Financing Micro and Small Firms in the Great Recession Megha Patnaik Job Market Candidate November 3, 2016 Department of Economics Stanford University Motivation I Small businesses high share of firms, employment, payroll I Face asymmetric information in credit markets I More financially constrained Constraints high in recessions: I I Great Recession - Kauffman firm survey (2008): I I I I A third of firms were denied or had trouble accessing credit 18% did not apply for funding fearing rejection Transmit shocks and amplify fluctuations (Bernanke, 1983) Linked to aggregate dynamics in the Great Recession I entry, exit and employment growth .15 Mean credit growth 0 .05 .1 -.05 2007 2008 2009 2010 Year Small businesses 2011 2012 2013 Compustat Annual QoQ growth in average total liabilities for small businesses in the sample and Compustat firms. Both firm-level datasets are filtered to keep only firms with at least 4 quarters of data and a moving average of three quarters taken over growth. Sample data is restricted to positive values of long-term liabilities for firm-quarters and winsorized at the 1% level. This Paper Question: Do credit-supply shocks affect credit for small businesses? How do these effects vary across firm size? Two shocks in the Great Recession: I Failure of insolvent banks with FDIC intervention I lending relationships (Petersen & Rajan 1994, Drexler & Schoar 2014) I Movements in house prices I owner’s personal collateral (Evans & Leighton 1990, Hurst & Lusardi 2004) Data from the leading online accounting software: I Very small firm sizes including micro firms (BLS: <10 employees) I Data on financials - credit I Measures of both shocks at firm level - links to banks and owner’s home address Challenge Demand side shocks: I County-quarter fixed effects as controls Bank failures: I FDIC insured banks are dispersed with many branches I Performance of firms doesn’t predict bank failure House price movements: I House prices determined at owner’s home ZIP I Tradable industries (Mian and Sufi, 2009) Result Differences in sensitivity to shocks across firm size I micro firms: 2-10 employees I small firms: 10-250 employees Credit: Micro Banking Housing ! Small ! Sensitivity of small business credit to banking shocks driven by small firms and to housing collateral shocks driven by micro firms Context Different sources for micro and small firmsSurvey of Small Business Finances (2003): Share collateralized 1221122112211221 Micro firms Small firms Housing collateral 27% 13% Business value 9% 29% Cost to banksI Lending relationships costly I Loan volume proportional to firm size Contribution I Two types of shocks to credit supply in the Great Recession I Banking shocks I I Greenstone, Mas, Nguyen (2014);Chodorow-Reich (2014); Nguyen (2015); Darmouni (2016); Duygan-Bump, Levkov, Montoriol-Garriga (2015), Amiti, Weinstein (2013) Housing shocks I Chaney, Sraer, Thesmar(2012); Adelino, Schoar, Severino (2012); Ersahin,Irani (2015); Schmalz, Sraer, Thesmar(2013); Fort, Haltiwanger, Jarmin, Miranda (2013) I Both banking shocks and housing shocks at the firm-level for small businesses I Sensitivity to banking driven by small firms, to housing driven by micro firms Overview I Data I Banking shock I Housing shock I Revenue and credit I Conclusion and future research Overview I Data I Banking shock I Housing shock I Revenue and credit I Conclusion and future research Firm-data Transactions-level data for &140,000 companies from the leading online accounting software: I Financial variables - short and long-term credit, revenue, expenses I Background characteristics - Employment, Address of owner and firm, industry and age I Companies linked to banks (&77,000 firms) Variable definitions I Credit: Sum of all positive long-term liabilities in a quarter I loans and credit lines, SBA loans, loans from family, transfers from personal bank account I Employment: Hiring and release dates of payroll employees I Location: Business address and the owner’s home address I Industry: home address NAICS industry from Dun & Bradstreet I Age: minimum of age from software and from Dun & Bradstreet Summary statistics aaaaaaaaaaaaaaaaa Meanui