Dynamism Diminished: The Role of Housing Markets and Credit Conditions

Transcription

1 Dynamism Diminished: The Role of Housing Markets and Credit Conditions Steven J. Davis and John Haltiwanger Chicago Booth & NBER U. of Maryland & NBER 5 December 2016 University of Chicago

2 Overview 1. Longer-term context A. Secular Declines in Business Formation Rates and Young-Firm Activity Shares B. Secular Declines in Business Dynamism and Labor Market Fluidity Large, pervasive declines in Job Reallocation Rates in past 35 years, accelerating after 2000 Large, pervasive declines in Worker Reallocation and Churn Rates since Young firms were hit especially hard by the Great Recession 3. Questions: What role for credit conditions in the fortunes of young firms? What role for the housing market boom and bust? 4. Our approach: Exploit spatial and industry time-series variation to assess the forces behind young-firm (and small-firm) activity firms. A. House price movements, bank lending conditions, local economic conditions B. Use IV estimation and locally exogenous bank loan supply shifts to identify effects 5. Use industry characteristics (output tradability, labor intensity, working capital, startup costs), firm age structure of local industries, local share of national banks, and distinction between young firms and young establishments to throw light on transmission channels. In progress 2

3 Share of Employees in Young Firms, , U.S. Nonfarm Private Sector Source: Annual Rates, Business Dynamic Statistics (BDS) Employment in firms less than five years old fell from about 18% of private sector employment in 1981 and to 8.2% in Young means < 60 months since the firm s first paid employee as of March in the indicated calendar year Firm age is set to age of its oldest establishment when the firm first becomes a legal entity, and increments by 1 each year thereafter. Establishment age is the number of years since 3 operations began at the location in the same narrowly defined industry.

4 3 The Young-Firm Share of Employment Exhibits Pronounced and Time Varying Cyclicality 2 1 Log Points For each expansion and contraction episode, the chart shows annualized deviations from the overall mean, which equals -2.3 log points per year. 4

5 (Local) Housing Prices and (Local)Young-Firm Activity: Potential Transmission Channels 1. Wealth and Risk Tolerance: Home equity up à greater willingness to take on risks of new/young business. (Khilstrom-Laffont, 1979, Guiso-Paiella, 2008) 2. Wealth Effect on Desire to Be Own Boss: Demand for being own boss is a normal good and increases in wealth. (Hurst and Pugsley, 2015) 3. Liquidity and Collateral Effects: Households tap home equity to relax liquidity constraints, increasing their ability to finance new/young businesses. Higher house prices à greater collateral value. (Evans and Jovanovic, 1989) 4. Local Credit Supply: Local housing conditions affect local banks lending capacity + young firms are relatively dependent on local bank credit. 5. Local Outlook and Credit Supply: Banks see local housing prices as indicators of (future) local business conditions, affecting their willingness to lend; and new and young firms are relatively dependent on bank credit. 6. Non-uniform Consumption Expenditure Responses: Young firms supply goods 5 and services whose demand is relatively sensitive to local income/wealth.

6 Our Data Sources Business Dynamic Statistics (BDS): Annual activity for firms with paid employees, with tabulations by firm age at State and MSA level. Quarterly Workforce Indicators (QWI): Similar to BDS in key respects, but includes tabs by age-industry-msa. Covers shorter time period than BDS. Local area unemployment rates (LAUS): This BLS program uses CPS data, UI claims, CES data and other sources to estimate local unemployment rates. Local house price measures (FHFA): Federal Home Loan Finance Agency data, available monthly at MSA and State Level. Saiz MSA housing supply elasticity: To instrument for house price changes Local bank loan supply (CRA): Community Reinvestment Act data. Banks with assets >1 billion report # and volume of loans to businesses with <$1 million in gross revenue. We use to construct local small business loan supply shocks. Quarterly Census of Employment and Wages (QCEW): Employment at MSAindustry level. To construct Bartik-like instruments for local cycle variables. 6

7 Young Firm Outcomes at the State-Year Level, Using Data from 1981 to

8 Log Difference in Young Firm Employment Share, Unemployment Rates and Housing Prices in State-by-Year Data from 1981 to 2013 Slope=-1.77 (0.13) Employment share of young firms rises when local housing prices increase Slope=0.30 (0.02) Simple bivariate relationships show great variation across states and time. Employment share of young firms declines when local economic conditions deteriorate (as measured by local unemployment rate). 8

