Really Uncertain Business Cycles

Size: px

Start display at page:

Download "Really Uncertain Business Cycles"

Ginger Tate
5 years ago
Views:

1 Really Uncertain Business Cycles Nick Bloom (Stanford & NBER) Max Floetotto (McKinsey) Nir Jaimovich (Duke & NBER) Itay Saporta-Eksten (Stanford) Stephen J. Terry (Stanford) SITE, August 31 st

2 Uncertainty as another driver of business cycles Many sources of business cycle fluctuations in the literature: Neutral technology shocks Investment-specific technology shocks Oil price shocks Monetary policy shocks Fiscal policy shocks Financial shocks News shocks All of these are first moment (levels) shocks But do second moment shocks matter? 2

3 Summary of what this paper does A. Provides empirics suggesting uncertainty is: 1. Counter-cyclical 2. Not driven by first moment (demand) shocks B. Builds a DSGE model generalized with time-varying uncertainty, heterogeneous firms and non-convex adjustment costs, finding: 1. Uncertainty shocks generate a moderate drop (about -2% GDP) & rebound in labor, capital, TFP & output; 2. Uncertainty shocks substantially reduce the impact of policies on the economy 3

4 Uncertainty is counter-cyclical Uncertainty is not driven by demand shocks Model Simulation of an uncertainty shock Policy experiment 4

5 Focus on census data to measure uncertainty Uncertainty is hard to measure We use Census based measures because allows us to have huge samples across many years But show Census based uncertainty measures very correlated with other popular uncertainty measures 5

6 The US Census data set Census data sets ASM matched to the CM ( ) LBD ( ) Data on output, inputs, capital stocks etc. Sample Primary manufacturing plants with 25+ years (to keep sample selection fixed) But results robust to using all manufacturing, or even all manufacturing, retail, wholesale and mining 6

7 TFP Shocks as a measure of uncertainty log(tfp) Plant fixed effect Year fixed effects Lagged log(tfp) TFP shock Use Census manufacturing establishment data to define log Total Factor Productivity (TFP) as real output less industry factor share weighted inputs. Note: because we only have 4-digit price deflators TFP will also include potential plant demand shocks (e.g. Foster, Haltiwanger & Syversson, 2008) 7

8 TFP shocks appear to measure uncertainty Correlate TFP shocks with other possible uncertainty measures Firm level Volatility of CRSP monthly and daily stock returns Volatility of Compustat quarterly sales Industry level Volatility of industrial production growth The regression equation: Averaged at the firm/industry level Volatility measures 8

9 Example: TFP Shocks correlate with firm stock vol Dependent Variable: mean of establishment absolute(tfp shocks) within firm year: S.D. of parent monthly firm stock returns within firm year S.D. of parent daily firm stock returns within firm year S.D of monthly growth of industrial production within industry year (1) (2) (5) 0.312*** (0.09) 0.326*** (0.099) 0.344*** (0.068) Establishments 9,823 9,823 14,385 Firms 1,761 1,761 10,059 Industries Micro observations 156, , ,839 Observations 23,321 23,321 15,443 9

We find TFP shocks have a higher cross-sectional standard-deviation in recessions Annual Standard deviation of plant TFP shocks Average Quarterly GDP Growth Rates 10 Notes: Constructed from the

10 We find TFP shocks have a higher cross-sectional standard-deviation in recessions Annual Standard deviation of plant TFP shocks Average Quarterly GDP Growth Rates 10 Notes: Constructed from the Census of Manufacturers and the Annual Survey of Manufacturing establishments using all establishments with 25+ years to address sample selection. Grey shaded columns are share of quarters in recession within a year.

