Policy Research Working Paper 11140 Do Formal Loans Boost SME Performance? Key Takeaways from a Meta-Analysis Miriam Bruhn Johan Andrey Ortega Hernandez Claudia Ruiz Ortega Development Economics Development Research Group June 2025 Policy Research Working Paper 11140 Abstract This paper conducts a meta-analysis of 24 studies evaluating larger when loans are issued by public rather than private the impact of formal loans on small and medium-sized banks, and the effects are broadly similar across firm size, enterprise performance. Using a Bayesian hierarchical country income levels, and guarantee structures. The larger model, the paper estimates that formal loans increase small impact of public bank loans suggests that private lenders’ and medium-sized enterprise employment by 12 percent, profit-maximizing incentives may not always align with sales by 18.3 percent, and profits by 17.6 percent. Subgroup providing funds to the most credit-constrained firms that analyses show that the effects of credit on employment are have the highest returns to capital. This paper is a product of the Development Research Group, Development Economics. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at mbruhn@worldbank.org and cruizortega@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Do Formal Loans Boost SME Performance? Key Takeaways from a Meta-Analysis Miriam Bruhn, Johan Andrey Ortega Hernandez, Claudia Ruiz Ortega* JEL Classification: G21, G28, O12, O16 Keywords: bank loans; SMEs; job creation * Miriam Bruhn: DECFP, The World Bank, mbruhn@worldbank.org (corresponding author). Johan Andrey Ortega Hernandez: DECFP, The World Bank, jortegahernandez@worldbank.org. Claudia Ruiz Ortega: DECFP, The World Bank, cruizortega@worldbank.org. We are thankful for the feedback we received from David McKenzie and Rachael Meager. All errors are our own. 1. Introduction Small and medium-sized enterprises (SMEs) represent over 90% of businesses and provide two- thirds of jobs worldwide (ILO 2022). At the same time, SMEs often face significantly greater credit constraints than larger firms (Veselin et al., 2014; Beck et al., 2006; Beck and Demirgüç-Kunt, 2006). The estimated gap between SME credit demand and supply stands at US$5.7 trillion (Singh et al. 2025). To address this shortfall, many governments and international organizations have introduced programs designed to expand access to finance for SMEs, with the goal of promoting firm growth and job creation. These initiatives commonly involve providing funds or offering credit guarantees to private or public banks to incentivize lending to SMEs. While a growing number of studies have evaluated the impact of loan programs on SME performance, findings are often mixed and context-specific, making it challenging to draw broad conclusions about their effectiveness in boosting job creation, revenues, or profits. Some studies report limited or no effects on average firm outcomes (e.g., Bryan, Karlan, and Osman, 2024), while others find substantial positive impacts (e.g., Cai and Szeidl, 2024). These differences may stem from variations in program design, implementation, or the characteristics of the firms and financial institutions involved. To address this issue, we conduct a meta-analysis that consolidates this expanding body of evidence to estimate the average effect of lending interventions on SMEs, and to examine how impacts vary across key dimensions—such as country income level, the type of implementing institution (public versus private bank), the presence of loan guarantees, and the size of targeted firms. By synthesizing evidence across diverse contexts, our meta-analysis provides a clearer picture of when and how these programs are most effective. We assess the effects of formal loans on SME employment, sales, and profits by combining results from different studies to produce average impact estimates. Starting from seven recent anchor studies and using a snowballing approach, we identified 1,025 potentially relevant papers, ultimately including 24 in our meta-analysis. Of these, 22 report impacts on employment, 14 on sales, and 9 on profits. To synthesize these findings, we use a Bayesian hierarchical model (BHM), which accounts for heterogeneity across studies while estimating average effects. 2 Our results show that formal loans increase SME employment by 12%, sales by 18.3%, and profits by 17.6%. The three effects are statistically significant at the 95% level. The larger number of studies on employment allows us to conduct subgroup analysis for this outcome. We find that the effect of formal loans on SME employment is similar across guaranteed versus non-guaranteed loans, across developing versus high-income countries, and for firms of different sizes. However, loans originated by public financial institutions have larger effects on employment than loans from private institutions. Our study builds on previous meta-analyses examining the effects of financial interventions on business outcomes but differs in important ways. Grimm and Paffhausen (2015) assess the impact of such programs on employment among microentrepreneurs and SMEs, finding overall negative effects. However, their analysis is largely driven by interventions targeting micro-entrepreneurs— 20 of the 26 included studies—and focuses primarily on group lending schemes, grants, subsidies, and unconditional cash transfers, rather than formal loans. Meager (2019, 2022) focuses on microcredit interventions and finds small average effects on household businesses, with large heterogeneity, particularly among households in the upper tail of the distribution. In contrast to these studies, we focus exclusively on formal SME loans, primarily delivered through banks. These loans involve larger sizes than microcredit and go to established firms that use the funds for business rather than household purposes. The effects may thus differ from those of microcredit. Indeed, our estimated effect of formal loans on firm profits (17.6%) is more than twice as large as the less than 8% increase due to microcredit found in Meager (2019). We also contribute to prior syntheses of SME support programs. Piza et al. (2016) examine a broad range of interventions, such as training, consulting, and innovation support, with predominantly grant-based financing mechanisms. Our study, by contrast, restricts attention to formal, repayable loan products—offered through regulated credit channels. Closer to our work, Kersten et al. (2017) conduct a meta-analysis on SME finance programs, finding positive effects on financing, investment, and employment. We extend their work in three ways: (i) by focusing on formal bank lending, enhancing comparability across studies; (ii) by harmonizing firm outcomes (employment, sales, and profits) and reporting standardized effect sizes for clearer interpretation; and (iii) by incorporating a larger set of recent studies, providing a more comprehensive view of the evidence. 3 A novel feature of our meta-analysis is that we conduct a detailed subgroup analysis, which provides new insights into when loan programs are most effective. Our finding that loans originated by public banks have a larger effect on SME employment than those from private banks suggests that private lenders’ profit-maximizing incentives may not align with providing funds to the most credit-constrained SMEs that have the highest returns to capital. We highlight that there is a tradeoff between channeling SME credit through public versus private banks. On the one hand, public banks may take more risks and provide loans to SMEs with higher returns than private banks would. On the other hand, public banks can be subject to inefficiencies and public capture, particularly in low-capacity environments (World Bank 2013). The rest of the paper is organized as follows. Section 2 describes the study selection and data extraction. Section 3 covers the methodology. Sections 4 and 5 report the estimated average effects, as well as the subgroup analysis and meta-regression results. Finally, Section 6 concludes. 