9 Log Difference in Young Firm Share, Log Difference in Startup Rate and Housing Prices in State-by-Year Data from 1981 to 2013 Slope=0.28 (0.01) Startup rate of new firms rises when local housing prices increase Slope=0.25 (0.03) Share of young firms rises when local housing prices increase 9

10 Identification, Estimation and Quantification 10

11 Identification Instrument local house price changes to identify effects on local young-firm activity shares. Three IV approaches that use differences in MSA housing supply elasticities: 1. Following Mian & Sufi, exploit national housing boom & bust episodes that affect MSA-level house prices differently because of differences in local housing supply elasticities 2. Exploit local area demand shifts interacted with local housing supply elasticities to instrument for local house price changes. 3. Build on Approach 2 by using Bartik-type measure to also instrument for local demand shift control variable. Supplement with local small business loan shocks by adapting approach of Greenstone, Mas and Ngyugen (2015). Fits well with IV approaches 2 and 3 above. A shorter times-series dimension, because CRA data are only available from 1996 The paper also considers panel VAR specifications, with shocks identified by Cholesky orderings. Not a focus of today s talk. Panel VARs yield similar results to our IV regressions. They let one answer some additional questions at the cost of different, less plausible, identifying assumptions. 11

12 IV Approach (1): Stacked Boom/Bust Episodes Y "# = λ # + λ " + β HP "# + ε "# (1) Outcome variable: Log change in MSA young-firm employment share HP = log change in MSA house price index Compute changes over Boom ( ) and Bust ( ) periods, indexed by s. This approach exploits large national shocks to housing demand and uses differential local house price responses due to different housing supply elasticities. It follows Mian and Sufi (2011, 2013, 2014), except we stack the boom and bust changes. We consider different outcomes than M&S and have many more areas but lack within-area controls for micro units. Our instruments for MSA Housing Price Growth: Cubic polynomial in Saiz housing supply elasticity for MSA interacted with period effects, λ # 444 observations: 222 Boom changes Bust changes Stacking boom and bust episodes lets us control for arbitrary differences in MSA-level trends in the 2000s, addressing concerns that MSA-level economic trends are correlated 12 with MSA-level housing supply elasticities, as argued by Davidoff (2015).

13 Bivariate MSA- Level Relationship of House Price Growth to Log Change in Young- Firm Share in Boom, Bust and Other Periods All three panels show annualized log changes Log Diff Share Young Emp Log Diff Share Young Emp Log Diff Share Young Emp Boom ( ) Growth in Real Housing Price Slope = 0.191, SE = Bust ( ) Growth in Real Housing Price Slope = 0.182, SE = Growth in Real Housing Price Slope = 0.218, SE = Other Periods 13

14 Response of Log Difference in Young-Firm Share to Housing Price Growth, IV Approach (1): Stacked Boom/Bust Growth in real housing price (1) (2) (3) (4) OLS (Boom/Bust) IV (Boom/Bust) OLS (Boom/Bust) IV (Boom/Bust) *** *** *** *** (0.042) (0.077) (0.051) (0.063) F-Test for Excluded Instruments Period Effects Yes Yes Yes Yes MSA Effects No No Yes Yes R Observations Controlling for Differential MSA- Level Trends Notes: Boom ( ), Bust ( ). Instruments are period effects interacted with cubic in (log) Saiz elasticity. Standard errors in parentheses. 14

15 Second-Stage Relationship between Log Change in Young-Firm Share and House Price Growth Column 2 in the Previous Table Bust Boom The specification includes period controls. See Column (2) in the table for standard errors that are adjusted for the two-stage nature of the estimation. Slope =.24 (.04) 15

16 IV Approach (2): Using Interaction between Local Cycle and Local Housing Supply Elasticity Y "0 = λ " + αcyc "0 + βhp "0 + X 0 4 A + ε "0 (2) Outcome: Annual log change in MSA young-firm share from Controls: MSA effects, annual change in MSA unemployment rate, and national cycle controls (either a cubic in national GDP growth or Year effects). Our instruments for MSA Housing Price Growth: Cubic in log(saiz housing supply elasticity) for MSA interacted with CYC "0 Advantages relative to IV Approach (1): Ten times as many observations and a longer time period Readily accommodates inclusion of local loan supply shocks (below) Also let s us investigate robustness to a different source of IV variation. 16