Std.Devui uaMinu aaaaaMax Median All firms: Size (employees) Revenue ($) Credit ($) Credit ($) >0 11.92 37.13 0 5103 3 1,557,643 430,244,608 0 9,570,071 317,640 62,521 6,943,580 0 842,454 0 335,579 16,083,934 40 4,053,002 39,544 2.72 2.54 0 9 2 872,260 55,399,336 0 7,040,920 228,486 Micro firms: Size (employees) Revenue ($) Credit ($) Credit ($) >0 55,375 8,097,182 0 652,750 0 347,510 20,281,938 43 3,919,374 35,328 Small firms: Size (employees) Revenue ($) Credit ($) Credit ($) >0 36.59 65.00 10 255 20 3,396,628 820,702,464 0 14,115,132 715,018 81,695 1,287,384 0 1,299,130 0 315,857 2,516,734 35 4,321,989 46,262 Sample is 844,882 firm-year observations for the 141,678 firms in the sample. Size distribution Employmenta1221k 0-4 5-9 10-14 15-19 20-24 25-49 50-99 100+ 1Share (Census)1 61.89 17.34 6.82 3.54 2.17 5.78 1.31 1.14 1a Share (Sample) 49.14 19.24 9.39 5.55 3.65 7.42 3.59 1.94 Mid-March employment shares in the population and the sample for 2010. Population statistics are sourced from the Statistics of U.S. Businesses published by the Census Bureau (total number of firms is 5,734,538). The number of employees is sourced from hiring and release dates of employees for 2010 (total number of firms is 76,918). Age distribution Age (years) 0 1 2 3 4 5 6-10 11-15 16-20 21-25 26+ 1Share (Census)1 8.93 6.67 5.50 5.13 5.29 4.96 20.17 14.04 10.05 7.91 11.36 1a Share (Sample) 1.51 7.97 10.87 8.18 7.99 7.99 25.25 12.09 6.02 3.69 8.43 Comparison of population for 2012 from Business Dynamics Statistics and the sample for 2012 March on age, with 4,577,659 firms in the population and 91,571 in the population. Sector distribution Sectoraaddad2222diii Service Retail Construction Manufacturing Mining Agriculture aa Share (Census) aa 70.91 11.97 11.44 4.87 0.43 0.38 aa Share (Sample) 77.00 7.85 9.01 4.68 0.24 1.19 Distribution of firms across 1 digit NAICS Sectors for March 2010. Population statistics from the Statistics of U.S. Businesses, US Census Bureau. The total number of firms is 5,734,538. Sample data uses the industry from matching to Dun and Bradstreet for 76,918 firms in 2010. Firms under “Unclassified” and “Public Administration” have been removed. Overview I Data I Banking shock I Housing shock I Revenue and credit I Conclusion and future research Bank failures I FDIC assisted bank failures from 2007 to 2013 I 530 banks were dissolved and merged into healthier institutions I 130 are matched to the banks firms in the data hold accounts with Estimation OLS at firm-quarter level I Separately for micro and small firms Log (Creditit ) = βFailit + θtc + fi + eit I Outcome: Log (Creditit ) I Failit : dummy for quarter of bank failure and upto 6 following quarters I θtc : county-quarter fixed effect I fi : 6 digit NAICS fixed effect / firm fixed effect Bank failure and firm credit Log Credit All All Micro Micro Small Small -0.606*** -0.296*** -0.192 0.005 -0.716*** -0.361*** (0.170) (0.099) (0.127) (0.091) (0.220) (0.129) 235,790 235,790 135,253 135,253 100,537 100,537 Qtr-County Yes Yes Yes Yes Yes Yes NAICS2 Yes Failure Obs Firm Yes Yes Yes Yes Yes The independent variable takes value 1 for the quarter the firm faces bank failure and the following 6 quarters. The sample is all firms with linked banks. Columns (1) and (2) is the sample of firms of all sizes, columns (3) and (4) is restricted to micro firms (with less than 10 employees), columns (5) and (6) is small firms (which have more than 10 employees). Credit is measured as the sum of all transactions categorized as long-term liabilities to a firm. The dependent variable is Log(Credit) measured as new long-term liabilities to the firm. Dependent variables are winsorized at the top and bottom 1%. Regressions are weighted by employment. All standard errors are clustered at the firm level. Cutoff: micro and small = 5 Log Credit Bank Failure Observations All All -0.606*** -0.296*** (0.170) (0.099) aMicro a aMicro a Small Small -0.149 0.052 -0.641*** -0.316*** (0.128) (0.116) (0.186) (0.108) 235,790 235,790 74,705 74,705 161,085 161,085 Qtr-County Yes Yes Yes Yes Yes Yes NAICS2 Yes Firm Yes Yes Yes Yes Yes The independent variable takes value 1 for the quarter the firm faces