Credit Market Imperfections, Labor Markets, and Leverage Dynamics in Emerging Economies∗ Alan Finkelstein Shapiro†

Andr´es Gonz´alez G´omez‡

Tufts University

International Monetary Fund

June 8, 2017

Abstract Emerging economies (EMEs) have different credit and labor market structures relative to advanced economies. We document that economies with larger self-employment shares tend to exhibit less countercyclical leverage dynamics. We build a model where formal credit markets, input credit relationships, and the structure of labor markets interact that (1) captures a comprehensive set of EME business cycle regularities and (2) rationalizes our new fact. The interaction between firms’ net worth, interfirm input credit, and self-employment underlying our framework is critical for explaining our fact and is supported by the data. JEL Classification: E24, E32, E44, F41 Keywords: Emerging economy business cycles, financial frictions, labor search frictions, self-employment, credit policies

∗

This paper was previously circulated as ”Credit Market Imperfections, Labor Markets, and Business Cycles in Emerging Economies” and is a substantially revised version of an older paper that was part of an Inter-American Development Bank (IADB) research initiative (IADB financial support for the original version of the paper is greatly acknowledged). † Corresponding author. Contact information: Department of Economics, Tufts University, Braker Hall, 8 Upper Campus Road, Medford, MA 02155. Email: Alan.Finkelstein [email protected]. ‡ Contact information: Institute for Capacity Development, International Monetary Fund, 1919 Pennsylvania Ave NW, Washington, D.C. 20431. Email: [email protected].

1

1

Introduction

The financial crisis of 2008-2009 highlighted the connection between financial imperfections and labor markets in advanced economies. While similar linkages exist in emerging economies (EMEs), the latter differ considerably in their credit and employment structures relative to their advanced counterparts.1 In particular, EMEs have large self-employment shares as well as a limited segment of (mainly salaried) firms with access to financing from formal financial intermediaries. The remaining (owner-only or micro) firms resort to internal funds and input-based (informal) credit relationships with suppliers. Despite recent progress in understanding how the link between labor and credit markets contributes to credit and aggregate dynamics, which is important for shedding light on the effectiveness of cyclical interventions in credit markets, little is known about the relevance of differences in credit and employment structures across economies and their implications for credit market dynamics. In this paper, we document a new stylized fact: economies with higher average selfemployment shares exhibit less countercyclical firm leverage dynamics (see Section 2). This link is robust and holds using different datasets and country samples. We present a smallopen-economy (SOE) RBC search model that establishes a novel link between salaried firms’ imperfect access to formal credit, their net worth and leverage, and self-employment through interfirm input credit relationships. Our main results are threefold. First, using Mexico as a prototypical EME, we show that our framework replicates several EME business cycle facts that encompass credit and labor markets (these include, among others, the cyclical dynamics of firm leverage, selfemployment, salaried employment, and unemployment). Notably, this takes place in a context where two well-known features of EMEs—the high volatility of consumption and the countercyclicality of the current account—are captured under a simple shock specification with aggregate productivity shocks as the sole driver of business cycles. Second, our model not only provides an empirically-supported economic rationale behind the empirical relationship between average self-employment shares and the cyclicality of leverage, but also quantitatively captures this relationship. Finally, to stress the wider applicability of our 1

We discuss existing evidence on the following facts in Section 2.

2

framework, we show that cyclical credit policies that limit formal-credit-market stress during recessions introduce a tradeoff between credit stability and unemployment fluctuations. This tradeoff traces back to the distinct employment and credit structure in EMEs—in particular, the prevalence of self-employment—and vanishes in the absence of self-employment. Our framework generates empirically-consistent cyclical labor market and aggregate dynamics: adverse aggregate shocks lead to a rise in unemployment, leverage, formal-credit external finance premia, and self-employment (via higher transitions from unemployment and into self-employment); and with a fall in firm net worth, salaried employment, consumption, investment, and output. The model’s mechanism is as follows. An adverse aggregate shock erodes the financial net worth of salaried firms—firms that rely on formal credit to operate—by reducing the value of their assets; it increases firms’ leverage and borrowing costs, and reduces these firms’ capital and salaried labor demand. The shock also leads to a reduction in these firms’ capital usage, leading to a rise in idle capital in the economy, lower salaried employment opportunities, and a fall in salaried employment. Households respond to deteriorating salaried labor market conditions by searching for resources to create new self-employment ventures, thereby partially offsetting the fall in salaried income. Despite the fact that the aggregate shock affects all firms, salaried firms are willing to supply their idle capital to potential self-employed individuals via new input credit relationships because (1) the opportunity cost of capital is lower and (2) supplying unused capital to the self-employed represents an additional source of revenue that becomes particularly attractive during recessions. To the extent that the self-employed depend on input credit relationships to produce, the rise in the availability of idle resources facilitates transitions out of unemployment and into self-employment, leading to countercyclical self-employment, as in the data (see the discussion in Section 2). However, such expansion is less forceful than the contraction in salaried employment, resulting in an increase in unemployment. How does interfirm input credit contributes to the link between leverage dynamics and the share of self-employment? The revenue that salaried firms obtain from supplying unused capital to potential self-employed individuals is, by definition, part of firms’ accounts receivables and therefore net worth. Importantly, the rise of such revenue due to the factual expansion of self-employment during recessions props up these firms’ net worth during 3

times of tight credit conditions. This limits the sharp rise in leverage that would otherwise occur during recessions, which brings the countercyclicality of leverage closer to the data. Of note, the mere fact of supplying idle capital to the self-employed is not sufficient: the countercyclicality of self-employment (which is linked to the demand for, and supply of, idle resources during recessions absent salaried opportunities) plays a key role in allowing the demand for capital by the self-employed to act as a buffer in salaried firms’ net worth. Such countercyclicality, which is supported by the data, only arises in the presence of input credit relationships (capital search), where the latter are a relevant source of financing for micro firms. All told, capital search frictions are important in rationalizing the cross-country link between self-employment and leverage dynamics. The mechanism just described becomes increasingly important in limiting net-worth contractions during recessions in economies with higher average self-employment shares because the equilibrium amount of resources needed to sustain such self-employment shares (via interfirm input credit) is larger, implying that salaried firms’ average revenue from supplying input credit to the self-employed in these economies is higher. Then, the higher average input-credit revenue in economies with more self-employment prevents firms’ net worth from contracting as much during downturns, which effectively lowers the countercyclicality of leverage in economies with higher average self-employment. Thus, the link between formal credit, input credit relationships, and selfemployment in our model provides a plausible explanation for the relationship between the countercyclicality of leverage and the average share of self-employment in the data. We show that additional predictions of the model, which underlie this mechanism, are supported by the data. Our main objective is to shed light on the mechanisms that may explain the link between labor market structures and cross-country differences in cyclical leverage dynamics we document in the data. Thus, our work is closest to three related literatures. First, to theoretical work on financial frictions and labor markets in the U.S. (Kiyotaki and Moore, 1997; Bernanke, Gertler, and Gilchrist, 1999, henceforth BGG; Smets and Wouters, 2007; Jermann and Quadrini, 2012; Monacelli et al., 2012; Chugh, 2013; Mumtaz and Zanetti, 2013; Mimir, 2016; among others).2 Second, to the literature on EME financial frictions and 2

Kiyotaki and Moore (1997), and BGG are canonical (one-sector) closed-economy models with financial

4

business cycles (Neumeyer and Perri, 2005; Gertler, Gilchrist, and Natalucci, 2007; among others).3 Third, to work on EME labor markets and business cycles (Finkelstein Shapiro, 2014; Fern´andez and Meza, 2015; Boz, Durdu, and Li, 2015; Finkelstein Shapiro and Mandelman, 2016; Epstein and Finkelstein Shapiro, 2017a,b).4 Closest to our work are Fern´andez and Gulan (2015) (henceforth FG, 2015), who are the first to characterize the average cyclical dynamics of private-sector leverage of EMEs in the data in a one-sector Walrasian labor market environment, and Finkelstein Shapiro (2014), who shows how input credit relationships in the form of capital search play a key role in generating the countercyclicality of self-employment in the data. Our contributions relative to existing work are fourfold. First, relative to FG (2015) who focus on average cyclical leverage dynamics and how the latter can rationalize the average countercyclicality of country interest rates in EMEs, we provide a new empirical fact that links the structure of credit and labor markets to differences in cyclical leverage dynamics across countries. Second, our work is the first to jointly capture factual cyclical credit and sectoral labor market dynamics in an environment that replicates a broad set of EME business cycle regularities. In contrast to existing work on financial frictions in EMEs, frictions. Jermann and Quadrini (2012) include financial frictions similar to Kiyotaki and Moore (1997) and show the importance of financial shocks for U.S. business cycles. Smets and Wouters (2007) estimate a medium-scale closed-economy DSGE model with nominal rigidities which includes, among other elements, a wedge that captures the external finance premium in a reduced-form way. Mimir (2016) characterizes the U.S. financial cycle, shows that bank leverage and lending-deposit rates in the U.S. are countercyclical, and finds that financial shocks in the banking sector play an important role for U.S. business and financial cycles. Chugh (2013) and Mumtaz and Zanetti (2013) expand one-sector versions of Kiyotaki and Moore (1997) and BGG to introduce standard labor search frictions and explore the importance of financial frictions for cyclical labor market dynamics (they abstract from heterogeneity in employment and external financing across firms). On the empirical front, Korajczyk and Levy (2003) present empirical evidence for the U.S. suggesting that firms’ financial strength (as measured by their constrained status) plays a relevant role in their capital structure as well as their timing and choices over external financing, implying that the cyclicality of firm leverage may be influenced by these factors. Levin, Natalucci, and Zakrajˇsek (2004) characterize the time series behavior of debt and credit spreads for U.S. publicly-traded firms, showing that risk premia (and, incidentally, leverage) are countercyclical, and that the data supports the presence of financial frictions. 3 Neumeyer and Perri (2005) highlight the role of financial shocks (embodied in foreign interest rate and country premia shocks) for EMEs, while Gertler, Gilchrist, and Natalucci (2007) use a SOE version of BGG to shed light on the Asian financial crisis. While these papers stress the relevance of financial frictions in EMEs, they abstract from the labor market structure and labor market dynamics. 4 These papers abstract from considering financial frictions or the joint behavior of labor and credit markets. Epstein and Finkelstein Shapiro (2017b) is the only exception. In contrast to our work (which is centered around the cyclicality of financial markets), theirs abstracts from characterizing the cyclical behavior of credit markets and focuses instead on how average differences in the degree of domestic financial development explains cross-country differences in labor market dynamics in the data.

5

heterogeneous labor markets are at the center of our analysis. Third, our model identifies a novel and empirically-supported channel that reconciles the link between average selfemployment shares and the cyclicality of leverage. Finally, we illustrate how the presence of self-employment has implications for unemployment dynamics amid cyclical credit-market interventions. Taken together, our results stress the relevance of the interaction between financial imperfections and the EME credit and labor market structure for better understanding aggregate fluctuations and policy outcomes in these economies. The paper is structured as follows. Section 2 provides a succinct overview of existing empirical evidence that supports our modeling approach. Section 3 describes the model. Section 4 presents the calibration. Section 5 discusses the quantitative results. Section 6 concludes.