2. Data 2.1 Study Selection: A Snowballing Approach We identified studies using a backward snowballing method, starting with the seven most recent papers identified by experts in the field (since 2022), as shown in Appendix Figure A1, and reviewing the references cited in these papers. In the first round of snowballing, all references cited by the seven root papers were considered, resulting in 279 unique papers. Among these papers, we retained those that had any of the following key terms in the title or abstract: • Lending, loan, credit, borrow, finance, financial instruments, • SME, firm, business, entrepreneur. Following this initial—somewhat broad—filter, we performed full-text screening, including both published and unpublished impact evaluations that met the following criteria: • Intervention: Formal loans, defined as credit provided by licensed financial institutions— such as commercial banks or development finance institutions— to businesses with fewer than 250 employees. These loans are typically governed by enforceable contracts, subject to financial regulation, and accompanied by repayment terms and interest rates reflective of market or semi-market conditions. We exclude informal finance, personal loans to 4 households, and traditional microcredit programs, particularly those involving joint- liability group lending mechanisms. 1 • Outcomes: Employment, profits, or sales at the firm level. • Methods: Rigorous identification strategies with a clearly defined control group, such as RCTs, natural experiments, or quasi-experimental approaches. Full-text screening led us to exclude several related studies for various reasons, as detailed in the Online Appendix. After this process, only 13 of the 279 papers were retained for the second round. In the second round, the process was repeated, considering all 594 references cited by the 13 "new" node papers. After applying the same inclusion and exclusion criteria as before, six additional papers were selected for a third round, where we reviewed all 152 references cited by these papers, identifying only 1 more relevant study. In total, we examined 1,025 references, yielding 27 relevant papers, shown in Appendix Figure A1. We include a total of 24 studies in the meta-analysis (6 root papers and 18 papers identified through snowballing). Three node papers are literature reviews without original impact evaluation and are thus not in our final sample. Table 1 summarizes the characteristics of the 24 studies in the sample. Among them, 22 looked at employment, 2 14 examined sales, and 9 investigated profits. Thirteen studies used difference-in- differences (DiD) with a matched or synthetic control group, three conducted randomized controlled trials (RCTs), and the others relied on variation in program eligibility or timing to identify effects. Notably, 9 of the 24 papers had yet to be published as of March 6, 2025. 3 The post-intervention effects were measured over 1 to 5 years, mostly 2 to 3 years. 2.2 Data Extraction 1 Given that the definition of a firm in the microcredit literature is blurry, we exclude loans targeting households and self-employed individuals seeking to start a business. Instead, we focus exclusively on studies where the primary outcomes are at the firm level. 2 Most studies use the number of workers as a measure of employment, except Bertoni et al. (2019 and 2023) who use employment cost. 3 We include unpublished papers to increase the sample size. In Section 5, we perform subgroup analysis by journal quality, grouping unpublished papers with those in lower quality journals, and do not find a systematic difference in effects. 5 For most studies, we included only one effect, from the authors’ preferred specification. 4 Some studies evaluate multiple loan programs and report effects for each. In these cases, we include the estimate for each program, i.e., more than one estimate per study. When studies report effects for multiple periods, we choose the one that is closest to two years after loan-issuance since this is the most common time horizon. In most cases, we use the average treatment effect on the treated (ATT) or local average treatment effect (LATE). Four studies only report intention to treat (ITT) effects of formal loans. Studies often measure firm outcomes in different ways. To aggregate these metrics, prior meta- analyses have used standardized mean differences to aggregate various performance indicators into a single index (Piza et al., 2016; Grimm & Paffhausen, 2015; Kersten et al., 2017). While this method enables comparability across studies with different metrics, it can obscure the interpretation of individual outcomes. In contrast, we report impact estimates of specific outcomes –employment, sales, and profits. By presenting the results as percentage changes, we enhance interpretability and provide insights into specific dimensions of firm performance. To measure all outcomes in percentage changes, we apply the following approach. When outcomes are reported in logarithmic form or as growth rates, we interpret the estimated coefficients directly as percentage changes. For outcomes reported in levels, we convert the effect into a percentage change by dividing the estimated coefficient by the mean of the control group. If the control group mean is not available, we use the overall sample mean as an approximation. 4 If the authors do not directly highlight the preferred specification, we use the estimates that correspond to the numbers cited by the authors in the introduction. 6 Table 1: Included Studies Outcomes Effect Authors Methodology Country Intervention measured Empl. Sales Profits after Large-scale public intervention providing guarantees through the Akcigit et DiD with matched Türkiye Credit Guarantee Fund (CGF) for Yes Yes No 1-2 years al. (2024) control group loans granted by private financial institutions. Loans under the Women Entrepreneurship Development Project (WEDP), a national Alibhai et government program, for growth- RCT Ethiopia Yes Yes Yes 1-3 years al. (2022) oriented women entrepreneurs with a good psychometric score, distributed through the microfinance institution Wasasa. European Investment Bank (EIB) Staggered DiD with 28 EU Amamou et funds provided to financial matched control member Yes No No 2 years al. (2022) intermediary institutions, which group countries lend under favorable conditions. Partial credit guarantees provided Staggered DiD with by the National Guarantee Fund Arraiz et al. matched control Colombia (NGF) to firms with insufficient Yes Yes No 2 years (2014) group collateral, for accessing bank loans. Bulgaria, European Investment Fund (EIF) Czechia, Asdrubali Staggered DiD with guarantees, on behalf of the Estonia, and Signore matched control European Commission, for loans Yes Yes Yes 2 years Poland, (2015) group granted by private financial Romania, institutions. Slovenia CODEVI program loans, DiD with not-yet subsidized by the French Bach eligible sectors as France government, for small wholesale No No Yes 1-4 years (2014) control group trade businesses, provided through banks. Priority sector credit directed from Banerjee DiD with not-eligible public banks to firms with plant and Duflo firm sizes as control India No Yes Yes 2 years and machinery valued between Rs. (2014) group 6.5 million and Rs. 30 million. DiD with continuous Cartão BNDES, a program of the treatment based on development finance institution Average Bazzi et al. intermediary bank Brazil BNDES, run through intermediary Yes No No across 5 (2023) presence and credit banks, offering credit lines with years supply shocks favorable terms. 7 Italy, the EIF guarantees, on behalf of the Staggered DiD with Benelux Bertoni et European Commission, for loans matched control and the Yes Yes Yes 2 years al. (2019) granted by private financial group Nordic institutions. countries EIF guarantees, on behalf of the Staggered DiD with Bertoni et European Commission, for loans matched control France Yes Yes No 2 years al. (2023) granted by private financial group institutions. Small Business Administration DiD with continuous Brown and (SBA) guarantees to private treatment based on United Earle lenders (typically banks) through Yes No No 3 years preferred lender States (2017) two credit lines: the 7(a) and 504 presence programs. Loans from financial institutions that in turn receive loans from the Average Bruhn et al. Staggered DiD Ecuador development bank Corporación Yes Yes Yes across 2 (2025) without never-takers Financiera Nacional (CFN), years supported by World Bank funds Lending program implemented by the Alexandria Business Bryan et al. Egypt, 20 and 30 RCT Association, issuing loans four Yes Yes Yes (2024) Arab Rep. months times larger than those previously offered. A new uncollateralized loan product offered by a major Cai and commercial bank, featuring Szeidl RCT China Yes Yes Yes 3 years monthly loan officer visits for (2024) information and application assistance. Bulgaria, Credit lines provided by the EBRD Georgia, Cassano et DiD with matched to private partner banks, which Russian Yes Yes Yes 2 years al. (2013) control group then issued collateral-based and Federation, cash-flow loans to MSMEs. Ukraine Credit lines for MSMEs provided through the Global Credit Clemente et DiD with matched Argentina Program, funded by the Inter- Yes Yes No 1 year al. (2023) control group American Development Bank (IADB). Three regional public programs Da Silva et DiD with matched Brazil (FCO, FNE, FNO) providing loans Yes No No 3 years al. (2009) control group to MSMEs via public banks. Credit lines managed by BNDES De Negri et DiD with matched and FINEP to public financial Brazil Yes No No 2-3 years al. (2017) control group institutions for MSME long-term financing. 