The Labor Market Effects of Loan Guarantee Programs

Abstract

Numerous countries facilitate bank lending to small businesses through loan guarantee programs, whereby a government agency underwrites a share of the notional of loans issued by banks to qualifying borrowers (such as the Small Business Agency [SBA] programs in the United States).¹ As banks retain skin in the game, loan guarantees are designed to address the mistargeting and rent-seeking that plague direct public lending (see, for instance, Khwaja and Mian 2005). Policy makers’ interest in these programs increased in the wake of the 2008 financial crisis because of concerns that small businesses might be prevented from accessing capital sufficient for them to be resilient, to grow, and to create jobs (Chen, Hanson, and Stein 2017; Bord, Ivashina, and Taliaferro 2021).² At the same time, changes in credit supply might lead to labor misallocation and ultimately hurt aggregate productivity (see, e.g., Bai, Carvalho, and Phillips 2018; Blattner, Farinha, and Rebelo 2023). These issues are particularly acute as SMEs represent 70% of employment in OECD countries.³ The question of the design and efficiency of these programs has become even more important to policy makers since the COVID-19 outbreak as a large number of governments, including the United States and the majority of European countries, have massively turned to this tool to address the sharp recession resulting from the pandemic.^{4  ,  5} Despite their large and growing implementation, we know surprisingly little about the long-term effects of these programs on workers’ employment and mobility across firms, and on their aggregate implications. While these programs have been shown to foster job growth at beneficiary firms (Brown and Earle 2017), assessing their effectiveness at mitigating net employment effects of financing frictions, as well as the labor allocation effects of such programs, calls for measuring the impact of these programs on both workers’ transitions in and out of unemployment and their job-to-job mobility following a downturn. If these programs prevent workers from experiencing lengthy periods of unemployment, and/or impairing their human capital, the benefit of these programs can be large. However, these programs might also create a barrier to the beneficial reallocation of workers after a recession by keeping workers in less productive firms. Such a concern is, for instance, motivated by the fact that during a major countercyclical loan guarantee program in France, firms taking up the program were 24% less productive than the average firm population, per Figure 1.⁶

Figure 1:

Labor productivity by guarantee status

This figure displays average labor productivity measured by value-added (in thousand euros) over employees (VA/Emp) in 2008 for the population of French firms. The first bar represents labor productivity of firms receiving a guarantee under the recovery plan, while the second bar represents the labor productivity of firms not receiving a guarantee. Confidence intervals at the 99.99% level are represented in red. Table A.1 in the Internet Appendix shows that the pattern is robust to controlling for industry fixed effects.

