Maikol Cerda et al., The Causal Impact of Covid-19 Government-backed Loans on MSMEs Liquidity and Earnings (2023)
Summary
Inter-American Development Bank Working Paper IDB-WP-1430, February 2023, by Maikol Cerda, Paul Gertler, Sean Higgins, Ana María Montoya, Eric Parrado and Raimundo Undurraga. With private banks, the authors ran two randomized trials offering government-guaranteed Covid-19 loans to 10,072 small businesses in Chile and 3,079 in Colombia, most of which banks had earlier rejected on risk grounds. Take-up was 27% and 29%. The loans raised the total liquidity treated firms could access from the formal banking system by 15.7%, significant at the 1% level. The authors say the results should inform loan guarantee programs across Latin America.
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
“Maikol Cerda et al., The Causal Impact of Covid-19 Government-backed Loans on MSMEs Liquidity and Earnings (2023).”
Original at web.archive.org. Archived at Pandemic Darlings, https://pandemicdarlings.com/source-documents/maikol-cerda-et-al-the-causal-impact-of-covid-19-government-backed-loa-2ed4bb8d80/
, SHA-256 60b6ec8fe40eee9ffbd56e37042d1f3fa91d5505e444db9b6c8f3f9d6f1e353d.Full text
IDB WORKING PAPER SERIES Nº IDB-WP-1430
The Causal Impact of Covid-19 Government-
backed Loans on MSMEs Liquidity and
Earnings
Maikol Cerda
Paul Gertler
Sean Higgins
Ana María Montoya
Eric Parrado
Raimundo Undurraga
Inter-American Development Bank
Department of Research and Chief Economist
February 2023
The Causal Impact of Covid-19 Government-backed
Loans on MSMEs Liquidity and Earnings
Maikol Cerda*
Paul Gertler**
Sean Higgins***
Ana María Montoya****
Eric Parrado*****
Raimundo Undurraga****
* IDB Invest
** University of California Berkeley, Haas School of Business
*** Northwestern University, Kellogg School of Management
**** Universidad de Chile
***** Inter-American Development Bank
February 2023
Cataloging-in-Publication data provided by the
Inter-American Development Bank
Felipe Herrera Library
The causal impact of Covid-19 government-backed loans on MSMEs liquidity and
earnings / Maikol Cerda, Paul Gertler, Sean Higgins, Ana Maria Montoya, Eric Parrado,
Raimundo Undurraga.
p. cm. — (IDB Working Paper Series ; 1430)
Includes bibliographic references.
1. Loans-Government guaranty-Chile-Evaluation. 2. Loans-Government guaranty-
Colombia-Evaluation. 3. Small business-Finance-Chile. 4. Small business-Finance-
Colombia. 5. Coronavirus infections-Economic aspects-Chile. 6. Coronavirus
infections-Economic aspects-Colombia. I. Cerda, Maikol. II. Gertler, Paul, 1955- III.
Higgins, Sean. IV. Montoya, Ana María. V. Parrado, Eric. VI. Undurraga, Raimundo.
VII. Inter-American Development Bank. Department of Research and Chief Economist.
VIII. Series.
IDB-WP-1430
http://www.iadb.org
Copyright © 2023 Inter-American Development Bank. This work is licensed under a Creative Commons IGO 3.0 Attribution-
NonCommercial-NoDerivatives (CC-IGO BY-NC-ND 3.0 IGO) license (http://creativecommons.org/licenses/by-nc-nd/3.0/igo/
legalcode) and may be reproduced with attribution to the IDB and for any non-commercial purpose, as provided below. No
derivative work is allowed.
Any dispute related to the use of the works of the IDB that cannot be settled amicably shall be submitted to arbitration pursuant to
the UNCITRAL rules. The use of the IDB's name for any purpose other than for attribution, and the use of IDB's logo shall be
subject to a separate written license agreement between the IDB and the user and is not authorized as part of this CC-IGO license.
Following a peer review process, and with previous written consent by the Inter-American Development Bank (IDB), a revised
version of this work may also be reproduced in any academic journal, including those indexed by the American Economic
Association's EconLit, provided that the IDB is credited and that the author(s) receive no income from the publication. Therefore,
the restriction to receive income from such publication shall only extend to the publication's author(s). With regard to such
restriction, in case of any inconsistency between the Creative Commons IGO 3.0 Attribution-NonCommercial-NoDerivatives license
and these statements, the latter shall prevail.
Note that link provided above includes additional terms and conditions of the license.
The opinions expressed in this publication are those of the authors and do not necessarily reflect the views of the Inter-American
Development Bank, its Board of Directors, or the countries they represent.
Abstract
We conducted two randomized controlled trials (RCTs) to evaluate the
impact of government-guaranteed loans offered by the Chilean and Colombian
governments. The public funds of these programs greatly expanded following
the start of the Covid-19 pandemic and offered loans to Micro, Small and
Medium Enterprises to mitigate the negative impact of the shock. Through a
collaboration with private banks, we launched two experiments which offered
loans to a sub-set of the 10,072 Chilean and 3,079 Colombian small businesses
that took part in our experiments. Most of these firms had previously applied
for a loan–during the pandemic but prior to the RCTs–but were rejected by
banks due to their risk analysis of the firms. With take-up rates of 27% and
29%, respectively, we find that Covid-19 loans had a significant positive impact
on the total liquidity that treated MSMEs could access: total liquidity with
the formal banking system increased by 15.7% (statistically significant at the
1% level). The results of our RCTs will inform Latin American governments
concerning their strategies to support MSMEs via government-backed loan
programs and will shape similar public policies in the future.
JEL classifications: J16, L26, P52
Keywords: RCT, Government-guaranteed loans, Covid-19 pandemic, Event
study, Impact evaluation
1 Introduction
Firms confronted a significant negative demand shock during the Covid-19 pandemic
due to mobility, gathering, and other social and economics restrictions imposed by
federal governments. Under these circumstances, many small businesses required
short-term financial relief to confront the pandemic, survive as businesses, and con-
tinue paying rent and meeting payrolls. Yet, when the economy is under stress,
financial institutions are more reluctant to lend money to the most negatively im-
pacted and vulnerable firms since the risk of default of these companies is higher.
Among the firms facing higher financial stress during economic slumps, micro, small
and medium-sized enterprises (MSMEs) are the type of businesses that may benefit
the most from more flexible and generous access to credit. Access to credit to MSMEs
in developing economies is critical, moreover, as they are the backbone of economic
growth and labor demand.1 Because of this, central governments implemented new
or greatly expanded well-established government-guaranteed loans relief programs in
order to reduce the financial risk faced by banks when lending and to promote fast
liquidity to small businesses by partially or fully guaranteeing financial loans in order
to avoid a larger economic decline and higher unemployment during the pandemic.
The Chilean and Colombian relief program expansions are two of the most no-
ticeable cases in Latin America where governments early on interceded in helping
small businesses during the Covid-19 pandemic. Early in June 2020, the Chilean
government significantly expanded the most popular and long-established economic
program offering government-backed loans to national firms, the Small Entrepreneurs
Guarantee Fund, FOGAPE.2 Likewise, the Colombian MSMEs economic program
1
According to the IDB, Micro, small and medium-sized enterprises (MSMEs) com-
prise 99.5 percent of businesses, 60 percent of the employed population, and about
25 percent of the gross domestic product (GDP) in Latin America and the Caribbean
(LAC). Source: https://publications.iadb.org/en/msme-financing-instruments-in-latin-america-
and-the-caribbean-during-covid-19
2
The financial program started to operate in 1980 and has been expanded over the years. Most of
FOGAPE’s financial aid is spent exclusively on MSMEs and, during economic slumps, the amount
of financial support has grown either by increasing the percentage of the loan that is guaranteed
by the government or relaxing the requirements that MSMEs must fulfill. FOGAPE’s historical
statistics show that coverage has expanded continuously, reaching a higher number of businesses
2
counterpart, the Fondo Nacional de Garantias, FNG, also greatly expanded in May
2020 as a result of the implementation of the Covid-19 Unidos por Colombia pro-
gram. The FGN has supported most of the financial emergency loans approved by
the Colombian banking system from mid-2020 to date. As a result of these two
programs’ expansions, the government-guaranteed share of loans’ principal consid-
erably increased, which allowed financial institutions to soften the requirements for
access to loans, and private banks ended up lending more money to riskier firms in
comparison to the most recent pre-Covid-19 years.
Given the significant and broad implementation of Covid-19 emergency economic
programs in developing economies, and the scarce economic literature evaluating the
effect of these programs, our research takes advantage of the emergency setup and the
unique financial datasets to which we had access. We evaluated the causal impact of
the unexpected FOGAPE and FNG expansions by designing and implementing two
randomized control trials (RCTs). Through a collaboration with two local and large
private banks, we were able to perform controlled experiments consisting of offering
a financial loan to a subset of the 10,072 and 3,379 MSMEs clients (experimental
universe) in Chile and Colombia, respectively. When designing our experiments,
we noticed that most of these 13,451 MSMEs had already applied for a loan at
the beginning the pandemic, prior to the expansion of the Covid-19 economic relief
program, but many applications were rejected by the banks as a consequence of
their risk assessments. Taking into account the type of riskier small firms that were
part of our RCTs and the contemporaneous economic slump, we expect that our
experiments better inform policymakers regarding the impact of economic programs
that facilitate short-term liquidity to businesses that would not have otherwise had
access to financial support during an economic deterioration due to credit constraints.
The loan campaign extended from March to June 2021, and from May to December
2021 in the Chilean and Colombian RCT, respectively.
The experimental design and identification strategy of our RCTs was robust.
First, we randomized our large experimental groups with public intervention as an
instrument. Second, we had access to micro high-frequency firm-level administrative
during Chile’s most recent economic recessions, and the Covid-19 pandemic was no exception.
3
and financial records provided by our private partners for all firms in our experiments
from January 2019 to June 2022. These two elements allow us to robustly measure
the causal impact of the two experiments in the context of the pandemic. In this
research, our main variables of interest are liquidity and earnings at the firm level.
We use access to credit in the banking system (total and commercial debt) to measure
firms’ liquidity and the total amount of monthly deposits to proxy firms’ earnings.
With a joint take-up rate of 27.9% (percentage of firms in the treatment group
that received and accepted our Covid-19 loan offer), we show that government-
guaranteed economic relief programs had a significant positive impact on liquidity.
Using as the outcome variable total liquidity and commercial liquidity (the main
type of debts that approximate MSMEs’ available funding), we find an intent-to-
treat effect of 15.7% and 26.4%, both increases being statistically significant at the
99% confidence level. Moreover, the treatment-on-the-treated effect suggests even a
larger statistically significant increase in total liquidity (101.4% impact).
Our expectation is that our results will provide insights about the relevance
of economic programs focused on small businesses–like FOGAPE and Unidos por
Colombia–which positively impact MSMEs’ economic performance and probability
of survival during economic deterioration by increasing their liquidity. Additionally,
the results of our RCTs should help to quantify better the cost-efficiency analysis
of expanding economic programs, so that governments in developing economies can
make more informed decisions to better mitigate the negative impacts of potential
future economic slumps. Lastly, we believe that this research takes into account the
unanticipated pandemic shock and the disproportionate negative impact that Covid-
19 mobility and gathering restrictions generated on small firms that, besides being
the backbone of the economy in countries like Chile and Colombia, are also family
businesses.3
Our main contribution to the literature consists of showing that relaxing short-
3
MSMEs represent 98.6% of the total number of businesses, and they employ more than two-
thirds of the workforce in Chile (OECD, 2020). Moreover, the pandemic disproportionately im-
pacted small businesses in Chile: 63% of the MSMEs reported a decrease in sales, amounting to
an average of -37%. Also, while the employment rate of medium enterprises declined by -6.1%, the
reduction observed for SMEs was -8.9% (Chilean Economic Department, 2021).
4
term financial constraints to firms under stress during economic recessions improves
their economic performance and increases survival rates, which generates a positive
impact on the economy as a whole. Despite the high prevalence and large size of
the government-guaranteed economic relief programs that targeted MSMEs during
the Covid-19 pandemic, the economic literature studying the causal impact of these
policies on relevant economic variables (such as revenues, employment, access to
credit) is still limited. The implementation of these emergency economic programs
to face the negative shock of the pandemic sets up the perfect scenario to perform
causal impact analyses by allowing commercial banks to extend loans to firms that
would not have had access to credit due to the higher risk of default.
The main piece of evidence measuring the effects of these type of credits as public
policy is Cai and Szeidl (2018) [6]. The authors quantify the impact of loan offers by a
government-subsidized lender running a randomized experiment with 3,100 Chinese
firms located in 78 different local markets. The experimental setting allowed the
authors to exploit variation within and across markets. Cai and Szeidl’s experiments
reached a slightly higher take-up than ours (33%) and found that easier access to
loans increased firms’ revenues by 9%. Although their setting seems to be similar to
ours, the economic context behind the experiment is different. In our setting, access
to credit is due to government-backed loan programs, which were almost exclusively
offered to MSMEs that had been previously rejected due to the high risk of default.
Additionally, our RCTs were implemented in the middle of a pandemic, so we should
expect a greater impact of the policy on outcomes such as sales, employment, and
firm survival, taking into consideration the economic scenario and the type of firms
that were part of our experiments. Regarding the cost-benefit analysis of these
government-backed relief programs, the only paper related to ours, to the best of
our knowledge, is Kaufmann (2020) [14]. Kaufmann estimates the benefits and costs
of the Swiss Covid-19 lending program by exploiting regional variation in a non-
experimental setting. The authors find that an increase of CHF 100,000 in the
amount of the loans saved between 0.22 and 0.29 jobs. Taking into consideration the
costs that some borrowers may face because of firms’ defaults, the paper finds that
the cost of saving these 0.22-0.29 jobs is about CHF 11,500. Since we lack studies
5
determining the cost-benefits of similar policies, a systematic review by Kersten et al.
(2017) [15] notes that ”given the lack of a clear picture of the costs of different SME
financing programs, it remains difficult to estimate cost effectiveness.” A systematic
review by Grimm and Paffhausen (2015) [13] notes that ”a significant shortcoming
of the literature is that almost nothing is known about cost effectiveness.”
The rest of our paper is organized as follows. Section 2 describes the experimental
setting. Here, we highlight the relevance of FOGAPE and the FNG as active and
well-established economic programs in Chile and Colombia. Section 3 describes the
randomization process and shows the results of the baseline balance tests, which allow
us to conclude that most of the MSMEs’ relevant covariates are uncorrelated with
the treatment assignment. Section 4 shows the main results of our joint RCT. We
show the ITT estimates, month-by-month regressions, and the event study results.
(In Appendix A.6, we also perform TOT regressions and run different robustness
checks that confirm our main conclusions.) Section 5 shortly describes the main
policy implications we extract from our RCT, the next steps, and how our results
may help to better understand not only the impact but also the mechanisms by
which a financial loan can support small businesses not only during but also ex post
economic declines. Section 6 concludes.