17 IV Approach (2): Identification Y "0 = λ " + λ 0 + αcyc "0 + βhp "0 + ε "0 (2) HP "0 = δ " + δ 0 + θcyc "0 + CYC "0 Z " γ + η "0 (3) CYC "0 affects Y "0 directly and indirectly (through its effect on HP "0 ). What assumptions let us identify β? Exclusion restriction: CYC "0 Z ", the interaction between local cycle and local supply elasticity, affects Y "0 only through its effect on local house price growth, HP "0, conditional on controls. Auxiliary orthogonality condition: Cov(CYC "0, ε "0 )=0. Suppose this condition fails for CYC "0 = x "0 + c "0 with Corr(x "0, c "0 )<1. If Cov(ε "0,x "0 ) 0 and Cov(η "0, x "0 ) 0, then variation in x "0 generates movements in both HP "0 and Y "0, biasing the estimate of β even if the exclusion restriction holds. 17

18 IV Approach (2): Identification IV also addresses concerns that OLS yields a (downwardly) biased estimate of β due to measurement error in HP a serious concern, in our view, given the difficulties of constructing good house price indices. Why might Cov(CYC mt, ε mt )=0 fail? Reverse causality: Not a serious concern in our view, given that our lhs variable is log change in the young-firm employment share. The small size of the young-firm share also mitigates any reverse causality. Omitted Controls: CYC might not adequately control for local cycle conditions that affect local house price growth and Y "0 -- a serious concern in our view. IV approach (3) aims to address this concern by instrumenting for CYC. We will also try adding variables to better control for local conditions. 18

19 IV Approach (3): Identification Now treat CYC, as well as HP, as endogenous Expand and modify the instrument set to now include: Cubic polynomial in Bartik-type local demand shifter Bartik-type local demand shifter interacted with cubic in local housing supply elasticity That is, we drop the auxiliary orthogonality condition and adopt the new exclusion restriction that national shifts in industry demand interacted with local legacy industry structure (i.e., the Bartik measure) affect the local young-firm employment share only through effects on the local cycle variable, conditional on controls. 19

20 Bartik Variable Construction Details 1. Use QCEW data at the MSA X 2-digit NAICS level. 2. Let S E,",0FG = industry i share of employment in MSA m at t Calculate national industry log employment changes from t-1 to t using data for all MSAs other than m, yielding a set of g E,0 quantities 4. Compute Bartik demand shifter at t for MSA m as the S E,",0FG weighted sum of the national industry-level g E,0 values for m. 20

21 IV Estimates for Sample, IV Approaches (2) and (3) Dependent Variable: Log Change in Young-Firm Employment Share, MSA by Year Data OLS IV2 IV3 OLS IV2 IV3 Growth in real *** ** 0.265** *** * 0.313** housing price (0.027) (0.147) (0.127) (0.031) (0.154) (0.158) F-test for Excl Instruments Multivariate F-test for Excluded Instruments MSA Effects Yes Yes Yes Yes Yes Yes Year Effects No No No Yes Yes Yes Notes: Standard errors in parentheses clustered at MSA level. All specs include the change in the MSA unemployment Rate. Specifications without year effects include a cubic in National GDP Growth. For IV estimates, overidentification tests show we cannot reject the null of instrument validity. * p < 0.1, ** p < 0.05, *** p <

22 How Important Are House Price Movements for National Changes in Young-Firm Activity Shares? We use IV coefficient estimate and actual state-level house price changes from 1981 to 2013 to quantify national effects of housing market developments. By using all house price changes, we capture the effects of exogenous house price changes and the role of house prices in transmitting shocks that originate elsewhere. We aggregate state-level changes to the national level using state-level employment shares. Given correctly identified causal effects of local house price changes, our quantification exercise may overstate or understate the role of housing market developments in national young-firm activity shares: Overstate? Spatial equilibration of young-firm activity across local areas may attenuate national responses relative to the aggregated local responses. Understate? (1) Positive spillovers of young-firm activity across local areas. (2) Entrepreneurs may own houses outside the area where they operate young firms, an effect not captured by our regression model or aggregation. 22

23 Contribution of Housing Price Changes to Log Changes in Young-Firm Employment Shares By Cycle Episode Based on IV1 (Boom/Bust) Estimate of Coefficient on HP 3 Log Points About 50 percent of the (trend-deviated) decline in young-firm activity share in Great Recession is due to decline in housing prices, according to this exercise Solid Bar is Actual. Striped and Dotted Bars are counterfactuals implied by IV (1) approach with, respectively, controls for Period Effects only and MSA & Period Effects. Both counterfactuals use actual state-level house price changes. Annualized deviations from overall means depicted. The mean decline is -2.3 log points per year. 23