2

Stylized Facts and Modeling Implications

2.0.1

Average Self-Employment and Self-Employment Countercyclicality

Table 1 shows that average self-employment in EMEs over the period 2000-2007 represent between 20 and 50 percent of non-agricultural employment.5 In contrast, advanced economies exhibit much lower (and less dispersed) shares. Similar patterns hold if we consider selfemployment as a share of the total labor force (OECD, 2009; World Development Report, 2013). Recent studies using available quarterly data for EMEs have shown that self-employment is on average countercyclical (see Bosch and Maloney, 2008; and Fern´andez and Meza, 2015). Furthermore, available evidence on labor flows suggests that this countercyclicality is explained by transitions from unemployment into self-employment (Bosch and Maloney, 2008).6 Given the scarcity of quarterly data on worker flows and sectoral employment in most EMEs, Loayza and Rigolini (2011) use annual data on self-employment and confirm the counter5

Considering a broader time frame does not change this fact. The cyclical correlation of output with self-employment in Mexico and Brazil, two EMEs with quarterly data on labor flows, are -0.42 and -0.22, respectively, using quarterly data. The cyclical correlation between output and the transition probability into self-employment from unemployment for Mexico (Brazil) is -0.43 (-0.60) (Bosch and Maloney, 2008). 6

6

cyclicality of self-employment in a vector correction model for a large sample of countries. Based on the same annual data, we find that the median cyclical correlation between selfemployment and real GDP in a broad sample of EMEs is -0.24.7 This stands in contrast to the average procyclicality of total salaried employment. While self-employment is generally categorized as informal employment in EMEs, our interest is not on informal employment per se but rather on the differences in the composition of employment across economies and leverage dynamics, irrespective of whether employment is covered by social security or subject to labor market regulations (a common definition of informal employment). Table 1: Share of Self-Employment (% of Non-Agricultural Employment) in Selected Economies Emerging Self-Employment Advanced Self-Employment Economies Share Economies Share Argentina 25.3 Greece 28.6 Brazil 35.7 Italy 25.6 Chile 32.0 Canada 15.6 Colombia 49.3 Australia 15.3 Ecuador 49.3 United Kingdom 12.7 Mexico 29.3 Ireland 12.6 Paraguay 41.7 The Netherlands 11.3 Peru 31.3 Germany 11.6 Uruguay 19.5 Japan 11.2 Venezuela 40.3 Finland 10.9 Korea 28.0 Austria 9.5 Malaysia 17.8 Sweden 9.2 Philippines 32.4 France 8.6 Thailand 36.0 Denmark 7.9 Turkey 25.6 United States 7.5 South Africa 11.6 Norway 6.7 Average 31.6 Average 12.8 Dispersion 9.45 (7.95) Dispersion 6.16 (2.72)

Source: OECD (2009). Notes: Self-employment is expressed as a percent of non-agricultural employment and is an annual average using years 2000 through 2007. The sample of countries follows the literature on EME business cycles. We use the standard deviation as a measure of dispersion. The numbers in parentheses present the dispersion of self-employment for the EME (advanced economy) country sample excluding South Africa (Greece and Italy). 7

Specifically, we consider an unbalanced panel with annual data on self-employment (specifically, the number of own-account workers, obtained from the ILO) for Argentina, Brazil, Chile, Colombia, Ecuador, Malaysia, Mexico, Peru, Philippines, South Africa, Thailand, Turkey, and Venezuela from 1990 to 2014. The cyclical correlation of self-employment is computed as the contemporaneous correlation of HP-filtered self-employment and HP-filtered real GDP (with smoothing parameter 100). The corresponding median cyclical correlation of self-employment in advanced economies is 0.08.

7

2.0.2

External Financing Among Large and Small Firms and Input Credit

As emphasized by several cross-country studies, while formal, salaried firms face credit frictions, they tend to use formal credit (credit from formal financial intermediaries) as their main source of external financing.8 In contrast, limited access to formal credit among micro firms is the norm in EMEs (Beck, 2007; Stein, Pinar Ardic, and Hommes, 2013; OECD, 2013). Therefore, these firms rely heavily on both household resources and input-based credit from suppliers (Table 2; Demirg¨ u¸c-Kunt and Maksimovic, 2001; Burkart and Ellingsen, 2004; IDB, 2005; Chavis et al., 2011; Global Financial Development Report, 2014, henceforth GFDR, 2014).9 This type of credit represents relationship finance and is fundamentally rooted in long-term relationships between suppliers and firms. Table 2: Consequences from Limited Access to External Formal Financing and Sources of Financing for Small Firms in Latin America Consequences from Limited Percent of Access to Formal Financing: Entrepreneurs Reduced Scale 56.0 Search for Support from Suppliers or Customers 51.0 Delay in Launching Enterprise 32.0 Sources of Financing Financing from Suppliers Financing via Purchase of Second-Hand Machinery and Equipment Financing from Customers

37.0 21.0 19.0

Source: IDB (2005, Tables 6.2 and 6.5). Notes: This evidence is based on a cross-country comparative survey of micro- and small-firm entrepreneurship in Latin America, East Asia, and Southern Europe. The percentage in each row denotes the share of firm owners surveyed in Latin America (1) for whom restricted access to financing has led to search for suppliers or partners and to delayed entry into the sector, and (2) the source of alternative financing due to restricted access to formal credit. Similar evidence holds for countries in East Asia. A similar table is presented in Epstein and Finkelstein Shapiro (2017a,b).

While comprehensive micro evidence on interfirm input credit relationships for many EMEs is limited, recent studies suggest that close to two thirds of external financing for 8

Larger firms tend to be formal. Table A6 and the accompanying discussion in the Appendix shows that the these firms tend to have access to formal (bank) credit, but also represent a small share of all firms in EMEs. Also, bond issuance is mostly available to the largest firms only. The latter represent a minuscule share of the universe of firms and account for a relatively small share of total employment in EMEs (de la Torre et al., 2012). 9 Despite the fact that trade credit is also important in advanced economies (Mateut et al., 2006), individuals with sole proprietorships in these economies have access to formal credit through a wider set of channels that are not available for most self-employed in EMEs, such as using credit cards and personal lines of credit with commercial banks to finance their ventures.

8

Mexican micro firms comes from input suppliers (Pav´on, 2010) (see Tables A4 and A5 in the Appendix). Interfirm input credit relationships are equally important for micro firms in Argentina, Brazil, and Costa Rica and in other regions (Navas-Alem´an, Pietrobelli, and Kamiya, 2012; IDB, 2005). The role of input credit as the primary source of external financing among micro firms holds for other EMEs (see Table A3 and the accompanying discussion in the Appendix). The literature also suggests that many of these relationships are established within similar industries or narrow production categories, which allows for better monitoring by input suppliers (Cu˜ nat and Garcia-Appendini, 2012). Finally, due to institutional deficiencies in many EMEs, interfirm credit generally takes the form of in-kind (or input) credit as opposed to cash-based credit (McMillan and Woodruff, 1999; Burkart and Ellingsen, 2004). The trade credit literature finds evidence of countercyclical movements in supplier trade credit (Cu˜ nat, 2007; Coulibaly, Sapriza, and Zlate, 2011). For example, Love, Preve and Sarria-Allende (2007) find that larger firms (with better access to formal credit) increased the provision of supplier credit during the Mexican and East Asian crises. Klapper and Randall (2011) find that many firms either continued to extend trade credit or increased their supply of trade credit to firms during the global financial crisis of 2008-2009 in several countries. Moreover, the economies that suffered the greatest recessions saw the largest share of firms extend trade credit, and a larger percent of firms that cited an expected reduction in their workforce extended trade credit relative to their counterparts that did not expect changes in their workforce. Cu˜ nat and Garcia-Appendini (2012) review existing literature and suggest that trade credit may have a liquidity-insurance component over the business cycle. Other studies have found that the leasing and renting of capital—another form of relationship-based credit—is also countercyclical (Gal and Pinter, 2017). To complement the above summary of existing studies on supplier credit, the Appendix presents a more comprehensive and extensive review of the literature on supplier and trade credit and firms’ access to external financing, as well as additional evidence for other countries, that broadly support our approach.

9

2.0.3

Cyclical Leverage Dynamics and Self-Employment Shares Across Countries

Measures of Leverage We consider two measures of leverage—one for non-financial corporate firms only (nf c) and a broader measure that considers all sectors (all)—and two different data sources that differ in their country sample and construction in order to confirm the robustness of our new fact. The first source for leverage is FG (2015) and only covers 12 EMEs, as most of the EME business cycle literature does. We focus on the period 1995Q3—the first year available in the FG (2015) leverage dataset—through 2007Q4 since the global financial crisis of 2008-2009 had heterogeneous effects across countries and regions.10 The second data source used to construct a leverage measure is the Bank for International Settlements (BIS). One benefit of using this source is that it allows us to consider a much larger sample of 24 countries that includes both advanced economies and EMEs. While we focus on EMEs given the prevalence of self-employment in these economies, including advanced economies provides additional variation and allows us to confirm that our new fact holds more comprehensively and beyond EMEs. Our country sample is now larger, but is still restricted by the availability of high-frequency data for the variables that are needed to construct a time series for leverage (hence the 24 countries). To construct a leverage time series for each country, we use country-level data from 1990Q1 through 2007Q4 on real credit to the non-financial sector from the BIS and a market capitalization measure, MSCI (obtained from Haver Analytics). Of note, restricting our country sample to EMEs and to 10

The patterns we present in Figure 1 below continue to hold if we extend the period under consideration to 2014, but the relationship becomes somewhat weaker. Despite this fact, the overall pattern remains significant. FG’s (2015) country-level leverage measures are based on weighted firm-level data from publiclytraded firms for each country, obtained from Bloomberg, where the firm weights are based on each firm’s relative market capitalization in their given country. Their country sample includes Argentina, Brazil, Chile, Colombia, Korea, Malaysia, Mexico, Peru, the Philippines, South Africa, Thailand, and Turkey, and is dictated by micro-level data availability and the literature on EME business cycles (Neumeyer and Perri, 2005; Uribe and Yue, 2006). The length of the time series on leverage varies by country. FG’s (2015) leverage measure is defined as the ratio of assets to equity, where equity is measured using market capitalization and assets are measured using a combination of the book value of debt and the market value of equity for each firm in the sample. While we follow existing studies and do not differentiate between short-term and longterm debt in our analysis due to data limitations, recent studies suggest that EMEs generally tend to borrow short-term (see, for example, Broner, Lorenzoni, and Schmukler, 2013, and Caballero, Fern´andez, and Park, 2016).

10

the period 1995Q1-2007Q4 yields a very similar pattern to those based on the FG (2015) dataset.11 Measures of Leverage Cyclicality and Self-Employment The cyclicality of leverage in a country is given by the contemporaneous correlation of the cyclical component of the log of each leverage measure and the cyclical component of the log of real GDP in that country, where the cyclical component of each series is obtained using a Hodrick-Prescott (HP) filter with smoothing parameter 1600 (real GDP is obtained from the IMF’s IFS database). In turn, we plot the cyclicality of leverage against average self-employment shares for each country (expressed as a percent of non-agricultural employment; OECD, 2009). Table 3 below provides a brief description of all relevant variables. Table 3: Variable Names, Descriptions, and Period Coverage Description Period FG Sample Leverage, Non-Financial Firms (nf c) 1995Q3-2007Q4 (Subfigures 1.1 and 1.3, Figure 1) levall Leverage, All (Financial and 1995Q3-2007Q4 Non-Financial) Firms (all) (Subfigures 1.2 and 1.4, Figure 1) Cyclical Corr(y,levj ) Contemporaneous Correlation 1995Q3-2007Q4 of HP-Filtered log Real GDP and log Leveragej , j ∈ (nf c, all) Self-Employment Average Self-Employment 2000-2007 (SE) (% of Non-Agric. Empl.) Variable levnf c

11

Period BIS Sample 1990Q1-2007Q4 1990Q1-2007Q4

1990Q1-2007Q4

2000-2007

Our BIS country sample includes Australia, Austria, Canada, Denmark, Finland, France, Germany, Greece, Italy, Japan, Netherlands, Norway, Portugal, Singapore, Spain, Sweden, United Kingdom, United States, Argentina, Brazil, China, India, Indonesia, Korea, Malaysia, Mexico, South Africa, and Turkey. The only countries that overlap with the FG (2015) sample are: Argentina, Brazil, Korea, Malaysia, Mexico, South Africa, and Turkey. The length of the time series available to compute leverage varies by country, with the shortest series being for Turkey (1998Q1-2007Q4). Table A1 in the Appendix presents more details regarding time coverage. The quarterly series for credit to the non-financial sector are based on the outstanding amount of credit supplied at the end of each quarter by domestic banks, non-residents, and the remaining sectors. This credit measure includes loans, debt securities, currency, and deposits, all at market value. all includes the private non-financial sector, comprised of non-financial corporations, households, and non-profit institutions. See http://www.bis.org/statistics/totcredit.htm for more details. Our leverage measure for each country is constructed using data on real credit to the non-financial sector (for either non-financial corporations or all sectors) expressed at market value, and the MSCI as follows: we normalize the series for real credit and the MSCI for each country so that the first observation of these two series for each country is 100. The quarterly leverage time series for each country is then constructed by taking the ratio of the normalized measure of real credit to the normalized MSCI measure in each quarter. Restricting our country sample to EMEs and to the period 1995Q1-2007Q4 yields a very similar pattern to those based on the FG (2015) dataset.