2 Institutional Context
2.1 Fogape Chile
We first conducted a randomized controlled trial to evaluate the impact of Covid-19
government-guaranteed commercial loans offered through the long-term and well-
established Chilean program called Fondo de Garantia para Pequenos Empresarios
(FOGAPE). FOGAPE has been one of the largest and most well-known Chilean
financial programs aiming to help and foster the nationwide growth and develop-
ment of micro, small and medium-sized businesses by facilitating access to private
credit to firms that otherwise may face significant credit constraints in the private
banking system. The loans (and other financial services), which are administered
6
and offered by private financial institutions, are partially or fully guaranteed by the
government. The program started operation in 1980 and is therefore the oldest credit
guarantee scheme in the country. To date, FOGAPE is still the main public program
supporting and facilitating access to credit for most of the MSMEs that require for-
mal funding. Before the Covid-19 pandemic, the program had a yearly application
process open to all national firms that could fulfill a number of standard financial
requirements. Historically, the loans were partially guaranteed by the government,
so commercial banks were also responsible for managing and ensuring the compliance
of each requirement and for the payment of the principal.
During pre-pandemic years, FOGAPE awarded an average of US81.25 billion per
year and provided financial support to more than 6% of the total number of Chilean
MSMEs. Although this percentage seems small, it is important to mention that
Chile experienced stable economic growth in recent years, so many small businesses
did not apply for FOGAPE-backed loans. Economic programs strongly expanded,
however, as a response to the Covid-19 pandemic, and FOGAPE awarded a total of
US813.9 billion and granted access to financial credit to more than 32% of MSMEs.
The share guaranteed by the government also substantially increased to 60-90%.
The final percentage guaranteed to a specific MSME depends on several factors,
including the firm’s size and the economic sector in which the firm sells its goods or
offers services. This financial expansion, together with the increase in the percentage
guaranteed, allowed many private banks to offer more loans to higher-risk businesses
than they would have otherwise been able to.
2.2 FNG and the Covid-19 Unidos por Colombia Program
The analogue to FOGAPE in Colombia is the Fondo Nacional de Garantı́as S.A.
(FNG). The FNG is a specialized department that facilitates access to credit for
MSMEs and independent workers through government-guaranteed financial loans
and other financial products. The FNG began operation in 1982 and, similar to
FOGAPE in Chile, has continuous growth regarding the number of guaranteed loans
and the geographic scope of the program; to date, the FNG has provided support
7
to small firms in all 32 Colombian departments. The total amount of government-
guaranteed approved loans in 2010 was 86.1 trillion COP and the total approved
amount in 2019 expanded to 815 trillion COP.
With the arrival of the Covid-19 pandemic, the FNG played an important role
in the economy: the Fondo became the main mechanism for providing financial
support to small businesses during the crisis. In fact, while the average total amount
of approved government-guaranteed loans was 810.9 trillion COP between 2010 and
2019 (distributed among 397,000 loans), 819.8 COP trillions were approved in 2020
alone (distributed among 623,000 loans).4 During the pandemic, the Fondo has given
support not only to MSMEs, but also to the self-employed and large firms. This is
the result of a large increase in public funds allocated by the Central Government to
help firms across all regions of the country.
In the Colombian case, the intervention in the randomized experiment is an offer
for a loan guaranteed by the Colombian government, as part of its pandemic re-
sponse plan. In April 2020, the Colombian Government launched the ”Unidos por
Colombia” program, which implemented a series of measures for MSMEs to access
loans through the National Guarantee Fund (Fondo Nacional de Garantı́as, hence-
forth FNG). Among the measures are loans which are guaranteed by the FNG for
80% of the principal and allow for a grace period of up to 4 months from the start
of the loan. Until December 2021, the Government made approximately COP 36
trillion available (about USD 89 billion) to guarantee loans to MSMEs, formal and
informal microenterprises, and self-employed individuals, to help them to cover pay-
roll and working capital needs in order to survive the pandemic (Fondo Nacional de
Garantı́as, 2021). By the end of 2021, more than 600,000 firms had benefited from
this guarantee program.
The relevant expansion of public funds was highlighted by the World Bank, which
ranked Colombia second only to Mexico in a survey studying the performance and
scope of economic programs that different Latin American governments implemented
4
Source: FNG Annual Reports. Available at: https://www.fng.gov.co/nosotros/rendicion-de-
cuentas/historico-de-indicadores
8
to face the Covid-19 pandemic.5
Our Colombian RCT was possible due to the implementation of the special Covid-
19 subprogram called Unidos por Colombia, which focused fully on helping MSMEs
and self-employers, giving them support to increase their liquidity in the short term.
According to FNG, Unidos por Colombia increased loan guarantees from 50% to 90%
and generated larger commissions (70-75%).
2.3 Government-backed Loan Programs in Latin America
Besides Chilean and Colombian programs, government-backed loans programs were
common and one of the most significant policy responses to the Covid-19 crisis in
Latin American economies. Many of them expanded programs already established
to help small businesses during mobility constraints, which allowed firms to survive,
pay rent, and meet payrolls during these times.
Before this research, we conducted a survey with more than 50 private banks in
different Latin American countries. Out of all banks, 54% of them mentioned that
they were offering government-backed loans at the start of the pandemic. Further-
more, the banks also reported that they would only lend to higher-risk MSMEs if
they had access to partial or full government guarantees of the principal. Thus, the
innovation is likely to generalize to other contexts in which banks issue government-
backed loans to higher-risk MSMEs, which occurred in many countries in response
to the Covid-19 crisis. It would also likely generalize to government-backed loan
programs implemented in response to other crises. However, it may be unlikely to
generalize to the effect of non-government-backed loans (as banks do not issue these
loans to higher-risk MSMEs) or to the effect of government-backed loans during
non-crisis times.
The United Nations published a report at the end of 2021 that describes and
analyzes the main policies to support MSMEs during the pandemic.6 The report
5
According to the survey, Colombia was ranked second in the region in terms of government-
guaranteed loans as a percentage of the total number of loans approved in the financial system.
6
Available here: https://www.cepal.org/en/publications/47145-analysis-policies-support-smes-
confronting-covid-19-pandemic-latin-america
9
highlights the relevance of the policies that were focused on increasing and maintain-
ing short-term liquidity and facilitating access to credit.
3 Experimental Design
3.1 Implementation of the RCTs
As a result of two collaboration agreements with private formal financial institutions,
one in each country, we conducted two controlled experiments including 10,072 and
3,379 MSMEs in Chile and Colombia, respectively. The initial number of micro,
small, and medium-sized businesses in Chile was 5,575 experimental firms. However,
we increased the number of participant firms by 4,497, as a result of the second
FOGAPE tender that the federal government launched early in January 2021. One
special characteristic of the Chilean experimental firms is that most of them had
already requested a loan at the beginning of the pandemic (before the execution of
the first emergency Covid-19 FOGAPE tender). The controlled intervention thus, as
a result of the extraordinary expansion of FOGAPE public funds, permitted high-
risk MSMEs to receive a loan offer despite the fact that they still had the same
insufficient credit score that did not allow them access to funding in the past under
normal economic conditions.
We randomized the 10,072 firms, splitting them evenly (50-50) into the control
and treatment groups. The randomization randomly offered government- guaranteed
loans to 5,069 MSMEs. The Chilean marketing campaign (loan offer) lasted six
months (starting in February and ending in July, 2021). However, the six loans
approved in July were assigned to the treated firms the first day of the month.
Similarly, the six loans that were approved in February were finally assigned during
the last five days of the month. Therefore, during our impact evaluation, we assume
that the Chilean RCT extended instead from March to June only (four months).
The Colombian RCT was implemented in two subsequent waves. Initially–i.e.,
immediately after the implementation of the first tender of Unidos por Colombia–we
were able to launch the experiment with only 984 micro, small, and medium-sized
10
enterprises (Wave 1). As in the Chilean case, the randomization of these 984 MSMEs
evenly split (50-50) the experimental the experimental firms into the control and
treatment groups. The random process assigned 472 firms to the treatment group.
Once again, as in the Chilean case and as a result of a second expansion of
the Colombian emergency Covid-19 program, we were able to increase the number of
firms participating in the experiment. However, the program extension occurred after
the implementation of the Wave 1. Therefore, we conducted a second experiment
(Wave 2) that included 2,375 firms. This time, however, the split of firms into
treatment and control groups was 80-20 (as required by the private bank) so that
1,903 MSMEs were assigned to the treatment group.
The final number of firms participating in the Colombian intervention (pooled
RGf) was then 3,379, of which 2,375 units were assigned to the treatment group.
Lastly, it is worth mentioning that each wave also extended for four months: while
Wave 1 extended from May to August 2021, Wave 2 extended from September to
December 2021.
Figure 1 shows the timelines of the Chilean (CH) and Colombian (CO) RCTs.
Figure 1: Timeline of the Chilean (CH) RCT and of the Two Waves of the Colombian
(CO) RCT
Table 1 summarizes the total number of experimental firms participating in our
joint RCT and describes the aggregates split between countries.
11
Table 1: RCTs, Countries, Total Number of MSMEs, and Experimental Groups
RCT’s Description Chile Colombia Joint
- Financial Partner Private Bank Private Bank
- Initial Number of MSME 5,575 984 6,559
- New MSMEs (program expansion) (1) 4,497 2,375 6,872
- Final Number of MSMEs 10,072 3,379 13,451
- Treatment Group 5,069 2,375 7,444
- Control Group 5,003 1,004 6,007
(1) In Chile (Colombia), the extension of FOGAPE (Unidos por Colombia) allowed
us to increase the total number of MSMEs before (after) the implementation of the former
experiment. The Colombian RCT was then performed in two subsequent waves.
Regarding the experimental design, it is worth mentioning that our RCTs did not
involve in-person interactions. The financial loans were offered to firms via either
direct phone calls or an online marketing campaign. In this paper, we evaluate the
causal impact of the joint RCT using as outcome variables the total amount of debt
that firms contracted with the formal financial system (total and commercial liquid-
ity), and the total amount of deposits that each firm in our experiment performed
on a monthly basis. The administrative financial data we use in our analyses–which
was directly provided by our two private financial partners–include all MSMEs in our
experiments and are at the firm level on a daily/weekly basis. The period extends
from January 2019 to the most recent available month (July and August 2022 for
the Chilean and Colombian cases, respectively).
3.2 Relevance of Our Experiment
Taking into account the contemporary state of the Chilean and Colombian economies
together with the evolution of the Covid-19 pandemic, the setting and the timing
of our experiments are relevant. First, in both countries MSMEs are, by far, the
most relevant type of firms in terms of both employment and economic growth
(there are approximately 900,000 and 1,500,000 MSMEs in Chile and Colombia,
12
respectively). According to the OECD [17], more than 98% of the total number of
Chilean businesses are classified as MSMEs, and they employ around two-thirds of
the Chilean workforce (5.8 million workers in total). Second, FOGAPE and the FNG
are the main public policy levers for fostering and providing financial support to the
growth of small businesses. In fact, the formal banking system has approved 283,000
FOGAPE-guaranteed loans since the Covid-19 pandemic hit in February 2020. In
other words, if we assume that a firm could get at most one loan during the pan-
demic, nearly 32% of the total number of MSMEs have received government-backed
loans due to the expansion of these programs.7
Using raw estimations, and given the actual size of the Chilean FOGAPE pro-
gram, we think that the enlargement of the program may directly impact 12% of
the Chilean workforce (1.1 million workers). If we assume that all these workers live
in different households, the FOGAPE Covid-19 program’s scope may impact 19%
of households in Chile. Although these estimates may be crude approximations, we
believe that our RCTs will help policymakers to widen the understanding of the im-
pact of the program, since our experiment studies high-risk small businesses during
a recession: firms face greater financial stress in difficult times.
If these programs helped MSMEs to survive during an economic crisis, the Chilean
economy could better face future recessions if the government decides to broaden FO-
GAPE again. So, taking into account the significance and relevance of the FOGAPE
expansion during the economic crisis caused by the pandemic, we believe that the
results we obtain from our RCTs will be key when deciding whether to broaden
or keep the magnitude and scope of these programs after the Covid-19 pandemic
is over. Additionally, performing RCTs that include high-risk businesses during an
economic slump will inform policymakers about the effectiveness of these programs,
allowing them to define better strategies so that the use of limited public resources
is efficiently rationalized.
Another relevant feature of our experiments is that they incorporate not only
MSMEs located in the capital but also in the rest of the country. In fact, more
than half of the firms that took part in the experiments were located outside of the
7
Although the share could be lower, since firms could receive more than one loan from FOGAPE.
13
capital. This is also relevant when taking into account the effort that the federal
governments have taken to decentralize the economy in both countries.
3.3 Randomization
3.3.1 Chilean RCT
The 10,072 MSMEs in the Chilean experiment were randomly assigned (50-50 split)
into the treatment group (firms that received the loan offer) and the control group
(firms that did not receive the offer). To stratify, we used three relevant economic
variables associated with each firm (two continuous and one categorical variable):
credit score, annual sales, and specific type of firm. For the first two variables, score
and sales, we stratified based on their quartiles. However, to avoid a sparse matrix
post-stratification, we finally decided to merge the bottom two quartiles, such that
our three-level modified quartiles are the following:
• Score Q1 (Sales Q1): MSMEs having a score (annual sales) below the median;
• Score Q2 (Sales Q2): MSMEs having a score (annual sales) above the median
but below the 75th percentile;
• Score Q3 (Sales Q3): MSMEs having a score above the 75th percentile
The type of firm is a three-category variable that distinguishes MSMEs according
to the loan and campaign decisions made by the bank during its last marketing loan
campaign. We classified experimental firms into the following three types:
• Type 1 : MSMEs that received a Covid-19 loan offer and finally accepted the
offer before our experiment.
• Type 2 : MSMEs that received a Covid-19 loan offer but did not accept the
offer before our experiment.
• Type 3 : MSMEs that were rejected by the bank and thus did not receive any
offer before our experiment.
14
The total number of strata was, thus, 27. Table 6 in Appendix A.1 shows the
distribution of the Chilean MSMEs by type of firm, and score and sales quartiles. The
final randomization resulted in 5,069 and 5,003 MSMEs assigned to the treatment
and control groups, respectively.
Lastly, it is important to mention that, before implementing our stratification, we
studied whether the correlation between sales and scores was relevant. Our concern
was that MSMEs having higher sales may also have, at the same, lower risk scores. If
both variables were strongly correlated, we should only have included one of them to
stratify firms. However, as Table 7 in Appendix A.1 shows, the correlation between
the modified score and sales quartiles was not significant.
3.3.2 Colombian RCT
In a similar fashion, we stratified the Colombian MSMEs using a similar three-
category variable that defines the type of firm. The type of firm, then, categorizes
MSMEs according to the decision made by the Colombian bank regarding the loan
petition/response that our financial partner received during its last marketing cam-
paign. We group the 3,379 micro, small, and medium-sized firms into the following
three categories:
• Type 1 : MSMEs that received and accepted a loan in the previous credit
campaign.
• Type 2 : MSMEs that did not receive an offer before based on their credit score.
• Type 3 : MSMEs that either received an offer in the previous credit campaign
but did not accept it or were not contacted in the previous campaign but should
have received the offer given their credit score.
The stratification process performed in Wave 1 also included the total number
of months in which the MSMEs made at least one deposit during 2020 (we decided
to include this variable once a large proportion of firms from the initial universe we
received from the bank did not make any deposit that year). The total number of
15
strata in this wave was 36 (3x12), and the 984 firms that took part of the experiment
were split evenly (50-50) into the treatment and control groups. The stratification
process for Wave 2, in contrast, only used the type of firm as a variable, and 80%
of the 2,395 firms were assigned to the treatment group (80-20 split). Table 8 in
Appendix A.1 shows the final distribution of the Colombian MSMEs in our RCT by
type of firm and type of experiment (Wave 1, Wave 2, and Pooled RCT).