8 Credit lines by the development Most are Staggered DiD with Eslava et al. bank Bancóldex to commercial between 1 matched control Colombia Yes Yes No (2014) banks issuing loans to firms with and 4 group at least 10 employees. years DiD with synthetic Credit program managed by Grimaldi control group made BNDES offering subsidized loans 24 and Ornelas up of firms that Brazil Yes No No to firms through commercial months (2024) received non- banks. subsidized loans Funding for Growth Scheme Horvath DiD with matched (FGS) by the Hungarian Central and Lang Hungary Yes No No 2 years control group Bank to subsidize interest rates of (2021) commercial bank loans. RDD based on applicants' credit Loans from the public bank Jiménez et scores around a Spain Instituto de Crédito Official during Yes Yes No 4 years al. (2018) bank's loan approval the 2010 financial crisis. cutoff DiD with previously eligible industries as Lelarge et Guarantee scheme (SOFARIS) for control (newly France Yes No No 2 years al. (2010) commercial banks loans. eligible industries as treatment) Loan guarantees for private financial institutions provided by Oh et al. DiD with matched Korea, the Korea Credit Guarantee Fund Yes Yes No 3 years (2009) control group Rep. (KCGF) and the Korea Technology Finance Corporation (KOTEC). Notes: For studies lacking complete information, we contacted the authors directly via email. In the case of Da Silva et al. (2009), program details were obtained from Pires et al. (2017). For Bertoni et al. (2019), program information was obtained from The European Investment Fund on CIP beneficiaries. 9 3. Methodology Meta-analysis synthesizes quantitative results from multiple studies that investigate the same research question but vary in sample sizes, methodologies, and settings. By pooling information, this method helps mitigate limitations of individual studies, such as small sample sizes or inconsistent findings, and offers a more reliable and comprehensive understanding of the effect of interest. A key modeling decision in meta-analysis is how to account for treatment effect heterogeneity. Fixed-effects models assume a single true effect size across studies. Under this framework, each study-specific estimate is treated as a noisy measurement of a common effect, and differences among estimates are attributed solely to sampling variation (Rubin 1981). While appropriate in some contexts, this assumption is often unrealistic, particularly when included studies differ substantially in design, context, or sample characteristics. To address heterogeneity across studies, we adopt a random-effects model, which allows for true variation in treatment effects rather than assuming a single common effect. This approach models the observed estimates as draws from study-specific distributions, with the study-level effects themselves drawn from a common distribution. This hierarchical structure better captures contextual variation and improves the generalizability of results. As our main estimation strategy, we implement a Bayesian Hierarchical Model (BHM), following Meager (2019 and 2022), where the hierarchical model is formalized as follows: � θ 2 ∼ (θ , σ ) (1) θ ∼ (μ, τ2 ) (2) � θ 2 2 ∼ (μ, σ + τ ) (3) � Equation (1) states that the estimated effect size θ from study k is normally distributed around the true (but unobserved) study-specific effect θ , with variance equal to σ2 (i.e., the study’s own estimation error). Equation (2) indicates that θ is drawn from a normal distribution with mean μ and between-study variance τ2 . This equation thus models the heterogeneity in effects across studies. Equation (3) combines Equations (1) and (2) to integrate out θ . This equation represents 10 � the distribution of observed estimates θ directly, accounting for both the within-study error and the between-study heterogeneity. Rather than simply aggregating study-level point estimates, the Bayesian approach uses priors on the distribution of the average effect (μ) and the across-study heterogeneity (τ2 ). The BHM � updates and—when appropriate—shrinks the estimated effects (θ ) toward a common mean. This shrinkage reflects the hierarchical structure of the model and improves inference, particularly for studies with large standard errors or small sample sizes (Meager 2019; Harrer et al. 2021). We estimate the BHM using the Baggr package in R with weakly informative normal priors. We confirm that the results hold under three other prior assumptions: narrower normal priors, uniform priors, and Cauchy priors. We also cross-validate our results using a random-effects estimation via Restricted Maximum Likelihood (REML). 4. Results 4.1 Employment Figure 1 presents a forest plot showing the estimated effects of formal loans on SME employment. Each row represents a different study, with the point estimate (in percentage terms) shown as a dot and the corresponding 95% and 90% confidence intervals extending as horizontal lines. The vertical dashed line indicates that the average effect of formal loans on SME employment across studies is 12%, with a 95% confidence interval of [8.1%, 16.3%]. While the estimates for all studies are positive, the magnitude and precision of estimates vary substantially, reflecting differences in sample size and contextual factors. The wide confidence intervals in some studies underscore the importance of synthesizing results across settings to obtain a more precise estimate of the average treatment effect. 11 Figure 1: Estimated Effects of Formal Loans on SME Employment Note: This forest plot displays the estimated effects of formal loans on SME employment—expressed as percentage changes—across individual studies. The dashed vertical line represents the average effect, estimated using a BHM with normally distributed priors (0, 1002 ) for both the average effect (μ) and the across-study heterogeneity (τ2 ). The point estimates for individual studies reflect the posterior means obtained from the BHM, rather than the original point estimates reported in each study. The horizontal lines indicate the 95% confidence intervals. EAP = East Asia & Pacific, ECA = Europe & Central Asia, LAC = Latin America & Caribbean, MENA = Middle East & North Africa, NAC = North America, SSA = Sub-Saharan Africa, SAS = South Asia. To assess the robustness of the average effect of formal loans on SME employment, we estimate the model using both an REML approach and a BHM under three alternative specifications. As shown in Appendix Figure A2, the estimated effect remains stable across specifications, suggesting that our findings are not sensitive to the choice of estimation method or prior assumptions. Next, we examine the predictive distribution, which models the expected effect size in a new, yet- to-be-observed study. This distribution combines the posterior estimate of the average effect with the estimated heterogeneity across studies, typically resulting in a wider interval than the posterior alone. The average expected effect of formal loans on SME employment in a new study is 12.1% (Figure 2), closely aligned with the average effect estimated from existing studies (12%, shown in 12 Figure 1). However, the 95% predictive interval is broader—ranging from -6.3% to 29.6%—and reflects greater uncertainty about the outcomes of future interventions. Despite this wider range, the probability of observing a negative effect is relatively low (8.6%), compared to a 91.4% probability of a positive effect. Figure 2: Predictive Distribution of the Effect of Formal Loans on SME Employment Note: This figure displays the predictive distribution of the effect of formal loans on SME employment, estimated using a BHM with normally distributed priors (0, 1002 ) for both the average effect (μ) and the across-study heterogeneity (τ2 ). The dashed line represents the mean expected effect. The shaded areas show the probability of a negative effect (dark) and a positive effect (light). 