KEYWORDS: Bank credit, credit crunch, job losses, Great Recession, Senior Loan Officer Opinion Survey.
JEL CLASSIFICATION: G21, G28, G30, J20, L25.
This paper investigates how a bank credit crunch--a dramatic worsening of firm and consumer access to bank credit, such as the one observed over the Great Recession--translates into job losses in U.S. manufacturing industries. To establish the ideas, Figure 1 shows four separate channels through which access to bank credit may affect employment in the manufacturing sector: (1) the supply of commercial and industrial (C&I) loans directly to firms, (2) the availability of home equity loans or home equity lines of credit (HELOCs) to small business owners to prop up their businesses, (3) the supply of consumer installment loans to households, and (4) the availability of HELOCs to consumer-households. The paper examines these four channels using data for U.S. manufacturing industries and the Senior Loan Officer Opinion Survey (SLOOS) over the 1993-2011 period.
There are three reasons behind our choice to study the linkages between access to bank credit and U.S. manufacturing employment. First, by studying the real effects of changes in the supply of bank credit, we account for the possible substitution of funding sources at the firm and household levels. In fact, the key to identifying these economic effects is that bank loans are not perfectly substitutable with other types of external finance.1 Second, over the past few decades, the manufacturing sector has had a relatively stable structure, and banks have continued to supply a significant share of C&I loans to the sector. In contrast, other industries, such as retail trade, have experienced a shift toward large multi-unit firms with access to national capital markets. This shift has likely weakened these non-manufacturing industries' reliance on local bank credit. Third, manufacturing industries' output--in particular, the output of industries that produce durable goods--is sensitive to changes in the supply of consumer credit. Indeed, most purchases of durable goods, such as cars or large appliances, tend to be financed. This feature allows us to judge the importance of consumer access to bank credit for U.S. manufacturing employment.
Our explained variables--growth in employment, number of establishments, and average establishment size--are from the Quarterly Census of Employment and Wages (QCEW). To our knowledge, this is a novel application of the QCEW data. One of the advantages of this data set is that it does not contain a structural break in the classification of industries due to the transition from the Standard Industrial Classification (SIC) to the North American Industry Classification System (NAICS) in the late 1990s. Hence, the data set covers two recessions, including the Great Recession, on a consistent basis. We focus on employment in manufacturing rather than output because employment at the industry-state level can be measured more precisely. While the U.S. Census makes industry-state level output data available, the data are noisy by the Census' own admission, and the data compilation approaches alternate between census and non-census years.
Our explanatory variables are from several sources. Based on bank-specific responses to questions in the SLOOS, changes in commercial banks' C&I lending standards and in their willingness to originate consumer installment loans proxy for changes in firm and household access to bank credit, respectively. Indeed, a sharp tightening of C&I lending standards is followed by a steep contraction in employment in the manufacturing sector, as shown in Figure 2. In addition, estimates of growth in home equity proxy for changes in the availability of home equity loans. We use the state- and national-level house price indices compiled by CoreLogic, state- and national-level mortgage debt per borrower taken from TransUnion's Trend Data, and the national level household balance sheet data from the Federal Reserve's Z.1 statistical release to construct proxies for growth in home equity.
Our identification approach takes advantage of differences in the presence of large banks across different U.S. states and is similar to that used in the literature.2 We posit that, in accordance with the questions in the SLOOS, changes in major banks' C&I lending standards, apportioned to a particular state, are exogenous to developments in a given industry in a given state and at a given point in time. In other words, tightening (or easing) C&I lending standards does not target a particular industry in a given state. Indeed, the vast majority of the SLOOS respondents cite risk aversion or changes in the broad economic outlook as important reasons for changes in these lending standards. Variation in the geographical presence of large banks and in their timing of tightening (or easing) generates variation in changes in C&I lending standards. Similarly, changes in commercial banks' willingness to originate consumer installment loans and growth in households' home equity are understood to be exogenous to developments in a given industry in a given state and at a given point in time.
Our identification approach also relies on large mature firms' use of external finance and asset structure. As the literature suggests, cash flows and balance sheets of these firms covered by Compustat reflect deep technological parameters of the financially-unconstrained firms and, hence, captures their true demand for external funds and the ability to access them.3 We thus achieve identification by measuring the differential effect of a change in C&I lending standards on employment growth in manufacturing industries that depend relatively more on external finance and, at the same time, have relatively more pledgable assets (that is, industries most likely to be affected by changes in bank credit supply to firms) compared with other manufacturing industries. This setup also allows to control for bias due to omitted variables--for example, shocks that drive changes in both lending standards and manufacturing employment.
This definition of the treatment group builds on the setup that has been widely used in the literature to tease out a differential impact of credit supply changes on industries that depend on external finance, as in (Cetorelli and Strahan (2006) ). The novelty is that we take into account not only the need to borrow to finance physical capital investment (captured by the (Rajan and Zinglaes (1998) ) measure of dependence on external finance), but also the ability of manufacturing firms to access C&I loans by pledging tangible assets (captured by the (Braun, 2002) and (Claessens & Laeven, 2003) measure of asset tangibility). We interact these industry-specific indicators with SLOOS-based state-level changes in banks' C&I lending standards to isolate the effects of changes in the supply of local bank credit on employment growth in manufacturing. For the home equity channel that potentially affects employment growth at small businesses, we interact only our measure of dependence on external finance with estimates of growth in state-level home equity, a proxy for home equity availability.
The final piece to identification is quite intuitive and novel in the context of the paper. Consumption of durable goods is likely to be financed rather than paid for outright. Therefore, changes in consumer access to credit, though affecting consumption of both durable and nondurable goods, should affect the former to a larger extent.4 As the locations of production and consumption of durable goods may not be the same, we interact the indicator for industries producing durable goods with SLOOS-based national-level changes in banks' willingness to originate consumer installment loans to identify the effects of changes in the supply of consumer loans, and with estimates of national-level growth in home equity to identify the effects of changes in home equity availability to households.5
Several caveats to our list of identification assumptions exist. First, we apportion banks' changes in C&I lending standards according to whether these banks have deposit-taking branches in a given state. Close linkages between deposits taking and loan originating at the local level have been supported, to a certain extent, by the Community Reinvestment Act (CRA) of 1977.6 Still, potential borrowers in a given state may borrow from banks located in another state. For example, (Petersen & Rajan, 2002) show that banks were much more likely to lend over long geographic distances in the 1990s than they were in the 1970s. However, as banks began to operate branches across state lines, in particular with the passage of the Riegle-Neal Interstate Banking and Branching Efficiency Act of 1994, CRA compliance was again emphasized.7 In the banking literature, the deposit footprint has been used as a proxy for the loan footprint as well. For example, (Peek & Rosengren, 1995) use deposits as a proxy for loans.8 Second, following the literature, for example, (Cetorelli & Strahan, 2006), we associate dependence on external sources of finance in the Rajan-Zingales sense with dependence on bank loans. However, because C&I loans, unlike other sources of external finance, are predominantly collateralized, we also take into account borrowers' ability to pledge tangible collateral (as in (Braun, 2002) and (Claessens & Laeven, 2003)) so that we identify industries that most likely depend particularly on C&I loans. Indeed, as the Census' data suggests, the manufacturing industries that depend relatively more on external finance and have relatively more tangible assets have bank loans to total liabilities ratios that are 5 to 7 percentage points higher than those of other manufacturing industries.9 Third, we assume that banks' changes in C&I lending standards reported for all industries reflect changes in such standards for manufacturing industries. However, according to the FR Y-14 supervisory data covering bank holding companies with assets of more than $50 billion, the share of outstanding C&I loans to the manufacturing sector was 23 percent of the total as of the third quarter of 2012, a nontrivial share.