Open in new tabDownload slide

In this paper, we use novel administrative micro data combined with geographic variation in program design to uncover and assess this trade-off. We estimate the long-term impact of a countercyclical loan guarantee program in France on workers’ employment and earnings trajectories, both at the initial firm and at subsequent employers. Our data tracks a representative sample of individual workers across firms over time, as well as their transitions between employment and unemployment and the associated welfare benefits they receive. Matched with firms’ balance-sheet information, these data allow us to study how such programs affect workers’ reallocation following a recession by observing counterfactual worker job-to-job transitions, how the productivity of their new employer differs from their initial employer, and which type of workers are particularly affected by the changes in mobility resulting from the program. Implemented in the midst of the financial crisis, the Recovery Loan Guarantee Program offers a public guarantee for French small and medium-sized enterprises (SMEs) to rollover and extend their short-term debt. This new program, administered by Bpifrance, the French equivalent to the SBA, was announced in the last quarter of 2008 and extended until the end of 2010.⁷ As French regions differently augment the funding of the national program, the treatment intensity varies geographically in a significant and plausibly exogenous manner. We exploit this heterogeneity and integrate it with a regional border discontinuity approach in order to estimate the causal impact of the program on workers at firms benefiting from a loan guarantee. The identifying assumption in our setting is that workers in firms located on each side of a regional border would have experienced similar labor market outcomes in the absence of the loan guarantee program. We first provide evidence that the regional intensity of the loan guarantee program translates into a higher take-up of loan guarantees at the firm level within the regional border area. Furthermore, higher treatment intensity is associated with both an increase in the quantity of bank debt on firms’ balance sheets and a decrease in their cost of borrowing, which supports that our measure of treatment intensity captures heterogeneity in guarantee supply and not in firms’ demand for loans across regional borders. We then leverage our longitudinal worker-level data to evaluate how this program affects worker employment, earnings, mobility, and matching with firms until 2015. We find that the program has a significant and persistent positive impact on workers’ employment and earnings trajectories. Quantitatively, when extrapolating our estimates to the average treatment at the worker level, we obtain that individual workers initially employed by a treated firm receive earnings that are 26% higher on average over the 2009–2015 period, when compared to a counterfactual set of workers initially employed by nontreated SMEs. This finding mostly reflects an employment margin: workers more exposed to the program are significantly less likely to separate from their initial employer, and in turn to be unemployed over the sample period. Their unemployment benefits are consequently lower, representing a significant reduction in the cost of the government intervention. Overall, the program preserved 487,000 job(-years) at a gross cost per job(-year) of €1,400 and €425 when accounting for the ex ante or ex post cost. Since the loan guarantee program reduced workers’ unemployment spells and the associated savings for the unemployment national fund amount to around 2.1 billion euros, the program actually exhibits a negative net cost. However, these employment gains need to be contrasted with labor allocation effects. The relative gain in employment and earnings from higher retention at initial employers is half offset by a reduction in worker mobility toward other firms in the economy. In tight labor markets worker mobility fully offsets the employment gains at initial employers, canceling the net employment benefits of the program in this context. Further, the reduced mobility resulting from the loan guarantee program dampens the reallocation of workers toward more productive firms, and particularly so for workers with skills in high demand. Such a result is consistent with loan guarantees supporting skilled labor hoarding, that is, higher worker retention for firms with high-skill workforce. By comparing the productivity of both the initial and new employers of the workers in our sample, we first show that workers more exposed to the program are significantly less likely to move to firms with higher productivity than workers from the counterfactual. Turning to the cross-section of workers and occupations, we find that this dampening effect is particularly pronounced for workers with high earnings capacity, for occupations for which firms report hiring difficulties, and for nonroutine/cognitive-analytical occupations. Taken together, these findings highlight an important counterpart to the employment benefit of countercyclical loan guarantee programs we previously document: by keeping workers in their current firms and thereby creating a barrier to beneficial worker reallocation, these programs might affect the trajectory of the economy following recessions. We conduct a battery of tests to ensure that these results are not driven by alternative mechanisms, such as confounding local shocks, spillovers, or other policies. Crucially, we conduct a placebo analysis on firms that have a low propensity to obtain a guarantee, and find no effects of the program on the set of workers initially employed by these firms, which confirms that our baseline estimates are caused by the loan guarantee program rather than other policies or demand-side factors that could confound our results. To conclude our study, we develop a simple theoretical framework that helps us interpret our empirical results and provides an estimate for the effect of the program on aggregate productivity. In the model, firms need to finance labor in advance, and we interpret the guarantee program as providing treated firms with a subsidy to their cost of financing. We show that the program leads to an increase in labor demand for treated firms, which in turn depresses labor demand for untreated firms through crowding out effects on the labor market. We then build on the general insight from the misallocation literature that the program would have negative effects on aggregate productivity if the marginal revenue products on labor are lower for treated than for untreated firms. Consistent with our evidence that treated firms have lower labor productivity than other firms preprogram, we find that the program had a negative impact on aggregate productivity, of around –1%. Such a net negative effect is likely to hold in economies where SMEs exhibit lower labor productivity than larger firms, as is the case in the European Union at large, for instance.⁸ Our research contributes to the burgeoning literature on loan guarantees (de Andrade and Lucas 2009; Beck, Klapper, and Mendoza 2010; Lelarge, Sraer, and Thesmar 2010; Mullins and Toro 2018; Brown and Earle 2017; D’Acunto, Tate, and Yang 2017; de Blasio, Mitri, D’Ignazio, Russo, and Stoppani 2018; Bachas, Kim, and Yannelis 2021).⁹ Two contemporaneous papers expand on Brown and Earle (2017) and measure the real effects of loan guarantee programs at the firm level: Bonfim, Custodio, and Raposo (2023) do so in Portugal, and Gonzalez-Uribe and Wang (2022) in the United Kingdom. The settings of these two papers differ in terms of scheme design and economic contexts, resulting in distinct channels dominating: lower cost of debt for the former, and mitigation of credit rationing for the latter. Both studies however document positive firm-level effects in terms of employment and firm growth, and cost-effective job preservation when contrasting eligible firms’ employment growth to the program direct costs. Our study, which centers on a worker-level analysis, complements their findings by estimating net employment effects as well as labor reallocation effects. Our analysis requires tracking workers when they transition in and out of firms or employment, and sheds a new light on the implications of loan guarantee programs for aggregate productivity. Relatedly, we add to the empirical debate on the effectiveness of public policies aiming to protect or stimulate employment in downturns, such as hiring credits (Cahuc, Carcillo, and Le Barbanchon 2019; Neumark and Grijalva 2017), and subsidies for short-time work (Cahuc, Kramarz, and Nevoux 2018; Giupponi and Landais 2022). We show that loan guarantees have a positive and persistent impact on workers’ employment and earning trajectories obtained at a relatively low cost, likely because of effective targeting resulting from the loan guarantee design (Philippon 2021), but also significantly dampen the reallocation of the workforce toward more productive firms. In doing so, we relate more broadly to the literature studying the consequences for the economy of size-dependent policies that frequently favor smaller firms (see, e.g., Restuccia and Rogerson 2008; Hsieh and Klenow 2009; Bartelsman, Haltiwanger, and Scarpetta 2013; Garicano, Lelarge, and Van Reenen 2016). Lastly, our work assesses a possible remedy to the significant employment effects of financing frictions documented by a large body of empirical studies, both at the firm level (Chodorow-Reich 2014; Duygan-Bump, Levkov, and Montoriol-Garriga 2015; Greenstone, Mas, and Nguyen 2020; Giroud and Mueller 2017; Bentolila, Jansen, and Jiménez 2018) and at the worker level (Berton, Mocetti, Presbitero, and Richiardi 2018; Caggese, Cuñat, and Metzger 2019; Baghai, Silva, Thell, and Vig 2021; Babina 2020; Acabbi, Panetti, and Sforza 2020; Gortmaker, Jeffers, and Lee 2020). Our study also contributes to the literature studying labor misallocation effects resulting from financial policies (e.g., Bai, Carvalho, and Phillips 2018; Barbosa, Bilan, and Celerier 2019; Fonseca and Van Doornik 2022; Blattner, Farinha, and Rebelo 2023).