24 Year-By-Year results show that the housing boom attenuated the secular decline in young-firm employment shares from from and accelerated the decline after IV(1) estimates 10 Log Points Cumulative increase from from Housing Prices = 10 log points Recall that the mean change in the young firm employment share is -2.3 log points per year Cumulative decrease from from Housing Prices = 8.3 log points -10 Actual Housing Prices (IV Results) 24

25 Local Small Business Bank Loan Supply Shocks 25

26 IV Specification for Housing Prices and Small Business Loan Supply Shocks Outcome variable: Annual log change in MSA young-firm employment shares Controls: MSA effects, MSA Unemployment Rate Change, and National Cycle controls (quadratic in national GDP Growth rate). Estimated effects of interest: Housing price changes Small business loan supply shocks Estimate this specification by OLS, IV (2) and IV (3) using MSA-year data from 1998 to

27 Dependent Variable: Log Difference of Young Employment Share, Using MSA by Year Data Growth in real housing price Local Small Business Loan Shock F-test for Excluded Instruments Multivariate F-test for Excl. Instruments OLS OLS IV2 IV2 IV3 IV *** *** * ** 0.192* 0.182* (0.027) (0.027) (0.110) (0.111) (0.107) (0.107) *** *** 0.036*** (0.013) (0.012) (0.012) Notes: Standard errors in parentheses clustered at MSA level. All include the Change in MSA level Unemployment Rate. Specifications without year effects (first columns of OLS, IV2, IV3) include a cubic in National GDP Growth. For IV estimates, overidentification tests show we cannot reject the null of instrument validity. * p < 0.1, ** p < 0.05, *** p < ESTIMATES FOR Sample: IV estimates yield positive and statistically significant impact of housing prices and small business loan supply shocks on young-firm activity shares. Using small business activity share as outcome yields much weaker effects -- e.g., no effect of Small Business Loan Supply shock. 27

28 Contribution of Housing Price Changes and Small Business Bank Loan Supply Shocks to Log Changes in Young-Firm Employment Share by Cycle Episode Taken together, the decline in housing prices and bank loan supply shocks account for 68 percent of decline (relative to trend) in Young-Firm activity share in Great Recession During period these effects tended to boost Young-firm activity shares working against forces leading to diminished dynamism over this period of time Sold Bar is Actual, Diagonal Striped Bar is Counterfactual (Housing Prices only), Dotted Bar is Counterfactual (Loan Supply only), Horizontal Striped Bar is (Housing Prices + Loan Supply). Using IV3 estimates from Table 3. Annualized deviations from overall means depicted. The mean decline is 2.3 log points per year 28

29 Young versus Mature Firms? Or Small versus Large Firms? 29

30 Small-Firm Share of Private Sector Employees, Small = Less than 50 Employees 34 Share of Employee at Small Firms, , U.S. Nonfarm Private Economy Share of employment at small firms also declines secularly Percentage

31 Small-Firm Employment Share Growth by Cycle Episodes Log Points Log Differences in Small Employment Share by Cyclical Episodes The cyclical pattern here is essentially the opposite of what we saw in the analogous chart for the young-firm share. The pattern here reflects the size distribution fallacy stressed in Davis, Haltiwanger, and Schuh (1996). In recessions, many firms migrate from large to small. 31

32 Estimation Results: IV(2) Regression with CRA Shocks Table : MSA Level Regression from Dependent variables: Log Difference in Young/Small Employment Share (1) (2) (3) (4) Log Difference in Log Difference in Young Small Employment Employment Share(IV) Share (OLS) Log Difference in Young Employment Share (OLS) Log Difference in Small Employment Share(IV) Growth in Real Housing *** ** 0.056*** 0.105*** Price (0.027) (0.111) (0.010) (0.040) Local Small Business *** *** Loan Shock (0.013) (0.012) (0.005) (0.005) F-test for Weak Instruments Standard errors in parentheses clustered at MSA level. * p < 0.1, ** p < 0.05, *** p < 0.01 Control for MSA Change in Unemployment, cubic in national GDP growth rate, and MSA fixed effects Small Business Loan Supply Shocks have no effect on small-firm share but a nontrivial, statistically significant effect on young-firm share. Why? Young firms are predominantly small But small firms are heavily weighted to mature firms. These results indicates that youth, not smallness, is the key firm characteristic that makes for high sensitivity to changes in bank lending conditions 32