11 Obvious question is what drives what? Do recessions drive uncertainty or uncertainty drive recessions? Using macro data very hard to distinguish these because everything moves together Industry level has a big advantage of providing more data to identify causality We look at relationship between recessions and uncertainty in industry data 11

12 Quick primer on Census industry definitions. 23 APPAREL AND OTHER FINISHED PRODUCTS MADE FROM FABRICS 232 MEN'S AND BOYS' FURNISHINGS, WORK CLOTHING, AND ALLIED GARMENTS 2321 MEN'S AND BOYS' SHIRTS, EXCEPT WORK SHIRTS 2322 MEN'S AND BOYS' UNDERWEAR AND NIGHTWEAR 2323 MEN'S AND BOYS' NECKWEAR 2325 MEN'S AND BOYS' SEPARATE TROUSERS AND SLACKS 2326 MEN'S AND BOYS' WORK CLOTHING 2329 MEN'S AND BOYS' CLOTHING, NOT ELSEWHERE CLASSIFIED 12

13 Uncertainty is also higher in industry recessions Measure uncertainty in an industry as the spread of TFP shocks within industry-year Regress industry uncertainty on industry growth Include full set of industry and year dummies, so remove all business cycle effects 13

14 Uncertainty is also higher in industry recessions Dependent Variable: iqr(tfp shocks) within industry year Median real output growth rates Mean real output growth rates (1) (2) (3) (4) *** *** (0.021) (0.022) *** *** (0.018) (0.017) Industry trends included N Y N Y Observations 15,497 15,497 15,497 15,497 Mean obs per industry year Median obs per industry year Underlying sample size 403, , , ,839 14

15 Uncertainty is counter-cyclical Uncertainty is not driven by demand shocks Model Simulation of an uncertainty shock Policy experiment 15

16 What drives what: does demand drives uncertainty?? Challenge is to find an instrument that predicts first moment shocks but does not drive the second moment (uncertainty) Need a first moment shock that is: (i) exogenous, & (ii) predicted Turns out there is one such shock we can use as an IV 16

17 China joining WTO is an almost ideal instrument The Multi Fiber Agreement (1974) restricted apparel and textile exports from developing countries The MFA was negotiated into GATT (WTO) as part of the Uruguay Round in 1994, with a 4 phase abolition When China entered the WTO in Dec 2001 it gained access to this phased abolition, occurring mainly in 2005 When Chinese products came off quota in 2005 there was huge surge of imports into the US (+280% on average!) Large, anticipated demand shock, almost random by industry 17

Quotas operated at HS 6-digit level HS6 codes we match against SIC2321 610510 Men's or Boys' Shirts of Cotton, Knitted or Crocheted 610520 Men's or Boys' Shirts of Man-made Fibers, Knitted or

18 Quotas operated at HS 6-digit level HS6 codes we match against SIC Men's or Boys' Shirts of Cotton, Knitted or Crocheted Men's or Boys' Shirts of Man-made Fibers, Knitted or Crocheted Men's or Boys' Shirts of Other Textile Materials, Knitted or Crocheted Men's or Boys' Shirts of Wool or Fine Animal Hair Men's or Boys' Shirts of Cotton Men's or Boys' Shirts of Man-made Fibers Men's or Boys' Shirts of Other Textile Materials 18

19 IV at SIC4 level: share of imports previous under quota note seemingly random 19

20 Uncertainty is not driven by Demand Shocks Dependent Variable: iqr(tfp shocks) within industry year (1) (2) (3) (4) Sample Baseline Textile Textile Baseline Estimation OLS OLS IV IV Median real output growth *** * (0.021) (0.219) (0.484) (0.275) 2005 Quotas effect First Stage F-test Years Observations 15, ,056 Underlying sample 403,839 8,077 8, ,090 20

21 Uncertainty is Counter-Cyclical Uncertainty is Not Driven by Demand Shocks Model Simulation of an uncertainty shock Policy experiment 21

22 Driving Processes 22

23 Firms 23

24 Households 24

25 General Equilibrium Solution Overview We have a recursive competitive equilibrium Solve numerically as no analytic solution Numerical solution approximates μ (the firm-level distribution over z, k and n) with moments, building particularly on Krusell and Smith (1998) and Khan and Thomas (2008) 25