Our results show that changes in access to C&I and consumer installment loans, and changes in the availability of home equity loans notably affect manufacturing employment growth over the sample period.10 Specifically, we show that, for employment, household access to loans matters more than firm access to local loans, and that changes in access to bank credit affects employment growth mostly through changes in the size of firms rather than through changes in their numbers. The latter finding appears to be consistent with the literature. Small firms' entry decisions may depend less on the availability of bank credit and more on local economic conditions. Indeed, in our data, while the smallest establishments employ a relatively small fraction of employees in the manufacturing sector, these establishments are numerous and introduce noise into the aggregate series for the number of establishments. As for larger firms, consistent with the "hysteresis" behavior modeled in the literature, it may be that, following a tightening of access to credit (an unfavorable shock), the sunk cost aspect of the firm entry decision in the presence of fixed per-period costs results in these firms continuing to serve the market, but perhaps at a smaller scale that requires fewer employees.11
Structural break tests support the notion that a significant portion of manufacturing employment losses over the Great Recession was the manifestation of an unusually large tightening in credit availability--a credit crunch--rather than a structural change in the linkages between access to bank credit and employment. Indeed, our back-of-the-envelope exercise shows that, over the crisis, dramatic tightening in access to both C&I and consumer installment loans may have contributed significantly to the drop in employment in the manufacturing sector. Considering only the differential effects, the drastic worsening of access to C&I and consumer installment loans between 2007 and 2009 suggests a 5.1 percent decline in employment, nearly a third of the actual drop of 17.4 percent. In addition, the decline in the availability of home equity loans suggests an extra 1.4 percent decline in employment.
The outline of the paper is as follows. After a short literature review in the second section, we describe our data sources and the ways we transformed the raw data in the third. The fourth section goes over our empirical strategy and econometric specification. The fifth section presents the estimation results. We then detail the economic significance of changes in access to bank credit by estimating employment losses in manufacturing industries attributable to tightening access to business and consumer loans over the Great Recession. We end with some concluding remarks and policy implications.
Our identification assumption takes advantage of the differences in bank presence across U.S. states and is similar to that in (Peek & Rosengren, 2000), (Garmaise & Moskowitz, 2006), and (Lee & Stebunovs, 2012). For example, (Peek & Rosengren, 2000) use the Japanese banking crisis to test whether a loan supply shock to branches and agencies of Japanese banks affected construction activity in the U.S. commercial real estate market in California, New York, and Illinois. Similarly, (Garmaise & Moskowitz, 2006) study the effects of large bank mergers on changes in crime at the MSA level, arguing that such merger activity instruments for changes in bank competition at the local level. (Lee & Stebunovs, 2012) use a similar setup to study the effects of bank balance sheet pressures, manifested through bank capital ratios apportioned to a given state, on employment in different manufacturing industries in that state.
Our paper also contributes to the nascent literature that investigates the real effects of worsening access to bank credit over the Great Recession. In this literature strand, using employee-specific data, (Duygan-Bump et al., 2010) find that workers in small firms were more likely to become unemployed during the 2007-2009 financial crisis if they worked in industries with significant needs for external financing. From a more international perspective, (Bijlsma et al., 2010) find evidence that the credit crunch in 2008 and 2009 resulted in lower industrial growth in industries that are more dependent on external finance in OECD countries. In addition, (Bentolila et al., 2013) provide evidence that, in Spain, employment at firms funded by weak banks fell considerably more than employment at firms funded by healthier banks. From a reduced form perspective, (Fort et al., 2013) suggest that the collapse in house prices accounts for a significant part of the large decline in young/small firms during the Great Recession. However, because of their VAR approach, their "financing channel" reflects both the associated credit demand and supply factors, whereas we attempt to identify the economic impact of disruptions in the supply of bank credit.12
We focus on manufacturing industries for a few reasons. First, U.S. manufacturing industries are often studied in the finance and banking literature--for example, as in (Cetorelli & Strahan, 2006) and (Kerr & Nanda, 2009). Second, in contrast to some other industries that have experienced a shift over time toward multi-unit firms, manufacturing industries have had relatively stable structures, and many manufacturing firms continue to rely on local bank loans.13 Third, because consumption of durable goods is predominantly financed, we can evaluate the importance of household access to consumer loans for manufacturing employment and contrast that with the importance of firm access to bank loans.
Our explained variables come from the QCEW. Explanatory variables are derived from the SLOOS, TransUnion's Trend Data, CoreLogic's house price data, the Federal Reserve's Z.1 statistical release, and other sources.
The QCEW program publishes quarterly employment and wages data by industry at the county, MSA, state, and national levels as reported by employers, accounting for 98 percent of U.S. jobs.14 The program's primary outlet is the tabulation of the employment and wages of establishments that report to the Unemployment Insurance (UI) programs. Employment covered by these UI programs represents about 99.7 percent of all wage and salary civilian employment in the country. The QCEW data are collected on an establishment basis. An establishment is an economic unit, such as a farm, mine, factory, or store, that produces goods or provides services. It is typically at a single physical location and engaged in one, or predominantly one, type of economic activity to which a single industrial classification may be applied.
Admittedly, if someone is interested in the number of firms rather than the number of establishments in a given industry, then there might be some measurement error in our dependent variable induced by the fact that large firms often operate multiple establishments. Nevertheless, the number of establishments from the QCEW is highly correlated with the economic quantity--the number of firms--for at least two reasons. First, according to the Bureau of Labor Statistics, most employers have only one establishment.15 Second, earlier research--for example, by (Black & Strahan, 2002)--has shown that the rate of creation of new businesses is correlated with the share of new establishments in a local economy.
The QCEW data are reported under a promise of confidentiality. The Bureau of Labor Statistics withholds the publication of data for any industry level when necessary to protect the identity of cooperating employers. In fact, at a low level of aggregation, many data points are not reported. However, totals at the industry level for the states and the country include the undisclosed data suppressed within the more detailed tables. We limit ourselves to studying growth in employment, the number of establishments, and the average establishment size (measured in employees) over the 1993-2011 period at the industry-state level.16
To examine how bank credit supply affects firms or small business owners, we first construct measures of dependence on external sources of finance for each of the 21 manufacturing industries in our sample. These measures are based on (Rajan and Zingales (1998) ) and are calculated as the fraction of total capital expenditures not financed by internal cash flows from operations.17 The measures are widely viewed as technologically determined industry characteristics that are innate to the manufacturing processes and exogenous from the perspective of individual firms. Each industry is classified as either relatively more dependent on external finance ( industry) or less dependent ( industry) based on whether its Rajan-Zingales measure lies above or below the median Rajan-Zingales measure for the 21 manufacturing industries. Although particular values of the Rajan-Zingales measures change over time, the relative ordering of the measures changes very little as we show in the appendix. The industries that our approach identifies as depending relatively more on external finance remain generally the same whether we use the 1980s data or more recent data.18 To sum up, as in (Cetorelli & Strahan, 2006), implicit in our identification strategy is the assumption that dependence on external finance is constant over time, or rather that the industry ordering is not altered substantially. Since we compute measures of dependence on external finance using mature firms, which are likely closer to industry steady state conditions, this assumption appears to be reasonable.