33 Growth Rate of Small-Firm Shares, Cycle Episodes 2.0 Log Points Housing price changes generate changes in small-firm shares that often go in the wrong direction. Why? Because the actual small-firm share changes are dominated by by cyclicality of firms that cross size-class boundaries. Loan supply shocks have no effect on small-firm employment shares -1.5 Actual HP only (IV, MSA only) CRA only (IV, MSA only) HP and CRA (IV, MSA only) Note: Counterfactuals from IV2 estimation 33

34 Estimation Results: IV Regression with CRA Shocks Table: MSA Level Regression from Dependent variables: Net Differential Young vs. Old /Small vs. Large (1) (2) (3) (4) Net Differential Net Differential Young vs. Old Small vs. Large (IV) (OLS) Net Differential Young vs. Old (OLS) Net Differential Small vs. Large (IV) Growth in Real *** *** *** *** Housing Price (0.026) (0.106) (0.013) (0.052) Local Small Business *** *** 0.010* 0.009* Loan Shock (0.011) (0.011) (0.006) (0.006) F-test for Weak Instruments Standard errors in parentheses * p < 0.1, ** p < 0.05, *** p < 0.01 Control for MSA Change in Unemployment, quadratic in national GDP growth rate, and MSA fixed effects A better way to measure the differential sensitivity of small firms is the net growth rate differential between small and large firms, where firms are assigned to a fixed size class in the initial and terminal year of the change interval. With this better measure, small business loan supply shocks have a small effect on the small-large net differential, but they have material and statistically significant effects on the young-old differential. 34

35 Exploring the Channels Through Which House Price Changes Affect Young-Firm Activity Shares 35

36 Exploring Channels Broad question: Through what channels do house price changes drive youngfirm activity shares? Immediate goal: Test the view that house price changes matter for young-firm shares only because they alter consumption demand patterns in ways that favor young firms. Idea: Test whether the local industry growth rate response to local house price changes depends on the local industry s age structure of employment. If house prices work entirely through consumption demand channels, then we expect the local industry response to be invariant to its firm age structure. The alternative view says that local industry response rises with its young-firm activity share due to wealth, collateral, and liquidity effects of house prices on relative propensity to start a new business or expand a young business. Implement using annual QWI employment data at 2-digit NAICS by MSA level. 36

37 Specifications for Implementing the Invariance Test GR K"0 = a + b G CYC "0 + b N HP "0 + b O empshare K",0FG +c V HP "0 empshare K",0FG + f 0 + f " + f K + ε K"0 (1) GR K"0 = a + b O empshare K",0FG +c V HP "0 empshare K",0FG +f "0 + f K + ε K"0 (2) GR K"0 = a + b O empshare K",0FG +c V HP "0 empshare K",0FG +f "0 + f K0 + ε K"0 (3) where j is industry, m is MSA, and t is time. Industry classifications based on 2-digit NAICS codes. GR K"0 is log employment change from t-1 to t for industry j in MSA m, and empshare K",0FG is the lagged young-firm share. 37

38 Detail on Industry Classifications for This Test 23 Construction Manufacturing 42 Wholesale Trade Retail Trade Transportation and Warehousing 51 Information 52 Finance and Insurance 53 Real Estate and Rental and Leasing 54 Professional, Scientific, and Technical Services 55 Management of Companies and Enterprises 56 Administrative and Support and Waste Management and Remediation Services 62 Health Care and Social Assistance 71 Arts, Entertainment, and Recreation 72 Accommodation and Food Services We omit the following industries because few MSAs have positive employment and/or QWI coverage limitations: Agricultural Services Mining Utilities Educational Services (Mostly non-profits in QWI) Other Services (Many religious organizations and other non-profits in QWI). The parts of 61 and 81 included in the QWI also have weak relationships to cyclical variables, including housing prices. 38

39 Dependent Variable: Annual Log Employment Change at Industry-MSA level OLS OLS OLS IV2 IV3 Changin *** *** *** Unemp. Rate (0.152) (0.185) (0.311) Housing Price *** *** *** Log Chang (0.011) (0.034) (0.035) Young-Firm *** *** *** *** *** Emp. Share (0.010) (0.010) (0.010) (0.010) (0.011) HP x Young_Sh *** *** *** *** *** (0.062) (0.078) (0.093) (0.112) (0.114) MSA FE Yes No No Yes Yes Year FE Yes No No Yes Yes Industry FE Yes Yes No Yes Yes MSA-by-year FE No Yes Yes No No Ind-by-year FE No No Yes No No N R Standard errors in parentheses. Clustered standard errors at MSA level. * p < 0.1, ** p < 0.05, *** p < 0.01