1 Background

Numerous governments, including the United States, provide loan guarantees to small firms. These programs are usually implemented through a specialized entity, such as the Small Business Administration (SBA) in the United States, or Bpifrance in France, which partners with banks.

1.1 Economic rationale of loan guarantees

Loan guarantee programs allow small businesses to mitigate their financing frictions, which are particularly pronounced during recessions. Access to credit for small firms might be limited in general by adverse selection (Stiglitz and Weiss 1981), moral hazard (Holmstrom and Tirole 1997), and transaction costs. Such financing frictions are typically amplified during recessions, since revenue shortfall worsens the pool of borrowers and increases debt overhang (Brunnermeier and Krishnamurthy 2020). While also acting as a subsidy to the cost of capital, loan guarantees by a government-backed entity have several advantages over direct subsidized public lending. First, this public intervention design typically delegates screening and monitoring to private banks. Relying on banks’ expertise and infrastructure mitigates the risk that political considerations drive the allocation of credit. Second, as the guarantees are partial, banks retain skin in the game when screening loans. Lastly, guarantees do not require the guarantor institution to disburse cash and raise capital at the time of their implementation, although they do create regulatory capital requirements. Theoretically, the employment effects of loan guarantee schemes are ambiguous, although prior work documents positive effects of relaxing financial constraints on employment growth at the firm level (see Chodorow-Reich 2014; Brown and Earle 2017), among others. During downturns, the main focus of our study, one important economic rationale for loan guarantee schemes is to support labor hoarding, as put forward during both the financial crisis and the COVID-19 pandemic. Financial constraints might prevent firms facing a temporary shock from optimally maintaining employment relationships with their workers, as argued in Giroud and Mueller (2017), creating an excess sensitivity of separations to business cycle fluctuations.¹⁰ While loan guarantees might allow recipient firms to retain their workforce, this does not necessarily translate into net aggregate positive effects on employment. In particular, in areas or periods with low unemployment rates, displaced workers might easily find a job in other firms of the economy, in which case the positive effect of guarantees on the employment of recipient firms does not translate into net employment gains in the aggregate. Moreover, worker retention, while usually beneficial to a given employer, can be detrimental to the economy at large. Downturns are indeed typically associated with significant reallocation of workers from low to high-productivity firms as less productive firms reduce the scale of their operations or exit, while more productive firms might better resist or even grow. In this context, loan guarantees might reduce efficiency on the labor market—that is, the aggregate quality of the matches between workers and firms—if they allow recipient firms with structurally low labor productivity to retain their employees, and thus dampen the reallocation of the workforce toward firms with higher labor productivity. We formalize the trade-off between the employment and productivity effect in the last section of the paper.