11

Notes: The time period for each country in the two country samples (FG or BIS) varies based on data availability. See Table A1 in the Appendix for more details regarding sources and specific time coverage. The cyclical component of the log of real GDP and the log of leverage are obtained using an HP filter with smoothing parameter 1600. See the footnote in Figure 1 for more details regarding country coverage in each sample and sources.

Main Results Figure 1 illustrates our new stylized fact: the countercyclicality of leverage is decreasing in the share of self-employment in the economy.12 Subfigures 1.1 and 1.2 present our new fact using the leverage data from FG (2015). In turn, subfigures 1.3 and 1.4 present the same fact using our constructed measure of leverage for the BIS sample of 24 countries, which includes both advanced economies and EMEs. In what follows, we focus on non-financial-firm leverage since input credit relationships generally take place between non-financial firms. The corresponding correlations in Figure 1 are statistically significant at conventional levels. One factor that could influence the cyclicality of leverage across economies is the average level of leverage. As Tables A2.1 and A2.2 in the Appendix confirm, the relationship between self-employment and the cyclicality of leverage continues to be significant even after controlling for the average level of leverage or the aggregate domestic credit share in both of our country samples. This validates our focus on self-employment as opposed to credit market development.13 Of note, it is well-known that self-employment is highly and negatively correlated with real GDP (a measure of the level of development). Then, our new fact could simply reflect a negative link between the level of development and cyclical leverage dynamics, with self-employment playing a minor role. 12 FG (2015) document that private-sector leverage is, on average, countercyclical in a sample of EMEs— the average cyclical correlation between private-sector leverage and output in their EME sample is -0.27 (the average correlation between Corporate Emerging Market Bond Index (CEMBI) spreads (a proxy for borrowing costs and domestic interest rates) and output is -0.61). However, they abstract from analyzing how this countercyclicality varies across countries, and more importantly, they do not consider the factors that may explain this variation, both of which are the main focus of our work. 13 Our main focus is on whether the correlation observed in the data holds even if we account for another plausible factor that may affect leverage dynamics rather than making any serious statistical inference since our sample of countries with available data (and therefore the number of observations) is relatively small. Figure A7 of the Appendix revisits our model-based analysis when we focus on differences in average leverage (and not on self-employment) across countries. As Figure A7 confirms, our model continues to outperform simpler model alternatives in replicating the change in the cyclicality of leverage across countries amid standard shocks.

12

Figure 1: Self-Employment and the Cyclicality of Leverage (1.1) Leverage

nfc

and SE, FG (2015) Sample

(1.2) Leverage

all

0.4 corr = 0.476

0.2 0 −0.2 −0.4 −0.6

0.2

all

Cyclical Corr(y,lev )

nfc

Cyclical Corr(y,lev )

0.4

0

10

20 30 40 Self−Employment (SE)

(1.3) Leverage

nfc

−0.2 −0.4 −0.6

50

and SE, BIS Data

10

20 30 40 Self−Employment (SE)

(1.4) Leverage

all

corr = 0.529

0.4 all

Cyclical Corr(y,lev )

nfc

0

50

and SE, BIS Data

0.6

0.4 Cyclical Corr(y,lev )

corr = 0.419

0

0.6

0.2 0 −0.2 −0.4 −0.6 −0.8

and SE, FG (2015) Sample

corr = 0.282

0.2 0 −0.2 −0.4 −0.6

0

10

20 30 40 Self−Employment (SE)

50

−0.8

0

10

20 30 40 Self−Employment (SE)

50

Sources: OECD (2009) (self-employment measured as a percent of non-agricultural employment; average using annual data from 2000 to 2007), FG (2015) (cyclical correlation of leverage with GDP), and BIS, IMF, and Haver Analytics (cyclical correlation of leverage with GDP). The list of economies in FG (2015) includes Argentina, Brazil, Chile, Colombia, Korea, Malaysia, Mexico, Peru, The Philippines, South Africa, Thailand, and Turkey. The period used to compute the cyclical correlation of leverage with output in this country sample is 1995Q3 to 2007Q4 (varies by country). The list of economies using the BIS data includes Australia, Austria, Canada, Denmark, Finland, France, Germany, Greece, Italy, Japan, Netherlands, Norway, Portugal, Singapore, Spain, Sweden, United Kingdom, United States, Argentina, Brazil, China, India, Indonesia, Korea, Malaysia, Mexico, South Africa, and Turkey. China, India, Malaysia, and South Africa do not have long-enough time series for non-financial-corporation leverage and are therefore excluded from Figure 1.3. The period used to compute the cyclical correlation of leverage with output in this country sample is 1990Q1 to 2007Q4 (varies by country; see the Appendix for details).

As part of our model-based quantitative analysis, we show that generating differences in output levels (say, via changes in exogenous productivity) in a model that abstracts from self-employment fails to quantitatively replicate the relationship between cyclical leverage 13

dynamics and output levels. A similar fact holds when we consider differences in average leverage alone. This suggests the presence of a deeper mechanism beyond simple differences in the level of economic development that potentially explain differing patterns in cyclical leverage dynamics, and supports our focus on the labor market structure. All told, Figure 1 confirms a robust negative relationship between self-employment and the countercyclicality of leverage. It is particularly noteworthy that the quantitative relationship between leverage dynamics and self-employment for non-financial firms (nfc) is very similar in subfigures 1.1 and 1.3, despite the differences in country samples and data sources for leverage. We interpret this as strong evidence for our new fact. 2.0.4

Modeling Implications

All told, we consider two types of financial structures. The first one is modeled after BGG to introduce leverage dynamics; it prevails among (larger) salaried firms, is rooted in formal credit markets, and is characterized by asymmetric information problems between borrowers and lenders. The second one is modeled after Finkelstein Shapiro (2014); it prevails among the self-employed, is characterized by a costly search process for input suppliers, and is reflected in capital search frictions to produce factual cyclical self-employment dynamics. Finally, we include standard labor search frictions to generate countercyclical unemployment (otherwise, the countercyclicality of self-employment by itself would produce procyclical unemployment). These ingredients are needed to explore the link between leverage dynamics and self-employment all while generating factual cyclical dynamics. If agency problems are important in formal credit markets, they may also play a role among self-employed firms, which tend to be more financially opaque. Our input credit structure abstracts from information asymmetries between the self-employed and their input suppliers since existing literature suggests that input credit relationships can reduce these information problems through closer monitoring and more effective enforcement mechanisms rooted in reputation costs, especially as input suppliers often operate in the same sector and therefore have better information than formal financial institutions (Burkart, Ellingsen, and Giannetti, 2011). In addition, the model is consistent with a focus on aggregate dynamics (and not on specific industry dynamics) while at the same time getting at the notion of 14

interfirm input-based credit relationships.

3

The Model

We incorporate frictional self-employment following Finkelstein Shapiro (2014) into a SOE RBC model with BGG frictions.14 There are two agents—salaried entrepreneurs and households— and four types of firms—(1) salaried firms (owned by salaried entrepreneurs); (2) selfemployed (one-person) firms (owned and operated by households); (3) capital producers (owned by households); and (4) matching firms (owned by households). Salaried and selfemployed firms produce the same tradable final good but differ in their production technologies and external financing sources. Output markets are perfectly competitive. Capital producers accumulate capital and sell it to salaried firms in frictionless markets. Salaried firms use salaried workers and capital to produce. They borrow external funds to finance the purchase of capital from capital producers—this is our notion of formal credit (i.e., credit provided by formal credit institutions). We assume that the supply of external funds comes solely from abroad. Asymmetric information between salaried firms and lenders imply that borrowed funds are subject to an external finance premium that is increasing in firms’ leverage. Importantly, in making input demand decisions, salaried firms also choose how much of their existing capital stock is used in production. Any remaining, unused capital becomes available as input credit for self-employed individuals via matching markets. Matching firms act as intermediaries between salaried firms and households in frictional capital and labor markets. Matching firms post vacancies to find salaried workers for salaried firms and purchase unused capital from these firms to match it with potential self-employment ventures. The self-employed use a single unit of matched capital to produce. Thus, the measure of matched capital also represents the measure of self-employed individuals. Labor and capital relationships are endogenously created but exogenously destroyed. A sketch of the model economy is presented in Figure A1 in the Appendix. 14

Our description of the BGG structure follows Gilchrist and Zakrajˇsek (2011). The microfounded optimal financial contract that underlies financial frictions is identical to BGG and is discussed in the Appendix for completeness. A key difference in our work is the inclusion of revenue from input credit in firms’ net worth. This does not imply a change in the nature of the optimal financial contract, and as such our setup is comparable to more standard models.

15

Capital Producers Capital producers choose investment it to maximize E0

P∞

t=0 Ξt|0 Πk,t

subject to profits Πk,t = [Qt (kt+1 − (1 − δ)kt ) − it ] and the capital production technology kt+1 = (1 − δ)kt + Φ (it /kt ) kt , where Ξt|0 is the discount factor (defined in the household’s problem below), Qt is the price of capital, and δ is the exogenous depreciation rate. Φ (it /kt ) is an investment adjustment cost function with Φ′ (·) > 0, Φ′′ (·) < 0. The solution to this problem yields the price of capital: Qt = [Φ´(it /kt )]−1 . Salaried Firms Salaried firms are operated by risk-neutral entrepreneurs who discount profits at rate β. They produce using a constant-returns-to-scale production function F (ns,t , λt kt ), where ns,t is salaried labor, kt is the firms’ existing capital stock, and λt is the fraction of kt used in production in period t. λt is a choice for firms. The remaining, unused capital (1 − λt )kt is sold to matching firms at price pk,t once λt is determined and becomes available as input credit for self-employed firms through a matching process between matching firms and households. Importantly, the revenue from selling unused capital to matching firms pk,t (1 − λt )kt is part of salaried firms’ total assets as an account receivable, and therefore a component of salaried-firm net worth (see below). Borrowed funds bt+1 = Qt kt+1 − nwt+1 are given by the difference between capital expenditures Qt kt+1 and firms’ net worth nwt+1 , where ∗ nwt+1 = φ[Rk,t Qt−1 kt − Ψt Rt−1 bt +

pk,t(1 − λt )kt {z } |

revenue from supply of unused

] + st , k

where φ denotes the exogenous survival probability of entrepreneurs, Rk is the return to ∗ capital, and Ψt Rt−1 is the ex-post cost of borrowed funds. Borrowed funds are subject to

an external finance premium Ψt+1 over the gross real foreign interest rate Rt∗ . Specifically, Ψt+1 = Ψ (Qt kt+1 /nwt+1 ) = (Qt kt+1 /nwt+1 )ν where ν > 0 is the elasticity of the external finance premium with respect to leverage Qk/nw, and Ψ(·) is taken as given by firms (Gilchrist and Zakrajˇsek, 2011). Finally, st is an endowment left behind by exiting entrepreneurs for entering entrepreneurs (this guarantees that the measure of entrepreneurs remains constant; BGG). Importantly, the above equation links net worth—and ultimately leverage—to selfemployment via input-credit revenue from selling unused capital, pk,t(1 − λt )kt . 16

At the end of period t − 1, firms have chosen the stock of capital kt , where this choice was based on the expected gross return on capital in t − 1, Et−1 Rk,t . At the beginning of period t, exogenous aggregate productivity is realized and firms choose salaried labor demand ns,t (purchased at price pn,t from matching firms), the fraction λt of the current capital stock kt that is used in period-t production, tomorrow’s capital kt+1 (purchased from P t capital producers at price Qt ), and borrowed funds bt+1 to maximize E0 ∞ t=0 β Πs,t where Πs,t = zt F (ns,t , λt kt ) − pn,t ns,t − Qt kt+1 + pk,t(1 − λt )kt   ∗ +Qt (1 − δ)λt kt + bt+1 − Et−1 Ψt Rt−1 bt .