4.2 Sales Figure 3 shows a forest plot of the estimated percentage changes in SME sales resulting from formal loans. As with employment, there is substantial variation in effect sizes across studies, with some interventions showing very large sales impacts of up to 71% (e.g., Akcigit et al., 2024) and others yielding more modest results. Two studies have null effects (Alibhai et al. (2022) and Bryan, Karlan, and Osman 2024). The dashed vertical line indicates the average effect of formal loans on SME sales across all studies, estimated at 18.3%, with a 95% confidence interval of [7.2%, 29.5%]. 13 The estimated average effect of formal loans on SME sales is robust across alternative estimation approaches. As shown in Appendix Figure A3, we obtain similar results estimating the random- effects model via REML, as well as relying on a BHM under three alternative prior distributions. Across all specifications, the average estimated effect remains close to 18.3%, with overlapping confidence intervals, reinforcing the validity of our findings. Figure 3 shows the expected size effect of formal loans on SME sales in a new, future study. The mean expected effect is 18.4%, but the 95% predictive interval is wide—ranging from –24.9% to 60.2%—indicating considerable uncertainty about the likely effect in new contexts. Notably, the probability of observing a negative effect is 17.6%, compared to an 82.4% probability of a positive effect. Figure 3: Estimated Effects of Formal Loans on SME Sales Note: This forest plot displays the estimated effects of formal loans on SME sales—expressed as percentage changes— across individual studies. The dashed vertical line represents the average effect, estimated using a BHM with normally distributed priors (0, 1002 ) for both the average effect (μ) and the across-study heterogeneity (τ2 ). The point estimates for individual studies reflect the posterior means obtained from the BHM, rather than the original point estimates reported in each study. The horizontal lines indicate the 95% confidence intervals. EAP = East Asia & Pacific, ECA = Europe & Central Asia, LAC = Latin America & Caribbean, MENA = Middle East & North Africa, NAC = North America, SSA = Sub-Saharan Africa, SAS = South Asia. 14 Figure 4: Predictive Distribution of the Effect of Formal Loans on SMEs Sales Note: his figure displays the predictive distribution of the effect of formal loans on SME sales, estimated using a BHM with normally distributed priors (0, 1002 ) for both the average effect (μ) and the across-study heterogeneity (τ2 ). The dashed line represents the mean expected effect. The shaded areas show the probability of a negative effect (dark) and a positive effect (light). 4.3 Profits Like the results for employment and sales, we find a positive average effect of formal loans on SME profits (Figure 5). The effect is statistically significant at the 95% confidence level, and the magnitude—a 17.6% increase in profits—is notably larger than the less than 8% profit effect of microcredit interventions reported in Meager (2019). Robustness checks, shown in Appendix Figure A4, indicate that the estimated effect is relatively stable across alternative prior specifications. Among all three outcomes, the profit estimates have the widest 95% confidence interval, ranging from 2.7% to 40.1%. The likely reasons for this are the smaller number of studies examining profit outcomes, compared to employment and sales, as well as the noise in profit data. While the estimated effect is positive in all studies, only four in ten have statistically significant results. 15 Figure 5: Estimated Effects of Formal Loans on SME Profits Note: This forest plot displays the estimated effects of formal loans on SME profits—expressed as percentage changes—across individual studies. The dashed vertical line represents the average effect, estimated using a BHM with normally distributed priors (0, 1002 ) for both the average effect (μ) and the across-study heterogeneity (τ2 ). The point estimates for individual studies reflect the posterior means obtained from the BHM, rather than the original point estimates reported in each study. The horizontal lines indicate the 95% confidence intervals. EAP = East Asia & Pacific, ECA = Europe & Central Asia, LAC = Latin America & Caribbean, MENA = Middle East & North Africa, NAC = North America, SSA = Sub-Saharan Africa, SAS = South Asia. The predictive distribution in Figure 6 reflects both the variability of estimated effects and the limited number of studies examining profits. The average expected effect of formal loans on SME profits in a new study is 17.6%, but the 95% confidence interval spans a wide range—from -35.4 to 77.2% —reflecting substantial uncertainty. The probability of a negative effect of formal loans on SME profits in a new study is 17%, lower than the corresponding probability reported for microcredit programs. Meager (2019) finds a 25% chance that microcredit leads to a negative impact on profits. 16 Figure 6: Predictive Distribution of the Effect of Formal Loans on SMEs Profits Note: This figure displays the predictive distribution of the effect of formal loans on SME profits, estimated using a BHM with normally distributed priors (0, 1002 ) for both the average effect (μ) and the across-study heterogeneity (τ2 ). The dashed line represents the mean expected effect. The shaded areas show the probability of a negative effect (dark) and a positive effect (light). 5. Effect Heterogeneity Subgroup analysis can help explain why some studies yield smaller or larger effects than others and explore how treatment effects vary across different contexts. However, the reliability of subgroup findings depends heavily on statistical power, which diminishes when the number of studies is limited. Methodological guidelines recommend a minimum of 10 studies overall (Schwarzer, Carpenter, and Rücker, 2015), with at least 5 studies per subgroup to ensure credible comparisons (Borenstein et al., 2009, p. 163). Given the limited number of studies available for sales and profit outcomes, we restrict our subgroup analysis to the employment effects of formal loans, where the sample size is sufficient to support meaningful comparisons. 17 5.1 Effect Heterogeneity by Study Context We conduct subgroup analysis along four key dimensions of study context: (i) guaranteed versus non-guaranteed loans, (ii) loans originated by public versus private institutions, (iii) borrower firm size, and (iv) country income group. For each dimension, we split the sample accordingly and estimate separate average effects, as shown in Figure 7. In Panel A, we examine whether the effect of formal loans on SME employment differs between guaranteed and non-guaranteed loans. We find the same effect size of about 12% for both groups. These results suggest that the positive employment effects of formal loans are present regardless of whether the loan is backed by a guarantee. 5 Panel B examines the type of financial intermediary, revealing a notably larger effect for loans issued by public financial institutions (29%) compared to private institutions (10.8%). However, the confidence interval around the public institution estimate is much wider. Panel C explores heterogeneity across firm size. We find a larger effect for firms with 10 or fewer employees (16.4%) compared to those with more than 10 employees (10%). Despite the difference in point estimates, the confidence intervals overlap, suggesting the difference is not statistically significant. In Panel D, we compare country income levels, where effects are similar between high-income countries (13.3%) and developing countries (10.8%), again with overlapping intervals. 5 We conduct a related subgroup analysis comparing loans financed with public versus private funds, which can also reflect differences in risk-sharing structures. Consistent with the loan guarantee results, we find broadly similar effects across funding sources (Appendix Figure A5). 