3.4 Balance Checks
To check the balance between MSMEs assigned to the control group and MSMEs
assigned to the treatment group, we ran a set of balance checks with a blocking
dummy control variable using the main economic variables that we associated with
the firms. Balance checks include: i) the total amount of deposits; ii) the liquidity
(total debt) that MSMEs contracted with the formal financial system; iii) the age
of the firm; iv) the nature of the firm: legal entity (recognized as a business) or
natural person; v) the geographical location (capital or rest of the country); and vi)
the gender of the owner. The total number of MSMEs included in a specific balance
check varies with the covariate we test. The reason explaining this variation is that
our partners did not fully characterize every single MSME in the experiment.
The main specification for our balance checks is the following:
J
X
yi = α + β · T reati + γj · Stratumi,j + εi (1)
j=1
where yi is the metric or MSME economic feature that we are interested in testing,
T reati is a dummy variable indicating the treatment status of firm i, and Stratumi,j
is a dummy variable indicating whether the i-th MSME belongs to the stratum j.
The parameter of interest is β. We estimate equation (1) with and without strata
fixed effects as a robustness checks.
To check the correct balance between MSMEs assigned to the control group and
MSMEs assigned to the treatment group, we ran a set of balance checks with a
blocking dummy control variable using the main economic variables that we could
16
associate with the firms. Balance checks include: i) the total amount of deposits; ii)
the liquidity (total debt) that MSMEs contracted with the formal financial system;
iii) the age of the firm; iv) the nature of the firm: legal entity (recognized as a busi-
ness) or natural person; v) the geographical location (capital or rest of the country);
and vi) the gender of the owner. Table 2 shows the results of our baseline balance
checks. The total number of MSMEs that were included in a specific balance check
varies with the covariate we test. The reason explaining this variation is that our
partners could not fully characterized every single MSME in the experiment.
The results of our balance checks indicate that there are no statistically significant
differences between the control and the treatment groups for almost all metrics we
test (for some specifications, either the liquidity or deposits are significant in the
Colombian case). Last, we run an F-Test and conclude that the covariates do not
jointly exhibit a statistically significant relationship with the treatment assignment.
17
Table 2: Baseline Balances, Chilean and Colombian (Pooled) RCTs
Chilean RCT Colombian RCT
Variable N Control Mean Mean Diff. N Control Mean Mean Diff.
Firms Mean Diff. w/strata FEs Firms Mean Diff. w/strata FEs
(1)
Deposits per month 10,072 1.906 0.007 0.007 3,379 2.165 -0.255*** -0.048
(0.016) (0.022) (0.020) (0.050) (0.060) (0.058)
[0.000] [0.758] [0.720] [0.000] [0.000] [0.412]
(1)
Liquidity per month 10,072 2.599 0.033 0.041 3,379 4.280 -0.097 -0.025
(0.023) (0.033) (0.029) (0.062) (0.074) (0.075)
[0.000] [0.317] [0.154] [0.000] [0.190] [0.739]
(2)
Firm Age 10,052 10.813 -0.018 -0.022 2,892 13.403 0.678* -0.447
(0.097) (0.137) (0.135) (0.308) (0.367) (0.372)
[0.000] [0.897] [0.852] [0.000] [0.065] [0.229]
(3)
Firm nature 10,069 0.746 0.001 0.002 3,379 0.387 -0.070*** 0.020
- Legal Entity = 1 (0.006) (0.009) (0.009) (0.015) (0.018) (0.016)
[0.00] [0.900] [0.852] [0.000] [0.000] [0.215]
Location 10,007 0.520 0.001 0.001 3,378 0.248 -0.038** 0.012
- Capital = 1 (0.007) (0.010) (0.010) (0.013) (0.016) (0.016)
[0.000] [0.890] [0.910] [0.000] [0.016] [0.452]
Gender Owner 9,390 0.324 -0.013 -0.012 3,300 0.242 -0.005 -0.013
- Female = 1 (0.007) (0.010) (0.010) (0.014) (0.016) (0.017)
[0.000] [0.179] [0.199] [0.000] [0.741] [0.457]
(1) Dependent variable: winsorized top 5%, with log(y+1). Variable measured as the mean per month taking into
account the six months pre-RCT implementation. Liquidity is defined as the total debt contracted with the formal
banking system.
(2) For the Chilean case, firm age is defined as the number of years that the small business has been a bank’s
client.
(3) Firm nature is a dummy variable equal 1 (0) when the SME is registered as a legal entity (natural person).
- Robust standard errors in parenthesis. P-values in brackets. *Sign. at 10%. **5%. ***1%
Table 9 in Appendix A.2 shows the baseline balance results for each of the two
18
Colombian waves separately. The table confirms that there are no statistically sig-
nificant differences between the control and treatment groups in either wave. On
the other hand, Tables 10-12, also in Appendix A.2, show the results of the main
baseline balances when using the variables we use to stratify in each of the RCTs.
As expected, all mean difference estimates are not statistically significant at the
99% confidence level. Therefore, we conclude that our randomization process is
well-designed and statistically valid.
Lastly, Tables 13 and 14 in Appendix A.3 compare the main metrics and firm
characteristics between experimental groups. The comparison includes the average
total amount of deposits and liquidity per month pre-RCT implementation, the gen-
der of the owner, location, firm size, and economic sector. The tables show that
there is no statistically significant difference between groups.
4 RCTs Ex post Main Results
4.1 Take-up Rates and Loan Amounts
Among the firms assigned to the treatment groups, 1,377 Chilean and 700 Colombian
businesses (2,077 in total) accepted and received the offer (27.2% and 29.5% take-up
rates, respectively, and 27.9% in the joint RCT -). Although MSMEs belonging to
the control group did not receive any offer, 107 non-treated Chilean firms had access
to the loan (2.1% of the total number of MSMEs assigned to the control group).8 We
consider the marketing financial campaigns a success taking into consideration the
ex ante estimates that our partners suggested, taking into account the take-ups from
the most recent previous marketing loan campaigns. Because of recent loan offers,
the Chilean bank was expecting a take-up rate closer to 20%, and the Colombian
bank was expecting a take-up of 16-18%.
Table 3 shows, by country (and the joint RCT), the total number of firms that
received the loan, the take-up rates, and the average amount of approved loans in
8
These 107 MSMEs were not contacted by us when implementing our RCT campaign. They
instead contacted the bank directly as business clients.
19
millions of local currency and thousands of US dollars. As the table shows, the total
amount approved by our partners summed up to US$ 824.2M and US$ 825.5M in
the Chilean and Colombian experiment respectively (taking into account only the
loans approved to treated firms). The average loan size was US$ 817.6K and US$
836.4K in the Chilean and Colombian RCT, respectively.
Table 3: Take-up Rates and Average Loan Amount
Variable Chile Colombia Pooled Joint
Treat Control Total Treat Control Total Treat Control Total
N MSMEs 5,069 5,003 10,072 2,375 1,004 3,379 7,444 6,007 13,451
N Compliers 1,377 107 1,484 700 - 700 2,077 107 2,184
Take-up 27.2% 2.1% 14.7% 29.5% - 20.7% 27.9% 1.8% 16.2%
(1)
Loan Amount 12.5 20.9 13.1 136.1 - 136.1 - - -
(2)
Loan Amount 17.6 29.4 18.4 36.4 - 36.4 23.9 29.4 24.2
(1) Average amount in millions of Chilean and Colombian pesos.
(2) Average amount in thousands of US dollars. We use the exchange rate of May 2021 in each.
country to transform local currencies to US dollars. Exchange rates: Chile = 711.6 CLP/US dollar;
Colombia = 3,735.7 COP/US dollar.
Table 15 in Appendix A.4 describes the take-up rates and average loan sizes for
each of the two Waves in the Colombian RCT. Figures 2 and 3 shows the distribution
of the assigned Covid-19 loans by month, treatment group, and country. As Figure
2 shows, most of the 1,377 loans (99.3% of the total) in the Chilean RCT were
approved between March and June 2021. Although the RCT campaign extended
from February to July, only 10 loans were assigned in these two months (the six
loans approved in July were assigned to the MSMEs on July 1). Given the timing
and distribution of approved loans, when running the ITT and TOT regressions, we
instead assume that the Chilean RCT extended from March to June only.
Lastly, Appendix A.5 performs the power calculations using the Duflo et al.
(2007) methodology [10] for the joint and country-specific RCTs.
20
Figure 2: Loans Approved by Month and Experimental Group, Chilean RCT
Figure 3: Loans Approved by Month, Treatment Group, Colombian RCT. *Note:
Wave 1: May-Aug ; Wave 2: Sep-Dec.
21
5 Estimating Loans Effects
5.1 Data Sources and Main Outcomes
Given our experimental design, the randomization strategy, and the results from our
balance baseline tests, we were able to track the potential causal impact of our RCTs
over time on relevant economic outcomes that approximate MSMEs’ performance
during the pandemic. In this paper, we study the impact of the Covid-19 loan
offers on liquidity (defined either as the total amount of debt or the total amount
of commercial debt that MSMEs contracted with the formal banking system) and
earnings (using as proxy the total amount of deposits they performed each month).
Liquidity is a relevant economic outcome for MSMEs since it tells us about the
capacity of high-risk small firms to obtain funding during economic recessions, which
will allow them to continue paying short-term commitments (payrolls and payments
to suppliers), and to have a higher probability of survival when mobility and gathering
restrictions are in place.
The liquidity data consist of monthly information provided directly by our part-
ners. The banks, however, get the information directly from the Chilean financial
regulator (Comision para el Mercado Financiero, CMF) and the second largest and
most recognized private credit bureau in Colombia respectively (both official sources
that compile and deliver trustworthy data on liquidity). In both countries, we are
able to disaggregate total liquidity into commercial, consumption and mortgage debt.
We are exclusively focused on commercial and total liquidity since the Covid-19 loans
were loans approved to businesses, so that new injection of liquidity may only affect
total debt through the commercial channel. However, we include the other two types
of liquidity in our regressions to perform additional robustness checks. We expect
either null or negative (substitution effect) impact when we take the consumption
and mortgage liquidity as an outcome of interest.
We proxy earnings by using direct transactional information from the banks’
internal systems, which provided us with high-frequency transactions (at the daily
and weekly basis for Chile and Colombia respectively) containing the total number
22
and the total amount of deposits made by each firm. However, in our analyses,
earnings are aggregated at the monthly level (liquidity is only available on a monthly
basis). Although the time period of the data is the same for each country, the
disaggregation of deposits is different. While we could distinguish in-person, online
and POS deposits in the Colombian case, the Chilean data do not contain information
regarding POS, and the bank was not able to disaggregate the information into in-
person and online deposits.
The period of analysis extends from January 2019 to July 2022, the most recent
update of information at the time of writing. However, we expect to update the
results once we receive new data. Liquidity and earnings are measured in real terms
using January 2019 as the base month and are then converted into US dollars. (How-
ever, when we perform the analysis for each country independently in Appendices
A.7 and A.8, the data are in millions of real Chilean (CLP) and Colombian (COP)
pesos, respectively).
Lastly, as a note of caution, we acknowledge that our earnings proxy (total
amount of deposits) may be partial and have measurement issues. To further study
the impact of our experiment, we continue working with our partners to incorporate
new relevant outcome variables that more thoroughly capture MSMEs’ economic
performance.
5.2 Methodology
Our experimental design and valid randomization strategy allows us to estimate
in reduced form the impact of the FOGAPE and Unidos por Colombia Covid-19
government-guaranteed loan relief economic programs. Our baseline intention-to-
treat (ITT) regression to estimate is the following:
K
X
yi,post = α + β · T reati + γk · Stratumi,j + δ · yi,pre + εi (2)
k=1
where yi,post is the outcome of interest (average liquidity or average earnings per
month post treatment) of M SM E i post treatment, T reati is an indicator variable
23
denoting treatment assignment, Stratumi,j is a strata FE, yi,pre is the lagged out-
come of interest, i.e., the average outcome of interest per month ex ante the RCT’s
execution), and εi are heteroskedastic-robust standard errors. Our parameter of
interest is β. For robustness, we estimate 4quation (2) using three different specifi-
cations: including or excluding both strata FEs and the lagged outcome of interest.
Our preferred specification, given the covariate regression adjustments, is the one
that includes both as explanatory variables. Dependent and lagged variables were
transformed to logarithms and upper-winsorized at the 5% level to avoid the impact
of outliers. The winsorizing was done at the monthly-treatment group levels. Lastly,
regardless of the specification of equation (2), the main results (sign and significance
of our parameter of interest) remain the same in most cases.
The regressions we show in this section endogenize the month in which the treated
firm i was exposed to the experiment. For treated firms that received and accepted
the Covid-19 loan offer, we use the month in which the loan was delivered as the
month in which they were exposed to the treatment. For those treated MSMEs that
did not take the offer, we use the month when they were contacted by phone as the
experiment’s exposed month. Lastly, for those treated firms that did not receive the
Covid-19 loan and could not be contacted (797 MSMEs out of 7,444) and the firms
assigned to the control group, we assume that they were exposed to the experiment
the first month of each RCT-wave implementation.
Additionally, we run alternative ITT regressions using three different scenarios
as additional robustness checks. The three alternative specifications, described in
Table 17 in Appendix A.6, assume different exogeneous and general ex ante and ex
post RCT periods. The results of these three robustness checks are also shown in
Appendix A.6.
Our final robustness check uses an instrumental variable (IV) strategy. Given
that there was imperfect compliance in both RCTs, we explore whether we obtain
different results by using an IV methodology to localize the impact of the loan offer
based on firms who took up the loan as a result of the offer. To this end, we run the
24
following specification for the two-stage least squares (TSLS):
K
X
yi,post = α + β · \
Compliancei + γk · Stratumi,j + δ · yi,pre + εi (3)
k=1
where \
Compliancei is the fitted value for the take-up dummy variable. The rest
of the variables are the same as those defined in equation (2). The first stage consists
of regressing the compliance dummy, Compliancei , using the treatment assignment
dummy variable as instrument. The first-stage equation is then the following:
Compliancei = θ · T reati + µi
The TSLS estimates for the specific country RCTs are also shown in Appendix
A.6. The results from the first stage show that the coefficient associated with the
treatment variable is positive and statistically significant at the 99% confidence level
(see Table 22). However, as Table 23 in the same Appendix shows, the second-stage
estimate impact from the fitted take-up variable remains not statistically significant
at the 99% confidence level when the variable of interest is MSME’s earnings. Con-
clusions do not change when using as dependent variable the total or commercial
liquidity: the impact is positive and statistically significant at the 99% confidence
level.
Lastly, Appendix A.6 also shows the results from the treated-on-treat (TOT)
regressions. The TOT regressions on liquidity show a positive but even larger statis-
tically significant impact of our experiments. Interestingly, the TOT regressions show
that the impact of our experiments on firms’ earnings are robust and statistically
significant at the 99% level of confidence (see Table 25).