40 How Large Are Departures from Age Invariance of Local Industry Reponses to Local House Price Changes? Dispersion in Young-Firm Employment Shares Industry-MSA Young-Firm Share Boom Period Bust Period 90th Percentile th Percentile Std Deviation Dispersion in Local Log House Price Changes Log MSA House Price Change Boom Period Bust Period 90th Percentile th Percentile Std Deviation Consider the young-firm share differential across MSA-industry cells. Compute the implied log employment change differentials, evaluated at the 90 th and 10 th percentiles of MSA log house-price changes. Use IV3 coefficient estimate on the interaction term. Boom Period Bust Period Percentile of MSA House-Price Log Change 90p 10p 90p 10p Implied industry-msa growth differential, log points per year

41 Exploring Channels (Next Steps in Progress) Wealth effects on desire to be own boss are especially relevant for industries with low startup costs and low working capital requirements (i.e., Hurst and Pugsley (2011, 2015) industries). Wealth effects on desire to boss others are especially relevant for laborintensive industries and possibly industries with low wages per worker. Use BDS to distinguish responses of young establishments operated by mature firms from responses of young firms. Except for consumption demand effects, the channels we stressed involve young firms, not young establishments per se. Investigate changes over time in size of HP effects, perhaps reflecting changes in banks willingness to extend loans collateralized by residential real estate. Are housing price effects larger in states with larger market shares for local banks? The Local Credit Supply Channel we articulated on Slide 5 says yes. 41

42 Additional Slides 42

43 Growth in Real Housing Prices (National and Employment-Weighted State) 0.1 Growth in Real Housing Prices ( ) Our counterfactual exercises use Employment-weighted predictions of State-level outcomes from the contribution Of housing prices. This figure shows the Employment-weighted state-level growth rate In housing prices is virtually identical to the National growth of real housing prices National Weighted State (Employment-Weighted) 43

44 Young Firm Share of Firms and Young-Old Growth Differential Also Exhibit Pronounced and Time Varying Cyclicality Cyclicality of Log Changes in the Young-Firm Share of Firms 2 1 Net employment growth of young fell relative to old, especially in and Great Recession. Log Points Annualized deviations from mean. Mean=-1.7 log points Early 1980s and Great Recession saw especially big declines in share of young firms Cyclicality of Net Growth Rate Differential Between Young and Mature Firms Annualized deviations from mean. Mean=21 percent 44

45 Greater Sensitivity of Young and Small Firms to Credit Conditions? Gertler and Gilchrist (1994) highlight greater responsiveness of small firms to monetary policy and credit shocks. Supporting evidence in Sharpe (1994) and Chari, Christiano, and Kehoe (2013) Fort et. al. (2013) find local housing price shocks adversely affect young/small businesses. Robb and Wolken (2002) and Robb and Robinson (2012) highlight role of home equity for young firms. Fairlie and Krashinsky (2012) and Corradin and Popov (2015) find local local housing price gains increase transition rates into self employment. Adelino et al. (2015) find that small businesses experience stronger employment growth (compared to large businesses) in regions with greater housing price appreciation, which they interpret as evidence for the importance of housing collateral in lending to small businesses. Mian and Sufi (2011) highlight a net worth/aggregate demand channel of local housing price shocks on local non-tradables. No direct implication for young and small firms, but these firms may be more sensitive to such shocks for several reasons, as we discuss. Greenstone, Mas and Nguyen (2015) show connection between small business bank lending and small business employment growth at local level They use an IV method to identify supply shock contribution. Morgan, Rime and Strahan (2004) argue that banking deregulation made small firms less sensitive to local banking conditions. 45

46 Re-Doing the Quantification Exercises Using IV(2) and IV(3) Coefficient Estimates 46

47 Contribution of Housing Price Changes to Log Differences in Young-Firm Employment Share by Cycle Episode (IV2 Estimates) 3 Log Points About 60 (with year effects) to 75 (w/o year effects) percent of decline in Young-firm shares in Great Recession due to housing price fall Solid Bar is Actual, Striped Bar is Counterfactual (MSA effects), Dotted Bar is Counterfactual (MSA and Year Effects). In both counterfactuals, actual housing price is used. Difference reflects coefficient estimates from alternative models. Annualized deviations from overall means depicted. The mean decline is -2.3 log points per year. 47