Above, total salaried output is ys,t = zt F (ns,t, λt kt ) and zt denotes exogenous aggregate productivity. Revenue from selling unused capital pk,t (1 − λt )kt becomes available immediately after λt is determined and prior to production. Once production has taken place, firms sell back the stock of capital that they used in production to capital producers and obtain Qt (1 − δ)λt kt in revenue. Finally, firms receive bt+1 in borrowed funds in period t but must   ∗ repay Et−1 Ψt Rt−1 bt as a result of last period’s borrowing.

The solution to this problem yields an optimal demand for salaried employment, pn,t =

zt Fns ,t , which states that the price of a unit of salaried labor is equal to its marginal product; an optimal decision to sell capital to matching firms, pk,t = zt Fλk,t + Qt (1 − δ). Thus, the firm equates the marginal benefit of selling unused capital to its marginal cost, given by the opportunity cost of capital (comprised of salaried firms’ marginal product of capital and the value of capital after production);15 an optimal choice over tomorrow’s capital kt+1 , which can be expressed as

(1)

Et [Rk,t+1 ] ≡

Et [zt+1 Fλk,t+1 + Qt+1 (1 − δ)] , Qt

where Rk,t+1 is defined as the ex-post gross return on capital and is comprised of the marginal product of (effective) capital and the market value of capital net of depreciation; 15

Fλk,t is the partial derivative with respect to the second argument of the production function, λt kt , which represents the effective capital used in production. Note that the price of selling capital to matching firms differs from Q since, in contrast to more standard models with financial frictions, firms can allocate their existing capital stock between in-house production and matching firms, implying that the price of capital supplied to matching firms must incorporate salaried firms’ opportunity cost of capital in the current period.

17

and finally an optimal choice over borrowed funds, which can be expressed as Et [Rk,t+1 ] ≡ Et [Ψt+1 Rt∗ ] ,

(2)

where the right-hand-side is the ex-post cost of borrowing, a function of the external finance premium Ψ and the gross foreign interest rate R∗ . Denoting Et [Rk,t+1 ] as the cost of formal borrowing, the external finance premium creates a wedge between the cost of borrowed funds for salaried firms and the foreign interest rate. Except for the presence of revenue from selling unused capital in net worth and the choice over λ, this setup is identical to BGG. Matching Firms The inclusion of matching firms is purely expositional and allows us to consider BGG and search frictions separately even though there is a natural connection via input credit.16 Matching firms receive profits Πm,t and are owned by households. On the salaried side, matching firms receive revenue pn,t ns,t from selling matched salaried labor ns,t to salaried firms; they pay a Nash wage wt per salaried worker and post vacancies vs,t , where ψ is the fixed flow cost of posting a vacancy. Matched salaried labor separates with exogenous probability ρs . On the capital side, matching firms receive revenue rt ne,t from supplying one unit of capital per self-employed individual to existing self-employed individuals ne,t , where rt is the Nash capital rental rate. They also spend pk,tkm,t to purchase capital km,t from salaried firms, where km,t is then devoted to creating new capital matches with potential self-employed individuals. Matched capital separates with exogenous probability ρe , in which case it returns to matching firms. We assume that matching firms cover the depreciation of capital δ for surviving capital matches. Thus, matching firms ultimately receive (ρe − δ)ne,t from net capital separations every period. Finally, at the end of each period, firms can sell the capital they originally obtained from salaried firms (net of depreciation) back to capital producers at price Qt , but must net out a fraction (1 − ρe )qe,t of it as it was matched with endogenous probability qe,t (defined below). This assumption is inconsequential and ensures that the capital producers’ problem is standard. 16

See Zhang (2012) for the use of matching firms in an environment with financial imperfections.

18

Matching firms choose vacancies vs,t , desired salaried workers ns,t+1 , capital km,t , and P desired capital in self-employment ne,t+1 to maximize E0 ∞ t=0 Ξt|0 Πm,t subject to Πm,t = [pn,t ns,t − wt ns,t − ψvs,t ] + [rt ne,t − pk,t km,t ] +(ρe − δ)ne,t + [Qt (1 − δ) − (1 − ρe )qe,t ] km,t , and the perceived evolutions of capital in self-employment, ne,t+1 = (1−ρe ) [ne,t + km,t qe,t ] , and salaried employment, ns,t+1 = (1 − ρs ) [ns,t + vs,t qs,t ] . qs,t is the endogenous probability of finding a salaried worker. Wages, capital rental rates, and the probabilities qe,t , and qs,t are taken as given by matching firms. The solution to this problem yields a standard salaried job creation condition:

(3)

  ψ ψ s = (1 − ρ )Et Ξt+1|t pn,t+1 − wt+1 + , qs,t qs,t+1 and an optimal decision to supply capital to self-employed individuals:

(4)

pk,t − Qt (1 − δ) + (1 − ρe )qe,t (1 − ρe )qe,t

= Et Ξt+1|t

(

rt+1 + (ρe − δ)

+

pk,t+1 −Qt+1 (1−δ)+(1−ρe )qe,t+1 qe,t+1

)

.

Condition (4) equates the expected marginal cost—given by the cost per unit of purchased capital pk net of the opportunity cost of matching a unit of capital Q(1 − δ), and the value of an idle unit of matched capital that becomes productive next period, all adjusted by the matching probability—to the expected marginal benefit of matching capital—given by the Nash rental rate, the net value of separated capital, and the continuation value of the capital match. This equation links salaried formal credit market conditions to self-employment via the availability of unused capital, where the latter is reflected in pk and Q (see (??)). Households and Self-Employment

Following the search literature, there is a represen-

tative household with a large number of family members and perfect risk-pooling across household members. It spends resources to find capital to send a share of its unemployed members to self-employment. Successful capital (labor) matches with matching firms allow unemployed individuals to become self-employed (salaried workers). Each self-employed in-

19

dividual uses a single unit of matched capital and a fixed (inelastically-supplied) unit of his own labor to produce. Formally, the household chooses consumption ct , capital search expenP t ditures ve,t , desired self-employment ne,t+1 , and foreign debt b∗t to maximize E0 ∞ t=0 β u(ct ) subject to

∗ ct + κ (ve,t ) + Tt + Rt−1 b∗t−1 = (zt − rt )ne,t + b∗t + wt ns,t + Πk,t + Πm,t ,

and the perceived evolution of self-employment ne,t+1 = (1 − ρe ) [ne,t + ve,t fe,t ] .17 κ (ve,t ) represents the cost of searching for capital (i.e., the startup cost of self-employed firms), where κ′′ (·) > 0, κ′′ (·) ≥ 0, and Tt are lump-sum taxes. The gross interest rate on foreign debt is Rt∗ = R∗ [Θ(b∗t − b∗ )] , where Θ(b∗t − b∗ ) is an adjustment cost function that induces stationarity in debt holdings and b∗ represents steady-state foreign debt holdings (SchmittGroh´e and Uribe, 2003). While in principle foreign debt could allow individuals to self-finance their self-employment ventures, a capital match is still required to successfully move into selfemployment. Total self-employment output is ye,t = zt ne,t where zt is exogenous aggregate productivity. Each self-employed individual pays the Nash rental rate rt per unit of matched capital. Thus, earnings per self-employed individual are given by (zt − rt ). Households also receive salaried labor income wt ns,t from salaried workers and lump-sum profits Πm,t and Πk,t from ownership of matching firms and capital producers. The household’s perceived endogenous probability of finding capital is given by fe,t (defined below). Prices, profits from capital producers and matching firms, and matching probabilities are all taken as given. A decision to send individuals to self-employment obtains:    κ′t+1 κ′t uc (ct+1 ) e = (1 − ρ )Et β zt+1 − rt+1 + , fe,t uc (ct ) fe,t+1

along with a standard Euler equation for foreign debt: uc (ct ) = Et βRt∗ uc (ct+1 ). The

self-employment decision equates the expected marginal cost of searching for capital—the resource cost of searching for capital, adjusted for the matching probability—to the expected marginal benefit of doing so. The latter is given by individual self-employment earnings if the new match survives next period as well as the continuation value of the capital relationship. The stochastic discount factor used by capital producers and matching firms is 17

The household is also subject to a perceived law of motion for salaried employment. Given the absence of an explicit salaried labor supply decision, we abstract from including such law of motion.

20

Ξt|0 = β t uc (ct )/uc (c0 ). Matching Unemployment is given by ut = 1 − ns,t − ne,t where ns,t and ne,t denote the equilibrium measures of salaried employment and self-employment, respectively. The matching functions for salaried labor and capital, ms,t = ms (vs,t ,ut ) and me,t = me (km,t ,ve,t ), are constant-returns-to-scale, increasing and concave in each of their arguments. The job-finding (job-filling) probability fs,t (qs,t ) for salaried employment is equal to the ratio of salaried matches ms,t to unemployment ut (to salaried vacancies vs,t ). Both probabilities depend on labor market tightness θs,t = vs,t /ut . Analogously, the household’s (matching firm’s) capital matching probability fe,t (qe,t ) is given by the ratio of capital matches me,t to the household’s capital search expenditures ve,t (to the matching firm’s capital km,t ). Both probabilities depend on capital market tightness θe,t = ve,t /km,t . Nash Wage and Capital Rental Rates Wages wt and the capital rental rate rt are determined via bilateral Nash bargaining. The wage and capital rental rate equations, which are presented in the Appendix, are identical to those in Finkelstein Shapiro (2014). For expositional purposes, they can be expressed implicitly as follows: wt = Θ(θs,t , θe,t , mpls,t, mpks,t ) and rt = Θ(θs,t , θe,t , mpks,t ). Intuitively, the Nash wage is not only a function of the marginal product of salaried labor (mpls ) and salaried labor market conditions (θs ), but also of the self-employment outside option. The latter is reflected in the likelihood of entering selfemployment (embodied in θe and in salaried firms’ marginal product of capital mpks ). All else equal, higher capital market tightness implies a lower probability of entering self-employment and thus a reduction in the self-employment outside option, which in turn exerts downward pressure on wages. Similarly, the Nash rental rate r is not only a function of labor market conditions in self-employment (embodied in θe ), but also of salaried labor market conditions (embodied in θs ). All else equal, higher salaried labor market tightness improves the self-employed’s outside option and puts downward pressure on the capital rental rate. Closing the Model In equilibrium, matching firms’ demand for capital km,t = (1 − λt )kt . The government’s budget constraint is gt = Tt , where gt is exogenous government spending. The economy’s resource constraint is given by 21

∗ yt = ct + gt + ψvs,t + it + b∗t−1 Rt−1 − b∗t ∗ + [Ψt − 1] Rt−1 bt + cs,t +κ (ve,t ) + ne,t+1 − (1 − δ)ne,t ,

(5)

∗ where [Ψt − 1] Rt−1 bt captures bankruptcy costs arising from credit market imperfections

and cs,t = (1−φ)st denotes the net resources from exiting salaried firms. The last three terms embody the resources devoted to frictional capital markets (the resource cost of search and investment in new capital matches). Total output is yt = ys,t + ye,t and the current account ∗ is cat = yt − ct − it − gt − (Rt−1 − 1)b∗t−1 .18