26 Real Options Effect Low Uncertainty 20% 25% 16% 20% Distribution of establishments (after the shock occurs) 12% 8% 4% 0% Hiring 15% 10% 5% 0% Hiring policy (low uncertainty) -4% Firing -5% -8% -10% Idiosyncratic productivity Note: For a given value of A, k and n 26

27 Real Options Effect High Uncertainty 20% 16% 12% 8% 4% 0% Distribution of establishments (after the shock occurs) Hiring policy, (high uncertainty) 25% 20% 15% 10% 5% 0% -5% -4% Hiring Firing -10% -8% Idiosyncratic productivity Note: For a given value of A, k and n 27

28 Uncertainty is Counter-Cyclical Uncertainty is Not Driven by Demand Shocks Model Simulation of an Uncertainty Shock Policy Experiment 28

29 Uncertainty's Effect on Output 29

30 Uncertainty's Effect on Other Aggregates Hours Investment % deviations from steady state % deviations from steady state Quarters Quarters 0.6 Labor weighted TFP Consumption Quarters Quarters 30 % deviations from steady state % deviations from steady state

31 Uncertainty is Counter-Cyclical Uncertainty is Not Driven by Demand Shocks Model Simulation of an Uncertainty Shock Policy Experiment 31

32 Policy Responsiveness 32

33 Differential policy effect 33

34 Conclusions Uncertainty is countercyclical (data) Uncertainty is of independent interest (data) Uncertainty generates a significant recession (model) Uncertainty decreases policy effectiveness (model) 34

35 Back up Slides (Empirics)

36 Table 1 Dependent Variable Sample (1) (2) (3) (4) (5) mean of establishment mean of establishment mean of establishment absolute(tfp shocks) within absolute(tfp shocks) within absolute(tfp shocks) within firm year firm year firm year mean of establishment absolute(tfp shocks) within firm year Manufacturing, 25+ year establishments, Compustat parent firm Manufacturing, 25+ year establishments, Compustat parent firm Manufacturing, 25+ year establishments, Compustat parent firm Manufacturing, 25+ year establishments, Compustat parent firm mean of establishment absolute(tfp shocks) within industry year Manufacturing, 25+ year establishments Dataset ASM, CRSP ASM, CRSP ASM, CRSP, COMPUSTAT ASM, COMPUSTAT ASM, FRB Regression panel dimension Firm by Year Firm by Year Firm by Year Firm by Year Industry by Year S.D. of parent monthly firm stock returns within firm year 0.312*** (0.09) S.D. of parent daily firm stock returns within firm year 0.326*** (0.099) S.D. of parent monthly firm stock returns within firm year, leverage adjusted 0.357*** (0.121) S.D. of parent quarterly firm sales growth rates within firm year 0.13*** (0.031) S.D of monthly growth of industrial production within industry year 0.344*** (0.068) Fixed effects firm firm firm firm industry Windsorized N Y N N N weights in regression N N N N Y Leverage adjustment N N Y N N Standard error adjustment Cluster by firm Cluster by firm Cluster by firm Cluster by firm Cluster by industry Establishments 9,823 9,823 9,823 9,823 14,385 Firms 1,761 1,761 1,761 1,761 10,059 Industries Micro observations 156, , , , ,839 Observations 23,321 23,321 23,321 23,321 15,443 All regressions include year dummies. Leverage adjustment is done using book value of equity and debt and windsorized at 10%. To match the timing in the tfp shock and real sales growth equations, S,D of monthly returns is the average of the S.D at t and the S.D. at t+1. Firms with less then 6 months of returns reported per year are dropped. The means at firm and industry level are weighted using plant's total value of shipment. Daily returns are normalized to monthly rate (by multiplying daily S.D by sqrt(21)) 36