bank failure and the following 6 quarters. The sample is all firms with linked banks. Credit is measured as the sum of all transactions categorized as long-term liabilities to a firm. The dependent variable is Log(Credit) measured as new long-term liabilities to the firm. Dependent variables are winsorized at the top and bottom 1%. Regressions are weighted by employment. All standard errors are clustered at the firm level. Cutoff: micro and small = 15 Log Credit Bank Failure Observations All All Micro Micro Small Small -0.606*** -0.296*** (0.170) (0.099) -0.241** -0.135 -0.746*** -0.382*** (0.116) (0.083) (0.253) (0.148) 235,790 235,790 166,019 166,019 69,771 69,771 Qtr-County Yes Yes Yes Yes Yes Yes NAICS2 Yes Firm Yes Yes Yes Yes Yes The independent variable takes value 1 for the quarter the firm faces bank failure and the following 6 quarters. The sample is all firms with linked banks. Credit is measured as the sum of all transactions categorized as long-term liabilities to a firm. The dependent variable is Log(Credit) measured as new long-term liabilities to the firm. Dependent variables are winsorized at the top and bottom 1%. Regressions are weighted by employment. All standard errors are clustered at the firm level. Monotonic in firm size Coefficient of Log Cred on closure -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 Coefficient across employment size bins 2-5 6-10 11-50 51-101 101-200 200+ Employee size bin 95% confidence interval beta Correlation of firm credit to bank failures across firm size. The x-axis is based on standard size categories followed by the US Census Bureau and the y-axis is the coefficient of the regression with dependent variable log credit and independent variable a dummy that equals 1 if the firm is impacted by bank closure in the current or the previous year. Fixed effects are at the NAICS 2 and Year-State levels. Standard errors are clustered at the firm level. Cutoff: Lag Employment (2 years) Log Credit Bank Failure Observations All All aMicro a aMicro a Small Small -0.606*** -0.296*** -0.183 0.049 -0.711*** -0.383*** (0.170) (0.099) (0.130) (0.092) (0.219) (0.129) 235,790 235,790 118,862 118,862 116,928 116,928 Qtr-County Yes Yes Yes Yes Yes Yes NAICS2 Yes Firm Yes Yes Yes Yes Yes The independent variable takes value 1 for the quarter the firm faces bank failure and the following 6 quarters. The sample is all firms with linked banks. Credit is measured as the sum of all transactions categorized as long-term liabilities to a firm. The dependent variable is Log(Credit) measured as new long-term liabilities to the firm. Dependent variables are winsorized at the top and bottom 1%. Regressions are weighted by employment. All standard errors are clustered at the firm level. Size and age 5 Mean Age 10 15 Average age across size bins 9.0 10.8 12.2 12.3 12.6 6-10 11-50 51-100 101-200 200+ 0 6.7 2-5 Average age for firms in different size bins. Standard size bins as used by the US Census Bureau. Robust to controlling for age Log Credit All All Micro Micro Small Small Bank Failure -0.626*** -0.305*** -0.185 -0.013 -0.739*** -0.372*** (0.173) (0.101) (0.131) (0.094) (0.222) (0.131) Log Age -0.087*** 0.046 -0.110*** 0.101** -0.142*** 0.025 (0.024) (0.063) (0.016) (0.045) (0.032) (0.089) 224,827 224,827 125,958 125,958 98,869 98,869 Qtr-County Yes Yes Yes Yes Yes Yes NAICS2 Yes Observations Firm Yes Yes Yes Yes Yes The independent variable takes value 1 for the quarter the firm faces bank failure and the following 6 quarters. The sample is all firms with linked banks. Credit is measured as the sum of all transactions categorized as long-term liabilities to a firm. The dependent variable is Log(Credit) measured as new long-term liabilities to the firm. firm age measured as the difference in years between the current year and the minimum of the first year of business recorded in Dun and Bradstreet of the firm and the registration date of the firm in the software. Dependent variables are winsorized at the top and bottom 1%. Regressions are weighted by employment. All standard errors are clustered at the firm level. Selection problem: Balance in 2006 Variable Failure No failure Credit