1.2 The French public guarantor and the post-GFC French recovery plan

Bpifrance¹¹ is the entity managing public loan guarantee programs in France and was created in 1982 as a French equivalent of the SBA. Bpifrance activities are mostly targeted toward SMEs and encompass, in addition to loan guarantees, direct lending, providing grants, and investing in equity. Bpifrance does not collect deposits, but funds itself in the wholesale market. Bpifrance works with a network of partner banks that include all major French banks and relies on them to source loan applications. As of 2017, Bpifrance possesses 48 local branches that process the loan guarantee applications provided by banks. Starting in the second half of the 2000s, French regions have been partnering with Bpifrance. This partnership takes the form of complementing Bpifrance intervention by guaranteeing an additional fraction of the loans that Bpifrance underwrites. This additional guaranteed fraction is capped and varies across regions in accordance with bilateral agreements between French regions and Bpifrance. The partnership is based on top-up financing independently provided by the regions through dedicated entities, the Fonds Regional de Garantie. The existence, timing, and generosity of such partnerships result from an idiosyncratic local political process conducted in the regional parliaments. For the purpose of our empirical analysis, we focus on a new loan guarantee program created at the end of 2008, which specifically aimed at allowing firms to access short- and medium-term debt in the wake of the financial crisis. The French recovery plan of 2009–2010 led to the creation of a large short-term credit guarantee program managed by Bpifrance (under the Oseo-Garantie name at that time). As illustrated in Figure 2, the plan guaranteed €5.3 bn of new bank debt between 2008Q4 and 2010Q4, which represented 0.2% of GDP. The plan targeted new lines of credit with a term between 12 and 18 months, as well as the restructuring of existing short-term debt into new loans with maturity between 2 and 7 years. Of firms that received guarantees on their new lines of credit, 4,000 received them for an amount of €1.8 bn, and 17,000 firms received guarantees on their medium-term new loans for €3.5 bn. Bpifrance charges an average insurance premium of around 1% per annum of the loan notional in exchange for such a guarantee. Ex post, the premiums represented a total of €126 M, while banks claimed guarantees for €333 M, which illustrates that the guarantee was subsidized on average.

Figure 2:

Yearly volume of guarantees of the Recovery Plan

This figure displays the total volume of guarantees by Bpifrance as part of the recovery plan.

Open in new tabDownload slide

2 Data

We use three complementary sources of administrative micro data, which we obtained from Bpifrance and the French Statistical Office (INSEE): an exhaustive file of individual loan guarantees, the exhaustive firm registry, and a worker panel covering 1/12th of the French workforce that tracks workers across jobs and employers, as well as in and out of unemployment. First, we use proprietary data provided by Bpifrance on the universe of firms benefiting from loan guarantee programs since 2002.¹² These data provide a unique firm identifier (SIREN), and information on the guarantee characteristics, including the date and amount of the loan, whether the guarantee was part of the recovery plan, the type of loan underlying the guarantee, and the fraction of the loan covered by the guarantee. Bpifrance data do not include information on interest rates but include information on default: whether the loan benefiting from the guarantee defaults over its life, and the loss amount. The data set does not include unsuccessful application data, as Bpifrance did not collect such data.

We also use administrative micro data extracted from tax files available until 2015. The data includes balance sheets as well as profit and loss statements for the universe of French firms.¹³ We track firms through time using their unique identifying number ascribed by INSEE. Lastly, we rely on worker longitudinal data (“DADS Panel”), built by INSEE from social security contribution declarations of firms and from unemployment benefits. The sample covers all individuals born in October of each year, that is, 1/12th of the French workforce. Each year firms declare the employment spells, the number of hours worked, and the associated wages for each worker. For workers who have multiple jobs in a given year, we aggregate earnings across all jobs and retain the identifier of the employer that accounted for the largest share of the worker’s earnings. Data on unemployment benefits is available since 2008.