18 Figure 7: Subgroup Analysis on SME Employment Notes: This figure presents subgroup analyses of the average effect of formal loans on SME employment across key study-level covariates. In Panel A, the sample includes 17 studies on non-guaranteed loans and 9 on guaranteed loans. In Panel B, there are 20 studies involving private financial institutions and 6 involving public financial institutions. By firm size (Panel C), 19 studies examine firms with more than 10 employees and 7 focus on firms with 10 or fewer employees. By country income level (Panel D), the analysis includes 12 studies in high-income countries and 14 from developing countries. To further investigate how effects vary across subgroups, we estimate a study-level meta-analysis regression model, incorporating indicator variables that correspond to the subgroup dimensions shown in Figure 7. This approach allows us to assess how much of the between-study variation in effect sizes can be attributed to observed covariates (study-level predictors), rather than treating 19 all heterogeneity as random. In this framework, instead of assuming a constant global mean effect (), each study-specific effect is modeled as a random draw conditional on its covariates . Formally, equations (2) and (3) are transformed into equations (4) and (5) as follows: θ ∼ ( β, τ2 ) (4) � θ 2 2 ∼ ( β, σ + τ ) (5) Here, is a vector of study-level covariates (e.g., the type of loan, firm size, country income level), β captures the marginal effect of each covariate on the conditional mean of θ , and τ2 represents the residual between-study variance not explained by the covariates. In this Bayesian specification, we also impose a prior distribution on β to incorporate uncertainty in the estimation. Since contains only categorical variables, the coefficient β represents the average difference in effect sizes between a given category and the reference group. The estimating equation for the meta-regression, shown in Equation 6, is conceptually like an OLS regression, but accounts for two sources of residual variation: the within-study sampling error ( ) and the between-study heterogeneity ( ). These components reflect the uncertainty of individual study estimates and the unexplained variability across studies, respectively. � θ = + + (6) The estimated coefficients from the meta-regression are displayed in Table 2. Column 1 reports indicator variables for loan type, lender type, and country income level as potential effect moderators. Column 2 adds an indicator for firm size category, although this reduces the sample size due to missing data for five estimates. Consistent with the patterns observed in Figure 7, the results in Column 1 indicate that the effect of formal loans on SME employment does not vary significantly with loan guarantees, lender type, or country income level. In contrast, Column 2 shows that when controlling for all other subgroup variables, firm size emerges as a significant moderator. Specifically, the estimated effect of formal loans is 4.8 percentage points higher for firms with 10 or fewer employees, compared to firms with more than 10 employees. This implies that the total estimated effect for smaller firms is 17.4% (12.6% + 4.8%), which is almost one-and-a-half times as big as the effect for larger firms (12.6%). 20 Table 2: Meta Regression Dependent variable: Estimated effect of � formal loans on SME employment (θ ) (1) (2) (3) (4) Average effect for the reference group 10.97*** 13.01*** 10.96*** 11.71*** (2.19) (2.84) (3.18) (3.86) Lender type: Public financial institutions 8.87 11.45* 12.24** 12.95* (6.15) (6.52) (6.22) (6.77) Income group: Developing 5.15 4.03 3.72 (4.32) (4.33) (4.76) Firm size: 10 or fewer employees 5.91 5.65 (4.72) (4.92) Loan type: Nonguaranteed loan 1.66 (4.49) Observations 26 26 26 26 Notes: All regressions are estimated using a BHM with normally distributed priors (0, 1002 ) for both the average effect (μ) and the across-study heterogeneity (τ2 ), and normally distributed priors (0, 202 ) for the regression coefficients . Standard errors are reported in parentheses. ***, **, * denote significance at the 99, 95, and 90 percent levels, respectively. Our results suggest that the employment effects of credit programs for SMEs are broadly consistent across guaranteed versus non-guaranteed loans, across countries with different income levels, and for firms of different sizes. But loans originated by public financial institutions have larger effects on employment than loans from private institutions. Loans increase SME employment by about 12% when issued by private banks, and by about 25% (12% + 13%) when coming from public banks. A caveat here is that five of the six estimates for public institutions come from Brazil. 6 The other estimate is for Spain. The only other paper on public banks in our sample—Banerjee and Duflo (2014) on firms in India—does not examine effects on employment. However, when looking at sales (Figure 3) and profits (Figure 5), the effects in India are also on the higher side. Public banks may not function well in all settings, particularly those with low capacity, since they are subject to inefficiencies and political capture, which can be mitigated by good governance (World Bank 2013). But SME lending by private banks also has limitations. For example, Ornelas et al. (2019) study earmarked credit in Brazil, where the government subsidizes funding costs for 6 Another paper on public banking in Brazil (Da Mata and Resende 2020) finds positive effects in the agriculture sector of the semiarid Northeast region. Low-income producers who became eligible for subsidized loans increased their climate resilient livestock. The credit expansion increased the income of poor households. 21 lenders, who then allocate credit to firms at regulated interest rates. The program had limited additionality as larger firms benefited most from it. For smaller firms, private banks used a “cross- selling” strategy offsetting lower profits from government-subsidized loans by increasing interest rates on free-market loans. In this case, credit did not reach the most constrained borrowers since banks’ profit maximizing incentives were not aligned with the goals of channeling funds to smaller and riskier borrowers. Bruhn et al. (2025) also find that, in a second-tier lending program in Ecuador, private banks were more likely to allocate loans to SMEs with prior credit access, although the program’s positive effects on job creation and sales were concentrated among firms without prior credit access. In line with the examples, our finding that the effect of loans on SME employment is larger for public banks suggests that loans channeled through private banks may not reach the most credit- constrained businesses. Profit-maximizing banks tend to prioritize less risky firms with existing credit histories, larger operations, or the ability to cross-subsidize other banking products, leading to less additionality and lower impact on firm performance. Our findings do not imply that loans should always be channeled through public banks. Instead, they highlight that there is a tradeoff between the risks of public banks (inefficiencies and political capture) and lower impacts of private banks due to mis-aligned incentives. Private banks maximize profits instead of SME growth and may thus avoid risky investments. Governments can lower the risk of lending to SMEs by strengthening credit infrastructure, such as credit bureaus and collateral registries. Regulators can also foster competition in the banking sector, which may provide incentives for banks to improve their technology for lending to SMEs. Finally, we do not have data on loan default and cannot say how much greater losses of public banks are relative to private banks, which is another element to consider in the tradeoff between lending through public versus private banks. 