Log(Credit) Log(Credit/Sales) Employment Log Employment 61,401 10.42 -2.57 12.26 1.86 51,166 10.36 -2.89 12.54 1.95 Diff. p-value (Raw) 0.284 0.727 0.072 0.819 0.145 Diff. p-value (with FE’s) 0.174 0.965 0.116 0.695 0.322 Balancing tests for firms that faced bank failures and firms which did not in 2006. Differences in p-values calculated directly as well as with county and 2 digit industry fixed effects Placebo: No effect 6 quarters prior to failure Log Credit All All Micro Micro Small Small -0.110 0.069 -0.147 -0.040 -0.102 0.098 (0.188) (0.116) (0.153) (0.098) (0.228) (0.144) 137,133 137,133 65,717 65,717 71,416 71,416 Qtr-County Yes Yes Yes Yes Yes Yes NAICS2 Yes aaaaaaaavaaaaaaaaaa Bank Failure - 6 Observations Firm Yes Yes Yes Yes Yes Placebo test for the response of firm credit to bank closure measured six quarters before bank failure. The sample is all firms with linked banks. Columns (1) and (2) is the sample of firms of all sizes, columns (3) and (4) is restricted to micro firms (with less than 10 employees), columns (5) and (6) is small firms (which have more than 10 employees). Credit is measured as the sum of all transactions categorized as long-term liabilities to a firm. The dependent variable is Log(Credit) measured as new long-term liabilities to the firm. Dependent variables are winsorized at the top and bottom 1%. Regressions are weighted by employment. All standard errors are clustered at the firm level. Credit around the event of bank failure Pattern of credit - quarters around bank failure: 1. 2. 3. 4. Can examine for the affected firm, need to control for trend Unaffected firm can act as control Do not have date of failure for unaffected bank Use matching Matching Firms with similar propensity score based on pre-failure variables I Propensity score based on naics4, state, log(age) and log(credit) in the previous year with caliper 0.01 for the distance in propensity score within a year I Difference in Log Credit for years leading up to closure and after between firms whose banks failed and similar firms whose banks did not fail I Difference significant for 2008 and 2009 during the Recession, when the highest number of bank closures occured. I Standard errors bootstrap the difference in log credit between the two types of firms Credit around bank failure -2-1.5-1 -.5 0 .5 1 1.5 2 2.5 3 Coefficient of Log Cred on closure Difference in log credit around closure (matched) -5 0 5 10 Quarters from closure 1 Standard Deviation beta Difference between log credit around bank closure of small firms whose banks failed and matched firms whose banks did not fail. Firms matched using propensity score based on 2 digit NAICS, state, log employment and log age a year before closure, with one match per affected firm and caliper for propensity score 0.01. Standard errors are bootstrapped with 500 draws from the sample. No effect 6 quarters after failure Log Credit aaaaaaaavaaaaaaaaaa Bank Failure - 6 Observations All All Micro Micro Small Small -0.353* 0.053 0.013 -0.084 -0.410* 0.075 (0.182) (0.140) (0.154) (0.120) (0.242) (0.190) 118,806 118,806 67,330 67,330 51,476 51,476 Qtr-County Yes Yes Yes Yes Yes Yes NAICS2 Yes Firm Yes Yes Yes Yes Yes Response of firm credit to bank closure measured six quarters after bank failure. The sample is all firms with linked banks. Columns (1) and (2) is the sample of firms of all sizes, columns (3) and (4) is restricted to micro firms (with less than 10 employees), columns (5) and (6) is small firms (which have more than 10 employees). Credit is measured as the sum of all transactions categorized as long-term liabilities to a firm. The dependent variable is Log(Credit) measured as new long-term liabilities to the firm. Dependent variables are winsorized at the top and bottom 1%. Regressions are weighted by employment. All standard errors are clustered at the firm level. Heterogeneity: banking relationships Number of relationships at time of bank failure I 75% have 1 linked bank, 20% have 2, 5% have more Log (Creditit ) = βFailit + Log (Empit ) + θtc + fi + eit I Outcome: Log (Creditit ) I Failit : dummy for quarter of bank failure + 6 quarters after I θtc : county-quarter