37 Table 2 Dependent Variable (1) (2) (3) (4) (5) S.D of shock to IQR of shock to IQR of sales log(tfp) log(tfp) growth rates S.D of shock to log(tfp) IQR of employment growth rates Sample Manufacturing, 25+ years establishments Manufacturing, 25+ years establishments Manufacturing, 25+ years establishments Manufacturing, 25+ years establishments Manufacturing, 25+ years establishments Dataset ASM ASM ASM ASM ASM Underlying observed entity Establishment Establishment Establishment Establishment Establishment Correlation with GDP Growth ** ** ** -0.36** Correlation with Industrial Prod ** ** * *** *** Mean of Dependant Variable Share of Quarters in Recession 0.05** 0.063*** 0.041** 0.051*** 0.037*** (0.023) (0.01) (0.017) (0.014) (0.009) Percentage Change in Recession 10.1% 12.8% 10.6% 26.6% 30.3% Time trend and census dummies N Y Y Y Y Industry FE in AR regressions Y Y Y Y Y Year FE in AR regressions Y Y Y Y Y De-meaned by Industry-year N N N N N Frequency Annual Annual Annual Annual Annual Years Observations Underlying sample size 403, , , , ,839 37

38 Table 2 cont. Dependent Variable (6) (7) (8) (9) (10) IQR of shock to IQR of monthly IQR of sales log(tfp) stock returns growth IQR of employment growth rates IQR of industrial production growth Sample Manufacturing, Mining, Wholesale and Retail, 25+ years establishments Manufacturing, 25+ years establishments Public firms, 25+ years Public firms, 25+ years Manufacturing Dataset LBD ASM CRSP Compustat FRB Underlying observed entity Establishment Establishment Firm Firm Industry Correlation with GDP Growth *** *** *** Correlation with Industrial Prod * *** *** *** Mean of Dependant Variable Share of Quarters in Recession 0.022*** 0.041** 0.025*** 0.033*** 0.044*** (0.004) (0.016) (0.004) (0.009) (0.006) Percentage Change in Recession 10.0% 11.1% 24.0% 17.5% 43.3% Time trend and census dummies Y Y Y Y Y Industry FE in AR regressions Y Y Y Y Y Year FE in AR regressions Y Y Y Y Y De-meaned by Industry-year N Y N N N Frequency Annual Annual Monthly Quarterly Monthly Years :1-2010:9 1962:1-1963:3 1972:1-2010:11 Observations Underlying sample size 11,951, , , ,414 70,487 38

39 Table 3 No Interactions (1) (2) (3) (4) (5) (6) (7) (8) Dependent Variable iqr(tfp shocks) iqr(tfp shocks) iqr(tfp shocks) iqr(tfp shocks) iqr(tfp shocks) iqr(tfp shocks) iqr(tfp shocks) iqr(tfp shocks) Sample Manufacturing, Manufacturing, Manufacturing, Manufacturing, Manufacturing, Manufacturing, Manufacturing, Manufacturing, 25+ year establishments 25+ year establishments 25+ year establishments 25+ year establishments 25+ year establishments 25+ year establishments 25+ year establishments 25+ year establishments Dataset ASM ASM ASM ASM ASM ASM ASM ASM Regression Panel Dimension Industry-year Industry-year Industry-year Industry-year Industry-year Industry-year Industry-year Industry-year Interaction variable Median real output growth rates *** *** (0.021) (0.022) Mean real output growth rates *** *** (0.018) (0.017) Median TFP shock *** *** (0.016) (0.018) Mean TFP shock *** *** (0.016) (0.017) Interaction of median real output growth rates with Weighting Y Y Y Y Y Y Y Y Industry trends included N Y N Y N Y N Y Years Observations 15,497 15,497 15,497 15,497 15,497 15,497 15,497 15,497 Mean obs per industry year Median obs per industry year Underlying sample size 403, , , , , , , ,839 See notes next slide 39