5.2 Effect Heterogeneity by Journal Rank and Methodology Our meta-analysis includes unpublished papers, which helps mitigate concerns about publication bias but may also introduce quality concerns for studies that have remained unpublished over time. To assess whether our results are influenced by differences in study quality, we conduct subgroup analyses based on journal rank (Appendix Figure A6). Specifically, we distinguish between papers published in higher-ranked journals and those in lower-ranked journals or still unpublished, using 22 two different approaches. The first approach divides the sample into studies from journals above and below the median Scimago Journal Rank (SJR5Y). As shown in Figure A6, Panel A, both groups have the same effect size of 12.5%. The confidence intervals are relatively wider for studies below the median, reflecting greater variability in effect estimates. Our second approach uses a stricter threshold, comparing studies published in journals with an SJR5Y above 1,500 versus those below 1,500. As shown in Panel B, we observe slightly more divergence: studies published in higher-ranked journals (SJR5Y > 1,500) show a somewhat lower average effect size (11.1%) compared to studies published in lower-ranked journals or unpublished papers (14%). However, the confidence intervals overlap substantially across both groups. In addition, we cover a range of methodologies, some of which may be more rigorous than others. While RCTs are considered the gold standard for causal inference, they remain relatively recent innovations in many fields, including SME finance, and their application is often constrained by the feasibility of experimental designs in real-world settings. As a result, we include both experimental and non-experimental impact evaluations in our analysis. Many of the papers use matching to create a control group, which may raise the concern that they do not rely on exogenous variation. We thus examine how the effects vary across papers that create control groups based on exogenous variation versus other methods. Appendix Figure A7 shows that the effect of formal loans on SME employment is positive and statistically significant in both groups. The point estimate is larger for studies with exogenous variation in program eligibility, but the confidence intervals overlap. Overall, these patterns suggest that our main findings are not systematically driven by lower- quality or unpublished studies, or by less rigorous methodologies. Effect sizes are broadly consistent across journal ranks and methodologies, providing reassurance about the robustness of our results. 6. Conclusions This study synthesizes evidence on the impact of formal loans on SME employment, sales, and profits using a BHM meta-analysis. We identified relevant studies through a snowballing approach, yielding 22 papers that look at employment, 14 at sales, and 9 at profits. The meta-analysis shows 23 that formal loans increase SME employment by 12%, sales by 18.3%, and profits by 17.6%. The 95% confidence intervals for all three outcomes exclude zero. We calculate predictive distributions for the effects expected in future studies, which combine the average estimates with information on the variation across studies. These predictions show that the probability of a new study finding a positive effect of formal loans is 91.4% for employment, 82.4% for sales and 83% for profits. That is, the variation across studies implies that there is an up to 17% probability that future studies could find a negative effect of formal loans on SME performance. For employment, we have enough studies to conduct subgroup analysis. The effect of formal loans on SME employment does not vary across guaranteed versus non-guaranteed loans, developing and high-income countries, or firm size. However, loans originated by public lenders have a larger effect on SME employment than loans from private banks. This finding suggests that profit- maximizing banks may not allocate funds to the most credit-constrained SMEs that have the highest returns to capital. In our analysis, most estimates for the impact of public bank loans come from Brazil. Future research can shed light on whether the effects of public loans on SME growth are similarly high in other countries, particularly since public banks may be subject to political capture and inefficiencies in low-capacity environments. Future projects and studies can also explore ways to align private banks’ incentives with the public goal of maximizing impact on SME growth, such as providing performance-based incentives. 24 Appendix Figures Figure A1: Top-down paths between root papers and papers in the meta-analysis Notes: This figure shows top-down relationships between root papers and those in the meta-analysis from the backward snowballing process. Down-to-top and horizontal relations are omitted for clarity. Flagged boxes indicate papers included in the meta-analysis. The non-flagged papers are literature reviews without original impact evaluation and are thus not in our final sample. 25 Figure A2: Average Effect of Formal Loans on SME Employment (Alternative Estimation Approaches) Notes: This figure reports the estimated average effect of formal loans on SME employment under alternative estimation strategies. The REML estimate corresponds to the random-effects model, with confidence intervals calculated using the Hartung-Knapp method. The Uniform model uses uniformly distributed priors— U(−100,100) for the average effect () and U(0,100) for the across-study heterogeneity ( 2 ). The Normal model uses normally distributed priors—(0, 102 ) for the average effect () and (0, 1002 ) for the across-study heterogeneity ( 2 ). The Cauchy model uses Cauchy distributions ℎ(0,25) for both the average effect () and the across-study heterogeneity ( 2 ), placing more weight on the likelihood of extreme values. The Base scenario uses normally distributed priors (0, 1002 ) for both the average effect () and the across-study heterogeneity ( 2 ), representing a weakly informative specification. 26 Figure A3: Average Effect of Formal Loans on SME Sales (Alternative Estimation Approaches) Notes: This figure reports the estimated average effect of formal loans on SME employment under alternative estimation strategies. The REML estimate corresponds to the random-effects model, with confidence intervals calculated using the Hartung-Knapp method. The Uniform model uses uniformly distributed priors— U(−100,100) for the average effect () and U(0,100) for the across-study heterogeneity ( 2 ). The Normal model uses normally distributed priors—(0, 402 ) for the average effect () and (0, 1002 ) for the across-study heterogeneity ( 2 ). The Cauchy model uses Cauchy distributions ℎ(0,25) for both the average effect () and the across-study heterogeneity ( 2 ), placing more weight on the likelihood of extreme values. The Base scenario uses normally distributed priors (0, 1002 ) for both the average effect () and the across-study heterogeneity ( 2 ), representing a weakly informative specification. 27 Figure A4: Average Effect of Formal Loans on SME Profits (Alternative Estimation Approaches) Notes: This figure reports the estimated average effect of formal loans on SME employment under alternative estimation strategies. The REML estimate corresponds to the random-effects model, with confidence intervals calculated using the Hartung-Knapp method. The Uniform model uses uniformly distributed priors— U(−100,100) for the average effect () and U(0,100) for the across-study heterogeneity ( 2 ). The Normal model uses normally distributed priors—(0, 402 ) for the average effect () and (0, 1002 ) for the across-study heterogeneity ( 2 ). The Cauchy model uses Cauchy distributions ℎ(0,25) for both the average effect () and the across-study heterogeneity ( 2 ), placing more weight on the likelihood of extreme values. The Base scenario uses normally distributed priors (0, 1002 ) for both the average effect () and the across-study heterogeneity ( 2 ), representing a weakly informative specification. 28 Figure A5: Subgroup Analysis of SME Employment Effects by Loan Funding Source Notes: This figure presents a subgroup analysis of the effect of formal loans on SME employment by the loan funding source (i.e., public vs. private funding). The sample includes 17 studies analyzing loans financed with public funds and 9 studies examining loan financed with private funds. Figure A6: Heterogenous Effects by Journal Rank Note: Panel A shows a subgroup analysis comparing studies published in journals above and below the median Scimago SJR5Y score (566). The SJR5Y is a cumulative impact score, weighted by the prestige of citing journals, over the last five years. Higher scores indicate greater journal influence. For working papers and non-ranked papers, we assign an SJR5Y value of 0. The sample includes 11 studies above the median and 15 below the median (9 unpublished and 4 non-ranked, 2 ranked below the threshold). Panel B presents a similar analysis using a threshold of 1,500 in the Scimago SJR5Y. The distribution of studies across this threshold is as follows: 10 studies with SJR5Y scores above 1,500 and 16 studies below (including 9 unpublished, 3 ranked, and 4 non-ranked studies). 