48 Year-By-Year results show that the housing boom attenuated the secular decline in young firm employment share from from and accelerated the decline after IV2 estimates 10 Log Points Cumulative increase from from Housing Prices = 15 log points Recall that the mean change in the young firm employment share is -2.3 log points per year Cumulative decrease from from Housing Prices = 12 log points -10 Actual Housing Prices (IV Results) 48

49 Contribution of Housing Price Changes to Log Differences in Young-Firm Employment Share by Cycle Episode (IV3 Estimates) 3 Log Points About 64 (with year effects) to 54 (w/o year effects) percent of decline in young firm activity in Great Recession due to decline in housing prices Solid Bar is Actual, Striped Bar is Counterfactual (MSA effects), Dotted Bar is Counterfactual (MSA and Year Effects). In both counterfactuals, actual housing price is used. Difference reflects coefficient estimates from alternative models. Annualized deviations from overall means depicted. The mean decline is -2.3 log points per year. 49

50 Year-By-Year results show that the housing boom attenuated the secular decline in young firm employment share from from and accelerated the decline after IV3 Approach. Log Points Cumulative increase from from Housing Prices = 13 log points Recall that the mean change in the young firm employment share is -2.3 log points per year Cumulative decrease from from Housing Prices = 11 log points -10 Actual Housing Prices (IV Results) 50

51 Contribution of Housing Price Changes and Small Business Bank Loan Supply Shocks to Log Changes in Young-Firm Employment Shares by Cycle Episode Log Points Taken together, the decline in housing prices and bank loan supply shocks account for 75 percent of decline (relative to trend) in Young-Firm activity share in Great Recession During period these effects tended to boost Young-firm activity shares working against forces leading to diminished dynamism over this period of time. -5 Sold Bar is Actual, Diagonal Striped Bar is Counterfactual (Housing Prices only), Dotted Bar is Counterfactual (Loan Supply only), Horizontal Striped Bar is (Housing Prices + Loan Supply). Using IV2 estimates from Table 3. Annualized deviations from overall means depicted. The mean decline is 2.3 log points per year 51

52 Some Robustness Checks and Alternatives Allowing for asymmetric responses to changes in unemployment and growth in housing prices. Allowing for lag effects in panel regression These results use data from

53 Log Difference in Young Employment Share (1) (2) Growth rate of the employment share for young firms Change in unemployment Indicator for Change in unemployment>0 (Change in unemployment) * (Ind Chg UR >0) Change in real housing price Indicator Change in real housing price >0 (Growth in real housing price) * (Ind HP >0) Growth rate of the employment share for young firms *** ** (0.377) (0.453) *** (0.004) 0.004) *** (0.423) (0.597) ** * (0.078) (0.077) *** * (0.004) (0.003) (0.120) (0.118) State Effects Yes Yes Year Effects No Yes Adjusted R Observations Standard errors in parentheses. Clustered at State Level. * p < 0.1, ** p < 0.05, *** p <

54 Log Difference in Young Employment Share (1) (2) Growth rate of the employment share for young firms Change in unemployment Change in real housing price Change in unemployment (1-yr lag) Growth in real housing price (1-yr lag) Change in unemployment (2-yr lag) Growth in real housing price (2-yr lag) Growth rate of the employment share for young firms *** *** (0.163) (0.270) *** *** (0.026) (0.034) *** (0.131) (0.191) (0.030) (0.031) (0.168) (0.199) ** (0.040) (0.029) State Effects Yes Yes Year Effects No Yes Adjusted R Observations Standard errors in parentheses. Clustered at the State Level. * p < 0.1, ** p < 0.05, *** p < 0.01 Unlike Panel VAR we don t lose 1981 and 1982 since lagged Change in unemployment rate and Growth rate in housing prices available Pre

55 Panel VAR Y is a vector of covariates (listed below) L is a lag operator of length L (in practice two years) A(L) is a matrix of lagged coefficients Y s,t = A(L)Y st + State s + Year t + ε st State and Year represent vector of state fixed and year fixed effects. One specification includes only state effects, the second includes state and year fixed effects. ε st is the residual innovation vector of shocks to each of the covariates. Convert to orthogonalized MA representation using Cholesky decomposition with ordering Change in State-Level Unemployment Rate State-level Housing Price Growth Young Firm Outcome Indicators (e.g., First Difference in Log Young Employment Share) Focus on the responses to the first two innovations: First: State-specific cycle shock; Second: State-specific housing price shock orthogonal to first innovation (purged of contemporaneous response of housing prices to national and local cyclical shocks).

