
407.5
405.0
402.5 Full Sample Regression: 9.74 (1.73)
Pre Jan 5 Regression: 27.22 (2.84)
400.0
397.5
0.25 0.50 0.75 1.00
Prob. Democrats Win Senate
After Election Before Election
Notes: this graph plots the end-of-day log dividend future for the S&P500 2 years ahead, against end of day probability of Democrat victory
from PredictIt. Standard errors, in brackets, are Newey-West with automatic lag length (Lazarus et al. 2018).
106
Figure C.23: 5 Year Nominal Interest Rate After 5 Years
Percentage Point Change in Nominal Interest Rates,5Y−5Y
0.4
0.2
0.0 Estimate at 2 PM on
Jan. 6: 0.161 (0.059)
Estimate at close of
Jan. 7: 0.202 (0.08)
First 2pm Last
Jan 5 Jan 6 Jan 7
−0.2 ARIMA(0,1,0)
Dec 21 Dec 28 Jan 04 Jan 11 Jan 18
Date
Notes: This plot shows the 5 year-5 year forward interest rate. Dashed lines indicate missing data from holidays and weekends. The green
line is the forecast if the policy had not taken place, and the gray shade is the 95% confidence interval. The dashed orange lines mark the first
observation on January 5th, 2 PM on January 6th, and the final observation on January 7th.
107
Figure C.24: Regression Specification—Year 1 Bonds
0.12
0.11
Year 1 Bonds
Full Sample Regression: 0 (0.01)
Pre Jan 5 Regression: −0.07 (0.05)
0.10
0.09
0.25 0.50 0.75 1.00
Prob. Democrats Win Senate
After Election Before Election
Notes: this graph plots the end-of-day 1 year nominal interest rate on government bonds; against end of day probability of Democrat victory
from PredictIt. Standard errors, in brackets, are Newey-West with automatic lag length (Lazarus et al. 2018).
108
Figure C.25: Regression Specification—5 Year Nominal Forward Rater After 5 Years
1.8
5 Year Nominal Forward Rate After 5 Year
1.7
1.6
Full Sample Regression: 0.4 (0.06)
1.5 Pre Jan 5 Regression: 0.48 (0.17)
1.4
1.3
0.25 0.50 0.75 1.00
Prob. Democrats Win Senate
After Election Before Election
Notes: this graph plots the end-of-day 5 year nominal interest rate on government bonds expected after 5 years; against end of day probability
of Democrat victory from PredictIt. Standard errors, in brackets, are Newey-West with automatic lag length (Lazarus et al. 2018).
109
Figure C.26: Federal Funds Futures - 3-month ahead
0.085 Jan 4 Jan 7
0.080
Expected Rate
0.075
0.070
0.065
Nov Dec Jan Feb
Date
Notes: this figure contains a plot of expected annual 3-month ahead EFFR rate from CME Federal Funds Futures. The expected rate is calculated
as 100-P, where P is price of the futures contract.
110
Figure C.27: Federal Funds Futures - 6-month ahead
0.080 Jan 4 Jan 7
0.075
Expected Rate
0.070
0.065
Nov Dec Jan Feb
Date
Notes: this figure contains a plot of expected annual 6-month ahead EFFR rate from CME Federal Funds Futures. The expected rate is calculated
as 100-P, where P is price of the futures contract.
111
Figure C.28: Federal Funds Futures - 9-month ahead
Jan 4 Jan 7
0.08
Expected Rate
0.07
0.06
Nov Dec Jan Feb
Date
Notes: this figure contains a plot of expected annual 9-month ahead EFFR rate from CME Federal Funds Futures. The expected rate is calculated
as 100-P, where P is price of the futures contract.
112
Figure C.29: Federal Funds Futures - 12-month ahead
0.09 Jan 4 Jan 7
0.08
Expected Rate
0.07
0.06
0.05
Nov Dec Jan Feb
Date
Notes: this figure contains a plot of expected annual 12-month ahead EFFR rate from CME Federal Funds Futures. The expected rate is calcu-
lated as 100-P, where P is price of the futures contract.
113
Figure C.30: Impulse response for price level
Notes: this graph plots impulse responses of the price level. All responses are to the shocks shown in Figure 8. Price level forecasts from the
data are shown with their 95% confidence interval.
114
File and source
- File
- do-deficits-cause-inflation-a-high-frequency-narrative-approach.pdf
- Size
- 2,037,320 bytes
- SHA-256
- e2b874cdda389e3485c76eb14eb2728c1eba8d36c8840df0719844ca600aa0c0
- Original
- www.lse.ac.uk