2.1 Data filtering and summary statistics

For the purpose of our identification strategy, we restrict the sample to firms with all employees in the same region and located within 10 miles of a regional border. Given that SMEs (defined as firms with less than 250 employees) represent virtually all the beneficiaries from the recovery plan, we also restrict the sample to SMEs. We then follow the literature and exclude firms from the financial and real estate sectors, as well as utilities, nonprofit, and regulated sectors. This filtering leaves us with 31,949 firms in our central sample. At the worker level, we restrict the sample to workers with high labor force attachment (as in, e.g., Autor et al. 2014, Yagan 2019), namely, workers with annual earnings above €10,000 in 2006, 2007, and 2008. To avoid measurement errors due to initial entry and exit from the workforce, we focus on workers that were at least 24 years old in 2008 and at most 58 years old in 2015, that is workers who were born between 1957 and 1984. We also restrict our analysis to French citizens in order to alleviate concerns over unobserved employment in foreign countries. Lastly, we only keep workers initially employed by establishments located within the region border zone. This filtering leaves 38,568 workers in the sample, which are, by construction, representative of 12 times more workers, or 462,816. For each of these workers, we then track their employment status (employed or unemployed), the source and magnitude of their earnings, labor earnings or unemployment benefits, as well as the economic performance (e.g., productivity and growth) of their new employers when they change jobs, over the sample period. Table 1 presents descriptive statistics for our filtered sample. Panel A provides information on the exposure to the loan guarantee program, both at the regional and at the firm level. Raw $Guaranteeregion,2009−2010$ corresponds to the average ratio of loan guarantees to firm assets, computed across all eligible firms in a given region, excluding firms within 10 miles of a regional border. The generosity of the program varies significantly across the 21 regions, with firms from the least generous region receiving on average 0.1% of their total assets in guarantees, while firms from the most generous region receive 7.3 times more. Panels B and C present descriptive statistics at the worker level. The average worker has worked for 6.5 years during the 2009–2015 period, received earnings equal to 6.5 times their initial annual earnings (average annual earnings over the 2006–2008 period), and received 0.2 times their initial annual earnings in unemployment benefits. The average worker is 38 years old, works 1,872 hours, and earns €23,836 per year.¹⁴  Table A.2 in the Internet Appendix presents the same characteristics separately by firm-level guarantee take-up. We note that 5.1% of workers in our sample are initially employed in a firm receiving a loan guarantee. Finally, in panel D, we present a number of firm characteristics measured in 2008. The average firm in our sample has 20 employees in 2008, is 18 years old, has assets of €3.04 million, and return on assets of 10%.

By construction, our main sample includes only SMEs and their associated workers initially located within a 10-mile distance to a regional border. One possible concern is that this sample is not representative of the whole universe of SMEs. To shed light on this potential issue, Internet Appendix Table A.3 displays firm and worker characteristics for our sample of SMEs located within a 10 mile distance of a regional border, and for the whole universe of French SMEs for which we observe worker-level data. We also present the distribution of firms in both groups across a list of 18 industries in Internet Appendix Table A.4. Overall, the characteristics of the two groups, and their distributions across industries, are comparable. Taken together, these statistics are supportive of our estimates being informative for the whole population of SMEs.

3 Measuring The Labor Market Effects of Loan Guarantee Programs

3.1 Empirical strategy

Figure 3:

Regional intensity of loan guarantee intervention

This figure displays the regional intensity of intervention by Bpifrance, Guarantee$region,09−10$⁠, estimated across SMEs outside the border area, see Table 1. Darker colors represent regions with higher treatment intensity. The gray area corresponds to municipalities within 10 miles of a regional border. Thin lines in gray represent department boundaries within regions.