4

Calibration

We assume that F (ns,t , λt kt ) = (ns,t )1−α (λt kt )α with 0 < α < 1 and u(ct ) = c1−σ /(1 − σ). t 1−ξ Matching is Cobb-Douglas: ms,t = Ms uξt vs,t and me,t = Me (km,t )ξ (ve,t )1−ξ , where Ms , Me ,

and ξ denote, respectively, each market’s exogenous matching efficiency and the common matching elasticity. Capital search expenditures are given by κ(ve,t ) = ψe (ve,t )ηe with ψe > 0 and ηe ≥ 1. The adjustment cost for foreign debt holdings is Θ(b∗t − b∗ ) = exp [ηb (b∗t − b∗ )] (Schmitt-Groh´e and Uribe, 2003). The investment adjustment cost is: Φ (it /k,t ) = it /kt − (ϕk /2) (it /kt − δ)2 , ϕk > 0. Aggregate productivity zt follows an AR(1) process: ln zt = iid

(1 − ρz ) ln z + ρz ln zt−1 + εzt , 0 < ρz < 1, where εzt ∼ N(0, σz ). Parameters from Existing Literature Mexico is our representative economy since it is one of the few EMEs with high quality data on labor flows. The time period is a quarter. The salaried capital share α is 0.32. The subjective discount factor β is 0.985. The capital depreciation rate δ is 0.025. The steady-state foreign interest rate R∗ is 1.015 and the utility function’s CRRA parameter σ is 2. The salaried and self-employment separation probabilities ρs and ρe are 0.05 and 0.02, respectively (Bosch and Maloney, 2008). Parameter ηb is set to 0.03, a value that ensures stationarity in debt holdings without affecting aggregate 18

Our results do not change if we assume a CES aggregator for total output (see Table A11 in the Appendix).

22

dynamics. Alternative values do not change our main conclusions. The convexity of capital search expenditures ηe is 1.1 (Finkelstein Shapiro, 2014). The search literature offers little guidance on the values of the bargaining powers and matching elasticities in EMEs due to data limitations that prevent an estimation of the matching function parameters. We initially set the bargaining power of salaried and self-employed workers, χs and χe , and the matching elasticity ξ to 0.50.19 The survival probability of entrepreneurs φ usually takes values above 0.97 in the advanced-economy literature. In an EME context, FG (2015) interpret φ as dividends transferred to shareholders, resulting in a value close to 0.91, which is a lower value relative to existing literature. We set φ to 0.93, consistent with existing evidence on the exit rate of Mexican firms (Bartelsman, Haltiwanger, and Scarpetta, 2009). Lower values of φ, as in FG (2015), do not change our main results and in fact improve the model’s ability to capture the cyclicality of leverage relative to our baseline calibration (see Table A8 in the Appendix). Calibrated Parameters We set the steady-state ratio of foreign debt to total (annual) output to 30 percent and the steady-state ratio of government spending to total output to 10.2 percent (Aguiar and Gopinath, 2007). The matching efficiency parameters replicate the average shares of salaried employment (0.72) and self-employment (0.23) based on Mexico’s employment survey (ENEU ).20 Given the absence of financial firms in our model’s formal credit structure and the evidence in Figure 1, we focus on leverage for non-financial firms.The elasticity of the external finance premium ν is then set to match an average leverage for non-financial firms for Mexico of 1.73 (FG, 2015). This target is consistent with evidence for Mexican non-financial firms as reported by the OECD. Following the financial frictions literature, we set the endowment of entering entrepreneurs s to a small number (equivalent to 1 percent of steady-state wages). The vacancy posting cost ψ represents 3.5 percent of wages (Levy, 2007). The capital 19

Table A8 in the Appendix shows that our results are robust to alternative and reasonable parameterizations for the bargaining powers. 20 The share of self-employment is somewhat lower relative to the one in OECD (2009). This is due to the fact that OECD (2009) reports self-employment as a share of non-agricultural employment, whereas we set ne to match the share of self-employment in the labor force. Our conclusions remain unchanged under a higher calibrated self-employment share.

23

search cost parameter ψe is set to three months of wages, in line with the average startup costs of Mexican microenterprises (McKenzie and Woodruff, 2006). Higher (lower) values for ψ (ψe ) do not change our conclusions. We consider data on the share of output from informal enterprises, which includes most of the output from self-employment and excludes most of the output from informal workers in formal firms, to set the steady state aggregate productivity such that the model matches a share of salaried output in total output in Mexico of 0.83.21 The implied value of z also generates a non-negligible differential in labor productivity between salaried (larger) and self-employed (smaller) firms, in line with the evidence. Finally, we use data from 1993Q1 to 2007Q4 for Mexico to obtain three secondmoment targets that we use to calibrate the investment adjustment cost and the volatility and the persistence of exogenous aggregate productivity. These targets are: the relative volatility of consumption and investment (1.13 and 3.06) and the volatility of total output (2.40).22 The resulting parameter values are: b∗ = 0.988, g = 0.084, Ms = 0.153, Me = 0.084, ν = 0.072, s = 0.006, ψ = 0.021, ψe = 0.597, z = 0.62, ρz = 0.964, σz = 0.0185, ϕk = 7.49. Of note, the resulting calibration implies that the share of total capital in the self-employment sector is 0.11. The corresponding share of total capital in micro firms based on Mexico’s census is 0.13 (Busso, Fazio, and Levy, 2012). Also, imperfect measurement of self-employment output in national accounts is inconsequential to our main conclusions since empirical measurements of unemployment explicitly take into account self-employment (ILO, 2013). Based on a standard SOE RBC model with BGG frictions, FG’s (2015) quantitative results suggest the need for a highly persistent transitory productivity process (ρz = 0.999) in order to generate an empirically-factual relative volatility of consumption greater than 1. Considering ρz as a calibrated parameter to match the relative volatility of consumption in our model is in line with their quantitative strategy.23 21 See http://wiego.org/sites/wiego.org/files/INEGI-Measuring-Informal-Economy-Mexico.pdf. Given our approach, the salaried-output share we choose represents a lower bound in the data. Our results would be even stronger if we consider a higher salaried output share. 22 All series are logged and HP-filtered using a smoothing parameter of 1600. The equilibrium conditions are log-linearized around the model’s steady state. We simulate the model for 2100 periods, remove the first 100 periods, apply the Hodrick-Prescott (HP) filter with smoothing parameter 1600 to the simulated series, and compute second moments as we would with real time series. 23 Given our framework’s features, our model does not need an extreme parameter value to match the volatility of consumption. Also, the resulting value of ρz = 0.964 in our baseline calibration is line with existing values used in the business cycle literature for Mexico.

24

5

Quantitative Results

Business Cycle Moments Table 4 presents selected business cycle statistics for the benchmark economy. Table 4: Business Cycle Statistics: Data vs. Benchmark Model and Alternative Models Targeted Data Benchmark Benchmark Self-Employment, Model Second Model Model, No No Capital Search without Moments Input Credit SE in nw σy 2.40 2.40 2.40 2.40 2.40 ∗ σc /σy 1.13 1.13 1.13 0.26 1.13 σi /σy 3.06 3.06 3.06 3.06 3.06 Non-Targeted Moments σu /σy 6.28 0.28 0.49 0.86 0.49 corr(ct , yt ) 0.94 0.99 0.99 0.97 0.99 corr(it , yt ) 0.94 0.98 0.99 0.57 0.98 corr(levt , yt ) −0.30 −0.64 −0.93 0.91 −0.94 corr(cat /yt , yt ) −0.79 −0.92 −0.85 0.41 −0.88 corr(ut , yt ) −0.85 −0.67 −0.68 −0.62 −0.73 corr(ne,t , yt ) −0.45 −0.40 −0.69 0.59 − corr(ns,t , yt ) 0.64 0.59 0.92 0.28 − corr(fu→e,t , yt ) −0.43 −0.96 −0.32 − − corr(Rt , yt ) −0.33 −0.37 −0.47 0.85 0.84 corr(yt , yt−1 ) 0.85 0.72 0.73 0.76 0.73 corr(ut , ut−1 ) 0.88 0.80 0.80 0.04 0.72 Note: A ∗ denotes a targeted second moment that cannot be captured with any plausible parametrization under our benchmark shock specification.

The labor and credit market facts on leverage and self-employment are similar for other EMEs (Bosch and Maloney, 2008; FG, 2015). The model successfully produces the countercyclicality of leverage and the current account-output ratio, of unemployment and selfemployment, and the transition probability from unemployment into self-employment (fu→e,t in Table 4), as well as the procyclicality of salaried employment and the high unemployment persistence in the data. Notably, recall that we only target three second moments and the only source of fluctuations are aggregate productivity shocks.24 The main goal of Table 4 is 24

The countercyclicality of leverage in the data is somewhat stronger if we expand the time frame to include the Great Recession. Also, the cyclical correlation between salaried employment and output in the data can range from 0.64 to 0.84 depending on the definition of salaried employment used (see, for example, Fern´andez and Meza, 2015). We focus on the lower bound for this second moment since our model does not differentiate between small salaried firms, where the procyclicality of salaried employment tends to be lower, and large firms, where the procyclicality of salaried employment tends to be higher. Our results remain unchanged if we consider imperfect measurement of self-employment output in the data.

25

to illustrate that our model captures the broad cyclical patterns in the data well, especially when it comes to the sign of the cyclical correlations of the labor and credit markets. Given the parsimonious shock specification and the rich labor market and credit structure, it is noteworthy that the model does well in quantitatively matching several second moments. There are three connections between formal credit frictions and labor markets in our model. First, the external finance premium influences salaried firms’ demand for borrowed funds and workers, and therefore labor market conditions via salaried labor demand. Second, salaried firms’ input demand decisions determines the availability of capital for new selfemployment ventures, and therefore labor market conditions via changes in the ease of entry into self-employment. Third, the availability of capital for self-employment affects firms’ net worth and ultimately the salaried firms’ leverage. To understand the model’s success in matching the countercyclicality of self-employment and leverage in the data, consider a negative shock to aggregate productivity z.25 The shock reduces salaried firms’ demand for capital and labor and capital usage, which lowers investment and the price of capital Q. The reduction in Q puts downward pressure on net worth and upward pressure on the external finance premium, causing borrowing costs to increase and leading to further reductions in salaried firms’ input demand and capital usage (to see this, consider expressions (1) and (2) jointly).26 This is the standard BGG mechanism. The temporary drop in z has two important and opposing effects on the availability of capital for self-employment. On the one hand, the shock reduces the marginal product of capital among salaried firms. This implies a reduction in the opportunity cost of supplying capital for matching (embodied in both zFλk and Q) and all else equal an increase in the availability of capital for household members to enter self-employment via a reduction in λ. On the other hand, the shock causes the Nash rental rate r to fall, which reduces the return from matching capital and all else equal lowers the incentive to supply capital to potential self-employed individuals. To see why self-employment ultimately expands during 25

Impulse responses are shown when discussing the policy implications in Figure 4 below. In principle, salaried firms could react to the rise in borrowing costs by varying the usage of their existing capital stock. Note, though, that more expensive external funds implies that the future capital stock will be lower, which reduces firms’ input demand, including capital usage λ. 26

26

recessions, note that vacancies and labor market tightness θs fall on impact. This worsens the self-employed’s outside options and puts upward pressure on the Nash rental rate such that in equilibrium r falls by less than salaried firms’ opportunity cost of capital. Then, in relative terms, salaried firms find it more attractive to supply unused capital for matching than to keep it in-house, despite the fact that aggregate productivity is lower across all firms. This results in a higher equilibrium transition probability from unemployment to self-employment and in an expansion of self-employment, as in the data (see Finkelstein Shapiro, 2014). Importantly, in the presence of financial frictions, the fact that salaried firms supply a larger portion of their existing capital stock k to matching firms relative to trend boosts salaried firms’ additional source of revenue from selling unused capital and props up these firms’ net worth nw. The relative boost in net worth from the rise in input credit revenue partially offsets the drop in net worth that initially resulted from a reduction in the price of capital Q, thereby making leverage less countercyclical than otherwise. This mechanism is tightly linked to: (1) the self-employed’s reliance on input credit relationships, (2) their countercyclicality, and (3) the inclusion of input-credit revenue in salaried firms’ net worth, and allows the model to generate cyclical leverage dynamics closer to the data relative to models that abstract from this mechanism and that generate more countercyclical leverage (see Table 4). A notable limitation of our framework, which is a well-known limitation of the majority of search models in general, lies in our framework’s inability to generate high unemployment volatility. This is a reflection of the well-known Shimer puzzle (Shimer, 2005). In fact, the presence of a financial accelerator hardly generates any amplification in unemployment when we discipline the model to match the steady-state leverage in the data. Of note, the main objective of our work is not to solve the Shimer puzzle. As such, exploring alternative mechanisms that might generate higher unemployment volatility in an EME context is left for future work.27 Table 4 also compares our benchmark model to: (1) a version of our benchmark model 27

As discussed in the Appendix, adopting a calibration similar to Hagedorn and Manovskii (HM) (2008), results for which are presented in Table A8, increases unemployment volatility but yields low consumption volatility and counterfactual self-employment dynamics, suggesting that a different approach is needed. For a plausible way to generate factual labor market and business cycle dynamics along with high unemployment volatility in an EME context, see Finkelstein Shapiro and Mandelman (2016) and Finkelstein Shapiro (2017).