5.3 ITT Estimates, Main Results
5.3.1 ITT Effects on Liquidity
Table 4 shows the ITT estimates for each of the four types of liquidity using three
different specifications of equation (2). The ex post RCT period goes from the month
25
after the treated MSME either received the Covid-19 loan or was contacted by phone
to June 2022. When we include strata FE and the lagged outcome, our results
show a positive and significant estimate of our parameter of interest when we use
as dependent variable either total liquidity (15.7%) or commercial liquidity (26.4%)
that MSMEs contracted with the formal banking system. As expected, we did not
find any positive significant increase in either consumption or mortgage liquidity.
26
Table 4: ITT Liquidity Effects, Joint RCTs (1), (2)
yi = log(Mean Debt per Month +1) ($US real)
(1) (2) (3)
Comm Cons Mort Total Comm Cons Mort Total Comm Cons Mort Total
Treat -0.011 0.194*** 0.089* 0.250*** 0.127*** -0.138*** -0.103*** 0.066* 0.264*** -0.042 -0.046*** 0.157***
(0.069) (0.063) (0.047) (0.058) (0.043) (0.026) (0.018) (0.036) (0.042) (0.026) (0.018) (0.036)
[0.868] [0.002] [0.056] [0.000] [0.003] [0.000] [0.000] [0.068] [0.000] [0.100] [0.010] [0.000]
C-Mean 8.142 2.013 0.785 8.876 8.142 2.013 0.785 8.876 8.142 2.013 0.785 8.876
Adj. R-Sq -0.000 0.001 0.000 0.001 0.603 0.819 0.838 0.611 0.634 0.827 0.841 0.640
N 13,451 13,451 13,451 13,451 13,451 13,451 13,451 13,451 13,451 13,451 13,451 13,451
Lag Debt No Yes Yes
27 Strata FE No No Yes
(1) Dependent variable: winsorized top 5%, with log(y+1).
(2) Post-RCT period goes from the month after the treated MSME either received the Covid-19 loan or was contacted by phone to June 2022.
- Robust standard errors in parenthesis. P-values in brackets. *Sign. at 10%. **5%. ***1%
- Comm=Commercial Debt; Cons=Consumption Debt; Mort=Mortgage Debt; Total=Comm+Cons+Mort Debts
5.3.2 Monthly ITT Effects on Liquidity
We now estimate equation (2) month by month to obtain monthly point estimates
of our parameter of interest (we only include strata FE in the regressions). The idea
here consists of inspecting whether significant differences are present in the estimates
before and after the implementation of the experiment, which extends from March
to December 2021 when we evaluate the impact of the joint RCT. Figures 4 and 5
show the monthly estimates from January 2019 to June 2022 for the log of total and
commercial liquidity, respectively. The figures show a strong and sustained positive
impact of the Covid-19 loan offer immediately after the execution of the Chilean
RCT (red vertical lines in the plots). Lastly, Figures 4 and 5 also show that the
increase stabilized beginning in January 2022 (end of the Colombian Wave 2) and
has since then slightly decreased over time.
Figure 4: Monthly ITT Estimates (Log-) Total Liquidity, Joint RCT (January 2019
to June 2022)
5.3.3 ITT Effects on Earnings
Although we have access to disaggregated data by type of deposits for the Colombian
MSMEs (in-person, online, and POS), we could not perform the analysis at that level
since the Chilean data are only at the aggregate level. While Appendix A.8 shows the
28
Figure 5: Monthly ITT Estimates (Log-) Commercial Liquidity, Joint RCT (January
2019 to June 2022)
disaggregated estimates for Colombia, here we perform the analysis using as outcome
variable the average of total earnings (total amount of deposits) per month that each
MSME performed ex post they were exposed to the experiment. Table 5 shows that
the parameter of interest is not statistically significant from zero at the 10% level
of confidence when strata FE and the lagged outcome are included. The conclusion
remains the same regardless of the alternative we use to define the ex post RCT
period (see Appendix A.6). As the table shows, the lack of significance is mainly due
to the almost negligible magnitude of the estimates. Although the ITT robustness
checks, the monthly estimates, and the event study approach confirm the lack of
significance, the conclusion changes when estimating TOT regressions. Table 25 in
Appendix A.6 shows that the Covid-19 loan offer led to a significant 45% increase in
treated-on-treat MSMEs’ earnings.
29
Table 5: ITT Earning Effects, Joint RCT (1), (2)
yi = log(Mean Total Deposits per Month + 1) ($US K real)
(1) (2) (3)
Treat -0.422*** -0.241*** -0.039
(0.045) (0.032) (0.031)
[0.000] [0.000] [0.209]
C-Mean 8.389 0.432 1.139
Adj. R-Sq. 0.006 0.484 0.532
N 13,451 13,451 13,451
Lag Debt No Yes Yes
Strata FE No No Yes
(1) Dependent variable: winsorized top 5%, with log(y+1).
(2) Post-RCT period goes from the month after the treated MSME either received
the Covid-19 loan or was contacted by phone to July 2022.
- Robust standard errors in parenthesis. P-values in brackets. *Sign. at 10%.
**5%. ***1%
5.3.4 Monthly ITT Effects on Earnings
Similar to the liquidity case, we next estimate equation (2) month by month to
obtain monthly estimates of our parameter of interest. Figure 6 confirms the results
obtain from our ITT estimations. The figure - which shows the estimates of equation
(2) from January 2019 to July 2022 - shows no impact of the Covid-19 loan offer
on firms’ earnings, whether immediately after implementation, during or after the
execution of the joint RCT.
5.3.5 Event Study
Once we had access to MSMEs’ liquidity over a long period of time, we were able to
estimate the effects of FOGAPE and Unidos por Colombia Covid-19 loan offers using
a dynamic framework (generalized difference-in-differences strategy). The dynamic
estimation allows us to incorporate into the analysis the fact that treated MSMEs
30
Figure 6: Monthly ITT Estimates (Log-) Total Earnings, Joint RCT (January 2019
to June 2022)
were exposed to the offer at different points in time. We exploit the panel data
structure and estimate a generalized diff-in-diff model which includes 42 periods in
total (from January 2019 to June 2022). The outcomes of interest, yit , corresponds
to either the liquidity (total or commercial) that the M SM Ei undertook with the
banking system during the month t or the total amount of deposits that the firm
performed during month t. The panel event regression specification is the following:
J
X K
X
yit = λi + δt + βj · Lagj,it + γk · Leadk,it + εit (4)
j=1 k=1
where λi and δt are MSMEs and months fixed effects, respectively, and εit is an
unobserved error term. Lagj,it and Leadk,it are dummy variables indicating that, at
time t, M SM Ei was j and k periods away from either the effective loan offer date.
The effective loan offer date is defined as the month when the treated M SM Ei either
received the Covid-19 loan or was reached by the bank regarding the loan offer. For
firms in the control group, Lagj,it = Leadk,it = 0 for all k, j, t and i. Also, since we
measure effects relative to the month before starting the experiment, we omit the
first Lag dummy in our estimations. Lastly, we drop the largest four lags and the
31
largest two leads estimates to avoid the influence of extreme points when plotting
the results of our estimations (since only a smaller number of firms did have those
extremes lags and leads).
As previously mentioned, we primarily use the date for when treated MSMEs
received the loan as the date for when they were exposed to the treatment. For
those treated firms that never received the loan, we instead use the date when they
were contacted by the bank. Lastly, there are 797 treated MSMEs (out of 7,444)
that ultimately could not be reached by the bank during the experiment. For those
uncontacted treated firms, we assume that they were exposed to the treatment in
the first month of the RCT-wave.
Figures 7 show the results of the event study estimation for total and commercial
liquidity when the dependent variable is measured in US dollars and in logs. Both
figures show the positive and significant impact on liquidity that the joint RCT
generates. Figure 8 shows the results when the dependent variable is total earnings.
As expected, given the ITT and monthly estimates analyses described above, the
figure shows that the experiment did have a statistically significant impact on firms’
earnings.
Figure 7: Event Study: Treatment Effects on Total Liquidity (left) and Commercial
Liquidity (right), Joint RCT,
Appendix A.9 reproduces the same analysis for each country independently. The
main results regarding the significance of the impacts on liquidity and earnings re-
main unchanged.
32
Figure 8: Treatment Effects on Total Earnings, Joint RCT, Event Study
6 Policy Implications and Limitations
This research provides one of the first rigorous impact evaluations of government-
guaranteed loan programs during the Covid-19 pandemic in Latin America. Our
randomized controlled trials were possible as a result of the significant expansions of
well-established economic policies that aim to support the development and growth
of micro, small, and medium-sized businesses in Chile and Colombia. The experimen-
tal design and the effective randomization technique allowed us to causally identify
the positive and statistically significant impact that the loan offers generated on
firms’ access to credit. The increase in short-term liquidity may potentially impact
positively MSMEs’ economic performance during the pandemic’s economic deterio-
ration. The channels by which the larger access to credit may help businesses face
better the pandemic include better chances of survival and being able to continue
paying rent, salaries to employees and short-term obligations with suppliers. The
statistically significant difference between treated and controlled firms regarding the
total liquidity they accessed during the Covid-19 pandemic also inform us that the
33
implementation of our experiments was relevant and precise in time since firms that
did not receive the loan offer could not obtain resources from other formal financial
institutions.
In this research, we study, as one of the many potential MSMEs’ economic out-
comes, whether the loan offers impacted firms’ performance by proxy earnings using
the total amount of deposits per month that each firm made with our partners. Al-
though we did not find statistically significant positive results, we were aware that
this proxy may be limited and biased downward. First, some small businesses may
still operate with cash so they do not make deposits on a regular basis. Second, firms
can have different bank accounts in which they may make deposits. A question that
remains open is to understand the contrasting results of our ITT regressions with
the estimates when considering as the main regressor the dummy variable identifying
only treated firms that effectively received the loan (taking into account the TOT
results, these MSMEs made significantly more deposits).
The results of our experiments, then, will directly inform policymakers concerning
the relevance of restructuring FOGAPE and FNG public budgets and the relaxation
of some criteria that MSMEs must fulfill to access to the programs’ benefits. At the
same time, the results of our research may inform private banks and the Department
of Finance about key aspects of the program and will help them to decide whether to
scale up or down the existing public funds accordingly to the economic phase. Lastly,
our policy implications may also be relevant taking into account that the Covid-19
pandemic may be close to an end, and the most negatively impacted Chilean and
Colombian industries are starting to enjoy economic growth again.
7 Conclusion
The implementation of policies that aimed to support and financially help micro,
small and medium enterprises in Latin America during the Covid-19 pandemic were
key to avoiding an even larger economic deterioration in these developing economies.
Longstanding economic programs were boosted after the implementation of mobility
and gathering constraints as one of the main tools implemented by governments.
34
The understanding and quantification of both direct and indirect impacts–and their
main economic channels–that these economic relief programs generated are key if
we want to explore the cost-effectiveness of programs whose goal is to help mitigate
negative shocks in the economy that impact MSMEs in these countries.
This research studied the implementation of two randomized control trials in
Chile and Colombia. These experiments were possible as a result of the large expan-
sion of well-established economic programs (FOGAPE and FNG, respectively), and
the partnerships with two private banks. We exploited rich high-frequency datasets
provided directly by our partners to quantify the causal impact of the Covid-19 emer-
gency relief programs on firms’ liquidity and earnings. We show that the financial
loans associated with our experiments had a positive and statistically significant im-
pact on liquidity (total and commercial), not only in our joint analysis but also in
both countries independently. These results may suggest that the Covid-19 guaran-
teed loans helped small businesses to face the economic crisis in a better position
since the short-term liquidity to which treated firms had access may have been used
productively. Although we did not find any positive result when using deposits as
a proxy for earnings as our outcome of interest, we were also aware that deposits
may not be a good proxy. We nonetheless still suspect that the new funds may be
used to support the business and confront short-term commitments such as rent and
salaries.
References
[1] Arráiz, I., Meléndez, M. and Stucchi, R. “Partial Credit Guarantees and Firm
Performance: Evidence from Colombia”. Small Bus Econ 43: 711–724 (2014).
https://doi.org/10.1007/s11187-014-9558-4
[2] Banerjee, A. V., and E. Duflo. “Do Firms Want to Borrow More? Testing Credit
Constraints Using a Directed Lending Program.” Review of Economic Studies
81, no. 2 (April 1, 2014): 572–607. https://doi.org/10.1093/restud/rdt046
35
[3] Banerjee, A., Duflo, E., Glennerster, R., & Kinnan, C. (2015a). “The Miracle of
Microfinance? Evidence from a Randomized Evaluation.” American Economic
Journal: Applied economics, 7(1): 22-53.
[4] Banerjee, A., Karlan, D., & Zinman, J. (2015b). “Six Randomized Evaluations
of Microcredit: Introduction and Further Steps.” American Economic Journal:
Applied Economics 7(1): 1-21.
[5] Benavente, José and Crespi, Gustavo and Figal Garone, Lucas & Maffioli,
Alessandro. (2012). The Impact of National Research Funds: A Regression Dis-
continuity Approach to the Chilean FONDECYT”. Research Policy. 41. 1461-
1475. 10.1016/j.respol.2012.04.007
[6] Cai, Jing, & Adam Szeidl, “Interfirm Relationships and Business Performance”,
Quarterly Journal of Economics, Volume 133, Issue 3, August 2018, Pages
1229–1282. https://doi.org/10.1093/qje/qjx049
[7] Cassano, Francesca, Karin Jõeveer, and Jan Svejnar. “Cash Flow vs. Collateral-
Based Credit: Performance of Micro, Small and Medium-Sized Firms in Tran-
sition Economies.” Economics of Transition 21, no. 2 (April 2013): 269–300.
https://doi.org/10.1111/ecot.12016
[8] Clarke D., and Tapia Schythe K. “Implementating the Panel Event Study”. The
Stata Journal: Promoting Communications on Statistics and Stata, January
2022
[9] De Mel, Suresh, David McKenzie, and Christopher Woodruff. “Enterprise
Recovery Following Natural Disasters.” The Economic Journal 122, no. 559
(March 1, 2012): 64–91. https://doi.org/10.1111/j.1468-0297.2011.02475.x
[10] Duflo, E., Glennerster R., and Kremer M. “Using Randomization in Develop-
ment Economics Research: A Toolkit” Handbook of Development Economics
Volume 4, 2007: 3895-3962
36
[11] Fondo Nacional de Garantı́as. (2021, December 30). Ası́ va-
mos, programa de garantı́a ”Unidos por Colombia” . Re-
trieved from Movilización Programa Unidos por Colombia:
https://www.fng.gov.co/ES/Paginas/Asi Vamos Unidos por Colombia.aspx
[12] Guerrero-Amezaga, Maria Elena, John Eric Humphries, Christopher A. Neil-
son, Naomi Shimberg, and Gabriel Ulyssea. “Small Firms and the Pandemic:
Evidence from Latin America.” Journal of Development Economics 155 (March
2022): 102775. https://doi.org/10.1016/j.jdeveco.2021.102775
[13] Grimm, Michael, and Anna Luisa Paffhausen. “Do Interventions Tar-
geted at Micro-Entrepreneurs and Small and Medium-Sized Firms Cre-
ate Jobs? A Systematic Review of the Evidence for Low and Mid-
dle Income Countries.” Labour Economics 32 (January 2015): 67–85.
https://doi.org/10.1016/j.labeco.2015.01.003
[14] Kaufmann, Daniel. “Does Government-Backed Lending Prevent Unemploy-
ment? An Assessment of the Swiss COVID–19 Lending Program,” n.d., 21.
https://www5.unine.ch/RePEc/ftp/irn/pdfs/WP20-10.pdf
[15] Kersten, Renate, Job Harms, Kellie Liket, and Karen Maas. “Small
Firms, Large Impact? A Systematic Review of the SME Fi-
nance Literature.” World Development 97 (September 2017): 330–48.
https://doi.org/10.1016/j.worlddev.2017.04.012
[16] Ministerio de Economı́a, Fomento y Turismo (2021), Boletı́n: Análisis Descrip-
tivo del Impacto de la Pandemia sobre las Empresas en Chile, Gobierno de Chile,
Santiago. https://www.economia.gob.cl/wp-content/uploads/2021/07/Boletin-
Analisis-descriptivo-del-impacto-de-la-pandemia-sobre-las-empresas-en-Chile-
1.pdf.