29 Figure A7: Heterogenous Effects by Methodology Notes: This figure presents a subgroup analysis of the effect of formal loans on SME employment by methodology (i.e., papers with exogenous variation in eligibility and other methodologies). The sample includes 9 studies with exogenous variation and 17 studies with other methodologies (i.e., using a matched control group). 30 References Aivazian, V. A., & Santor, E. (2008). Financial constraints and investment: assessing the impact of a World Bank credit program on small and medium enterprises in Sri Lanka. Canadian Journal of Economics. doi:https://doi.org/10.1111/j.1540-5982.2008.00471.x Akcigit, U., Seven, Ü., Yarba, İ., & Yılmaz, F. (2024). Firm-level impact of public credit guarantees. European Economic Review, 170, 104861. https://doi.org/10.1016/j.euroecorev.2024.104861. Alibhai, A. S., Cassidy, R. M., Goldstein, M. P., & Papineni, S. (2022). Evening the credit score? Impact of psychometric loan appraisal for women entrepreneurs. Policy Research Working Paper Series, 10230. The World Bank. Amamou, R., Gereben, Á., & Wolski, M. (2023). Assessing the impact of the EIB’s intermediated lending to SMEs during funding shocks. Small Business Economics, 3. Arráiz, I., Meléndez, M., & Stucchi, R. (2014). Partial credit guarantees and firm performance: Evidence from Colombia. Small Business Economics, 43(3), 711-724. Asdrubali, P., & Signore, S. (2015). The economic impact of EU guarantees on credit to SMEs – Evidence from CESEE countries. EIF Working Paper Series, 2015/29. European Investment Fund (EIF). Bah, E.-h., Brada, J. C., & Yigit, T. (2011). With a little help from our friends: The effect of USAID assistance. Journal of Comparative Economics, 202-220. Bach, L. (2014). Are small businesses worthy of financial aid? Evidence from a French targeted credit program. Review of Finance, 18(3), 877–919. https://doi.org/10.1093/rof/rft022. Banerjee, A. V., & Duflo, E. (2014). Do firms want to borrow more? Testing credit constraints using a directed lending program. The Review of Economic Studies, 81(2), 572-607. Bartik, A. W., Cullen, Z. B., Glaeser, E. L., Luca, M., Stanton, C. T., & Sunderam, A. (2020). The Targeting and Impact of Paycheck Protection Program Loans to Small Businesses. NBER Working Papers 26989, National Bureau of Economic Research, Inc. Bazzi, S., Muendler, M.-A., Oliveira, R. F., & Rauch, J. E. (2023). Credit supply shocks and firm dynamics: Evidence from Brazil. NBER Working Paper No. w31721. Available at SSRN: https://ssrn.com/abstract=4582337. Beck, T., & Demirgüç-Kunt, A. (2006). Small and Medium-Size Enterprises: Access to Finance as a Growth Constraint. Journal of Banking and Finance, 30, 2931–2943. Beck, T., Demirgüç-Kunt, A., Laeven, L., & Vojislav, M. (2006). The Determinants of Financing Obstacles. Journal of International Money and Finance, 25(6): 932–52. Benavente, J. M., Crespi, G., & Maffioli, A. (2007). Public Support to Firm-Level Innovation: An Evaluation of the FONTEC Program. OVE Working Papers 0507, Inter-American Development Bank, Office of Evaluation and Oversight (OVE). Bentolila, S., Jansen, M., & Jiménez, G. (2018). When Credit Dries Up: Job Losses in the Great Recession. Journal of the European Economic Association, 16(3), 650-695. doi:https://doi.org/10.1093/jeea/jvx021 31 Bertoni, F., Brault, J., Colombo, M. G., Quas, A., & Signore, S. (2019). Econometric study on the impact of EU loan guarantee financial instruments on growth and jobs of SMEs. EIF Working Paper Series, 2019/54. European Investment Fund (EIF). Bertoni, F., Colombo, M. G., & Quas, A. (2023). The long-term effects of loan guarantees on SME performance. Journal of Corporate Finance, 80, 102408. https://doi.org/10.1016/j.jcorpfin.2023.102408. Binelli, C., & Maffioli, A. (2007). A Micro‐econometric Analysis of Public Support to Private R&D in Argentina. International Review of Applied Economics, 339-359. Blasio, G. d., Mitri, S. D., D’Ignazio, A., Russo, P. F., & Stoppani, L. (2018). Public guarantees to SME borrowing. A RDD evaluation. Journal of Banking and Finance, 73-86. Borenstein, M., Hedges, L. V., Higgins, J. P. T., & Rothstein, H. R. (2009). Introduction to meta-analysis. John Wiley & Sons. https://doi.org/10.1002/9780470743386 Brault, J., & Signore, S. (2019). The real effects of EU loan guarantee schemes for SMEs: A pan- European assessment. Brown, J. D., & Earle, J. S. (2017). Finance and growth at the firm level: Evidence from SBA loans. Journal of Finance, 72(3), 1039-1080. Brown, J. D., Earle, J. S., & Morgulis, Y. (2015). Job Creation, Small vs. Large vs. Young, and the SBA. National Bureau of Economic Research Working Paper Series. Retrieved from http://www.nber.org/papers/w21733 Bruhn, M., Diaz Kalan, F., Fraccaroli N., Ruiz Ortega C. (2025). Access to Finance for MSMEs in Ecuador: A Firm-Level Impact Evaluation. World Bank Policy Research Working Paper No. 11121. Bryan, G., Karlan, D., & Osman, A. (2024). Big loans to small businesses: Predicting winners and losers in an entrepreneurial lending experiment. American Economic Review, 114(9), 2825– 2860. https://doi.org/10.1257/aer.20220616. Cai, J., & Szeidl, A. (2024). Indirect effects of access to finance. American Economic Review, 114(8), 2308–2351. https://doi.org/10.1257/aer.20220711. Carlin, B. P., & Louis, T. A. (2008). Bayesian Methods for Data Analysis. CRC Press. Retrieved from https://books.google.com/books?id=GTJUt8fcFx8C&lpg=PP1&hl=es&pg=PR4#v=onepage&q&f=f alse Cassano, F., Jõeveer, K., & Svejnar, J. (2013). Cash flow vs. collateral-based credit: Performance of micro, small, and medium-sized firms in transition economies. Economics of Transition, 21(2), 193– 380. https://doi.org/10.1111/ecot.12016. Clemente, A., González Rozada, M., Támola, A., & Fernández Díez, M. C. (2023). The Impact of Credit Programs on the Post-Pandemic Reactivation of the Fabric of Production and Employment: The Case of Argentina. https://doi.org/10.18235/0005122 Chodorow-Reich, G. (2014). The Employment Effects of Credit Market Disruptions: Firm-level Evidence from the 2008–9 Financial Crisis. The Quarterly Journal of Economics, 129, 1–59. 32 Chudnovsky, D., López, A., Rossi, M., & Ubfal, D. (2006). Evaluating A Program of Public Funding of Private Innovation Activities. An Econometric Study of FONTAR in Argentina. OVE Working Papers 1606, Inter-American Development Bank, Office of Evaluation and Oversight (OVE). Crépon, B., El Komi, M., & Osman, A. (2024). Is it who you are or what you get? Comparing the impacts of loans and grants for microenterprise development. American Economic Journal: Applied Economics, 16(1), 286–313. https://doi.org/10.1257/app.20210683. Da Mata, D., & Resende, G. 2020. “Changing the climate for banking: the economic effects of credit in a climate-vulnerable area.” Journal of Development Economics, 146, p. 102459. Da Silva, A. M. A., Resende, G. M., & Silveira Neto, R. M. (2009). Eficácia do gasto público: uma avaliação do FNE, FNO e FCO. Estudos Econômicos, 39(1). https://doi.org/10.1590/S0101- 41612009000100004. Eslava, M., Maffioli, A., & Meléndez, M. (2014). Credit constraints and business performance: Evidence from public lending in Colombia. Documento CEDE No. 2014-37. Available at SSRN: http://dx.doi.org/10.2139/ssrn.2516517. Gelman, A. (2006). Prior distributions for variance parameters in hierarchical models. Bayesian Analysis, 1(3), 515–534. Grimm, M., & Paffhausen, A. L. (2015). Do interventions targeted at micro-entrepreneurs and small and medium-sized firms create jobs? A systematic review of the evidence for low and middle income countries. Labour Economics, 67-85. doi:https://doi.org/10.1016/j.labeco.2015.01.003 Grimaldi, D., & Ornelas, J. R. H. (2024). Do firms need cheaper credit to grow? Investigating the effectiveness of subsidized earmarked loans. Working Papers Series, 599. Central Bank of Brazil, Research Department. Haas, R. D., & González-Uribe, J. (2024). Public Policies for Private Finance. EBRD Working Paper No. 297. doi:http://dx.doi.org/10.2139/ssrn.4989844 Harrer, M., Cuijpers, P., Furukawa, T. A., & Ebert, D. D. (2021). Doing Meta-Analysis in R: A Hands-on Guide. Chapman & Hall/CRC. Available at: https://bookdown.org/MathiasHarrer/Doing_Meta_Analysis_in_R/ Higgins, J. J. (2019). Cochrane Handbook for Systematic Reviews of Interventions. Cochrane. Retrieved from www.training.cochrane.org/handbook Horvath, A., & Lang, P. (2021). Do loan subsidies boost the real activity of small firms? Journal of Banking & Finance, 122, 105988. https://doi.org/10.1016/j.jbankfin.2020.105988. Huber, K. (2018). Disentangling the Effects of a Banking Crisis: Evidence from German Firms and Counties. American Economic Review, 108(3): 868–898. ILO. (2022). Guide to Recommendation 189: Job creation in small and medium sized enterprises Recommendation. International Labour Organization (ILO). Jiménez, G., Peydró, J., Repullo, R., & Saurina, J. (2018). Burning money? Government lending in a credit crunch (CEPR Discussion Paper No. 13267). CEPR Press. https://cepr.org/publications/dp13267 33 Kapoor, M., Ranjan, P., & Raychaudhuri, J. (2012). The impact of credit. doi: http://dx.doi.org/10.2139/ssrn.2023068 Kersten, R., Harms, J., Liket, K., & Maos, K. (2017). Small Firms, large Impact? A systematic review of the SME Finance Literature. World Development, 330-348. Kuntchev, V. R.