27

where revenue from input credit is absent in net worth; (2) a version of our model without capital search (the self-employed continue to use capital, but the latter is obtained in frictionless markets); and (3) a model without self-employment. These comparisons are meant to highlight the importance of all the ingredients in our benchmark model for providing a better overall fit with the data. Excluding revenue from input credit from net worth still provides a good fit with the data but implies a countercyclicality of leverage that is much higher relative to both the data and our benchmark model for the reasons outlined above. Also, as shown below, the presence of input-credit revenue in firms’ net worth is critical for replicating the change in leverage dynamics as average self-employment increases (recall Figure 1). Column 5 in Table 4 compares the benchmark model to a framework that abstracts from capital search frictions: this column presents clear evidence that supports the inclusion of long-lasting input credit relationships. Indeed, without capital search, the model fails to deliver many of the facts in the data, including the countercyclicality of self-employment, leverage, the current account-output ratio; the persistence of unemployment; the relatively high procyclicality of salaried employment in the data; and a relative volatility of consumption greater than one.28 Intuitively, in the absence of capital search, the marginal cost of entering self-employment is unaffected by changes in the relative ease of entry into self-employment as in the benchmark model and depends more strongly on aggregate productivity, leading to procyclical self-employment. At the same time, salaried firms can readily reallocate their capital without facing any frictions, which makes net worth less procyclical and leverage procyclical. This ultimately explains the poor fit of the model without capital search. For completeness, we also show the results from a simpler model with a homogeneous frictional labor market that abstracts from self-employment (column 6 in Table 4). The model performs similarly well to the benchmark model, except for the fact that it generates stronger countercyclical leverage dynamics relative to both the data and our benchmark model. To 28

Without capital search, households spend resources to send individuals to self-employment, but they do not require a capital match. Instead, each self-employed individual rents a single unit of capital from salaried firms in frictionless capital markets. The only friction preventing unemployed individuals from instantaneously transitioning into self-employment is a timing assumption (required to be consistent with the timing in our benchmark model). There is no plausible parametrization for this alternative framework that can generate high consumption volatility (hence the ∗ for σc /σy in column 5).

28

be consistent with the presence of capital usage (λ) in the benchmark model and to have a more appropriate comparison, we also experiment with a version of this simpler model that allows for variable capital utilization (see, for example, Gertler, Gilchrist, and Natalucci, 2007). This model version generates procyclical leverage dynamics, which is inconsistent data.29 All told, Table 4 confirms that our framework successfully captures the general cyclical behavior of credit and labor markets alongside the cyclical patterns of more conventional macro variables in EMEs well. Notably, this takes place within a parsimonious calibration strategy and shock specification.30 In what follows and given the fact that our benchmark model performs best best relative to other plausible alternatives in capturing key features of the data we are interested in, we use the model to explore the relationship between cyclical leverage dynamics and self-employment documented in Figure 1. Leverage Dynamics and Self-Employment Shares To operationalize our experiment, we compute the cyclicality of leverage for different steady-state self-employment equilibria. We follow Finkelstein Shapiro (2014) and introduce a salaried-firm productivity parameter as such that salaried production becomes zt as F (ns,t , λt kt ), where as = 1 in our baseline calibration. In our model, changes in as affect the steady-state allocation of the capital stock between salaried production and capital available for the self-employed: increasing as lowers steady-state self-employment. Parameter as can be broadly interpreted as capturing the (normalized) level of institutional quality in a our baseline economy relative to other economies. Institutional quality is not only strongly associated with differences in average self-employment shares across economies but also has a more decisive influence on the productivity and operations of (larger) salaried firms (Akyol and Athreya, 2009).31 Of note, we 29

Results available upon request. As shown in Table A10 in the Appendix, adding net worth shocks improves the fit of our model with respect to leverage while maintaining the good overall fit with the data. The Appendix presents additional results where we explore the inclusion of other shocks in the benchmark model. 31 As supporting evidence that this parameter can be interpreted as embodying institutional quality, we use the Rule of Law measure from the World Bank’s Worldwide Governance Indicators as a proxy for institutional quality and find that the correlation between self-employment and Rule of Law in the data is -0.75 and strongly significant. As an alternative approach, since as is ultimately a productivity parameter in the model, we use data on TFP from the Penn World Tables and find a strong negative correlation between TFP and average self-employment (-0.65) in our sample of advanced economies and EMEs. We consider TFP prior to year 2000 such that our measure of TFP precedes the years we use to compute average self30

29

focus on the implications of differences in average self-employment for cyclical leverage dynamics and do not intend to explain the underlying causes of differences in self-employment across economies. As such, the way in which we engineer changes in steady-state selfemployment is not critical for our results. Indeed, changing other parameters that can also affect the size of self-employment (the cost of posting salaried vacancies, which can proxy for labor market regulations, or the bargaining power parameters for workers) does not change any of our conclusions. In what follows, we change parameter as to generate steady-state self-employment shares consistent with each of the countries in our EME sample, and then compute the cyclicality of leverage for each steady-state self-employment share. The model prediction is then based on a regression of the model-generated measure of leverage cyclicality on the average selfemployment shares in our country sample. Figure 2 shows that the model does surprisingly well in capturing the change in the countercyclicality of leverage as the average self-employment share changes for the sample of EMEs used by FG (2015). Since, as shown in Figure 1, the quantitative relationship between non-financial-firm leverage dynamics and self-employment is very similar in the larger BISbased set of countries, it is not surprising that our model performs equally well in matching the patterns if we include both advanced economies and EMEs. This is confirmed in Figure A3 in the Appendix. Since the model generates higher leverage countercyclicality relative to the data in our baseline calibration (compare corr(levt , yt ) in columns 2 and 3 in Table 4), the benchmark model does not quantitatively match the intercept of the regression in the data. However, this is not a problem per se given our main goal of explaining crosscountry differences in leverage dynamics. More important is the fact that the model traces the change in leverage cyclicality exceedingly well. Moreover, allowing for i.i.d. net worth shocks which, as already mentioned improves the quantitative fit of the model with respect to the cyclicality of leverage (see Table A10 in the Appendix), provides a better overall fit with the data without changing our main conclusions. Importantly, assuming that input-credit employment (2000-2007), but the same conclusions hold if we use average TFP for years 2000-2007. For related evidence linking productivity differences and the composition of employment specifically as it relates to informal employment (which most of of the self-employed belong to), see D’Erasmo and Moscoso Boedo (2012).

30

revenue is not a component of salaried firms’ net worth implies a virtual complete disconnect between leverage dynamics and the share of self-employment (see the dashed black line in Figure 2). Figure 2: Self-Employment and the Cyclicality of Leverage: Data vs. Model

0.3 0.2 0.1

Cyclical Corr(y,levnfc)

0 −0.1 −0.2 −0.3 −0.4 −0.5

Benchmark Model Benchmark with nw Shocks Model No Input Credit Model No SE Reg. Line FG (2015) Data Countries

−0.6 −0.7 −0.8 −0.9 10

15

20

25 30 35 Self−Employment

40

45

50

It is well-known that self-employment and the level of development are negatively correlated (Loayza and Rigolini, 2011). Generating differences in steady-state output across economies via the same parameter as in a model that abstracts from self-employment produces a similar disconnect, in this case between output levels and cyclical leverage dynamics (see the bold black line in Figure 2, where for this particular model, higher average selfemployment corresponds to a lower output level since by construction self-employment does not exist). This confirms that the level of economic development (as proxied by the level of total output) is not behind the link between self-employment and cyclical leverage dynamics, despite the strong (negative) relationship between self-employment and the level of development in the data. This is consistent with the results in Tables A2.1 and A2.2 in the Appendix. The inability of alternative models to replicate the facts in Figure 1 further 31

validates our focus on frictional self-employment and therefore our benchmark framework. Intuition and Empirical Validation Sustaining higher average self-employment shares requires allocating more salaried-firm (and economy-wide) resources towards input credit for the self-employed in steady-state. Hence, the contribution of input-credit revenue to salaried firms’ net worth is larger in economies with higher average self-employment. Importantly, this holds despite the fact that these economies have smaller capital stocks. Coupled with the countercyclicality of self-employment (which is key for input-credit revenue to limit the fall in net worth during recessions), the larger influence of input-credit revenue on firm net worth implies that firms acting as input suppliers have an alternative source of revenue that not only contributes to these firms’ financial health (as measured by net worth), but is also increasingly relevant in limiting the deterioration of salaried firms’ net worth during recessions in economies where self-employment is more prevalent. This provides a rationale behind the fact that economies with higher average self-employment exhibit less countercyclical leverage. Our model points to a critical connection between net worth, input credit, and selfemployment. While Figure 2 confirmed that the model can capture the relationship between leverage dynamics and self-employment in the data well, the unavailability of firm-level data on input credit and net worth for a large sample of countries prevents us from confirming whether the mechanism that explains the data in our framework is in fact true. However, our model does provide an additional, testable prediction that provides indirect evidence in favor of the channel through which self-employment affects net worth (and therefore leverage) in our framework: the latter predicts that economies with higher self-employment shares exhibit higher steady-state net worth-output ratios. In turn, these higher ratios are a direct reflection of the higher input-credit revenue that salaried firms receive in economies with larger average self-employment shares. This implies that net worth-output ratios provide critical information on the link between self-employment, input-credit revenue, net worth, and leverage dynamics. To determine whether our model’s prediction with respect to net worth-output ratios holds, we use publicly-available data on the share of financial net worth of non-financial corporations as a percent of GDP for a sample of 21 OECD countries. Our country sample

32

is restricted by the availability of data on financial net worth-output ratios for non-financial firms.32 Despite the fact that this country sample does not include EMEs due to data availability on net worth-output ratios, there is enough variation in the self-employment shares in the sample of countries we have to be able to characterize a pattern in the data. We do not consider this to be a severe limitation since, as shown in the bottom part of Table A2.2 in the Appendix, the link between leverage dynamics and self-employment is present among advanced economies as well. Figure 3: Self-Employment and Net Worth-GDP Ratios SE and Net Worth/GDP, 2000

SE and Net Worth/GDP, 2003 −40 Net Worth/GDP, 2003

Net Worth/GDP, 2000

−50 −100 −150 −200 −250

0

5

10 15 20 Self−Employment (SE)

25

−80 −100 −120 −140 −160

corr = 0.19 −300

−60

30

−180

corr = 0.165 0

SE and Net Worth/GDP, 2000−2007

25

30

2 No Input Credit Revenue

−60

1.9

−80

Net Worth/GDP

Net Worth/GDP, 2000−2007

10 15 20 Self−Employment (SE)

SE and Net Worth/GDP: Models

−40

−100 −120 −140

1.8 1.7 1.6

−160 −180

5

Benchmark Model

corr = 0.13 0

5

10 15 20 Self−Employment (SE)

25

1.5

30

32

0

5

10 15 20 Self−Employment (SE)

25

30

The countries in our sample include: Australia, Austria, Belgium, Canada, Chile, Denmark, Finland, France, Germany, Greece, Hungary, Ireland, Italy, Japan, Netherlands, Norway, Portugal, Spain, Sweden, United Kingdom, and United States. Self-employment shares in this country sample range from 6.7 percent of non-agricultural employment in Norway to 29 percent in Greece. The data is available at: https://stats.oecd.org, under National Accounts¿Financial Dashboard¿Financial Indicators – Stocks¿Financial Net Worth of Non-Financial Corporations, as a percent of GDP. To the best of our knowledge, similar, comparable data does not exist for EMEs. We consider data from 2000 to 2007, consistent with the sample period used above, to compute the average share of financial net worth of non-financial corporations for each country, restricting the country sample to countries for which we have data on selfemployment.