[17] OECD (2020), Financing SMEs and Entrepreneurs 2020: An OECD Scoreboard,
OECD Publishing, Paris, https://doi.org/10.1787/061fe03d-en.
37
[18] OECD (2021), OECD Economic Surveys: Chile, OECD Publishing, Paris,
http://www.oecd.org/economy/chile-economic-snapshot/.
[19] Romano, Joseph P., and Michael Wolf. “Balanced Control of General-
ized Error Rates”. The Annals of Statistics 38, no. 1 (2010): 598–633.
http://www.jstor.org/stable/25662253.
[20] Tan, Hong. 2009. Evaluating SME Support Programs in Chile Us-
ing Panel Firm Data. Impact Evaluation series ; no. IE 39,Pol-
icy Research Working Paper 5082. World Bank. © World Bank.
https://openknowledge.worldbank.org/handle/10986/4274 License: CC
BY 3.0 IGO.
[21] World Bank (2015), “SME Finance in Chile: Enhancing Efficiency
of Support Programs”, Finance and Private Sector Development,
https://openknowledge.worldbank.org/bitstream/handle/10986/34954/SME-
finance-in-Chile-enhancing-efficiency-of-support-programs.pdf?sequence=1&isAllowed=y.
A. Appendix
A.1 MSMEs Distribution by Stratification Variables
Table 6: MSMEs Distribution by Stratification Variables, Chilean RCT
Firms’ Type Qs Score Qs Sales Total
Q1 Q2 Q3 Q1 Q2 Q3
Type 1 1,987 1,338 1,282 1,980 1,257 1,370 4,607
Type 2 2,593 909 830 2,410 1,008 914 4,332
Type 3 5,036 271 406 646 253 234 1,133
Total 5,036 2,518 2,518 5,036 2,518 2,518 10,072
38
Table 7: Correlation Matrix: Q-Scores and Q-Sales
Qs Sales/Qs Score Q1 Score Q2 Score Q3 Score
Q1 Sales 0.47 0.26 0.28
Q2 Sales 0.51 0.26 0.23
Q3 Sales 0.56 0.23 0.21
Table 8: MSMEs Distribution by Stratification Variables, Colombian RCT
Firm Wave 1 (Number of Months with deposits) Wave 2 Pooled
Type 1 2 3 4 5 6 7 8 9 10 11 12 Total Total Total
Type 1 38 18 26 19 19 25 14 12 18 24 31 40 284 233 517
Type 2 16 11 17 14 10 12 13 10 8 7 6 19 143 422 565
Type 3 90 68 63 41 38 33 25 32 26 31 40 70 557 1,740 2,297
Total 144 97 106 74 67 70 52 54 52 62 77 129 984 2,395 3,379
39
A.2 Baseline Balances and Variables Used to Stratify
Table 9: Baseline Balances by Waves, Colombian RCT
Wave 1 Wave 2
Variable N Control Mean Mean Diff. N Control Mean Mean Diff.
Firms Mean Diff. w/strata FEs Firms Mean Diff. w/strata FEs
(1)
Deposits per month 984 2.487 -0.051 -0.045 2,395 1.830 -0.051 -0.049
(0.071) (0.102) (0.089) (0.070) (0.078) (0.075)
[0.000] [0.619] [0.614] [0.000] [0.518] [0.513]
(1)
Liquidity per month 984 4.489 -0.118 -0.221* 2,395 4.063 0.091 0.095
(0.092) (0.133) (0.130) (0.087) (0.097) (0.093)
[0.000] [0.158] [0.090] [0.000] [0.351] [0.308]
Firm Age 873 11.326 0.118 0.198 2,019 15.646 -0.863* -0.865*
(0.385) (0.550) (0.542) (0.452) (0.507) (0.493)
[0.000] [0.830] [0.716] [0.000] [0.089] [0.079]
(2)
Firm Nature 984 0.518 0.054* 0.048 2,395 0.252 0.002 0.003
- Legal entity=1 (0.022) (0.032) (0.030) (0.02) (0.022) (0.019)
[0.000] [0.087] [0.115] [0.000] [0.917] [0.864]
Location 983 0.323 0.031 0.024 2,395 0.171 0.004 0.005
- Capital =1 (0.021) (0.030) (0.031) (0.017) (0.019) (0.019)
[0.000] [0.306] [0.433] [0.000] [0.825] [0.801]
Gender of the Owner 905 0.222 -0.010 -0.003 2,395 0.260 -0.018 -0.018
- Female = 1 (0.019) (0.027) (0.028) (0.019) (0.022) (0.022)
[0.000] [0.716] [0.916] [0.000] [0.397] [0.399]
(1) Dependent variable: winsorized top 5%, with log(y+1). Variable measured as the mean per month taking into
account the six months pre-RCT-wave implementation. Liquidity is defined as the total debt contracted with the
formal banking system.
(2) Firm nature is a dummy variable equal 1 (0) when the SME is registered as a legal entity (natural person).
- Robust standard errors in parenthesis. P-values in brackets. *Sign. at 10%. **5%. ***1%
40
Table 10: Baseline Balance for Variables Used to Stratify, Chilean RCTs (1)
Variable Control Mean Variable Control Mean
Mean Diff. Mean Diff.
Risk Score Q1 0.501 -0.003 Sales Q3 0.251 -0.003
(0.007) (0.010) (0.006) (0.009)
[0.000] [0.765] [0.000] [0.739]
Risk Score Q2 0.249 0.001 Firm Type 1 0.461 -0.008
(0.006) (0.009) (0.007) (0.010)
[0.00] [0.899] [0.000] [0.433]
Risk Score Q3 0.249 0.002 Firm Type 2 0.425 0.010
(0.006) (0.009) (0.007) (0.010)
[0.000] [0.827] [0.000] [0.318]
Sales Q1 0.503 -0.007 Firm Type 3 0.114 -0.002
(0.007) (0.010) (0.014) (0.006)
[0.000] [0.511] [0.000] [0.742]
Sales Q2 0.245 0.009
(0.006) (0.009)
[0.000] [0.274]
(1) In all balance checks, the number of firms is 10,072.
41
Table 11: Baseline Balance and Variables Used to Stratify, Colombian RCTs (1)
Wave 1 Wave 2 Joint
Variable Control Mean Control Mean Control Mean
Mean Diff. Mean Diff. Mean Diff.
Firm Type 1 0.270 0.038 0.098 -0.000 0.185 -0.046***
(0.020) (0.029) (0.013) (0.015) (0.011) (0.014)
[0.000] [0.193] [0.000] [0.982] [0.000] [0.001]
Firm Type 2 0.146 0.006 0.179 -0.003 0.162 0.009
(0.016) (0.023) (0.017) (0.019) (0.012) (0.014)
[0.000] [0.790] [0.000] [0.862] [0.000] [0.542]
Firm Type 3 0.584 -0.044 0.724 0.004 0.652 0.038**
(0.022) (0.032) (0.020) (0.023) (0.015) (0.018)
[0.000] [0.167] [0.000] [0.870] [0.000] [0.032]
(1) In all balance checks, the number of firms are 984 in Wave 1, and 2,395 in Wave 2.
42
Table 12: Baseline Balance and Variables Used to Stratify, Colombian Wave 1
Wave 1
N Control Mean N Mean Control N Control Mean
months Mean Diff. months Diff. Mean months Mean Diff.
1 0.144 0.006 5 0.068 -0.004 9 0.044 0.016
(0.016) (0.023) (0.011) (0.016) (0.010) (0.014)
[0.000] [0.785] [0.000] [0.822] [0.000] [0.260]
2 0.088 0.024 6 0.068 0.005 10 0.066 -0.004
(0.013) (0.019) (0.011) (0.017) (0.011) (0.016)
[0.000] [0.221] [0.000] [0.763] [0.000] [0.812]
3 0.120 -0.023 7 0.058 -0.009 11 0.078 -0.001
(0.014) (0.020) (0.010) (0.015) (0.012) (0.017)
[0.000] [0.243] [0.000] [0.553] [0.000] [0.966]
4 0.072 0.007 8 0.058 -0.004 12 0.134 -0.014
(0.012) (0.017) (0.010) (0.015) (0.015) (0.022)
[0.000] [0.664] [0.000] [0.769] [0.000] [0.520]
43
A.3 Experimental Groups and Main Firms’ Features: A Com-
parison
Table 13: MSMEs Features and Metrics: Groups’ Comparison, Chilean RCT
Variables Category Missing Overall Control Treat P-Value
[N] [Mean (SD) / /N (% of Total)]
Deposits Pre RCT (1) Mean 0 1.9 (1.1) 1.9 (1.1) 1.9 (1.1) 0.758
Max 0 2.4 (1.2) 2.4 (1.2) 2.4 (1.2) 0.464
Min 0 1.1 (1.1) 1.1 (1.1) 1.1 (1.1) 0.876
Last Pre-Month 0 1.7 (1.3) 1.7 (1.2) 1.7 (1.3) 0.878
Deposits Post RCT Mean 0 2.0 (1.1) 2.0 (1.1) 2.0 (1.1) 0.862
Max 0 2.8 (1.2) 2.8 (1.2) 2.8 (1.2) 0.766
Min 0 0.9 (1.0) 0.9 (1.0) 0.9 (1.0) 0.807
Liquidity Pre RCT (1) Mean 0 2.6 (1.6) 2.6 (1.6) 2.6 (1.7) 0.317
Max 0 2.7 (1.6) 2.7 (1.6) 2.7 (1.6) 0.373
Min 0 2.4 (1.7) 2.4 (1.7) 2.4 (1.7) 0.265
Last Pre-Month 0 2.6 (1.7) 2.6 (1.6) 2.6 (1.7) 0.396
Liquidity Post RCT Mean 0 2.6 (1.6) 2.6 (1.6) 2.7 (1.6) <0.001
Max 0 2.8 (1.6) 2.8 (1.6) 2.9 (1.6) <0.001
Min 0 2.3 (1.7) 2.2 (1.7) 2.3 (1.7) <0.001
Econ Sector Agriculture 2,706 448 (6.1) 218 (6.0) 230 (6.2) 0.700
Construction 652 (8.9) 313 (8.6) 339 (9.1)
Education 91 (1.2) 45 (1.2) 46 (1.2)
Finance 714 (9.7) 347 (9.5) 367 (9.9)
Healthcare 308 (4.2) 164 (4.5) 144 (3.9)
Manuf. 632 (8.6) 297 (8.1) 335 (9.0)
Others 1,126 (15.3) 563 (15.4) 563 (15.2)
Retail 2,881 (39.1) 1,441 (39.4) 1,440 (38.8)
Social 99 (1.3) 55 (1.5) 44 (1.2)
Transp-Tourm 415 (5.6) 211 (5.8) 204 (5.5)
Gender Owner Female 682 2,986 (31.8) 1,516 (32.4) 1,470 (31.2) 0.186
Male 6,404 (68.2) 3,156 (67.6) 3,248 (68.8)
Capital Flag Capital 65 5,206 (52.0) 2,579 (52.0) 2,627 (52.1) 0.906
No Capital 4,801 (48.0) 2,385 (48.0) 2,416 (47.9)
MSME Size Medium 3,996 70 (1.2) 34 (1.1) 36 (1.2) 0.940
Micro 1909 (31.4) 957 (31.6) 952 (31.2)
Small 4097 (67.4) 2,038 (67.3) 2,059 (67.6)
Sample Size 10,072 5,003 5,069
- Mean per-month using a pre-RCT period of six months. Variable winsorized top 5%, with log(y+1). Liquidity here
is defined as the total debt in the banking system.
44
Table 14: MSMEs Features and Metrics: Groups’ Comparison, Colombian RCT
Variables Category Missing Overall W1 Control W1 Treat W2 Control W2 Treat P-Value
[N] [Mean (SD) //N (Percent)]
Deposits Pre RCT (1) Mean 0 2.0 (1.6) 2.5 (1.6) 2.4 (1.6) 1.8 (1.6) 1.8 (1.5) <0.001
Max 0 2.6 (1.8) 3.1 (1.7) 3.1 (1.7) 2.5 (1.8) 2.4 (1.8) <0.001
Min 0 1.0 (1.5) 1.4 (1.6) 1.3 (1.6) 0.9 (1.4) 0.9 (1.4) <0.001
Last Pre-Month 0 1.8 (1.8) 2.1 (1.8) 2.0 (1.8) 1.7 (1.8) 1.6 (1.8) 0.056
Deposits Post RCT Mean 0 2.0 (1.6) 2.5 (1.7) 2.4 (1.6) 1.9 (1.6) 1.8 (1.6) <0.001
Max 0 2.7 (1.8) 3.3 (1.8) 3.2 (1.7) 2.5 (1.8) 2.4 (1.8) <0.001
Min 0 1.0 (1.5) 1.3 (1.7) 1.2 (1.6) 0.9 (1.4) 0.9 (1.5) <0.001
Liquidity Pre RCT (1) Mean 0 4.2 (2.0) 4.5 (2.1) 4.3 (2.1) 4.1 (2.0) 4.2 (1.9) <0.001
Max 0 4.4 (1.9) 4.7 (2.0) 4.5 (2.1) 3.7 (2.1) 3.8 (2.1) 0.001
Min 0 3.8 (2.2) 4.2 (2.3) 4.0 (2.3) 3.7 (2.1) 3.8 (2.1) 0.001
Last Pre-Month 0 4.1 (2.1) 4.4 (2.2) 4.2 (2.2) 3.9 (2.1) 4.0 (2.1) 0.001
Liquidity Post RCT Mean 0 4.2 (2.0) 4.4 (2.2) 4.2 (2.1) 3.9 (2.0) 4.3 (2.0) 0.001
Max 0 4.5 (2.0) 4.8 (2.1) 4.6 (2.1) 4.2 (2.1) 4.5 (2.0) <0.001
Min 0 3.5 (2.4) 3.6 (2.5) 3.6 (2.4) 3.1 (2.3) 3.5 (2.3) 0.001
45 Econ Sector Agriculture 1,169 110 (5.0) 7 (1.6) 9 (2.2) 22 (7.6) 72 (6.6) <0.001
Construction 54 (2.4) 14 (3.3) 14 (3.5) 5 (1.7) 21 (1.9)
Education 84 (3.8) 3 (0.7) 3 (0.7) 18 (6.2) 60 (5.5)
Finance 57 (2.6) 13 (3.0) 12 (3.0) 4 (1.4) 28 (2.6)
Healthcare 48 (2.2) 8 (1.9) 12 (3.0) 2 (0.7) 26 (2.4)
Manuf. 260 (11.8) 79 (18.5) 67 (16.7) 20 (6.9) 94 (8.6)
Others 155 (7.0) 39 (9.1) 40 (10.0) 12 (4.1) 64 (5.9)
Retail 1,280 (57.9) 232 (54.3) 219 (54.5) 187 (64.3) 642 (58.9)
Social 23 (1.0) 7 (1.6) 3 (0.7) 5 (1.7) 8 (0.7)
Transp-Tourm 139 (6.3) 25 (5.9) 23 (5.7) 16 (5.5) 75 (6.9)
Gender Owner Female 79 785 (23.8) 105 (22.2) 92 (21.2) 128 (26.0) 460 (24.2) 0.300
Male 2,515 (76.2) 367 (77.8) 341 (78.8) 364 (74.0) 1,443 (75.8)
Capital Flag Capital 1 749 (22.2) 165 (32.3) 167 (35.4) 84 (17.1) 333 (17.5) <0.001
No Capital 2,629 (77.8) 346 (67.7) 305 (64.6) 408 (82.9) 1,570 (82.5)
MSME Size Medium 22 220 (6.6) 51 (10.0) 32 (6.8) 38 (7.8) 99 (5.2) <0.001
Micro 623 (18.6) 120 (23.6) 123 (26.2) 68 (13.9) 312 (16.5)
Small 2,514 (74.9) 338 (66.4) 314 (67.0) 383 (78.3) 1,479 (78.3)
Sample Size 3379 512 472 492 1903
- Mean per-month using a pre-RCT period of six months. Variable winsorized top 5%, with log(y+1). Liquidity here is defined as the total debt in
the banking system.