We associate dependence on external sources of finance in the Rajan-Zingales sense with dependence on bank credit.19 Although, in aggregate, U.S. firms' dependence on bank loans may be somewhat limited, the dependence is nevertheless larger for some firms (such as relatively small privately held firms) than for others (such as large publicly traded firms). For example, the Census' data suggest that, for firms with less than $25 million in total assets, the ratio of bank loans to total liabilities is more than 20 percent, while that for firms with more than $25 million in total assets is less than 10 percent.20 Using a more direct measure to capture dependence on bank loans would subject our analysis to endogeneity concerns, as a given industry's low dependence on bank loans could simply indicate financing constraints. In this vein, (Cetorelli & Strahan, 2006) argue that the Rajan-Zingales measures computed for only mature firms provide a "powerful instrument for small firms' demand for bank credit," but a direct measure of bank credit dependence--based on bank loans to assets ratios of small businesses from the 1998 Survey of Small Business Finance (SSBF)--does not.21 Ultimately, as shown in (Colla et al., 2013), even larger firms may specialize in a certain debt type and may not be able to quickly substitute away from scarce bank loans; hence, they may be forced to downsize or shut down. In other words, if small and large firms alike were able to substitute away from scarce bank loans completely, we would not be able to identify any real effects of a tightening in credit supply using our regression models.
To sharpen our identification approach, we also consider firms' ability to pledge collateral in obtaining bank C&I loans. As suggested by the Survey of Terms of Bank Lending (the Federal Reserve's E.2 statistical release), C&I loans tend to be secured by collateral, such as equipment and machinery. To reflect this particular feature of C&I loans, we consider asset tangibility by industry. We reason that firms in manufacturing industries with a relatively larger share of tangible assets relative to total book-value assets should have the ability to access C&I loans because such collateral is easy to pledge and to evaluate. Following the guidelines in (Braun, 2002) and (Claessens & Laeven, 2003), we compute such tangibility ratios at the three-digit NAICS level using the data for large U.S.-based firms over the sample period. Again, we are not interested in the exact value of the asset tangibility metric for each industry as such. Similar to our dependence on the external finance measure, the asset tangibility measure shows little difference in relative ordering regardless of which time period is used.22 Therefore, based on whether the measure for a given industry is below or above the median, we sort industries into those with a smaller share of tangible assets ( industries) and those with a larger share of tangible assets ( industries).23
Finally, we can define a treatment group: the industries that depend relatively more on external sources of funding in the (Rajan and Zingales (1998) ) sense and that have a relatively higher ability to pledge collateral to secure access to C&I loans in the sense of (Braun, 2002) and (Claessens & Laeven, 2003). For a given industry, the first part of the definition tells us the need to borrow, and the second part the ability to do so.24 Indeed, as the Census' data suggests, the manufacturing industries that depend relatively more on external finance and have relatively more tangible assets have bank loans to total liabilities ratios that are 5 to 7 percentage points higher than those of other manufacturing industries.25 In addition, to stay consistent with the literature, we consider the treatment group separately to gauge the effects of changes in home equity availability on employment growth at small businesses.
We define another treatment group by recognizing that the degree of consumer reliance on bank credit for consumption of durable goods is different than that of consumption of nondurable goods. Consumption of durable goods is more likely to be financed with consumer or home equity loans (rather than paid for outright) than consumption of nondurable goods. Hence, to a certain extent, the producers of durable goods are at the mercy of lenders to consumers. We follow the U.S. Census Bureau's breakdown of manufacturing industries into industries that produce durable goods ( industries) and those that do not ( industries).
Table 1 shows the breakdown of three-digit NAICS manufacturing industries into those that depend relatively more on external finance ( ), those that have relatively larger shares of tangible assets ( ), and those that produce durable goods ( ). Some industries do not have any of these characteristics, while others have one, two, or all three, which helps with our identification. In aggregate, manufacturing industries account for nearly 10 percent of non-farm employment and about 20 percent of the country's output.
Having defined the control and treatment groups, we look into the growth in employment, number of establishments, and average establishment size (measured in employees) in each of the groups. Figures 3 to 5 plot these measures for the entire economy. The figures suggest that growth in employment and the growth in the average establishment size in the treatment group are more procyclical than those in the control group. However, for the growth in the number of establishments, the business cycle pattern for the treatment group relative to the control group is less clear.26
To identify how access to different types of bank loans affects employment in manufacturing industries, we focus on three types of loans: C&I loans, consumer installment loans, and HELOCs.27 C&I loans include loans for commercial and industrial purposes to sole proprietorships, partnerships, corporations, and other businesses, whether secured or unsecured, and whether single payment or installment. C&I loans exclude the following: loans secured by real estate; loans to financial institutions; loans to finance agricultural production and other loans to farmers; loans to individuals for household, family, and other personal expenditures; and other miscellaneous loan categories. Typically, the interest rate for C&I loans is set as a spread over the prime rate or Libor and adjusts with movement in the benchmark rate over the loan term. Consumer installment loans are loans to individuals--for household, family, and other personal expenditures--that are not secured by real estate, such as auto loans. Typically, the interest rate for consumer installment loans is set as a spread over the prime rate or Libor and remains fixed over the loan term.
Available data sources suggest that banks provide a significant share of C&I loans to the manufacturing sector over our sample period. According to the Federal Reserve's G.27 statistical release, as of December 1982, about 28 percent of C&I loans outstanding at large commercial banks (with assets of more than $1 billion) were to the manufacturing sector. According to the FR Y-14 supervisory data covering bank holding companies with assets of more than $50 billion, as of the third quarter of 2012, the share of outstanding C&I loans to the manufacturing sector was 23 percent of the total.
In recent years, the popularity of HELOCs--revolving, open-ended lines of credit secured by residential properties--has overshadowed the use of non-collateralized consumer installment loans. HELOCs are typically secured by junior liens and are usually accessible by check or credit card. The rate on new home equity loans is often set as a spread to the prime rate or Libor. Lenders typically offer home equity loans up to 100 percent of the appraised property value, less the amount of any outstanding first mortgage liens.
Changes in C&I lending standards and in the willingness to originate consumer installment loans are based on bank-specific responses to questions about changes in lending standards and terms from the Federal Reserve's SLOOS.28 Papers such as (Bassett et al., 2014), (Gilchrist & Zakrajšek, 2012), and (Lown & Morgan, 2006) have studied aggregated responses of the survey results to examine how bank credit shocks affect or are associated with economy-wide real and financial variables.
The survey is usually conducted four times per year by the Federal Reserve Board, and up to 60 banks participate in each survey. The survey is voluntary; it typically includes the largest banks in each Federal Reserve district and is roughly nationally representative. All surveyed banks are considered relatively large: no bank in the survey has assets of less than $3 billion.