33

As Figure 3 illustrates, the correlation between self-employment and the net worth-output ratio in the data is positive and ranges from 0.20 using data from the early 2000s to 0.13 if we consider average net worth-output ratios over the 2000-2007 period. More importantly, as also shown in Figure 3, in a version of our model where revenue from input credit is not a component of firm net worth, the correlation between steady-state self-employment and the net worth-output ratio is negative and therefore inconsistent with the data (see the red line in the subfigure on the bottom right-hand corner of Figure 3). Taken together, this evidence, while based on a limited set of countries due to data limitations, provides indirect support for the mechanism in our model that explains the differences in cyclical leverage dynamics across economies with different self-employment shares. Cyclical Credit Policies and Unemployment

Our model also sheds light on the impli-

cations of cyclical credit policies that limit credit-market stress for unemployment dynamics (for evidence on the use of these policies, see Federico, Vuletin, and V´egh, 2014). Figure 4 compares the response of the benchmark economy (dashed red line) after a negative aggregate shock to an identical economy under a policy τt that limits the rise in the external finance premium during recessions (solid blue line). Specifically, τt is such that   ∗ the cost of borrowing Rk,t = Ψt Rt−1 τt where τt = exp [η (bt /b − 1)], η > 0. This policy captures the effect of credit-market interventions on firm borrowing premia. For illustrative purposes, we set η so as to halve the volatility of salaried-firm borrowed funds. Under the policy, the contraction in borrowed funds, salaried capital and labor demand, and the price of capital are more subdued. Thus, net worth is initially more resilient and investment exhibits a smaller contraction on impact and recovers earlier. While the policy benefits the salaried sector by limiting its contraction, the policy also pushes salaried firms to keep more of their capital in-house. As such, the rise in the availability of capital for self-employment during the downturn is smaller, leading to a smaller expansion in selfemployment.

34

Figure 4: Response to a Negative Aggregate Productivity Shock

% Dev. from SS

Total Output

Salaried Output

0

0

−0.5

−0.5

−1

−1

−1.5

−1.5

0 −0.5 −1

−2

0

10

20

30

−2

−1.5 0

Investment % Dev. from SS

−2

−4

−6

0

10

20

10

20

−2

30

0

Consumption

0

30

0.5

−0.5

0

−1

−0.5

−1.5

−1

−2

−1.5

−2.5

0

10

20

−2

30

0

Self−Employment

−0.1

10

20

30

Net Worth

0

Salaried Employment 0 % Dev. from SS

Self−Employment Output 0.5

10

20

30

Unemployment

1

0.8

0.8

0.6

0.6

0.4

0.4

0.2

−0.2 −0.3 −0.4

0

10 20 Quarters

30

0.2

0

0

−0.2

0

10 20 Quarters No Policy

30

0

10 20 Quarters

30

With Policy

Ultimately, the smaller contraction in salaried employment is insufficient to offset the smaller expansion in self-employment, resulting in a larger and more persistent rise in unemployment relative to an economy without policy. As such, there is a tradeoff between stabilizing output (as well as consumption and investment) and unemployment. As shown in Figure A5 in the Appendix, in the absence of self-employment, the policy would not lead to a sharper increase in unemployment. This suggests that policy prescriptions in environments where self-employment is prevalent may differ from those in economies where self-employment rates are smaller. Rather than downplaying the benefits of cyclical credit policies, our results point to the usefulness of our framework for, among other things, gaining a deeper and more comprehensive understanding of the combination of credit-market and

35

active labor market policies in economies with non-negligible self-employment shares.33 A Note on Small Firms in EMEs For tractability, we assumed that all salaried firms are able to borrow external funds. In reality, only a fraction of salaried firms in EMEs (mostly medium and larger firms) are able to access formal financing (GFDR, 2014). Separating salaried firms with access to formal financing from (smaller) salaried firms without access (but dependent on informal financing from input suppliers, as the evidence suggests) would not change our conclusions.34 Such modification would bring the cyclicality of leverage in the model closer to the data by expanding the input-credit-revenue base of salaried firms acting as input suppliers. Additional Results and Robustness Checks The Appendix presents results for a number of modifications to our benchmark model, including household heterogeneity, imperfect substitutability between sectoral output, the inclusion of additional shocks, and alternative parameterizations of the model (including a calibration in the spirit of Hagedorn-Manovskii (2008)), among others. Our main conclusions remain unchanged. For completeness, Figure A7 in the Appendix presents an additional experiment where we engineer changes in steady-state leverage and compare our model’s predictions regarding the negative link between average leverage and the cyclicality of leverage to the data. This experiment confirms that, amid standard productivity shocks, our model can quantitatively replicate the change in leverage cyclicality as average leverage changes.35

6

Conclusion

Emerging economies (EMEs) differ from advanced economies in three important ways: only a fraction of (salaried) firms have access to formal credit; self-employed firms account for an 33

The Appendix discusses the implications of a similar experiment with household heterogeneity. See Epstein and Finkelstein Shapiro (2017a,b) for possible ways to introduce firm heterogeneity in a setting with sectoral employment dynamics consistent with EMEs in a tractable way. 35 FG (2015) conduct a similar experiment by varying the death rate of BGG entrepreneurs in their model. For our purposes, a natural way to engineer changes in steady-state leverage is to change ν (the elasticity of the external finance premium). As shown in Figure A7, a simpler model without self-employment qualitatively captures the empirical link between average leverage and leverage cyclicality, but faces limitations in replicating the quantitative change in cyclicality as average leverage changes. 34

36

important share of total employment but lack access to formal credit; these firms depend on input-based credit relationships with other firms to operate. We document a new empirical fact: economies with larger self-employment shares exhibit less countercyclical private-sector leverage. We build a small-open-economy business cycle search model where formal credit frictions, input credit relationships, and the EME labor market structure interact that rationalizes this fact. Our framework jointly captures important stylized facts of cyclical credit and labor market dynamics in an environment that replicates a broad set of EME business cycle regularities, and successfully generates the relationship between the share of self-employment and leverage dynamics in the data. Using the model, we identify a plausible and empirically-supported channel that reconciles such relationship and highlight the importance of informal financing for understanding credit fluctuations across countries. Our framework also offers relevant policy implications regarding the tradeoffs between stabilizing credit markets and unemployment in economies where self-employment is prevalent. In order to have a tractable environment where we explore the role of the labor market structure and self-employment amid financial frictions in a transparent way, our framework abstracted from, among other things, heterogeneity in access to different types of formal external finance across salaried firms, endogenous salaried-firm entry, imperfect competition in the goods market, a richer input-output structure that allows for intermediate-goods trade credit supply chains, and nominal rigidities. These features can be incorporated into our framework in order to deepen our understanding of trade credit supply chains, labor market fluctuations, and price dynamics in economies where informal and formal finance interact, to characterize the interaction between conventional monetary policy and macroprudential policies that promote financial stability in EMEs, and to explore possible complementarities between cyclical labor market interventions and sectoral credit policies that promote financial deepening. We plan to pursue these issues in future work.

7

Acknowledgements

We thank We thank Joshua Aizenman, two anonymous referees, Jorge Rold´os, Pau Rabanal, Alejandro Izquierdo, Andrew Powell, Andr´es Fern´andez, Carlos Viana de Carvalho, Javier 37

Garc´ıa-Cicco, Ryan Chahrour, Susanto Basu, Fabio Schiantarelli, Jaromir Nosal, participants in IADB-JIMF 2014, Boston College, the World Bank, and the 12th Dynare Conference for comments and suggestions. The opinions expressed are our own and do not necessarily reflect the views of the International Monetary Fund, its Board of Directors, or the countries they represent. Any errors are our own.

38

References [1] Aguiar, Mark, and Guita Gopinath. 2007. “Emerging Market Business Cycles: The Cycle is the Trend,” Journal of Political Economy, Vol. 115, No. 1, pp. 69-102. [2] Akyol, Ahmet, and Kartik Athreya. 2009. “Self-Employment Rates and Busines Size: the Roles of Occupational Choice and Credit Frictions,” Annals of Finance, Vol. 5(3–4), pp. 495–519. [3] Allen, Franklin, Elena Carletti, Jun Qian, and Patricio Valenzuela. 2012. “Financial Intermediation, Markets,and Alternative Financial Sectors,” Working Paper 12-05, Wharton Financial Institutions Center. [4] Bartelsman, Eric, John Haltiwanger, and Stefano Scarpetta. 2009. “Measuring and Analyzing Cross-Country Differences in Firm Dynamics,” in Producer Dynamics: New Evidence from Micro Data, Ed. Timothy Dunne, J. Bradford Jensen, and Mark J. Roberts: University of Chicago Press. [5] Beck, Thorsten H.L. 2007. “Financing Constraints of SMEs in Developing Countries: Evidence, Determinants, and Solutions,” In Financing Innovation-Oriented Businesses to Promote Entrepreneurship. Unknown Publisher. [6] Bernanke, Ben S., Mark Gertler, and Simon Gilchrist. 1999. “The Financial Accelerator in a Quantitative Business Cycle Framework,” in J.B . Taylor and M. Woodford, Eds., Handbook of Macroeconomics, Vol. 1, Elsevier, Chapter 21, pp. 1341-1393. [7] Bosch, Mariano, and William Maloney. 2008. “Cyclical Movements in Unemployment and Informality in Developing Countries,” mimeo. [8] Boz, Emine, Ceyhun Bora Durdu, and Nan Li. 2015. “Emerging Market Business Cycles: The Role of Labor Market Frictions,” Journal of Money Credit and Banking, Vol. 47(1), pp. 31-72. [9] Broner, Fernando A., Guido Lorenzoni, and Sergio L. Schmukler. 2013. “Why Do Emerging Economies Borrow Short Term?,” Journal of the European Economic Association, Vol. 11, pp. 67–100. [10] Burkart, Mike, and Tore Ellingsen. 2004. “In-Kind Finance: A Theory of Trade Credit,” American Economic Review, Vol. 94, No. 3, pp. 569-590. [11] Burkart, Mike, Tore Ellingsen, and Mariassunta Giannetti. 2011. “What You Sell Is What You Lend? Explaining Trade Credit Contracts,” Review of Financial Studies, Vol. 24(4), pp. 1261-1298. [12] Busso, Mat´ıas, Mar´ıa Victoria Fazio, and Santiago Levy. 2012. “(In)formal and (Un)Productive: The Productivity Costs of Excess Informality in Mexico,” IDB Working Paper Series No. IDB-WP-341.