A.4 Take-up Rates and Loan Sizes, Colombian Waves
Table 15: Take-up Rates and Average Loan Amount, Colombian Waves
Variable Colombia Wave 1 Colombia Wave 2
Treat Control Total Treat Control Total
N MSMEs 472 512 984 1,903 492 2,395
N Compliers 80 - 80 620 - 620
Take-up (%) 16.9 - 8.1 32.6 - 25.9
(1)
Loan Amount 118.2 - 118.2 138.4 - 138.4
(2)
Loan Amount 31.6 - 31.6 37.0 - 37.0
(1) Average Amount in millions of Colombian pesos.
(2) Average Amount in thousands of US dollars. We use the exchange rate of May 2021.
to transform local currencies to US dollars. Exchange rate: 3,735.7 COP/dollar
46
A.5 Power Calculations
We measure our MDE in standard deviation, so are able to perform ex-post power
calculations using different outcomes of interest (such as access to credit, earnings,
employment, wages, and profits). We follow Duflo, Glennerster, and Kremer (2007)
[10] to compute standardized minimum detectable effects (M DEs) when imperfect
compliance is present. The M DE is defined as follows:
s r
1 σ2 1
M DE = (t(1−κ) + tα ) × (5)
p(1 − p) N (c − s)
where t(1−κ) , tα are the standard critical values for the standard 80% power and
5% significance l evels ( t(1−κ) = 0 .84 a nd t α = 1 .96 w hen u sing t he s tandard normal
distribution), respectively. The parameter σ is equal to 1, since the effects we want to
estimate are standardized to allow comparability with other studies, N is the sample
size, p is the proportion of treatment firms, p , a nd (c − s ) i s t he t ake-up r ate. The
M DE decreases (and thus the power increases) exponentially with N and p (1 − p),
meaning drops in the experimental sample size, or a larger difference b etween the
proportion of treated and controlled firms, w ill d isproportionately a ffect th e power
of the study. The M DE also decreases proportionally with the take-up rate.
Table 16 shows the power calculations for the joint RCT, but also by country and
wave (Colombian RCT). As a result of the larger number of MSMEs taking part in
the experiment, the joint RCT delivers a M DE of 0.17 standard deviations. For the
Colombian case, Wave 1 is highly underpowered due to the small sample size. The
results indicate that more than a full standard deviation impact is required to reliably
identify the regression results on the outcome. Merging both waves give us a slightly
better results since the required impact to detect statistical significance falls to 0.35
standard deviations, which is still high. The Chilean experiment is more powered
(M DE = 0.21) due to the larger number of small businesses that participated in
the RCT. Lastly, it is important to mention that (as we will see in the next section),
since we have not yet been able to find robust and statistically significant impact on
earnings, we may need deeper insights from the power of the sample to make sure
47
our conclusions regarding firms’ earnings are valid.
Table 16: Power Calculations: Joint, Chilean and Colombian RCTs
RCT / Wave N MSMEs N Treatment N Control Take-up MDE
Joint 13,451 7,444 (55%) 6,007 (45%) 27.9% 0.17
Chile 10,072 5,069 (50%) 5,003 (50%) 27.2% 0.21
Colombia, Pooled 3,379 2,375 (70%) 1,004 (30%) 29.5% 0.35
Colombia, Wave 1 984 472 (48%) 512 (52%) 16.9% 1.06
Colombia, Wave 2 2,395 1,903 (79%) 492 (21%) 32.6% 0.43
A.6 Additional Results: Joint RCT
Table 17: Alternative Specifications, ITT Regressions, Robustness Checks
Alternative Pre-RCT Period Post-RCT Period
A Jan-19 to the month before the RCT First month of the RCT - onwards
B Jan-19 to the last month of the RCT First month after the RCT - onwards
C Jan-19 to the month before the RCT First month after the RCT - onwards
48
Table 18: ITT Liquidity Effects, Joint RCT, Alternative A (1), (2)
yi = log(Mean Debt per Month +1) ($US real)
(1) (2) (3)
Comm Cons Mort Total Comm Cons Mort Total Comm Cons Mort Total
Treat 0.012 0.229*** 0.105** 0.27*** 0.211*** -0.098*** -0.081*** 0.123*** 0.342*** -0.016 -0.026 0.215***
(0.068) (0.064) (0.047) (0.057) (0.043) (0.025) (0.018) (0.036) (0.041) (0.025) (0.018) (0.035)
[0.864] [0.000] [0.026] [0.000] [0.000] [0.000] [0.000] [0.001] [0.000] [0.512] [0.135] [0.000]
C-Mean 8.190 2.031 0.790 8.920 8.190 2.031 0.790 8.920 8.190 2.031 0.790 8.920
Adj. R-Sq. -0.000 0.001 0.000 0.002 0.591 0.831 0.84 0.607 0.625 0.838 0.842 0.64
N 13,451 13,451 13,451 13,451 13,451 13,451 13,451 13,451 13,451 13,451 13,451 13,451
Lag Debt No Yes Yes
49 Strata FE No No Yes
(1) Dependent variable: winsorized top 5%, with log(y+1).
(2) Post-treat period starts the month of the RCT implementation (Mar-19). Pre-treat period goes from Jan-19 to the month previous to the RCT
(Feb-19).
- Robust standard errors in parenthesis. P-values in brackets. *Sign. at 10%. **5%. ***1%
- Comm=Commercial Debt; Cons=Consumption Debt; Mort=Mortgage Debt; Total=Comm+Cons+Mort Debts
Table 19: ITT Liquidity Effects, Joint RCT, Alternative B (1), (2)
yi = log(Mean Debt per Month +1) ($US real)
(1) (2) (3)
Comm Cons Mort Total Comm Cons Mort Total Comm Cons Mort Total
Treat 0.020 0.194*** 0.090** 0.283*** 0.112*** -0.130*** -0.089*** 0.100*** 0.278*** -0.018 -0.024 0.191***
(0.071) (0.063) (0.046) (0.06) (0.044) (0.027) (0.018) (0.038) (0.043) (0.026) (0.018) (0.038)
[0.777] [0.002] [0.050] [0.000] [0.010] [0.000] [0.000] [0.008] [0.000] [0.497] [0.174] [0.000]
C-Mean 8.015 1.969 0.757 8.762 8.015 1.969 0.757 8.762 8.015 1.969 0.757 8.762
Adj. R-Sq. -0.000 0.001 0.000 0.002 0.623 0.809 0.829 0.606 0.651 0.817 0.833 0.632
N 13,451 13,451 13,451 13,451 13,451 13,451 13,451 13,451 13,451 13,451 13,451 13,451
Lag Debt No Yes Yes
50 Strata FE No No Yes
(1) Dependent variable: winsorized top 5%, with log(y+1).
(2) Post-treat period starts the month after the RCT ended (Jan-20). Pre-treat period goes from Jan-19 to the last month of the RCT (Dec-19).
- Robust standard errors in parenthesis. P-values in brackets. *Sign. at 10%. **5%. ***1%
- Comm=Commercial Debt; Cons=Consumption Debt; Mort=Mortgage Debt; Total=Comm+Cons+Mort Debts
Table 20: ITT Liquidity Effects, Joint RCT, Alternative C (1), (2)
yi = log(Mean Debt per Month +1) ($US real)
(1) (2) (3)
Comm Cons Mort Total Comm Cons Mort Total Comm Cons Mort Total
Treat 0.020 0.194*** 0.09** 0.283*** 0.215*** -0.12*** -0.087*** 0.138*** 0.368*** -0.016 -0.022 0.237***
(0.071) (0.063) (0.046) (0.06) (0.048) (0.028) (0.02) (0.041) (0.046) (0.028) (0.019) (0.041)
[0.777] [0.002] [0.050] [0.000] [0.000] [0.000] [0.000] [0.001] [0.000] [0.576] [0.255] [0.000]
C-Mean 8.015 1.969 0.757 8.762 8.015 1.969 0.757 8.762 8.015 1.969 0.757 8.762
Adj. R-Sq. -0.000 0.001 0.000 0.002 0.534 0.788 0.803 0.539 0.574 0.796 0.806 0.576
N 13,451 13,451 13,451 13,451 13,451 13,451 13,451 13,451 13,451 13,451 13,451 13,451
Lag Debt No Yes Yes
51 Strata FE No No Yes
(1) Dependent variable: winsorized top 5%, with log(y+1).
(2) Post-treat period starts the month after the RCT ended (Jan-20). Pre-treat period goes from Jan-19 to the month previous to the RCT (Feb-19).
- Robust standard errors in parenthesis. P-values in brackets. *Sign. at 10%. **5%. ***1%
- Comm=Commercial Debt; Cons=Consumption Debt; Mort=Mortgage Debt; Total=Comm+Cons+Mort Debts
Table 21: ITT Liquidity Effects, Joint RCT, IV Strategy (1), (2)
yi = log(Mean Debt per Month +1) ($US real)
(1) (2) (3)
Comm Cons Mort Total Comm Cons Mort Total Comm Cons Mort Total
Compliance -0.044 0.744*** 0.342* 0.956*** 0.483*** -0.533*** -0.395*** 0.255* 1.044*** -0.166* -0.181*** 0.619***
(0.266) (0.242) (0.179) (0.219) (0.163) (0.101) (0.069) (0.139) (0.163) (0.1) (0.069) (0.141)
[0.868] [0.002] [0.056] [0.000] [0.003] [0.000] [0.000] [0.066] [0.000] [0.097] [0.009] [0.000]
C-Mean 8.143 2.000 0.779 8.859 8.143 2.000 0.779 8.859 8.143 2.000 0.779 8.859
Adj. R-Sq. -0.002 0.002 -0.002 0.036 0.615 0.815 0.836 0.618 0.652 0.826 0.841 0.649
N 13,451 13,451 13,451 13,451 13,451 13,451 13,451 13,451 13,451 13,451 13,451 13,451
Lag Debt No Yes Yes
52 Strata FE No No Yes
(1) Dependent variable: winsorized top 5%, with log(y+1).
(2) Post-RCT period goes from the month after the treated MSME either received the Covid-19 loan or was contacted by phone.
- Robust standard errors in parenthesis. P-values in brackets. *Sign. at 10%. **5%. ***1%
- Comm=Commercial Debt; Cons=Consumption Debt; Mort=Mortgage Debt; Total=Comm+Cons+Mort Debts
The results of the first stage are shown in the following table. The parameter of
interest is statistically significant at the 99% confidence level, and its magnitude is
similar to the take-up rate. Note also that these first-stage estimates are the same
regardless of whether we are interested in measuring the RCT impact on liquidity or
earnings.
Table 22: First Stage, IV Strategy, Joint RCT
yi = Compliance Dummy Variable
Treat 0.279***
(0.005)
[0.000]
N 13,451
Adj. R-Sq. 0.265
53
Table 23: ITT Earning Effects, Joint RCT, Robutsness Checks (1), (2)
yi = log(Mean Total Deposits per Month + 1) ($US real)
Alternative A Alternative B Alternative C IV Strategy
Treat -0.385*** -0.202*** 0.007 -0.411*** -0.192*** 0.011 -0.411*** -0.225*** 0.014 -1.615*** -0.919*** -0.154
(0.044) (0.032) (0.030) (0.047) (0.034) (0.032) (0.047) (0.036) (0.034) (0.177) (0.126) (0.122)
[0.000] [0.000] [0.813] [0.000] [0.000] [0.736] [0.000] [0.000] [0.688] [0.000] [0.000] [0.208]
C-Mean 8.416 8.416 8.416 8.287 8.287 8.287 8.287 8.287 8.287 8.418 8.418 8.418
Adj. R-Sq. 0.006 0.475 0.531 0.006 0.493 0.534 0.006 0.427 0.486 -0.081 0.451 0.529
N 13,451 13,451 13,451 13,451 13,451 13,451 13,451 13,451 13,451 13,451 13,451 13,451
Lag Debt No Yes Yes No Yes Yes No Yes Yes No Yes Yes
Strata FE No No Yes No No Yes No No Yes No No Yes
54 (1) Dependent variable: winsorized top 5%, with log(y+1).
(2) Alternative A: Post-treat period starts the month of the RCT implementation. Pre-treat period goes from Jan-19 to the month previous to the
RCT. Alternative B: Post-treat period starts the month after the RCT. Pre-treat period goes from Jan-19 to the last month of the RCT. Alternative
C: Post-treat period starts the month after the RCT. Pre-treat period goes from Jan-19 to the month previous to the RCT. IV Strategy: post-RCT
period goes from the month after the treated MSME either received the Covid-19 loan or was contacted by phone.
- Robust standard errors in parenthesis. P-values in brackets. *Sign. at 10%. **5%. ***1%
Table 24: TOT Liquidity Effects, Joint RCT (1), (2)
yi = log(Mean Debt per Month +1) ($US real)
(1) (2) (3)
Comm Cons Mort Total Comm Cons Mort Total Comm Cons Mort Total
Compliance 2.790*** 0.529*** 0.012 2.019*** 1.792*** 0.067* -0.047* 1.164*** 1.691*** 0.055 -0.017 1.014***
(0.047) (0.092) (0.064) (0.040) (0.045) (0.039) (0.028) (0.031) (0.048) (0.039) (0.028) (0.032)
[0.000] [0.000] [0.853] [0.000] [0.000] [0.089] [0.090] [0.000] [0.000] [0.156] [0.533] [0.000]
C-Mean 7.682 2.035 0.832 8.686 7.682 2.035 0.832 8.686 7.682 2.035 0.832 8.686
Adj. R-Sq. 0.066 0.003 -0.000 0.050 0.630 0.819 0.838 0.628 0.655 0.827 0.841 0.651
N 13,451 13,451 13,451 13,451 13,451 13,451 13,451 13,451 13,451 13,451 13,451 13,451
Lag Debt No Yes Yes
55 Strata FE No No Yes
(1) Dependent variable: winsorized top 5%, with log(y+1).