In the survey, banks are asked to report whether they have changed their credit standards over the past three months on six categories of core loans, including C&I loans. Both the series indicating changes in credit standards on C&I loans and the series capturing changes in banks' willingness to originate consumer loans are available from the May 1990 survey. (Questions regarding changes in standards on credit card loans and other consumer loans were added to the survey in February 1996 and May 1996, respectively, and were not used in the paper.) The SLOOS asks banks to report changes in their lending practices over the previous three months, and the survey is conducted so that it coincides with regular meetings of the Federal Open Market Committee. Hence, the January SLOOS refers to the period from October to December of the prior year.
We aggregate bank responses concerning changes in C&I lending standards to the state level in two steps. First, we map individual bank responses to indicator variables. Second, we sum these responses across banks for each U.S. state at an annual frequency. The following two paragraphs describe the two steps in detail.
The question about changes in C&I lending standards reads, "Over the past three months, how have your bank's credit standards for approving applications for C&I loans or credit lines--other than those to be used to finance mergers and acquisitions--to large and middle-market firms and to small firms changed?" Banks respond to that question using a categorical scale from 1 to 5: 1 = eased considerably, 2 = eased somewhat, 3 = remained about unchanged, 4 = tightened somewhat, and 5 = tightened considerably. We use the answers based on banks' responses with respect to small firms because the QCEW data is predominantly composed of small businesses and we attempt to capture how local bank credit supply affects firms.29 In addition, though banks were extremely unlikely to characterize their changes in lending standards as having "eased considerably" or "tightened considerably," we use all five of the classifications available to survey respondents. Letting index the respondent banks and index time, we define an indicator variable as follows: if bank reported considerable easing of standards at time , if bank reported some easing, if bank reported no change in standards, if bank reported some tightening, and if bank reported considerable tightening.30
We aggregate individual bank responses across banks for each U.S. state and convert those from quarterly to annual frequency. Using the indicator variables, we construct a composite of changes in lending standards for a particular state , weighted by total business loans (C&I loans plus commercial real estate loans) for each year , in part to strengthen the exogeneity assumption that banks with exposure to a broader economy affect industry-state employment growth dynamics. In other words, the largest of banks with branches in multiple states get weighted the most; this is desirable because a large fraction of small business loans are originated by the largest banks.31 More specifically, the tightening measure we use is calculated as follows: , where denotes a quarter of the year. Of the banks that participate in the SLOOS, we select only those that have deposit taking branches in state according to the Summary of Deposits. Hence, the total number of banks, , in a given state may be below 60 for a particular state. We limit the coverage to 32 states (including the District of Columbia) where the selected banks have a material cumulative share of deposits in every year of our sample.32 These filters ensure that our state-level tightening measure is, in fact, representative for a given state.
Figure 6 shows the weighted average tightening of C&I lending standards, --that is, the weighted average of individual bank's responses, --for three states: California, New York, and Texas. It shows a drastic tightening of C&I lending standards around the past three recessions as well as a notable loosening of the standards in the mid-2000s. Across the 32 states in our sample, we observe adequate cross-sectional variation in the measures of changes in C&I lending standards, which helps with identification.
Changes in banks' willingness to originate consumer installment loans proxy for the changes in standards on consumer loans. The question about changes in consumer installment loans reads, "Please indicate your bank's willingness to make consumer installment loans now as opposed to three months ago." Among available options, the respondents indicate whether they are more or less willing to originate these loans.33 By analogy with the construction of , we construct a national composite measure of changes in willingness to make consumer installment loans, , weighted by total consumer loans (excluding residential real-estate loans) as the location of production and consumption of durable goods are usually not the same. Figure 7 plots the weighted average change in banks' willingness to originate consumer installment loans at the national level, .34
Because the SLOOS only recently began including questions about the availability of HELOCs, we rely on estimates of growth in (unencumbered) home equity to proxy for changes in the availability of HELOCs. To construct these estimates at the state level, we use house price data from CoreLogic and mortgage debt data from TransUnion's Trend Data.35 Although these estimates are only available starting from 1993, there is notable heterogeneity across states in the timing and magnitude of house price changes to our advantage. Some areas experienced strong decreases in home values over the recent crisis, while other areas avoided the housing boom and experienced no significant house price depreciation. As for the estimates at the national level, we rely on CoreLogic for house price data and the Federal Reserve's Z.1 statistical release for mortgages, HELOCs, and home equity loans secured by junior liens.
We construct the proxy for changes in the availability of HELOCs as follows. We start with the premise of (Avery et al., 2011) that the difference between house prices and outstanding mortgage debt (as well as junior liens and HELOCs, if available) should approximate home equity. Since we cast our regression models in growth rates, we construct a proxy for the growth rate of home equity (using the inverse of the loan-to-value ratio) for a given state as: , where is the growth rate of home equity in state at time , is the growth rate of the house price index in state at time , and is the growth rate of mortgage debt in state at time .36 Admittedly, this might be a noisy proxy for growth in home equity, but we believe it is the best available state-level measure. Similarly, we construct a proxy for the growth rate in home equity at the national level, denoted by . However, in this instance, because of the richness of the Federal Reserve's Z.1 statistical release, we can estimate unencumbered home equity a bit more precisely by accounting for outstanding HELOCs and junior liens.37 The measures, and , appear to be satisfactory proxies for changes in the availability of HELOCs, that is, growth in unencumbered home equity appears to be reflective of changes in the bank credit supply of HELOCs. For example, in the past six years, the correlation between and an aggregate measure of tightening of credit standards for HELOCs based on the SLOOS responses was -0.77 on a quarterly basis.
We include into our regression models growth in aggregate household net worth ( ), constructed using the Federal Reserve's Z.1 statistical release, to control for household wealth-driven demand. The inclusion of growth in aggregate household net worth in the regression sharpens the interpretation of growth in home equity as a proxy for changes in access to home equity loans.
The unit of observation for employment growth is at the NAICS three-digit manufacturing industry level in a given state and a given year. To ensure more robust identification, we could have worked with county- or MSA-level data, but at such a low level of aggregation, there would have been too many missing observations due to confidentiality and non-disclosure issues. In contrast, the QCEW industry data at the state level are available over a long period and include the undisclosed data suppressed within the detailed disaggregated tables. Hence, working with state-level data appears to strike a balance between exogeneity concerns and data quality. Although the QCEW is a quarterly frequency data set, we choose to work with its annual averages for a few reasons. We are interested neither in immediate responses of employment growth to changes in access to bank credit, which might later be reversed, nor in the seasonality of manufacturing employment growth and changes in credit provision.38
We examine how changes in credit supply conditions for both firms and households affect growth in manufacturing employment. To isolate these effects and to control for omitted variable bias (for example, from technological changes that drive both changes in lending standards and employment), we exploit the variation in industries' dependence on external finance and the availability of tangible assets as well as the variation in households' dependence on consumer loans. Specifically, we examine whether changes in C&I lending standards, the ability of small business owners to extract home equity to prop their businesses, banks' willingness to originate consumer installment loans, and the ability of consumer-households to extract home equity matter for changes in employment in manufacturing industries.