56 Housing prices fall in response to local unemployment shock, as seen in the chart below on the left. However, there remains substantial residual orthogonal variation in housing prices when we place the local unemployment rate first in the causal ordering, as seen in the chart below on the right. Solid lines depict Impulse Response Functions (shocks to one standard deviation orthogonalized innovations). Confidence interval (dashed lines) at 5 th and 95 th percentiles. 56

57 In response to adverse cyclical shock, employment at young businesses falls relative to total employment. In response to orthogonal increase in housing prices, employment at young businesses rises relative to total employment 57

58 Year-By-Year Patterns of Panel VAR results 10 Despite the rather different methodologies, these results are broadly similar to the preferred IV estimates Cumulative increase from from Housing Prices = 9 log points These panel VAR counterfactuals isolate the contribution of the orthogonalized housing price shocks Log Points Cumulative decrease from from Housing Prices = -8 log points -10 Actual Housing Prices (Controlling for State/National Cycle, State Effects) 58

59 More Background 59

60 Job Reallocation Rate, U.S. Private Nonfarm (Quarterly) Source: Business Employment Dynamics (BED) Job Reallocation (JC+JD) Job Reallocation (Trend) Hodrick-Prescott Trends The secular decline in the pace of job reallocation accelerated after 2000 and continued after the Great Recession. Job Reallocation Rate, U.S. Private Nonfarm (Annual) Source: BDS Job Reallocation Job Reallocation (Trend)

61 Quarterly Rates of Worker Reallocation, Job Reallocation & Churn, As a Percent of Employment, U.S. Nonfarm Private Sector, Worker Reallocation (H+S) "Excess" Churning (H-JC=S-JD) Job Reallocation (JC+JD) Source: BED and Job Openings and Labor Turns Survey (JOLTS) data using methods developed in Davis, Faberman and Haltiwanger (JME, 2012). Our methods adjust for undercounts of hires and separations in JOLTS data induced by weaknesses in the JOLTS sample design Worker Reallocation = Job Reallocation + Churn (Hires + Separations) (Creation + Destruction) 61

62 200,000 Code of Federal Regulations Page Count, 1949 to ,000 Truman Eisenhower Kennedy-Johnson Nixon-Ford Carter Reagan Bush 41 Clinton Bush 43 Obama 160, ,000 Page counts do not include executive memoranda, regulatory guidance, and other regulatory dark matter. 120, ,000 80, ,000 pages = 133 King James Bibles! 60,000 40,000 20,000 0 Updated from Davis (2015), who draws on Dawson and Seater (2013) and Crews (2016) In recent years, the CFR contains more than one million instances of must, shall, may not, required and prohibited.

63 200,000 Code of Federal Regulations Page Count, 1949 to ,000 Truman Eisenhower Kennedy-Johnson Nixon-Ford Carter Reagan Bush 41 Clinton Bush 43 Obama 160, , , ,000 80, ,000 pages = 133 King James Bibles! 60,000 40,000 20,000 0 Simply describing the US Tax Code takes Another 70,000 pages = 52 more bibles! Updated from Davis (2015), who draws on Dawson and Seater (2013) and Crews (2016)

64 Are Reduced Business Dynamism And Labor Market Fluidity Cause for Concern? 1. Beneficial and benign aspects of reduced fluidity: A. Less job reallocation means fewer layoffs and smaller unemployment inflows. This effect is large see Davis et al. (AEJ Macro, 2010). B. Reduced fluidity is partly a by-product of developments that raised productivity and improved welfare: The shift away from small, independent stores to big box retailers (e.g., Wal-Mart) raised productivity, lowered prices, and increased product selection, while bringing lower reallocation. 2. Reasons for concern: A. Reallocation plays a key role in prominent theories of innovation and growth. B. Factor reallocation flows are an important source of medium-term productivity growth according to many empirical studies. C. Fluidity facilitates job mobility, wage growth and career advancement. D. Fluidity promotes high employment, especially among young and less educated see Davis and Haltiwanger (2014) E. Increasing prevalence of policies that directly curtail dynamism and fluidity: Erosion of employmentat-will, occupational licensing, laws that create protected worker classes F. General increase in regulatory complexity raises compliance costs. Sunk and fixed components of compliance costs à regulatory burdens fall more heavily on younger and smaller businesses. 64