(2) Post-RCT period goes from the month after the treated MSME either received the Covid-19 loan or was contacted by phone.
- Robust standard errors in parenthesis. P-values in brackets. *Sign. at 10%. **5%. ***1%
- Comm=Commercial Debt; Cons=Consumption Debt; Mort=Mortgage Debt; Total=Comm+Cons+Mort Debts
Table 25: TOT Earnings Effects, Joint RCT (1), (2)
yi = log(Mean Total Deposits per Month + 1) ($US real)
(1) (2) (3)
Compliance 0.456*** 0.387*** 0.450***
(0.053) (0.038) (0.037)
[0.000] [0.000] [0.000]
C-Mean 8.082 8.082 8.082
Adj. R-Sq. 0.004 0.485 0.536
N 13,451 13,451 13,451
Lag Debt No Yes Yes
Strata FE No No Yes
(1) Dependent variable: winsorized top 5%, with log(y+1).
(2) Post-RCT period goes from the month after the treated MSME either received
the Covid-19 loan or was contacted by phone.
- Robust standard errors in parenthesis. P-values in brackets. *Sign. at 10%. **5%. ***1%
A.7 Chilean RCT, Main Results
Since the Chilean RCT was performed in just one round and we could not access dis-
aggregated information by type of deposits, we included in the analysis different post
treatment periods to discern whether the FOGAPE Covid-19 loan offers impacted
firms d ifferently in th e sh ort te rm vs . th e me dium te rm. We es timate IT T regres-
sions using three alternative post-RCT ending periods: September 2021 (short-term
impact), January 2022, and July 2022 (medium-term impact). As the tables and fig-
ures highlight in the following subsections, the parameters of interest remain similar
and the conclusions regarding their significance and sign do not change regardless of
the post-treatment period extension we use.
A.7.1 ITT Effects o n Liquidity
Table 26 shows the results for the ITT regressions for each of the four different types
of liquidity. In all regressions, we use the firm-specific date in which each treated
56
MSME was exposed to the experiment as the first month post treatment. Our
estimates show a positive and significant estimate for our parameter of interest, treat,
when we use as dependent variable either the total or the commercial liquidity. As
expected, we did not find significant positive results for consumption and mortgage
liquidity. Using the medium-term ex post RCT period and the full specification, i.e.,
including the outcome lagged and the dummies by strata variables), our estimates
show that the Chilean RCT increased total and commercial liquidity by 10.1% and
10.5%, respectively.
57
Table 26: ITT Liquidity Effects, Chilean RCT (1), (2)
yi = log(Mean Debt per Month +1) ($MM real CLP)
(1) (2) (3)
Comm Cons Mort Total Comm Cons Mort Total Comm Cons Mort Total
A. Post-treat period: Approval/Contact Month - Sep21
Treat 0.136*** 0.004 0.023 0.126*** 0.088*** -0.014** -0.009*** 0.081*** 0.096*** -0.014** -0.009*** 0.087***
(0.032) (0.014) (0.015) (0.032) (0.017) (0.006) (0.004) (0.017) (0.016) (0.006) (0.004) (0.016)
[0.000] [0.802] [0.128] [0.000] [0.000] [0.019] [0.009] [0.000] [0.000] [0.018] [0.008] [0.000]
C-Mean 2.391 0.308 0.191 2.568 2.391 0.308 0.191 2.568 2.391 0.308 0.191 2.568
B. Post-treat period: Approval/Contact Month - Jan22
Treat 0.150*** 0.004 0.023 0.139*** 0.103*** -0.014** -0.009** 0.096*** 0.111*** -0.014** -0.009** 0.096***
58 (0.032) (0.014) (0.015) (0.032) (0.017) (0.006) (0.004) (0.017) (0.016) (0.006) (0.004) (0.017)
[0.000] [0.793] [0.123] [0.000] [0.000] [0.025] [0.025] [0.000] [0.000] [0.022] [0.021] [0.000]
C-Mean 2.376 0.313 0.188 2.555 2.376 0.313 0.188 2.555 2.376 0.313 0.188 2.555
C. Post-treat period: Approval/Contact Month - Jun22
Treat 0.143*** 0.005 0.018 0.133*** 0.098*** -0.012* -0.012*** 0.09*** 0.105*** -0.013* -0.013*** 0.101***
(0.031) (0.014) (0.014) (0.032) (0.018) (0.007) (0.004) (0.018) (0.017) (0.006) (0.004) (0.016)
[0.000] [0.702] [0.212] [0.000] [0.000] [0.057] [0.002] [0.000] [0.000] [0.051] [0.002] [0.000]
C-Mean 2.354 0.316 0.185 2.536 2.354 0.316 0.185 2.536 2.354 0.316 0.185 2.536
N 10,072 10,072 10,072 10,072 10,072 10,072 10,072 10,072 10,072 10,072 10,072 10,072
Lag Debt No Yes Yes
Strata FE No No Yes
(1) Dependent variable: winsorized top 5%, with log(y+1).
(2) Post-RCT period starts the month after the treated MSME either received the Covid-19 loan or was contacted by phone.
- Robust standard errors in parenthesis. P-values in brackets. *Sign. at 10%. **5%. ***1%
- Comm=Commercial Debt; Cons=Consumption Debt; Mort=Mortgage Debt; Total=Comm+Cons+Mort Debts
A.7.2 Monthly ITT Effects on Liquidity
Figure 9 shows the monthly estimates for the Chilean RCT from January 2019 to
June 2022. The figure shows a strong and sustained positive impact of the Covid-
19 loans program immediately after the RCT implementation (red vertical line) in
commercial and total liquidity (the same does not happen when we study the impact
on consumption liquidity). Figure 9 also shows that the increasing positive impact
stabilized in January 2022 and since then has slightly decreased over time.
Figure 9: Monthly ITT Estimates by Type of Debt in the Banking System, Chilean
RCT (January 2019 - June 2022)
A.7.3 ITT Effects o n Earnings
Table 27 shows the estimates when running the ITT regressions using as depen-
dent variable the mean of the total amount of deposits per month that each MSME
performed ex post exposition to the RCT. The parameter of interest, treat, is not
statistically significant from zero at the 90% confidence le vel. The conclusion remains
the same regardless of the specification a nd t he l ength o f t he e x p ost R CT period
we use. The lack of significance is mainly due to the almost negligible magnitude of
the estimate.
59
Table 27: ITT Earnings Effects, Chilean RCT (1), (2)
yi = log(Mean Total Deposits per Month + 1) ($MM real CLP)
A. End post-treat period: Sep21 B. End post-treat period: Jan22 C. End post-treat period: Jul22
Treat -0.013 -0.021 -0.018 -0.008 -0.015 -0.012 -0.012 -0.019 -0.016
(0.023) (0.014) (0.014) (0.022) (0.014) (0.014) (0.021) (0.014) (0.013)
[0.562] [0.152] [0.208] [0.715] [0.270] [0.360] [0.577] [0.165] [0.221]
C-Mean 2.005 2.005 2.005 2.044 2.044 2.044 1.993 1.993 1.993
N 10,072 10,072 10,072 10,072 10,072 10,072 10,072 10,072 10,072
Lag Debt No Yes Yes No Yes Yes No Yes Yes
Strata FE No No Yes No No Yes No No Yes
(1) Dependent variable: winsorized top 5%, with log(y+1).
(2) Post-RCT period starts the month after the treated MSME either received the Covid-19 loan or was contacted
by phone.
- Robust standard errors in parenthesis. P-values in brackets. *Sign. at 10%. **5%. ***1%
A.7.4 Monthly ITT Effects o n Earnings
Similarly to the liquidity case, Figure 10 shows the monthly Chilean estimates for the
outcome variable total earnings from January 2019 to July 2022. Figure 10 confirms
the conclusions we obtain when running the ITT estimations. The figure s hows no
impact of the FOGAPE Covid-19 loans program after either the implementation or
the completion of the RCT.
A.7.5 TOT Effects on Liquidity and Earnings
Tables 28 and 29 shows the results of the TOT regressions using as dependent variable
the mean of liquidity and total earnings post-RCT, respectively. Regardless of the
specification, the parameter of interest is always positive and statistically significant
from zero at the 99% confidence level for total liquidity, commercial liquidity, and
total earnings. Using the medium-term post-treat period and the full specification,
we find a statistically significant FOGAPE Covid-19 loan impact of 57.6% and 19.4%
in total liquidity and total earnings, respectively.
60
Figure 10: Monthly ITT Estimates Total Earnings, Chilean RCT (January 2019 -
June 2022)
61
Table 28: TOT Liquidity Effects, Chilean RCT (1), (2)
yi = log(Mean Debt per Month +1) ($MM real CLP)
(1) (2) (3)
Comm Cons Mort Total Comm Cons Mort Total Comm Cons Mort Total
A. Post-treat period: Approval/Contact Month - Sep21
Compliance 1.130 0.088*** -0.002 1.051*** 0.758*** 0.035*** 0.004 0.727*** 0.603*** 0.016* 0.001 0.572***
(0.030) (0.022) (0.021) (0.031) (0.019) (0.009) (0.005) (0.018) (0.019) (0.009) (0.005) (0.019)
[0.000] [0.000] [0.923] [0.000] [0.000] [0.000] [0.409] [0.000] [0.000] [0.074] [0.921] [0.000]
C-Mean 2.293 0.297 0.202 2.477 2.293 0.297 0.202 2.477 2.293 0.297 0.202 2.477
B. Post-treat period: Approval/Contact Month - Jan22
Compliance 1.106*** 0.092*** -0.001 1.031*** 0.746*** 0.040*** 0.005 0.717*** 0.595*** 0.020** 0.001 0.566***
62 (0.030) (0.022) (0.021) (0.031) (0.019) (0.009) (0.005) (0.019) (0.019) (0.010) (0.006) (0.019)
[0.000] [0.000] [0.944] [0.000] [0.000] [0.000] [0.387] [0.000] [0.000] [0.035] [0.911] [0.000]
C-Mean 2.288 0.301 0.200 2.473 2.288 0.301 0.200 2.473 2.288 0.301 0.200 2.473
C. Post-treat period: Approval/Contact Month - Jun22
Compliance 1.084*** 0.100*** -0.001 1.015*** 0.735*** 0.048*** 0.005 0.735*** 0.562*** 0.028*** -0.000 0.588***
(0.030) (0.023) (0.021) (0.031) (0.020) (0.010) (0.006) (0.020) (0.020) (0.010) (0.006) (0.020)
[0.000] [0.000] [0.953] [0.000] [0.000] [0.000] [0.416] [0.000] [0.000] [0.005] [0.999] [0.000]
C-Mean 2.266 0.304 0.195 2.454 2.266 0.304 0.195 2.454 2.266 0.304 0.195 2.454
N 10,072 10,072 10,072 10,072 10,072 10,072 10,072 10,072 10,072 10,072 10,072 10,072
Lag Debt No Yes Yes
Strata FE No No Yes
(1) Dependent variable: winsorized top 5%, with log(y+1).
(2) Post-RCT period starts the month after the treated MSME either received the Covid-19 loan or was contacted by phone.
- Robust standard errors in parenthesis. P-values in brackets. *Sign. at 10%. **5%. ***1%
- Comm=Commercial Debt; Cons=Consumption Debt; Mort=Mortgage Debt; Total=Comm+Cons+Mort Debts
Table 29: TOT Earnings Effects, Chilean RCT (1), (2)
yi = log(Mean Total Deposits per Month + 1) ($MM real CLP)
A. End post-treat period: Sep21 B. End post-treat period: Jan22 C. End post-treat period: Jul22
Compliance 0.259*** 0.210*** 0.166*** 0.277*** 0.229*** 0.184*** 0.268*** 0.222*** 0.177***
(0.030) (0.020) (0.020) (0.029) (0.019) (0.019) (0.028) (0.018) (0.018)
[0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000]
C-Mean 1.960 1.960 1.960 1.999 1.999 1.999 1.947 1.947 1.947
N 10,072 10,072 10,072 10,072 10,072 10,072 10,072 10,072 10,072
Lag Debt No Yes Yes No Yes Yes No Yes Yes
Strata FE No No Yes No No Yes No No Yes
(1) Dependent variable: winsorized top 5%, with log(y+1).
(2) Post-RCT period starts the month after the treated MSME either received the Covid-19 loan or was contacted
by phone.
- Robust standard errors in parenthesis. P-values in brackets. *Sign. at 10%. **5%. ***1%
A.8 Colombian RCT, Main Results
For the Colombian RCT, besides running the ITT and TOT regressions using the
full sample (pooled), we also estimate separately the causal regressions for each
wave. The goal of performing regressions by waves consists of studying whether
significant d ifferences ar e pr esent in ea ch wave so th at th ey ex plain th e significance
and magnitudes of the estimates when estimating the causal impact of the loans
under the total universe of 3,379 MSMEs.
To make the analysis concise, we only include in this appendix the results when
using the full post-RCT period, i.e., all regressions use data that extend to July 2022).
Lastly, similarly to the Chilean case, the parameter of interest remains similar and
the conclusions regarding its significance a nd s ign d o n ot c hange r egardless o f the
specification and experimental universe we use.
63
A.8.1 ITT Effects on Liquidity
Table 30 shows the results of the ITT specifications for each of the four types of
liquidity. Once again, we conclude that the Covid-19 loans program did not have
positive and significant impacts on consumption and mortgage liquidity. Table 30
also shows that our experiment positively impacted the total and commercial liquid-
ity when we pool both waves (10.7% and 32.1% both statistically significant at the
99% confidence level). These results are driven by the significant positive impact
that the Unidos por Colombia emergency program generated on treated firms that
were part of Wave 2. As the table shows, our RCT did not significantly impact any
type of liquidity after the implementation of the first wave.
64
Table 30: ITT Liquidity Effects, Colombian RCT (1), (2)
yi = log(Mean Debt per Month +1) ($MM real COP)
(1) (2) (3)
Comm Cons Mort Total Comm Cons Mort Total Comm Cons Mort Total
Pooled (N = 3,379)
Treat 0.180* 0.003 -0.025 0.069 0.283*** -0.056* -0.077*** 0.041 0.321*** -0.039 -0.052** 0.107***
(0.096) (0.063) (0.048) (0.078) (0.051) (0.034) (0.022) (0.04) (0.053) (0.037) (0.024) (0.042)
[0.062] [0.956] [0.604] [0.375] [0.000] [0.099] [0.001] [0.306] [0.000] [0.292] [0.029] [0.009]
C-Mean 2.976 1.09 0.481 4.162 0.135 0.075 0.065 -0.108 0.367 0.138 0.074 0.108
Wave 1 (N = 984)
Treat 0.007 -0.212** 0.019 -0.179 0.013 -0.075 0.002 -0.024 0.01 -0.081 -0.000 -0.033
65 (0.163) (0.105) (0.085) (0.138) (0.073) (0.054) (0.038) (0.062) (0.075) (0.056) (0.038) (0.064)
[0.964] [0.043] [0.823] [0.195] [0.856] [0.163] [0.954] [0.705] [0.899] [0.146] [0.998] [0.603]
C-Mean 3.384 1.082 0.485 4.378 -0.079 0.054 0.024 -0.052 0.02 0.084 0.002 0.006
Wave 2 (N = 2,395)
Treat 0.545*** 0.051 -0.033 0.302*** 0.515*** -0.007 -0.083*** 0.196*** 0.518*** -0.006 -0.083*** 0.201***
(0.126) (0.082) (0.063) (0.101) (0.072) (0.05) (0.03) (0.055) (0.071) (0.05) (0.03) (0.055)
[0.000] [0.531] [0.603] [0.003] [0.000] [0.889] [0.005] [0.000] [0.000] [0.911] [0.005] [0.000]
C-Mean 2.552 1.097 0.477 3.937 0.075 0.051 0.071 -0.223 0.32 0.079 0.064 0.062
Lag Debt No Yes Yes
Strata FE No No Yes
(1) Dependent variable: winsorized top 5%, with log(y+1).