Given a high degree of persistence in the explained variables over the sample period, as well as the nature of the measure of changes in C&I lending standards and in the willingness to originate consumer installment loans, we work with an empirical model cast in growth rates. This model is stationary and allows us to control for aggregate trends in levels and growth rates (because of included fixed effects). The growth rates of the explained variables are not persistent, with very low autoregressive coefficients; lagged dependent variables are therefore omitted from our regression models.
Our identification assumption is that changes in banks' lending standards, apportioned to a particular state, and growth in home equity are exogenous to developments in a given manufacturing industry in a given state and at a given point in time. In accordance with the questions in the SLOOS, we postulate that banks tighten C&I lending standards broadly across the country rather than targeting a particular state and/or a particular industry. Variation in the geographical presence of banks and in the timing of tightening generates variation in our explanatory variables.39
Besides omitted variables, we control for aggregate credit, state, and national economic conditions. In addition to stand-alone SLOOS measures, growth in home equity, and growth in net worth, aggregate credit conditions are proxied by the change in the realized real interest rate calculated by the difference in the three-month Libor and the PCE inflation rate. As a proxy for national economic conditions, we include the growth rate of U.S. real GDP. With the exception of growth in real GDP and the change in the real interest rate, the controls for contemporaneous economic conditions, all other explanatory variables are lagged one year. To address the potential endogeneity of industry location choices and industry-state-specific trends, industry-state fixed effects are included.
Putting all the pieces together, the benchmark specification is:
We compute errors clustered separately in several ways: clustering by , clustering by , and double clustering by and . The multiple clustered errors are calculated using the (Cameron et al., 2011) code.
The 32 states used in our regressions appear to be representative of the population of manufacturing industries in the entire country. We checked the data breakdown by employment, the number of establishments, and the average establishment size for two years, 2007 and 2010. The population measures are shown in Table 4 and the same measures for the 32 states in our sample are shown in Table 3. In percentage terms, the breakdown of employment and number of establishments in our sample is very similar to that in the population. In addition, the average establishment size in the sample is nearly identical to that in the population as our sample accounts for more than 80 percent of the total employment in the United States as of 2010.41
First, we determine whether each of the credit channels mentioned in Figure 1 exists separately from the others. For the effects of the supply of credit to firms, we find that a percentage point increase in the tightening of C&I lending standards ( ) in the prior year leads to a 0.04 percentage point decrease in employment growth in industries that depend relatively more on external finance and have a relatively higher ability to pledge tangible assets ( industries), as can be seen in the first column of Table 4. Next, a percentage point increase in the availability of home equity loans to small business owners ( ) in the prior year drives up employment growth in industries that depend relatively more on external fiance ( industries) by 0.06 percentage point, as shown in the second column. As for the effect of the supply of credit to households, we find that employment growth in industries that produce durable goods ( industries) increases by 0.06 percentage point with a percentage point increase in additional willingness to originate consumer installment loans ( ) in the prior year, as described in the third column. Finally, a percentage point increase in the availability of home equity credit ( ) for household-consumers in the prior year has a positive effect of nearly 0.11 percentage point on employment growth in the treatment industries, as stated in the fourth column. All of these coefficients are statistically significant at conventional levels.
Next, Tables 5 to 7 present empirical results with all four channels included simultaneously: Table 5 shows the results for growth in employment, Table 6 shows those for growth in the number of establishments (the extensive margin), and Table 7 shows those for growth in the average establishment size (the intensive margin). Given that the growth rate of employment is just a sum of the growth rates of the number of establishments and the average establishment size, the regression coefficients in the employment growth regression in Table 5 are nearly exact sums of the corresponding coefficients in Tables 6 and 7. Each of the these three tables shows the estimation results for models with industry-state fixed effects and various specifications of error clustering. In each table, in the first column, errors are clustered by industry-state; in the second, by year; and, in the third, double-clustered by industry-state and year.42
As the first row in Table 5 shows, for the credit supply to firms, a percentage point increase in the tightening of C&I lending standards reduces the growth rate in employment in the treatment industries by 0.04 percentage point, a similar result to that in Table 4. Note that this result is statistically significant at conventional levels even in the regression with double-clustered errors. (The economic significance of this and other coefficients will be explored later in section 6.) The regression coefficient in the second row, capturing the impact of the ability of small business owners to extract home equity to finance their businesses, is positive, but no longer robustly statistically significant. As for the credit supply to households (shown in the third row), a percentage point increase in additional willingness to originate consumer installment loans boosts the growth rate of employment in the treatment industries by 0.06 percentage point. In addition, as seen in the fourth row, a percentage point increase in growth in home equity propels the growth rate in employment in the treatment industries by 0.15 percentage point. In contrast, growth in households' net worth--a proxy for other wealth-driven consumption of durable goods included as a demand side control--does not appear to have a robustly statistically significant impact on employment growth in the treatment industries. As for the other explanatory variables, all the estimated coefficients, with the exception of the coefficient for state-level home equity, are of the expected signs. However, only the coefficient to real GDP growth is estimated to be robustly statistically significant across all our error-clustering specifications.
Table 6 shows the regression results for growth in the number of establishments. For the credit supply to firms, the results show that the tightening of C&I lending standards does not have a statistically significant effect on the growth in the number of establishments in any of the specifications. Changes in the availability of HELOCS, on the other hand, appear to have a statistically significant effect on the growth in the number of establishments for industries depending relatively more on external finance. As for the credit supply to households, the results show that an increase in banks' willingness to originate consumer installment loans has a statistically significant effect on growth in the number of establishments only in the first specification, that is, in the model with the least strict error specification. Changes in the availability of HELOCs do not appear to have any statistically significant effect on the growth in the number of establishments for durable goods industries.
The lack of evidence that the supply of C&I loans has an effect on the number of establishments appears to be consistent with the literature. Small firms' entry decisions may depend less on the availability of bank credit and more on local economic conditions. Setting up a firm may not be that costly. For example, according to (Djankov et al., 2002), entrepreneurs' average cost of starting a firm (including the time to start up a firm) was 1.7 percent of per-capita income in the United States in 1999, or $520. In addition, layoffs by firms that are induced by stricter lending standards may spur some creation of firms (counted as establishments in our data), which may boost the number of establishments in times of distress. For example, (Aaronson et al., 2004) document the increase in the number of firms, which was accompanied by a fall in employment at the aggregate level. Indeed, in our data, while the smallest establishments employ a relatively small fraction of employees in the manufacturing sector, these establishments are numerous and introduce noise into the aggregate series for the number of establishments. As for larger firms, consistent with the "hysteresis" behavior modeled in other strands of the literature, it may be that, following a tightening of access to credit (an unfavorable shock), the sunk cost aspect of the firm entry decision in the presence of fixed per-period costs results in these firms continuing to serve the market, but perhaps at a smaller scale that require fewer employees, similar to (Alessandria & Choi, 2007).
Although analyzing employment growth separately at small and large firms would be desirable, our data do not have this categorization at the desired industry-state level. However, we can infer from another data set--the Census' County Business Patterns (CBP)-- that, for the entire country, changes in the number of establishments in the small establishment size classes do introduce noticeable noise to the aggregate number of establishments.43 However, because these small establishments employ, in aggregate, a small number of workers, changes in employment at these establishments have a very limited impact on aggregate employment growth.