Open in new tabDownload slide

Our empirical strategy is akin to a difference-in-differences setting with continuous treatment, as areas are differentially exposed to the short-term loan guarantee program. The exclusion restriction relies on the regional loan guarantee exposure only affecting workers’ outcomes through the subsidized access to new lines of credit and bank loans offered by the program to their employers in 2009 and 2010. In particular, regional exposure to the program needs to be orthogonal to other unobserved regional shocks that could otherwise affect workers’ outcomes over the sample period. In Section 7, we conduct a placebo-like analysis on firms that have a low propensity to obtain a guarantee, and find no effects of the program on the set of workers initially employed by these firms, which mitigates the concern that other unobserved economic or policy shocks could confound our results.

3.2 First-stage evidence

To ensure that our first stage is driven by firms targeted by the program, we regress an indicator variable for guarantee take-up, Guarantee (1/0), on Guarantee$region,09−10$⁠, interacted with the firm-level predicted take-up propensity. We predict take-up propensity by first estimating a linear probability model on the whole firm population (excluding firms in the border area), and then using observable firm characteristics (logarithm of assets, ROA, logarithm of firm age, dividend/sales, PPE/assets, debt/assets, credit risk and two-digit industry fixed effects) measured in 2008. The regression coefficients of the interaction terms are displayed in Figure 4. The figure illustrates that a higher regional treatment intensity translates into a higher likelihood to obtain a guarantee for the firms that exhibit the characteristics generally targeted by the program. In contrast, there is no effect for firms that do not have such characteristics, which we later exploit to conduct a placebo analysis.

Figure 4:

Regional treatment intensity, firm-level take-up propensity, and actual take-up

This figure plots regression coefficients and 95% confidence intervals of regressing actual guarantee take-up, Guarantee (1/0), on the regional exposure to the 2009–2010 loan guarantee program, Guarantee$region,09−10$⁠, interacted with quintiles of firm-level predicted take-up propensity. We estimate take-up propensity using observable firm characteristics (logarithm of assets, ROA, logarithm of firm age, dividend/sales, PPE/assets, debt/assets, credit risk and two-digit industry fixed effects) measured in 2008.

Open in new tabDownload slide

In addition, we also check that our first stage is driven by firms facing high financial constraints, consistent with the program target. To do so, we split our sample along proxies for firm financial constraints widely used in the literature: credit risk, measured as the inverse of the interest coverage ratio, dividend payout, and cash flows. We run our first-stage specification on each of these subsamples and present the regression results in Internet Appendix Table A.6. We confirm that the relationship between $Guaranteeregion,2009−2010$ and $Guaranteefirm,2009−2010$ is indeed driven by firms with above-median credit risk, not paying dividends, and below-median cash flows, as of 2008.

3.3 Effects on financial constraint

To further strengthen the validity of our first stage, we test whether a higher regional treatment intensity is associated with an increase in the quantity of bank debt combined with a similar or lower cost of debt, as predicted by a relaxation of the credit constraint for treated firms. We first run a specification similar to our first stage where the dependent variable is the firm-level growth rate of bank debt over 2008–2009.¹⁹ As shown in column 3 of Table 2, higher exposure to the loan guarantee program is indeed associated with a significantly higher growth in bank debt, which is consistent with a relaxation of financial constraints for the treated group. Second, we run a similar specification where the dependent variable is the change in the average interest rate paid on outstanding debt between 2008 and 2010.²⁰ Column 4 of Table 2 displays the regression coefficient. Treated firms exhibit a significantly lower interest rate, even when controlling for firm characteristics, which is consistent with the program driving the increase in debt, rather than higher local demand for credit.

4 Impact of Countercyclical Loan Guarantees on Employment

We now turn to analyzing the impact of exposure to the loan guarantee program during a downturn on workers’ employment and earnings.

4.1 Firm-level effects

We first study firm-level effects on employment and productivity and keep the same specification as our first stage. We start with the short-term effects on firm employment and labor productivity: in column 1 of panel B of Table 2, we use the logarithm of the change of firm employment over 2008–2010, $Δ08−10Emp$⁠, as dependent variable and find that being more exposed to the loan guarantee program results in significantly higher firm employment in the short run, consistent with the literature. Column 2 shows that this employment effect coincides with a decrease in labor productivity, measured by the change in value-added per employee, $Δ08−10VA/Emp$⁠. In the last section of the paper we formally show that such a decrease in labor productivity is consistent with the program relaxing financial constraints for treated firms. Turning to the long-term effects, we observe in column 3 that the positive employment effect weakens over time, while column 4 shows that loan guarantees do not translate into increased total factor productivity for treated firms in the long run, as measured by the change in TFP over 2008–2015, $Δ08−15TFP$⁠.²¹