fixed effect I fi : 2 digit NAICS fixed effect Higher effect for fewer linked banks Log Credit All Bank Failure Obs 1 bank 2 banks -0.646*** (0.110) Small 3-5 banks 1 bank 2 banks 3-5 banks -0.485*** -0.203 -0.758*** -0.562*** -0.202 (0.129) (0.148) (0.142) (0.166) (0.188) 234,274 231,923 231,065 99,806 99,011 98,666 Qtr-County Yes Yes Yes Yes Yes Yes NAICS2 Yes Yes Yes Yes Yes Yes Placebo test for the response of firm credit to bank closure measured six quarters before bank failure. The sample is all firms with linked banks. Columns (1) and (2) is the sample of firms of all sizes, columns (3) and (4) is restricted to micro firms (with less than 10 employees), columns (5) and (6) is small firms (which have more than 10 employees). Credit is measured as the sum of all transactions categorized as long-term liabilities to a firm. The dependent variable is Log(Credit) measured as new long-term liabilities to the firm. Dependent variables are winsorized at the top and bottom 1%. Regressions are weighted by employment. All standard errors are clustered at the firm level. Overview I Data I Banking shock I Housing shock I Revenue and credit I Conclusion and future research House price data Zillow House Price Index: I Monthly time series of house price index at the zipcode level for the US from 2007 to 2013 I Index uses all houses with estimated price even without sale (single family + condominium + cooperative) I Advantage over other indices: at the zipcode level - highly correlated across indices (Guerrieri et al 2013) I Average over months to create quarterly series Estimation OLS at firm-quarter level I Separately for micro and small firms Log (Creditit ) = βLog (HPIzt ) + θtc + fi + eit I Outcome: Log (Creditit ) I Log (HPIzt ): log of the house price index at the owner’s zipcode (quarterly) I θtc : county-quarter fixed effect I fi : 6 digit NAICS fixed effect/firm fixed effect House prices and firm credit Log Credit Log HPI Observations All All Micro Micro Small Small 0.090 -0.043 0.225*** 0.339*** 0.083 0.281* (0.055) (0.185) (0.031) (0.087) (0.065) (0.151) 448,866 448,866 245,522 245,522 203,344 203,344 Qtr-County Yes Yes Yes Yes Yes Yes NAICS2 Yes Firm Yes Yes Yes Yes Yes The correlation between firm credit and house prices. The dependent variable is the log of credit determined by aggregating all transactions which are long-term liabilities to the firm, and winsorized at the top and bottom 1%. Standard errors are clustered at the firm level. Monotonic in firm size Coefficient of Log Cred on log HPI -.8 -.6-.5-.4-.3-.2-.1 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1 Coefficient across employment size bins 2-5 6-10 11-50 51-100 100+ Employee size bin 95% confidence interval beta Correlation of firm credit with house price index across firm size. The x-axis is based on standard size categories followed by the US Census Bureau and the y-axis is the coefficient of the regression with dependent variable log credit and independent variable the ZIP code level house price index measured at the owner’s address. Fixed effects are at the NAICS 2 and Year-State levels. Standard errors are clustered at the firm level. Tradability I County-Quarter FE’s to control for demand may be insufficient I Tradable industries - demand is less localized Tradable sectors (Mian and Sufi 2009) I I I I 4 digit NAICS industry as tradable if it has the sum of imports and exports to be higher than $10,000 per employee or exceeding $500 million. Retail industries, restaurants and grocery are non-tradable. Strategy follows Adelino, Schoar, Severino (2014) Tradability - micro firms Log Credit dad Log HPI Observations 22Allda All - Construction All - Non-Tradables Manufacturing 0.211*** 0.220*** 0.216*** 0.171 (0.027) (0.029) (0.030) (0.142) 259,742 233,776 214,359 17,929 Qtr-County Yes Yes Yes Yes NAICS2 Yes Yes Yes Yes The correlation between firm credit and house prices. The dependent variable is the log of credit determined by aggregating all transactions which are long-term liabilities to the firm, and winsorized at the top and bottom 1%. Standard errors are clustered at the firm level. Overview I Data I Banking shock I Housing shock I