(2) Post-RCT period starts the month after the treated MSME either received the Covid-19 loan or was contacted by phone.
- Robust standard errors in parenthesis. P-values in brackets. *Sign. at 10%. **5%. ***1%
- Comm=Commercial Debt; Cons=Consumption Debt; Mort=Mortgage Debt; Total=Comm+Cons+Mort Debts
A.8.2 Monthly ITT Effects on Liquidity
Similarly to the Chilean case, we proceed to estimate equation (2) month by month
to obtain monthly estimates of our parameter of interest. Figure 11-13 show the
monthly estimates from January 2019 to June 2022 for all four types of liquidity.
The figures confirm the results shown in Table 30. The positive and significant
impact of the Covid-19 loans program is only present in the commercial and the
total liquidity. Figures 12 and 13 also confirm that the positive and statistically
significant impact is mostly totally driven by the results from Wave 2, since we do
not find significant impacts in any liquidity when including in our estimations only
the MSMEs that participated in Wave 1.
Figure 11: Monthly ITT Estimates by Type of Liquidity, Colombian RCT Pooled
(Jan-19 - Mar-22)
A.8.3 ITT Effects o n Earnings
We next study the impact of our Colombian RCT on firms’ e arnings. The Colombian
financial partner was able to disaggregate deposits into deposits performed in person
and online, and POS. We exploit this categorization to study whether heterogeneous
impacts are present since the Covid-19 pandemic may have differentiated impacts
among these three types of earnings. Taking into account the specifications that
66
Figure 12: Monthly ITT Estimates by Type of Liquidity, Colombian RCT Wave 1
(Jan-19 - Mar-22)
Figure 13: Monthly ITT Estimates by Type of Liquidity, Colombian RCT Wave 2
(Jan-19 - Mar-22)
67
include both the strata fixed e ffects an d th e pr e-RCT ea rnings me an as covariates,
we did not find any statistically significant impact regardless of the type of earnings
or wave we study. Table 31 shows the results for the pooled sample and each wave
separately. Similarly to the Chilean results, the parameters of interest remain non-
statistically significant at the 90% confidence level.
68
Table 31: ITT Earnings Effects, Colombian RCT (1), (2)
yi = log(Mean Deposits +1) ($MM real COP)
(1) (2) (3)
In person Online POS Total In person Online POS Total In person Online POS Total
Pooled (N = 3,379)
Treat -0.116** -0.235*** -0.036 -0.238*** -0.083** -0.093*** 0.027 -0.09** -0.038 -0.042 0.048* -0.027
(0.052) (0.06) (0.047) (0.065) (0.035) (0.033) (0.024) (0.039) (0.037) (0.034) (0.025) (0.04)
[0.026] [0.000] [0.448] [0.000] [0.019] [0.006] [0.270] [0.020] [0.306] [0.221] [0.060] [0.508]
C-Mean 1.344 1.434 0.491 2.271 1.344 1.434 0.491 2.271 1.344 1.434 0.491 2.271
Wave 1 (N = 984)
Treat -0.02 -0.187* 0.084 -0.123 -0.017 -0.034 0.046 -0.018 -0.015 -0.05 0.042 -0.03
69 (0.09) (0.105) (0.09) (0.111) (0.059) (0.06) (0.043) (0.065) (0.061) (0.061) (0.043) (0.065)
[0.825] [0.075] [0.349] [0.267] [0.777] [0.574] [0.277] [0.784] [0.801] [0.407] [0.328] [0.643]
C-Mean 1.476 1.757 0.600 2.604 1.476 1.757 0.600 2.604 1.476 1.757 0.600 2.604
Wave 2 (N = 2,395)
Treat -0.034 0.010 0.022 -0.002 -0.052 -0.039 0.049 -0.028 -0.051 -0.038 0.049 -0.027
(0.068) (0.073) (0.057) (0.084) (0.046) (0.041) (0.031) (0.051) (0.046) (0.04) (0.031) (0.051)
[0.613] [0.895] [0.701] [0.982] [0.264] [0.332] [0.115] [0.585] [0.266] [0.352] [0.116] [0.596]
C-Mean 1.206 1.098 0.377 1.925 1.206 1.098 0.377 1.925 1.206 1.098 0.377 1.925
Lag Debt No Yes Yes
Strata FE No No Yes
(1) Dependent variable: winsorized top 5%, with log(y+1).
(2) Post-RCT period starts the month after the treated MSME either received the Covid-19 loan or was contacted by phone.
- Robust standard errors in parenthesis. P-values in brackets. *Sign. at 10%. **5%. ***1%
A.8.4 Monthly ITT Effects on Total Earnings
Figure 14 shows the monthly estimates when using as outcome variable the total
earnings by wave and the pooled RCT from January 2019 to July 2022. The plots
confirm the results obtained when running the ITT regressions. The impact of the
Unidos por Colombia Covid-19 loans did not generate any positive significant impact
during or after the implementation of either wave. Figures 15-17 reproduce the same
analysis for each of the three specific typse of deposits we were able to identify
(in-person, online, and POS). All three figures show no significant impact.
Figure 14: Monthly ITT Estimates, Total Deposits Colombian RCT, by Wave (Jan-
19 - Jul-22)
A.8.5 TOT Effects o n L iquidity a nd Earnings
Tables 32 and 33 show the results of the TOT regressions using as dependent variable
the post-RCT mean of either the liquidity or earnings. Regardless of the specification,
the parameter of interest is always positive and statistically significant f rom z ero at
the 99% confidence level for total liquidity and total e arnings. We find a statistically
significant p ositive i mpact o f 8 4.9% a nd 2 1.8% i n t otal l iquidity a nd t otal earnings,
respectively, when we pooled both waves.
70
Figure 15: Monthly ITT Estimates, Deposits In-person, Colombian RCT, by Wave
(Jan-19 to Jul-22)
Figure 16: Monthly ITT Estimates, Online Deposits, Colombian RCT, by Wave
(Jan-19 to Jul-22)
71
Figure 17: Monthly ITT Estimates, POS, Colombian RCT, by Wave (Jan-19 to Jul-
22)
72
Table 32: TOT Liquidity Effects, Colombian RCT (1), (2)
yi = log(Mean Debt per Month +1) ($MM real COP)
(1) (2) (3)
Comm Cons Mort Total Comm Cons Mort Total Comm Cons Mort Total
Pooled (N = 3,379)
Compliance 2.128*** 0.267*** -0.011 1.285*** 1.691*** 0.012 -0.031 0.847*** 1.744*** 0.004 -0.017 0.889***
(0.070) (0.073) (0.052) (0.059) (0.050) (0.042) (0.042) (0.027) (0.053) (0.043) (0.028) (0.036)
[0.000] [0.000] [0.839] [0.000] [0.000] [0.781] [0.781] [0.000] [0.000] [0.922] [0.537] [0.000]
C-Mean 2.662 1.037 0.466 3.944 2.662 1.037 0.466 3.944 2.662 1.037 0.466 3.944
Wave 1 (N = 984)
Compliance 2.120*** -0.126 0.016 1.392*** 0.805*** -0.049 0.009 0.592*** 0.799*** -0.073 -0.004 0.577***
73 (0.181) (0.193) (0.161) (0.164) (0.100) (0.091) (0.046) (0.090) (0.107) (0.096) (0.051) (0.099)
[0.000] [0.515] [0.919] [0.000] [0.000] [0.588] [0.846] [0.000] [0.000] [0.448] [0.933] [0.000]
C-Mean 3.215 0.991 0.493 4.179 3.215 0.991 0.493 4.179 3.215 0.991 0.493 4.179
Wave 2 (N = 2,395)
Compliance 2.399*** 0.300*** -0.004 1.360*** 1.925*** 0.049 -0.012 0.981*** 1.882*** 0.025 -0.019 0.939***
(0.078) (0.080) (0.056) (0.065) (0.056) (0.047) (0.030) (0.039) (0.056) (0.047) (0.030) (0.038)
[0.000] [0.000] [0.944] [0.000] [0.000] [0.296] [0.684] [0.000] [0.000] [0.597] [0.529] [0.000]
C-Mean 2.380 1.060 0.452 3.824 2.380 1.060 0.452 3.824 2.380 1.060 0.452 3.824
Lag Debt No Yes Yes
Strata FE No No Yes
(1) Dependent variable: winsorized top 5%, with log(y+1).
(2) Post-RCT period starts the month after the treated MSME either received the Covid-19 loan or was contacted by phone.
- Robust standard errors in parenthesis. P-values in brackets. *Sign. at 10%. **5%. ***1%
- Comm=Commercial Debt; Cons=Consumption Debt; Mort=Mortgage Debt; Total=Comm+Cons+Mort Debts
Table 33: TOT Earnings Effects, Colombian RCT (1), (2)
yi = log(Mean Deposits +1) ($MM real COP)
(1) (2) (3)
In person Online POS Total In person Online POS Total In person Online POS Total
Pooled (N = 3,379)
Compliance 0.250*** 0.198*** 0.028 0.297*** 0.099** 0.123*** 0.079*** 0.176*** 0.126*** 0.143*** 0.094*** 0.211***
(0.058) (0.064) (0.053) (0.070) (0.039) (0.038) (0.021) (0.041) (0.041) (0.039) (0.022) (0.042)
[0.000] [0.002] [0.605] [0.000] [0.012] [0.001] [0.000] [0.000] [0.002] [0.000] [0.000] [0.000]
C-Mean 1.211 1.228 0.460 2.043 1.211 1.228 0.460 2.043 1.211 1.228 0.460 2.043
Wave 1 (N = 984)
Compliance 0.682*** 0.734*** 0.136 0.828*** 0.217* 0.157 0.185*** 0.253* 0.238** 0.133 0.176*** 0.215
74 (0.173) (0.190) (0.188) (0.180) (0.117) (0.139) (0.058) (0.133) (0.122) (0.144) (0.063) (0.137)
[0.000] [0.000] [0.471] [0.000] [0.064] [0.260] [0.001] [0.058] [0.051] [0.356] [0.005] [0.116]
C-Mean 1.411 1.608 0.630 2.478 1.411 1.608 0.630 2.478 1.411 1.608 0.630 2.478
Wave 2 (N = 2,395)
Compliance 0.271*** 0.274*** 0.078 0.394*** 0.128*** 0.176*** 0.081*** 0.235*** 0.110** 0.145*** 0.081*** 0.211***
(0.062) (0.067) (0.055) (0.076) (0.043) (0.039) (0.024) (0.044) (0.043) (0.039) (0.024) (0.044)
[0.000] [0.000] [0.156] [0.000] [0.003] [0.000] [0.001] [0.000] [0.011] [0.000] [0.001] [0.000]
C-Mean 1.109 1.035 0.374 1.821 1.109 1.035 0.374 1.821 1.109 1.035 0.374 1.821
Lag Debt No Yes Yes
Strata FE No No Yes
(1) Dependent variable: winsorized top 5%, with log(y+1).
(2) Post-RCT period starts the month after the treated MSME either received the Covid-19 loan or was contacted by phone.
- Robust standard errors in parenthesis. P-values in brackets. *Sign. at 10%. **5%. ***1%
A.9 Generalized Difference-in-Differences (Event Study)
A.9.1 Chilean RCT
We use the total and the commercial debts as main metrics to measure liquidity,
and estimate equation (4) using the full available period (from January 2019 to June
2021). For the Chilean RCT, the final specification of Equation (4) contains 30 lags
(from Jan-19 to Jun-21) and 15 leads (from Feb-21 to Jun-22).
Figures 18 and 19 show the results of the event study estimation for these two
types of debt when the dependent variable is measured in millions of real Chilean
pesos, and its logarithmic transformation. We use the date for when treated MSMEs
received the loan as the date for when they were exposed to the treatment. For those
treated firms that eventually did not receive the loan, we instead use the date when
they were contacted by the bank. There are 631 treated MSMEs (out of 5,069) that
ultimately could not be reached by the bank during the experiment. For these 631
firms, we assume that they were exposed to the treatment during the month of the
RCT implementation (March, 2021).
Figure 18: Event Study, Chilean RCT: Treatment Effects on Total Liquidity
Figure 20 replicates the methodology described above but uses instead the total
amount of deposits as dependent variable.
75
Figure 19: Event Study, Chilean RCT: Treatment Effects on Commercial Liquidity
Figure 20: Event Study, Chilean RCT: Treatment Effects on Total Earning
76
A.9.2 Colombian RCT
Similarly to the Chilean case, we use the total debt and the commercial debt as main
metrics to proxy for liquidity. We also estimate equation (4) using the full period
which, in this case, extends from January 2019 to March 2022. The total number of
lags and leads included in the final specification of equation (4) are function of the
RCT-wave: while Wave 1 has 31 lags (from Jan-19 to Jul-21) and 11 leads (from
May-21 to Mar-22), Wave 2 has 35 lags (from Jan-19 to Nov-21) and 7 leads (from
Sep-21 to Mar-22)
Figures 21-24 show the results of the event study estimation for the two types
of liquidity by wave when the dependent variable is measured in millions of real
Colombian pesos, and its logarithmic transformation. Here, we also use the specific
date for when the treated MSMEs were reached by the bank regarding the loan offer.
For those treated firms that eventually did not receive the loan, we use this date as
the date for when they were exposed to the treatment. As a note of caution, we still
have 166 treated MSMEs (out of 2,375) that ultimately could not be reached during
the experiment (37 and 129 MSMEs belong to Wave 1 and Wave 2, respectively).
For those 166 treated MSMEs, we assume that they were exposed to the treatment
during the first month of the RCT-wave implementation.
Figure 21: Event Study, Colombian Wave 1: Treatment Effects on Total Liquidity
77
Figure 22: Event Study, Colombian Wave 1: Treatment Effects on Commercial
Liquidity
Figure 23: Event Study, Colombian Wave 2: Treatment Effects on Total Liquidity
Figure 24: Event Study, Colombian Wave 2: Treatment Effects o n Commercial
Liquidity
78
File and source
- File
- cerda-2023-the-causal-impact-of-covid-19-government-backed-loans-on-msmes-liquidity-and-ea.pdf
- Size
- 2,425,555 bytes
- SHA-256
- 60b6ec8fe40eee9ffbd56e37042d1f3fa91d5505e444db9b6c8f3f9d6f1e353d
- Our copy
- cerda-2023-the-causal-impact-of-covid-19-government-backed-loans-on-msmes-liquidity-and-ea.pdf
- Original
- web.archive.org