4.2 Effects on worker employment and earnings

In Internet Appendix Table A.7 we show that the worker-level employment effects are driven by firms facing high financial constraints, that is, are above-median credit risk, not paying dividends, and below-median cash flows, as of 2008. Together with Table A.6, these results are reassuring, as we find employment effects only for workers in firms targeted by the program, a point we will return to later when conducting a placebo analysis. Lastly, we study the year-to-year impact of the loan guarantee program on worker outcomes. We plot the estimated effect of exposure to the loan guarantee program for each year from 2004 to 2015 on annual worker earnings in any firm in panel A of Figure 5.

Figure 5:

Dynamics: Effect on earnings

This figure plots regression coefficients and 95% confidence intervals from 12 regressions of earnings that a worker obtains in the year indicated on the x-axis, normalized by average annual earnings in 2006–2008, on our measure of regional exposure to the 2009–2010 loan guarantee program, Guarantee$region,09−10$⁠. All regressions include department-pair fixed effects, the distance from the regional border, changes in regional controls from 2008 to 2010 (public spending, local taxes, public equipment expenditures, public debt, state contribution, value-added of non-SMEs, and regional bank lending), as well as firm- and worker-level controls measured in 2008.

Open in new tabDownload slide

Exposure to the loan guarantee program is associated with a large and statistically significant effect on annual earnings for the whole sample period following the treatment. The point estimates for 2004 to 2008 are all insignificant, which supports the absence of pre-trends and our interpretation of a causal impact of the guarantees on workers’ earnings trajectories. As annual earnings are higher post treatment for the treated group, the cumulative effect on earnings keeps growing over that period. Overall, the effects of the policy on earnings are immediate and strikingly persistent until the end of the sample period.²⁶

4.3 Effect on unemployment insurance

In developed economies, earning losses due to involuntary unemployment are partly mitigated by unemployment insurance. In France, unemployment benefits cover a fraction of the initial wage, are subject to eligibility criteria, and are earned for up to 2 years. In our data set, we can isolate earnings coming from unemployment benefits, which allows us to estimate the fraction of earning losses in the counterfactual offset by unemployment insurance. We use the cumulated amount of unemployment benefits (scaled by initial earnings) during 2009–2015 as the dependent variables in our baseline specification. Table A.9 of the Internet Appendix displays the results. We find that treated workers collect significantly lower amounts of unemployment benefits over the study period. In economic terms, this point estimate indicates that workers from the average treatment region receive lower unemployment benefits over the 2009–2015 period, representing 1.3% of their initial annual income. This magnitude indicates that unemployment insurance offsets around 15% of the gap in earnings between the treated group and the counterfactual documented in Table 3.

4.4 Cost-per-job estimate

To conclude the analysis on employment effects, we calculate the cost-per-job(-year) resulting from the policy. We start with estimating the total number of job-years preserved by the policy. As our empirical analysis is conducted at the worker level, we multiply the average treatment of 0.29 (% of total assets) by the coefficient estimated in our baseline specification (0.240, see column 2 of Table 3) to calculate the average effect by worker. This calculation corresponds to an average gain of 0.07 years of employment for the average worker in our sample. As the full-time employee equivalent employment at SMEs in 2008 in France was 7.0 million, we obtain an estimate of around 487,000 job(-years) preserved over the period 2009–2015 (⁠$7.0m×0.29×0.240$⁠).²⁷ The ex ante gross cost to the French government was the provision of a €683 M fund, which translates into an estimate for the gross cost-per-job(-year) of around €1,400.²⁸ The ex post gross cost of the guarantee program can be estimated as the difference between the amount of Bpifrance payments to the banks of defaulting firms less the premiums paid to Bpifrance. Banks have claimed guarantee payments for an aggregate amount of €333 M, and Bpifrance has received premiums for an aggregate amount of €126 M. The ex post cost is therefore €207 M, which translates into an estimate for the gross cost-per-job(-year) around €425.²⁹ This gross cost-per-job(-year) ignores savings in unemployment and social benefits resulting from the loan guarantee program. We can easily adjust for the savings in unemployment benefits that we estimate in the previous subsection, which correspond to around €300 per worker, or €2.1 bn in aggregate. This calculation therefore yields a negative net cost for the program and the jobs it helps preserve, which would be even more pronounced if accounting for avoided social contributions. The positive employment effects need however to be contrasted with the effect of the program on labor allocation and aggregate productivity.