-M. (2014). What Have We Learned from the Enterprise Surveys Regarding Access to Credit by SMEs? World Bank Policy Research Working Paper No. 6670. Lelarge, C., Sraer, D., & Thesmar, D. (2010). Entrepreneurship and credit constraints: Evidence from a French loan guarantee program. En J. Lerner & A. Schoar (Eds.), International differences in entrepreneurship (pp. 243-273). University of Chicago Press. http://www.nber.org/chapters/c8218 Lopez-Acevedo, G., & Tinajero, M. (2016). Mexico: Impact Evaluation of SME Programs Using Panel Firm Data. World Bank Policy Research Working Paper No. 5186. Maffioli, A., Negri, J. A., Rodriguez, C. M., & Vázquez-Bare, G. (2017). Public credit programmes and firm performance in Brazil. Development Policy Review, 35(5), 675-702. Martín-García, R., & Santor, J. M. (2021). Public guarantees: a countercyclical instrument for SME growth. Evidence from the Spanish Region of Madrid. Small Business Economics, 427-449. Meager, R. (2019). Understanding the Average Impact of Microcredit Expansions: A Bayesian Hierarchical Analysis of Seven Randomized Experiments. American Economic Journal: Applied Economics, 11(1): 57–91. doi:10.1257/app.20170299 Meager, R. (2022). Aggregating Distributional Treatment Effects: A Bayesian Hierarchical Analysis of the Microcredit Literature. American Economic Review, 112(6): 1818–1847. doi:https://doi.org/10.1257/aer.20181811 OECD. (2024). Financing SMEs and Entrepreneurs 2024: An OECD Scoreboard. Paris: OECD Publishing. Retrieved from https://doi.org/10.1787/fa521246-en Oh, I., Lee, J. D., Heshmati, A., et al. (2009). Evaluation of credit guarantee policy using propensity score matching. Small Business Economics, 33(3), 335-351. https://doi.org/10.1007/s11187-008-9102-5 Ornelas, Jose Renato Haas, Alvaro Pedraza, Claudia Ruiz-Ortega, and Thiago Christiano Silva. 2024. “Market Power and the Transmission of Loan Subsidies.” The Review of Corporate Finance Studies 13 (4): 931–965. Pires, Jose Claudio Linhares, and Nathaniel Russell. (2017). Assessing Firm-Support Programs in Brazil. https://doi.org/10.18235/0010693 Piza, C., Cravo, T. A., Taylor, L., Gonzalez, L., Musse, I., Furtado, I., . . . Abdelnour, S. (2016). The Impact of Business Support Services for Small and Medium Enterprises on Firm Performance in Low- and Middle-Income Countries: A Systematic Review. Campbell Systematic Reviews. Retrieved from https://doi.org/10.4073/csr.2016.1 Ru, H. (2017). Government Credit, a Double-Edged Sword: Evidence from the China Development Bank. Journal of Finance, 275-316. doi:https://doi.org/10.1111/jofi.12585 Rubin, D. B. (1981). Estimation in Parallel Randomized Experiments. Journal of Educational and Behavioral Statistics. doi:https://doi.org/10.3102/10769986006004377 34 Sanguinetti, P. (2004). Determinants of Technology and Innovation Activities in Argentina: Evidence from a firm level survey. Econometric Society 2004 Latin American Meetings 202, Econometric Society. Singh, S, Niazi, Z.A., Gottschalk, M., Kotei, M.A., and Wimpey, J.S. 2025. MSME Finance Gap. International Finance Corporation. World Bank Group. Stucchi, R., Rojo, S., Maffioli, A., & Castillo, V. (2014). Knowledge Spillovers of Innovation Policy through Labor Mobility: An Impact Evaluation of the FONTAR Program in Argentina. IDB Publications (Working Papers) 4779, Inter-American Development Bank. Schwarzer, G., Carpenter, J. R., & Rücker, G. (2015). Meta-analysis with R. Springer. https://doi.org/10.1007/978-3-319-21416-0 Veselin, K., Ramalho, R., & Rodríguez-M, J. (2014). What Have We Learned from the Enterprise Surveys Regarding Access to Credit by SMEs? World Bank Policy Research Working Paper No. 6670. World Bank, 2013. Global financial development report 2013: Rethinking the Role of the State in Finance. World Bank Publications. Zia, B. H. (2008). Export incentives, financial constraints, and the (mis)allocation. Journal of Financial Economics, 498–527. 35 Online Appendix: Rationale Behind Exclusions As described in Section 2, we reviewed the text of papers identified via snowballing and determined that several of these did not meet the criteria for inclusion in the meta-analysis, for the following reasons. • Syndicated loans: Other papers (e.g., Chodorow-Reich, 2014) study syndicated loans, which typically involve large amounts of capital, complex structures, and are extended to large firms. • Negative credit shocks: Papers like Huber (2018) and Bentolila, Jansen, & Jiménez (2018) analyze the adverse effects of negative credit supply shocks on firm employment, providing suggestive evidence of the importance of access to credit for employment. However, we drop these papers from our sample as we cannot assume that positive credit shocks would have effects of the same magnitude in the opposite direction. Similarly, Zia, (2008) examines changes to Pakistan's Export Finance Scheme and finds that removing a subsidy led to a 29% decline in firm exports, particularly affecting privately owned firms. This study was excluded because it assessed a negative credit supply shock rather than an expansion of formal credit. • State-owned enterprises: We exclude studies focusing on loans to state-owned enterprises (SOEs), such as Ru (2017), as these firms are not private SMEs and may respond differently to financial interventions. For example, Ru (2017) finds that government credit enhanced SOE performance but crowded out private firms operating in the same industries. • Unit-of-analysis problem: Some papers evaluate the same intervention on the same outcomes for different subgroups or using different methodologies. In the meta-analysis, including more than one estimate from the same intervention can violate the assumption of independence across effect sizes, leading to a unit-of-analysis problem (Higgins, 2019; Borenstein et al., 2009). To deal with such cases, we exclude duplicated estimates by retaining the study with the simplest specification. Therefore, we excluded Brown, Earle, & Morgulis (2015), which analyzes the same SBA loan program as Brown & Earle (2017), as the former presents disaggregated results by firm size and age that show no significant heterogeneity. Similarly, we exclude Kapoor, Ranjan, & Raychaudhuri (2012), whose findings on a 1998 policy change overlap with Banerjee & Duflo (2014), which uses the same underlying experiment. • Different outcome variables: We exclude papers that focus on outcomes different to the ones we study —namely employment, sales, and profits. For example, we drop Aivazian and Santor 36 (2008) from the sample, as their primary outcome of interest is value added, which is not directly comparable to our selected performance measures. • Interventions different from conventional formal loans: We exclude studies evaluating programs such as the U.S. Paycheck Protection Program (PPP) (e.g., Bartik et al., 2020). PPP loans were issued during the COVID-19 pandemic to support firms in retaining employees and covering operating expenses such as payroll, rent, mortgage, and utilities. These loans were forgivable if firms maintained their workforce and used the funds for eligible business expenses, making them more akin to conditional grants than conventional repayable loans. • Credit guarantees: We included some studies of guaranteed loans but exclude those where the difference between the treatment and control groups is whether the loans are guaranteed or not, such as Blasio et al. (2018) and Martín-García & Santor (2021). 37