39

[13] Caballero, Juli´an, Andr´es Fern´andez, and Jongho Park. 2016. “On Corporate Borrowing, Credit Spreads, and Economic Activity in Emerging Economies: An Empirical Investigation,” IDB Working Paper Series IDB-WP-719. [14] Chavis, Larry, Leora Klapper, and Inessa Love. 2011. “The Impact of the Business Environment on Young Firm Financing,”World Bank Economic Review, Vol. 25 (3), pp. 486-507. [15] Christiano, Lawrence J., Mathias Trabandt, and Karl Walentin. 2011. “Introducing Financial Frictions and Unemployment in a Small Open Economy Model,” Journal of Economic Dynamics and Control, Vol. 35, Issue 12, pp. 1999-2041. [16] Chugh, Sanjay. 2013. “Costly External Finance and Labor Market Dynamics,” Journal of Economic Dynamics and Control, Vol. 37, Issue 12, pp. 2882-2912. [17] Coulibaly, Brahima, Horacio Sapriza, and Andrei Zlate. 2011. “Trade Credit and iIternational Trade during the 2008-09 gGlobal Financial Crisis,” Board of Governors of the Federal Reserve System International Finance Discussion Papers Number 1020. [18] Cull, Robert, David McKenzie, and Christopher Woodruff. 2008. “Experimental Evidence on Returns to Capital and Access to Finance in Mexico,” The World Bank Economic Review, Vol. 22, No. 3, pp. 457-482. [19] Cu˜ nat, Vicente. 2007. ”Trade Credit: Suppliers as Debt Collectors and Insurance Providers,” Review of Financial Studies, Vol. 20, pp. 491–527. [20] Cu˜ nat, Vicente, and Emilia Garcia-Appendini. 2012. “Trade Credit and its Role in Entrepreneurial Finance,” in Cumming, D. (Ed.), Oxford Handbook of Entrepreneurial Finance, Oxford University Press, New York, pp. 526-557. [21] De la Torre, Augusto, Alain Ize, and Sergio L. Schmukler. 2012. “Financial Development in Latin America and the Caribbean: The Road Ahead,” the World Bank Group: Washington D.C. [22] Demirg¨ u¸c-Kunt, Asli, and Vojislav Maksimovic. 2001. “Firms as Financial Intermediaries: Evidence from Trade Credit Data,” Policy Research Working Paper 2696, The World Bank: Washington, D.C. [23] D’Erasmo, Pablo, and Hernan Moscoso Boedo. 2012. “Financial Structure, Informality, and Development,” Journal of Monetary Economics, Vol. 59(3), pp. 286-302. [24] Epstein, Brendan, and Alan Finkelstein Shapiro. 2017a. “Employment and Firm Heterogeneity, Capital Allocation and Countercyclical Labor Market Policies,” Journal of Development Economics, Vol. 127, pp. 25-41. [25] Epstein, Brendan, and Alan Finkelstein Shapiro. 2017b. “Financial Development, Employment Heterogeneity, and Sectoral Dynamics,” mimeo.

40

[26] Epstein, Brendan, Alan Finkelstein Shapiro, and Andr´es Gonz´alez G´omez. 2017. “Global Financial Risk, Domestic Financial Access, and Unemployment Dynamics,” mimeo. [27] Federico, Pablo, Guillermo Vuletin, and Carlos V´egh. 2014. “Reserve Requirement Policy over the Business Cycle,” NBER Working Paper No. 20612. [28] Fern´andez, Andr´es, and Adam Gulan. 2015. “Interest Rates and Business Cycles in Emerging Economies: The Role of Financial Frictions,” American Economic Journal: Macroeconomics, Vol. 7(3), pp. 153-188. [29] Fern´andez, Andr´es, and Felipe Meza. 2015. “Informal Employment and Business Cycles in Emerging Economies: The Case of Mexico,” Review of Economic Dynamics Vol. 18, pp. 381-405. [30] Finkelstein Shapiro, Alan. 2014. “Self-Employment and Business Cycle Persistence: Does the Composition of Employment Matter for Economic Recoveries?” Journal of Economic Dynamics and Control, Vol. 46, September 2014, pp. 200-218. [31] Finkelstein Shapiro, Alan, and Federico S. Mandelman. 2016. “Remittances, Entrepreneurship, and Employment Dynamics over the Business Cycle,” Journal of International Economics, Vol. 103, pp. 184-199. [32] Finkelstein Shapiro, Alan. 2017. “Labor Force Participation, Interest Rate Shocks, and Unemployment Dynamics in Emerging Economies” mimeo. [33] Gal, Peter N., and Gabor Pinter. 2017. “Capital over the Business Cycle: Renting versus Ownership,” Journal of Money, Credit and Banking (forthcoming). [34] Gertler, Mark, Simon Gilchrist, and Fabio M. Natalucci. 2007. “External Constraints on Monetary Policy and the Financial Accelerator,” Journal of Money, Credit and Banking, Vol. 39, Issue 2-3, pp. 295-330. [35] Ghani, Ejaz, William Kerr, and Alex Segura. 2015. “Informal Tradables and the Employment Growth of Indian Manufacturing,” World Bank Policy Research Working Paper 7206. [36] Gilchrist, Simon, and Egon Zakrajˇsek. 2011. “Monetary Policy and Credit Supply Shocks,” IMF Economic Review, Vol. 59, pp. 195-232. [37] Global Financial Development Report. 2014. “Financial Inclusion,” The World Bank Group: Washington D.C. [38] Hagedorn, Marcus, and Iourii Manovskii. 2008. “The Cyclical Behavior of Equilibrium Unemployment and Vacancies Revisited.” American Economic Review, Vol. 98(4), pp. 1692-1706. [39] IDB. 2005. “Developing Entrepreneurship: Experience in Latin America and Worldwide,” Edited by Hugo Kantis, with collaboration of Pablo Angelelli and Virginia Moori Koenig, Inter-American Development Bank: Washington D.C. 41

http://idbdocs.iadb.org/wsdocs/getdocument.aspx?docnum=834797 (accessed June 5, 2017). [40] ILO. 2013. “Measuring Informality: a new Statistical Manual on the Informal Sector and Informal Employment,” International Labour Organization, ISBN 9789221273882. [41] Jermann, Urban, and Vincenzo Quadrini. 2012. ”Macroeconomic Effects of Financial Shocks,” American Economic Review, Vol. 102(1), pp. 238-271. [42] Kiyotaki, Nobuhiro, and John Moore. 1997. ”Credit Cycles,” Journal of Political Economy, Vol. 105, pp. 211-248. [43] Klapper, Leora F., and Dougles Randall. 2011. “Financial Crisis and Supply-Chain Financing,” in Trade Finance During the Great Trade Collapse, The World Bank Group. Ed. Jean-Pierre Chauffour and Marien Malouche. ISBN: 978-0-8213-8748-1. [44] Korajczyk, Robert A., and Amnon Levy. 2003. “Capital Structure Choice: Macroeconomic Conditions and Financial Constraints,” Journal of Financial Economics, Vol. 68, pp. 75-109. [45] Krause, Michael, and Thomas Lubik. 2010. “On-the-Job Search and the Cyclical Dynamics of the Labor Market,” Federal Reserve Bank of Richmond Working Paper WP 10-12. [46] Levin, Andrew T., Fabio M. Natalucci, Egon Zakrajˇsek. 2004. “The Magnitude and Cyclical Behavior of Financial Market Frictions,” Finance and Economics Discussion Working Paper Series 2004-70, Divisions of Research & Statistics and Monetary Affairs, Federal Reserve Board of Governors. [47] Loayza, Norman V., and Jamele Rigolini. 2011. “Informal Employment: Safety Net or Growth Engine?” World Development, Vol. 39, Issue 9, pp. 1503-1515. [48] Love, Inessa, Lorenzo A. Preve, and Virginia Sarria-Allende. 2007. “Trade Credit and Bank Credit: Evidence from Recent Financial Crises,” Journal of Financial Economics, Vol. 83, Issue 2, pp. 453-469. [49] Mateut, Simona, Spiros Bougheas, and Paul Mizen. 2006. “Trade Credit, Bank Lending and Monetary Policy Transmission,” European Economic Review, Vol. 50(3), pp. 603629. [50] Mateut, Simona, Spiros Bougheas, and Paul Mizen. 2009. “Corporate Trade Credit and Inventories: New Evidence of a Trade-Off from Accounts Payable and Receivable,” Journal of Banking & Finance, Vol. 33(2), pp. 300-307. [51] McKenzie, David J., and Christopher Woodruff. 2006. Entry Costs Provide an Empirical Basis for Poverty Traps? Evidence from Mexican Microenterprises,” Economic Development and Cultural Change, Vol. 55,No. 1, pp. 3-42.

42

[52] McMillan, John, and Christopher Woodruff. 1999. “Interfirm Relationships and Informal Credit in Vietnam,” Quarterly Journal of Economics, Vol. 114 (4), pp. 1285-1320. [53] Mimir, Yasin. 2016. “Financial Intermediaries, Credit Shocks and Business Cycles,” Oxford Bulletin of Economics and Statistics, Vol. 78(1), pp. 42-74. [54] Monacelli, Tomasso, Vincenzo Quadrini, and Antonella Trigari. 2012. ”Financial Markets and Unemployment,” Marshall Research Paper Series Working Paper FBE 01.13, University of Southern California Marshall School of Business. [55] Mumtaz, Haroon, and Fransesco Zanetti. 2013. “Labor and Financial Frictions and Aggregate Fluctuations,” Macroeconomic Dynamics (forthcoming). [56] Navas-Alem´an, Lizbeth, Carlo Pietrobelli, and Marco Kamiya. 2012. “Inter-Firm Linkages and Finance in Value Chains,” IDB Working Paper Series No. IDB-WP-349. [57] Neumeyer, Pablo Andr´es, and Fabrizio Perri. 2005. “Business Cycles in Emerging Economies: the Role of Interest Rates,” Journal of Monetary Economics, Vol. 52(2), pp. 345-380. [58] OECD. 2009. “Overview: Data on Informal Employment and Self-Employment,” in Is Informal Normal? Towards More and Better Jobs in Developing Countries, OECD Development Centre, Ed. Johannes P. J¨ utting and Juan R. de Laiglesia. [59] OECD. 2013. “Firm Size,” in OECD Science, Technology and Industry Scoreboard 2013: Innovation for Growth, OECD Publishing. dx.doi.org/10.1787/sti scoreboard-2013-62en. [60] Pav´on, Lilianne. 2010. “Financiamiento a las Microempresas y las Pymes en M´exico (2000-2009),” Serie Financiamiento del Desarrollo 226, CEPAL. [61] Schmitt-Groh´e, Stephanie, and Martin Uribe. 2003. “Closing Small Open Economy Models,” Journal of International Economics, Vol. 61, Issue 1, pp. 163-185. [62] Shimer, Robert. 2005. “The Cyclical Behavior of Equilibrium Unemployment and Vacancies,” American Economic Review, Vol. 95(1), pp. 25-49. [63] Smets, Frank, and Rafael Wouters. 2007. “Shocks and Frictions in US Business Cycles: A Bayesian DSGE Approach,” American Economic Review, Vol. 97 (3), pp. 586-606. [64] Stein, Peer, Oya Pinar Ardic, and Martin Hommes. 2013. “Closing the Credit Gap for Formal and Informal Micro, Small, and Medium Enterprises,” IFC Advisory Services, Access to Finance 949911, International Finance Corporation. [65] Uribe, Martin, and Vivian Z. Yue. 2006. “Country Spreads and Emerging Countries: Who Drives Whom?” Journal of International Economics, Vol. 69, pp. 6-36. [66] Zhang, Yahong. 2012. “Financial Shocks, Two Sector Labour Markets, and Unemployment,” mimeo.

43