Table 7 shows the estimation results for the growth rate in the average establishment size. For the credit supply to firms, the results show that a percentage point increase in the tightening of credit standards leads to a 0.04 percentage point decrease in the growth rate of the average size of the establishments in the treatment groups. The results also show that an improvement in small business owners' access to HELOCs does not have a statistically robust impact on the average establishment size in any of the three specifications. As for the credit supply to households, the results show that a percentage point increase in additional willingness to originate consumer installment loans leads to a nearly 0.05 percentage point increase in the growth rate of the average establishment size in the treatment industries. In addition, a percentage point increase in the availability of home equity boosts the growth rate of the average establishment size by 0.15 percentage point. These results suggest that most of the industry-wide employment growth dynamics driven by changes in bank credit availability are explained by changes in average establishment size (the intensive margin) rather than by changes in the number of establishments (the extensive margin).
To be consistent with the traditional approach, we check whether dependence on external finance or the availability of pledgable assets matter separately for employment growth in manufacturing industries. As Table 8 shows, it is the former that is statistically significant. However, we believe that it is correct to study the intersection of and industries for at least two reasons.44 First, the C&I loan definition and many survey results suggest the importance of pledgable assets for firms' access to bank credit. Indeed, as the Census' data suggests, the manufacturing industries that depend relatively more on external finance and have relatively more tangible assets have bank loans to total liabilities ratios that are 5 to 7 percentage points higher than those of other manufacturing industries. Second, the joint consideration of and industries, as a comparison of Tables 5 and 8 reveals, strengthens the effects of tightening of C&I lending standards on employment growth in the treatment industries. Specifically, the coefficient on the term in the benchmark specification in Table 5 is almost twice as large as the coefficient on the changes in C&I credit standards interacted only with .
Next, we check whether our results are driven by the developments specific to the Great Recession. To do so, we estimate a regression model that allows a break in the coefficients on the measures of changes in access to C&I, consumer installment, and home equity loans after 2007. In the regression model in Table 9, the proxies for changes in access to various types of credit are interacted with both respective industry-type dummies and a crisis indicator ( ) that captures the period from 2007 to 2011. The estimation results suggest no statistically robust evidence of structural breaks in the relationship between changes in access to the four types of bank lending and employment growth across manufacturing industries. That said, there is some indication that, over the crisis, for growth in manufacturing employment, the channel that links changes in small business owners' access to home equity loans to employment gains may have become impaired, which is a reason we do not see this channel at work in our benchmark model. However, controlling for the crisis, this channel appears to be statistically significant as shown by the coefficients to without the interaction. In addition, in the first specification, changes in households' access to consumer installment and home equity loans may have become less important (the triple interaction terms and have negative coefficients). However, this result is not statistically significant in the other error-clustering specifications.
We perform a back-of-the-envelope exercise to gauge the potential contribution of a dramatic worsening of the availability of C&I and consumer installment loans to the drop in employment in the manufacturing sector over the Great Recession. We only take into account the differential effect of tightening access to bank credit on employment growth in the treatment industries.45 The differential effects of a percentage point tightening in C&I lending standards ( ) and additional willingness to originate consumer installment loans ( ) may be inferred from the coefficients estimated in Table 5, lines 1 and 3. The values of and are inferred from state-level data and national level data, respectively, similar to what is shown in Figures 6 and 7. Over the 2007-2009 period, ranged from 38 to 56 percent, on average, per year, across the 32 states, and was about negative 34 percent, on average, per year, at the national level.46 As a benchmark, to judge the importance of the bank credit supply factors, we rely on actual declines in manufacturing employment. For the manufacturing sector in our 32-state sample, the pre- and post-crisis breakdowns in employment are shown in Table 3. Over the Great Recession, employment in the manufacturing sector declined about 17.4 percent, and the treatment groups ( and ) experienced notably larger declines in employment than the control groups ( and ).
Considering only the differential effects, keeping all other explanatory variables fixed and ignoring numerous fixed effects, we perform the back-of-the-envelope exercise as follows. For industry in state , let be employment in 2007 and be the change in employment over the 2007 to 2010 period. Then our back-of-the envelope exercise consists of two steps. The first is simply the following: , where and , which are estimated coefficients from Table 5. The second step is aggregation of industry-state specific across states and industries by industry type.
The back-of-the-envelope exercise shows that the dramatic tightening of access to both C&I and consumer installment loans may have contributed significantly to the drop in employment in the manufacturing sector. As the top panel in Table 10 shows, worsening access to C&I and consumer installment loans over the financial crisis suggests a 5.1 percent decline in employment, nearly a third of the actual drop of 17.4 percent. For the industries that depend relatively more on external sources of funding, have a larger share of tangible assets, and produce durable goods ( and ), the suggested impact on employment growth is the largest. For the industries that depend relatively more on external sources of funding, have a larger share of tangible assets, and produce nondurable goods ( and ), the impact on employment growth was the smallest. The bottom panel of Table 10 shows declines in employment in absolute terms. The exercise associates about a 588,000 decline in employment with the worsening in access to C&I and consumer installment loans, while the actual decline was about 2 million in the 32 states in our sample. Given the estimation results shown in Tables 6 and 7, the employment declines were likely driven primarily by a shrinkage in the establishment sizes rather than by a fall in the number of establishments.
For industries that produce durable goods, the regression results suggest that there is an additional channel at work--the availability of home equity loans to consumer-households. The proxy for the growth in home equity declined, on average, about 5.7 percent per year during the 2007-2009 period. Per similar calculations to those above, the reduction in the availability of home equity loans suggests an additional 1.4 percent decline in manufacturing employment.
The back-of-the-envelope exercise has several caveats. First, it is merely an illustration of the potential economic significance of the worsening availability of C&I and consumer installment loans to the fall in employment. It is not a formal in-sample forecasting exercise, as we do not take into account the potentially offsetting impact of other variables, such as lower real interest rates or certain fixed effects. Second, the s of, at most, 30 percent suggest that the estimated models leave a sizeable variation in employment growth at an industry-state level unexplained. Presumably, a large fraction of the unexplained variation is related to a fall in demand. In fact, (Mian & Sufi, 2012) point to a large negative demand shock, in part, attributable to household overindebtedness, underlying employment losses in the non-traded goods sector over the Great Recession. Third, we caution that any generalization of the back-of-the-envelope exercise to the entire economy is limited because we only consider job losses in manufacturing industries and do not account for displaced workers that may be absorbed by firms in other sectors. However, we believe that our findings are indicative of developments in other sectors reliant on bank credit as well.
This paper investigates how a bank credit crunch--a dramatic worsening of firm and consumer access to bank credit, such as the one observed over the Great Recession--translates into job losses in in U.S. manufacturing industries. To isolate these effects, we exploit variation in changes in C&I lending standards, in changes in banks' willingness to originate consumer installment loans, and in growth in home equity. To control for omitted variable bias, we rely on differences in the degree of industries' dependence on external finance and in the sensitivity of these industries' output to changes in consumer credit. We show that changes in C&I lending standards, in willingness to originate consumer installment loans, and in the availability of home equity loans notably affect employment growth in manufacturing industries over the 1993-2011 period. In particular, the results highlight the adverse effects that disruptions in the supply of bank credit had on manufacturing employment growth over the Great Recession.