5 Impact of Loan Guarantees on Labor Allocation

5.1 Worker retention and adjustment margin

Given our ability to follow workers over time, and observe their job-to-job transitions across firms, we turn to precisely measuring both the impact of the loan guarantee program on employment in initial firms, and how much of this effect is offset by the adjustment margin at other firms. We therefore follow prior work (e.g., Autor et al. (2014)) to decompose the overall effect on years employed and cumulated earnings in Table 4. Column 1 displays the net effect, which corresponds to the results in column 2 of Table 3. Column 2 presents the share coming from the firm at which the worker is initially employed as of 2008 and column 3 the share coming from their subsequent employment at other firms. The point estimate of column 2 captures the differences in employment and earnings obtained by workers at their initial employer.³⁰ The baseline coefficients of column 1 represent less than half of these effects at the initial firm, and reflect the fact that the relative employment and earning gains at the initial firm for treated workers are partially offset by counterfactual workers’ mobility to other firms. Indeed, as shown in column 3, workers less exposed to the loan guarantee program are more likely to subsequently work and receive earnings from other employers over the sample period.

This exercise highlights the benefit of using worker-level panel data to accurately assess the net employment and earnings effects of loan guarantee programs, and evidences the significant dampening effect of the program on worker mobility following a downturn. It raises the question whether loan guarantee programs might be preventing the beneficial reallocation of the workforce toward more productive firms. This concern is particularly relevant given the persistence of the effects on both worker retention and mobility across firms. Panels B and C of Figure 5 displays in a longitudinal setting the breakdown between the effects on worker earnings at the initial firm and at other firms, and evidences significant effects 5 years after the end of the program.

5.2 Low versus high unemployment areas

To explore the heterogeneity of the effects of loan guarantee programs according to labor market conditions, we split our sample between low and high unemployment municipalities and run our central specification to measure the net effect and the retention effect.³¹ Theoretically, the program should have less impact on net employment in tight labor markets, as workers that are displaced in the absence of support to their employers can more easily find a new job in another firm. Regression coefficients are displayed in Table 5, separately for high unemployment areas (columns 1 and 2) and low unemployment areas (columns 3 and 4). In low unemployment areas, although the retention effect is large and significant (column 4), the net effect of the loan guarantee program on workers’ cumulative employment over the sample period is instead low and statistically insignificant. However, in high unemployment areas, while exposure to the program is associated with a large and significant retention effect by initial employers as in low unemployment areas (as shown in column 6 comparing the coefficients presented in columns 2 and 4), the net effect on workers cumulative employment remains large, and of similar magnitude compared to the effect on retention. In high unemployment areas, higher retention rates at recipient firms thus benefit the employment trajectories of individual workers, as it prevents them from experiencing lengthy periods of unemployment.

5.3 Labor hoarding

We further investigate retention policies of treated firms by exploring heterogeneity across workforce and individual worker skill level. We first classify workers in the highest tertile of the overall worker distribution as high-skill for three different skill measures: earnings within age cohorts, hiring difficulty for a given occupation, and cognitive-analytical task content. Based on this classification, we compute the firm-level fraction of high-skill workers for each measure.³² We then split the sample of firms at the median and implement our baseline firm-level specification to measure firm employment effects on these split samples. Regression coefficients are displayed in panel A of Table 6. For each measure of skill, we observe that the positive firm employment effect previously documented in Table 2 results from firms with a high-skill workforce. Such heterogeneity is consistent with higher incentives to hoard workers for such firms, as high-skill workers might be costlier to rehire in the future. We then split the worker sample at the median of the respective skill measure, and implement our baseline worker-level specification, using years employed and cumulative earnings at the initial firm as the dependent variable. Such analysis yields consistent results with the firm-level one. Worker retention resulting from the program is higher for high-skill workers, although retention is also increased for lower-skill workers. This analysis suggests that firms prioritize retaining high-skill workers as they might be particularly costly to replace. While beneficial to the firm, such retention policies might however have negative aggregate implications, as the economy could benefit from a reallocation of these workers to more productive firms, an issue that we explore in the following section.³³