40 Table 3 Interactions (9) (10) (11) (12) (13) (14) (15) Dependent Variable iqr(tfp shocks) iqr(tfp shocks) iqr(tfp shocks) iqr(tfp shocks) iqr(tfp shocks) iqr(tfp shocks) iqr(tfp shocks) Sample Manufacturing, 25+ Manufacturing, 25+ Manufacturing, 25+ Manufacturing, 25+ Manufacturing, 25+ Manufacturing, 25+ Manufacturing, 25+ year establishments year establishments year establishments year establishments year establishments year establishments year establishments Dataset ASM ASM ASM ASM ASM ASM ASM Regression Panel Dimension Industry-year Industry-year Industry-year Industry-year Industry-year Industry-year Industry-year Interaction variable iqr(mean(tfp)) iqr(mean(te)) iqr(mean(m/y)) iqr(mean(k/n)) iqr(mean(sales growth)) iqr(mean(i/y)) Geographic Dispersion Median real output growth rates *** *** *** *** *** *** *** (0.037) (0.023) (0.029) (0.029) (0.051) (0.023) (0.029) Interaction of median real output growth rates with (0.07) (0.021) (0.147) (0.328) (0.95) (0.274) (0.132) Weighting Y Y Y Y Y Y Y Industry trends included N N N N N N N Years Observations 15,497 15,497 15,497 15,497 15,497 15,497 15,497 Mean obs per industry year Median obs per industry year Underlying sample size 403, , , , , , ,839 Notes: Constructed from the Census of Manufacturers and the Annual Survey of Manufacturing establishments matched to Compustat and CRSP using SSEL-Compustat bridge. Sample includes all establishments with 25+ years. All regressions include year dummies and FE at the industry level. An observation is industry by year for all regressions. Industries are weighted by the number of establishments. Standard errors are clustered by industry. Total employment (in column 10) and k/n (column 12) are divided by a 1,000 40

41 Table 4 (1) (2) (3) (4) (5) (6) Dependent Variable iqr(tfp shocks) iqr(tfp shocks) iqr(tfp shocks) iqr(tfp shocks) iqr(tfp shocks) iqr(tfp shocks) Sample Manufacturing, 25+ year establishments textile plus (5 years window) textile plus (5 years window) textile plus (5 years window) Manufacturing, 25+ year establishments. Manufacturing, 25+ year establishments Dataset ASM ASM ASM ASM ASM ASM Estimation OLS OLS IV Reduced form IV Reduced form Median real output growth rates *** * (0.021) (0.219) (0.484) (0.275) Industry ex. rate Industry ex. rate at t Quotas effect (0.067) (0.038) (0.039) First Stage 2005 Quotas effect *** (0.027) Industry ex. rate *** (0.024) Industry ex. rate at t *** (0.021) F-test Weighting Y Y Y Y Y Y Years Observations 15, ,056 15,056 Mean obs per industry year Median obs per industry year Underlying sample size 403,839 8,077 8,077 8, , ,090 41

42 Summary of what this paper (currently) does not do Does not endogenize uncertainty Modeled as exogenous, like first moment shocks Endogenous uncertainty could be an amplification mechanism Does not analyze other potentially important uncertainty channels: Consumer durables Credit Risk 42

43 Back up Slides (Model)

44 Equilibrium 44

45 Numerical Method Kahn and Thomas (2008) 45

46 Micro versus Macro Uncertainty 46

47 Robustness to Different Depreciation Rates 47

48 First and Second Moment Shocks 2% Output % Deviations 1% 0% -1% -2% -3% -4% -5% Quarter Baseline 1st and 2nd Moment 48

49 First and Second Moment Shocks % Deviations Consumption Baseline 1st and 2nd Moment 8% 6% 4% 2% 0% -2% -4% -6% -8% -10% Quarter Hours 2% 1% % Deviations 0% -1% -2% -3% -4% -5% Baseline 1st and 2nd Moment Quarter 49

50 Calibration 50

51 Calibration Moments 51

52 Business Cycle Statistics 52

53 Plant Level Investment Rates 53

Skewed Business Cycles

Skewed Business Cycles Sergio Salgado Fatih Guvenen Nicholas Bloom University of Minnesota University of Minnesota, FRB Mpls, NBER Stanford University and NBER SED, 2016 Salgado Guvenen Bloom Skewed Business