These results have a few implications for understanding the recovery in the economy and the labor market going forward. To some extent, the tightening of lending standards and the decrease in willingness to originate loans reflect commercial banks' efforts to deleverage. By changing the composition of their balance sheets from business and consumer loans toward U.S. Treasury securities, commercial banks may improve their standing with both bank regulators and investors. These adjustments may temporarily hold back employment growth in manufacturing industries, and, thus, contribute to persistently weak conditions in the labor market more generally. Moreover, the housing market may need to improve more substantially to have a more noticeable effect on manufacturing employment growth. In the longer term, the displaced workers in the industries affected the most may be absorbed by firms in other sectors of the economy.
When dealing with the unusually low levels of employment growth in manufacturing in an environment like the Great Recession, a policy prescription that follows from our back-of-the-envelope exercise is that household access to bank credit appears to matter more for manufacturing employment than firms' access to bank loans. This conclusion is consistent with some stylized facts about the U.S. economy. First, the share of consumption in the country's GDP, at about 70 percent, by far dominates the share of physical capital investment. Second, while households rely more on banks for credit, many large firms may have the ability to access capital markets directly. However, one should bear in mind that prolonged periods of underinvestment in physical capital may have implications on productivity growth and potential output in the long run.
We thank William Bassett, Thorsten Beck, Nicholas Bloom, Christopher Carroll, John Driscoll, Lucca Guerrieri, John Haltiwanger, and Jonathan Rose as well as seminar participants at the Federal Reserve Board for helpful comments and suggestions. In addition, we thank discussants Adonis Antoniades, Oscar Arce, Chiara Banti, Jin Cao, Enrique Martinez-Garcia, Anthony Murphy, Leonard Nakamura, Francisco Rodríguez-Fernández, and Pei Shao and participants at numerous conference and meetings for valuable comments and suggestions. We thank Shaily Patel for excellent research assistance. Viktors Stebunovs thanks the Office of Financial Stability Policy and Research for its hospitality.
In the appendix, we provide details on the Rajan-Zingales measures of dependence on external finance and the Braun-Claessens-Laeven measures of asset tangibility. We also discuss several robustness checks: two models estimated using the scale dependence and tangibility measures, rather then the industry-type dummies based on these measures; a model estimated using the ratios of bank loans to total assets computed for the 1998 Survey of Small Business Finance; and a model with industry-specific trends. Our final significant robustness check concerns the usage of different measures of the weighted average tightening of C&I lending standards and the weighted average change in banks' willingness to originate consumer installment loans. Finally, we touch upon briefly less significant robustness checks.
As in (Cetorelli & Strahan, 2006), implicit in our identification strategy is the assumption that dependence on external financial is constant over time, or rather that the industry ordering (in terms of dependence on external finance) is not altered substantially. Indeed, as Table A1 shows, the Rajan-Zingales measures for mature firms change somewhat over time, but the industry ordering remains relatively stable.47 The first column shows these measure for different three-digit NAICS codes from the 1980s period used in (Rajan and Zingales (1998) ); the second column the measures based on the 1980-1997 period used in (Cetorelli & Strahan, 2006), and the third column the measures based on the 1990-2011 period used in our paper. Note that the columns show the measures that we calculated, even those for the periods used in (Rajan and Zingales (1998) ) and (Cetorelli & Strahan, 2006), to present the data on a more comparable basis. While (Rajan and Zingales (1998) ) relies on the ISIC and (Cetorelli & Strahan, 2006) on the SIC, all our calculations are based on NAICS. However, as the Census' information suggests, these classifications may not be directly comparable.48 In the table, a measure in bold indicates whether the corresponding industry depends relatively more on external finance than others based on whether its measure is above or equal to the median Rajan-Zingales measures for the 21 industries in the sample. A comparison across the columns points to a remarkable stability in the industries that our approach identifies as relatively more dependent on external finance. For example, only one manufacturing industry (325 - Chemical Manufacturing) that is relatively more dependent on external finance in our sample period was not classified as such based on the 1980s period used in (Rajan and Zingales (1998) ).
In addition, implicit in our identification strategy is the assumption that industries asset tangibility is constant over time, or rather that the industry ordering is not altered substantially. Again, as Table A2 shows, asset tangibility measures change over time, but a comparison across the columns points to a remarkable stability in the industries that we identify as those with relatively more tangible assets. For example, only two industries (312 - Beverage and Tobacco Product Manufacturing and 314 - Textile Product Mills) that has relatively more more asset tangibility in our sample period were not so classified in the 1980s period used in (Rajan and Zingales (1998) ); whereas only one industry (312 - Beverage and Tobacco Product Manufacturing) that has relatively more asset tangibility in our sample period was classified differently in the 1980-1997 period used in (Cetorelli & Strahan, 2006). Such findings support our identification approach which relies on the ordering of the industries rather than on the usage of particular measures of dependence on external finance and availability of tangible assets that change over time. Our results are robust to using other periods in calculations of the Rajan-Zingales and Braun-Claessens-Laeven measures for construction of our and industry-type dummies. Moreover, our results hold if we identify our treatment industries as those in the top third of the joint distribution of the Rajan-Zingales and Braun-Claessens-Laeven measures.
For further robustness, albeit with significant shortcomings, we estimate two models with dependence on external finance and asset tangibility captured by scale Rajan-Zingales and Braun- Claessens-Laeven measures, rather then the industry-type dummies based on these measures. Recall that to identify the impact of tightening C&I credit standards ( ) on employment growth we define a treatment group comprising the industries that depend relatively more on external finance ( ) and have relatively more tangible assets ( ) than other industries. By analogy, when using scale Rajan-Zingales ( ) and Braun-Claessens-Laeven ( ) measures, the triple interaction term may be . Because both and may assume positive and negative values, it is difficult to interpret the regression coefficient for this new triple interaction term. (Note that such interpretation ambiguity is not an issue for some other papers because they interact the Rajan-Zingales measure with a categorical variable--for example, with the dummy indicating whether a state allows for interstate banking or not as in robustness checks in (Cetorelli & Strahan, 2006). Moreover, these papers tend to have less complex treatment groups, making the triple interaction terms unnecessary.) To circumvent this ambiguity, we map the Rajan-Zingales measures into a interval. (The mapping formula is as follows: , where and .) While this transformation fixes the issue with negative s, it does not address a possibility of an outlier driving the estimation results one way or another.49 In contrast, our and approach is more resilient to such outliers. Finally, the difficulties with the interpretation of the coefficient and the identification approach still remain. In our benchmark model, the interaction term, , has a straightforward interpretation when it equals one. Specifically, these are the industries that depend relatively more on external finance and have relatively more tangible assets than other industries; this breakdown clearly identifies the industries that are potentially affected the most by changes in the availability of C&I loans. In contrast, the interaction term , a product of the transformed Rajan-Zingales measures and the Braun-Claessens-Laeven measures, is much less intuitive. In particular, this interaction term may be high if is high, or is high or both, and it is the combination of both high and high that should identify the treatment group, that is, the industries that have the need to seek C&I loans and have the ability to obtain them by pledging tangible assets. In short, the interaction term obscures identification. These arguments notwithstanding, the first two columns of Table A3 show the estimation results. The first column includes the double interaction terms and to our benchmark specification. Not surprisingly, these new additional double interaction terms appear to be collinear with the triple interaction term , as the results in the second column illustrate. In the model with only the triple interaction term included, its regression coefficient is negative and statistically significant. Again, the meaning of the value of the coefficient, however, is difficult to interpret.