Conclusion and future research Aggregate effects House price changes - 1. Zillow price index: highest $196K (Feb-June 2007) to lowest $152K (Nov 2011-Apr 2012) = decline of 22.44% 2. Micro firms have coefficient β from Log(Credit) & Log(HPI) = 0.33: corresponds to 7.4% change in credit 3. Mean annual credit volume for micro firms is & $350K this gives a change of & $25K 4. Median annual revenue for micro firms is & $200K which means a loss of credit of as a share of revenue & 12.5% Bank failures - 1. Bank shock led to 30% decline in credit level (1-exp(-0.36))*100 2. Mean annual credit for small firms is & $316K: a decline of & $95K 3. Median annual revenue for small firms is & $715K: a loss of credit as a share of revenue of & 13.25% Conclusion I Sensitivity of credit to bank closure - small firms I I I I not driven by selection robust to age controls decreases with more banking relationships Sensitivity of credit to housing collateral - micro firms I holds for tradable industries I Both effects monotonic and not sensitive to cutoff I Aggregate effects of shocks to credit supply: I I House prices: average micro firm credit decline&12.5% of revenue Bank failures: average small firm credit decline ˜13.25% of revenue Future research I Existing financial constraints I I I Response of firms to shock I I I I I I Leverage before credit supply shock Response of net credit (involves repayments for existing loans) Employment Wages Revenue and Profitability Expenses - different types Investment Broader banking shocks Thank you Megha Patnaik [email protected] Back-Up Slides Small businesses in the US economy Employees <500 <50 <20 Firms 99% 95% 90% Share Emp Payroll Sales Source: Statistics of US Businesses (2013) Back Banking relationships Banking and housing shocks Log Credit Bank Failure All All Micro Micro Small Small -0.629*** -0.339*** -0.123 -0.043 -0.765*** -0.399*** (0.142) (0.201) (0.111) (0.139) (0.105) (0.260) 0.145** -0.046 0.218*** 0.286 0.146* -0.109 (0.069) (0.297) (0.042) (0.244) (0.085) (0.373) 180,679 180,679 105,748 105,748 74,931 74,931 Qtr-County Yes Yes Yes Yes Yes Yes NAICS2 Yes Log(HPI) Observations Firm Yes Yes Yes Yes Yes Relationship between log credit and housing and banking shocks. Bank failure is a dummy that equals 1 for the quarter the firm faces bank failure and the following 6 quarters. House price measure is log of the Zillow monthly index at the ZIP code of the owner’s address, averaged over months in a quarter. The sample is all firms with linked banks. Columns (1) and (2) is all firms, columns (3) and (4) is micro firms, columns (5) and (6) is small firms. Credit is measured as the sum of all transactions categorized as long-term liabilities to a firm. All regressions have been weighted by the number of employees. Standard errors are clustered at the firm level. Back Borrowing frequency Number of months in a year that firms borrow. The sample is for 141,678 firms restricted to those with at least one year of borrowing in the dataset. House prices and firm credit - Banking sample Log Credit aaaaaaaaaaaaaa Log HPI Observations All All Micro Micro Small Small 0.090 -0.043 0.225*** 0.339*** 0.083 0.281* (0.055) (0.185) (0.031) (0.087) (0.065) (0.151) 448,866 448,866 245,522 245,522 203,344 203,344 Qtr-County Yes Yes Yes Yes Yes Yes NAICS2 Yes Firm Yes Yes Yes Yes Yes The correlation between firm credit and house prices. The dependent variable is the log of credit determined by aggregating all transactions which are long-term liabilities to the firm, and winsorized at the top and bottom 1%. Standard errors are clustered at the firm level. Tradable firms - all Log Credit dad 22Allda All - Construction All - Non-Tradables Log HPI 0.165*** 0.171*** 0.176*** 0.182 (0.023) (0.024) (0.025) (0.128) 448,877 401,785 367,345 22,930 Observations Manufacturing Qtr-County Yes Yes Yes Yes NAICS2 Yes Yes Yes Yes The correlation between firm credit and house prices. The dependent variable is the log of credit determined by aggregating all transactions which are long-term liabilities to the firm, and winsorized at the top and bottom 1%. Standard errors are clustered at the firm level.
© Copyright 2026 Paperzz