As another robustness check, we re-estimate the benchmark model using the ratios of bank loans to total assets computed for the 1998 Survey of Small Business Finance to identify our treatment groups. The estimation results are shown in the third column of Table A3.50 It appears that both easing access to C&I loans and easing of small business owners' access to HELOCs have a positive, statistically significant impact on growth in manufacturing employment. This finding appears to be broadly consistent with the literature that the smallest (youngest) firms rely more on funds supplied by the owners than on bank loans. We are cautions about promoting this finding because industries' sorting based on the ratios of bank loans to total assets likely introduces endogeneity into the estimation. In contrast to the Rajan-Zingales measures, that reflect deep technological parameters of the least financially-unconstrained firms, the ratios of bank loans to total assets are computed for undoubtedly financially constrained firms. And a low ratio may simply indicate firms' severe financial constraints.
As yet another robustness check, we re-estimate the benchmark model with separate time trends for each of the 21 manufacturing industries in the sample. Our results, shown in the fourth column in Table A3, are robust to accounting for these trends. However, we believe the inclusion of these regressors in the model is redundant, as only a few industries have statistically significant trends. We emphasize that our benchmark regressions are cast in growth rates, and this should take into account any trends in the level of employment across manufacturing industries. In addition, we have industry-state fixed effects that control for (a) the trend in growth rates that differ by state and industry and (b) industries' self-selected locations across the states.
Our final significant robustness check concerns the usage of different measures of the weighted average tightening of C&I lending standards and the weighted average change in banks' willingness to originate consumer installment loans. We noted in the main text that banks were extremely unlikely to characterize their changes in lending standards as having "eased considerably" or "tightened considerably," or to characterize their changes in willingness to originate consumer installment loans as "much more willing" or "much less willing." Indeed, over the past two decades, with the exception of 2008, no more than a handful out of usually close to 200 bank responses over the course of a year have reported such extreme responses as can be seen in Tables A4 and A5. Similarly, if we use a more simple method of categorizing the survey results, namely, when we define the indicator variable as follows: if bank reported considerable or some easing of standards at time , if bank reported no change in standards, and if bank reported some or considerable tightening (and similarly for responses to the changes in willingness to originate consumer installment loans), we get the results shown in Table A6 that are quantitatively similar to those in Table 5.
Finally, we conduct a few more robustness checks (not shown). Our results are robust to exclusion of the growth rate of real GDP, inclusion of state-level GDP, alternative home equity definitions, and inclusion of lagged dependent variables. Our results also hold when we exclude from the sample "bank-friendly" states, such as Delaware and South Dakota, or large states, such as California. We find some evidence that the trade credit channel--captured by the interaction of an indicator of dependence on trade credit and the three-month Libor--was operational, but it is not statistically robust.
Duygan-Bump, B., A. Levkov, and J. Montroiol-Garriga (2010): Financing Constraints and Unemployment: Evidence from the Great Recession,Quantitative Analysis Unit Series Paper QUA10-6, Federal Reserve Bank of Boston.
Jarmin, R. S., S. D. Klimek, and J. Miranda (2009): The Role of Retail Chains: National, Regional and Industry Results, in Producer Dynamics: New Evidence from Micro Data, ed. by T. Dunne, J. B. Jensen, and M. J. Roberts, pp. 237-262. University of Chicago Press.
Lee, S. J., and V. Stebunovs (2012): Bank Capital Ratios and the Structure of Nonfinancial Industries, Finance and Economics Discussion Series 2012-53, Board of Governors of the Federal Reserve System (U.S.).
|NAICS||Description||Empl. Share (Percent)||Output Share (Percent)|
|312||Beverage and Tobacco Product Manufacturing||0.0||0.0|
|314||Textile Product Mills||0.0||0.0|
|316||Leather and Allied Product Manufacturing||0.0||0.0|
|321||Wood Product Manufacturing||0.4||0.4|
|323||Printing and Related Support Activities||0.4||0.4|
|324||Petroleum and Coal Products Manufacturing||0.1||2.3|
|326||Plastics and Rubber Products Manufacturing||0.5||0.8|
|327||Nonmetallic Mineral Product Manufacturing||0.4||0.5|
|331||Primary Metal Manufacturing||0.3||1.0|
|332||Fabricated Metal Product Manufacturing||1.1||1.3|
|334||Computer and Electronic Product Manufacturing||0.9||1.6|
|335||Electrical Equipment, Appliance, and Component||0.3||0.5|
|336||Transportation Equipment Manufacturing||1.2||2.9|
|337||Furniture and Related Product Manufacturing||0.4||0.3|
|Num. of observations||10062||10062||10062||10062|
|Num. of clusters||559||559||559||559|
|Num. of observations||10062||10062||10062|
|Num. of clusters||559||18||559×18|
|Num. of observations||10062||10062||10062|
|Num. of clusters||559||18||599×18|
|Num. of observations||10062||10062||10062|
|Num. of clusters||559||18||559×18|
|Num. of observations||10062||10062||10062|
|Num. of clusters||559||18||559×18|
|Num. of observations||10062||10062||10062|
|Num. of clusters||559||18||559×18|
|NAICS||Rajan and Zingales (1998):1980s||Cetorelli and Strahan (2006):1980-1997||Our sample:1990-2011|
|NAICS||Rajan and Zingales (1998): 1980s||Cetorelli and Strahan (2006): 1980-1997||Our sample:1990-2011|
|Num. of observations||10062||10062||10062||10062|
|Num. of clusters||559||559||559||559|
Note: The dependent variable is growth in employment for a given industry in a given state. denotes the Rajan-Zingales measures of dependence on external finance scaled to the unit interval, the Braun-Claessens-Laeven measures of asset tangibility, a dummy for industries that depend relatively more on external finance, a dummy for industries that have relatively more tangible assets, and a dummy for industries that produce durable goods. Coefficients are reported with * if significant at the 10% level, ** at the 5% level, and *** at the 1% level. Column (1) and (2) use the raw Rajan-Zingales measures transformed to lie on a interval, column (3) uses an measure based on the bank loans to assets ratio based on the 1998 SSBF, and column (4) includes industry-specific time trends. All specifications include industry-state fixed effects. Standard errors are clustered by . t-statistics are reported below the coefficients. Industry-specific trend coefficients are not shown.
|Year||Number of bank responses: Tightened considerably||Number of bank responses: Tightened somewhat||Number of bank responses: No change||Number of bank responses: Eased somewhat||Number of bank responses: Eased considerably|
|Year||Number of bank responses: Much more willing||Number of bank responses: Somewhat more willing||Number of bank responses: About unchanged||Number of bank responses: Somewhat less willing||Number of bank responses: Much less willing|
|Num. of observations||10062||10062||10062|
|Num. of clusters||559||18||559×18|