Frictional Wage Dispersion in Search Models: A Quantitative Assessment

Transcription

1 Frictional Wage Dispersion in Search Models: A Quantitative Assessment Andreas Hornstein Federal Reserve Bank of Richmond Per Krusell Princeton University, IIES-Stockholm and CEPR Gianluca Violante New York University and CEPR University of Maryland, May 2nd 2007 Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 1/4

2 Residual wage distribution for janitors in Philadelphia Density Mean / first percentile = Wage residual Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 2/4

3 Why are similar workers paid differently? Unobserved heterogeneity Human capital (e.g., innate ability) Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 3/4

4 Why are similar workers paid differently? Unobserved heterogeneity Human capital (e.g., innate ability) Compensating differentials Job characteristics (e.g., non-pecuniary amenities) Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 3/4

5 Why are similar workers paid differently? Unobserved heterogeneity Human capital (e.g., innate ability) Compensating differentials Job characteristics (e.g., non-pecuniary amenities) Labor market luck Search/matching frictions frictional wage dispersion Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 3/4

6 Outline of the talk 1. New tool to study frictional wage dispersion in search models Mean-min ratio Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 4/4

7 Outline of the talk 1. New tool to study frictional wage dispersion in search models Mean-min ratio 2. Calibrate the textbook model frictional dispersion is very small 3. Data detour: residual wage dispersion is larger by a factor of 20! 4. Extensions of the textbook model: risk-aversion, volatile wages, OJS, OJS with endogenous search effort 5. Relation to empirical search literature and to Shimer-Hall puzzle Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 4/4

8 McCall-Mortensen search model (continuous time) Homogeneous workers, infinitely lived, risk-neutral Discount rate r, flow value of unemployment b Wage remains constant on the job Exogenous job separation (into unemployment) at rate σ Search only during unemployment At rate λ u, unemployed workers encounter wage offers drawn from the exogenous distribution F(w) Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 5/4

9 Solution of the model Flow values of employment and unemployment rw (w) = w σ [W (w) U] w max ru = b + λ u w [W (w) U]dF (w) Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 6/4

10 Solution of the model Flow values of employment and unemployment rw (w) = w σ [W (w) U] w max ru = b + λ u w [W (w) U]dF (w) The reservation wage equation is: w = ru = b + λ u r + σ w max w [w w ] df (w) Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 6/4

11 Solution of the model WLOG, let b ρ w, with w = E[w w w ], then: w = ρ w + λ u[1 F(w )] r + σ w max w [w w ] df (w) 1 F(w ) = ρ w + λ u r + σ [ w w ] where λ u λ u [1 F(w )] Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 7/4

12 Solution of the model WLOG, let b ρ w, with w = E[w w w ], then: w = ρ w + λ u[1 F(w )] r + σ w max w [w w ] df (w) 1 F(w ) = ρ w + λ u r + σ [ w w ] where λ u λ u [1 F(w )] Last step eliminates F(w) which is unobservable through the job finding rate λ u Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 7/4

13 Solution of the model WLOG, let b ρ w, with w = E[w w w ], then: w = ρ w + λ u[1 F(w )] r + σ w max w [w w ] df (w) 1 F(w ) = ρ w + λ u r + σ [ w w ] where λ u λ u [1 F(w )] Last step eliminates F(w) which is unobservable through the job finding rate λ u Equation relates average wage w and lowest wage paid w only through observables (r, σ, λ u, ρ) Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 7/4

14 Mean-min ratio (Mm) Mm w λu w = r+σ + 1 λ u r+σ + ρ Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 8/4

15 Mean-min ratio (Mm) Mm w λu w = r+σ + 1 λ u r+σ + ρ New measure of wage dispersion in search models Distribution-free measure [i.e., does not depend on F(w)] Derived naturally from reservation wage equation Increasing in (σ, r), decreasing in (ρ, λ u) Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 8/4

16 Equilibrium search models Lucas-Prescott (1974) island-model Random search across islands Competitive labor market on each island Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 9/4

17 Equilibrium search models Lucas-Prescott (1974) island-model Random search across islands Competitive labor market on each island Pissarides (1990) matching model Aggregate matching function Free entry of firms Nash bargaining Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 9/4

18 Equilibrium search models Lucas-Prescott (1974) island-model Random search across islands Competitive labor market on each island Pissarides (1990) matching model Aggregate matching function Free entry of firms Nash bargaining Both models imply that same expression for Mm Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 9/4

19 Quantitative implications for M m Mm = λ u r+σ + 1 λ u r+σ + ρ Calibration of model to U.S. economy (monthly frequency) Interest rate: r = (Cooley, 1995) Separation rate: σ = (Shimer, 2005a) Job finding rate: λ u = 0.39 (Shimer, 2005a) Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 10/4

20 Quantitative implications for M m Mm = λ u r+σ + 1 λ u r+σ + ρ Calibration of model to U.S. economy (monthly frequency) Interest rate: r = (Cooley, 1995) Separation rate: σ = (Shimer, 2005a) Job finding rate: λ u = 0.39 (Shimer, 2005a) Flow value of unemployment: ρ = 0.40 (Shimer, 2005b) Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 10/4

21 Quantitative implications for M m Mm = λ u r+σ + 1 λ u r+σ + ρ Calibration of model to U.S. economy (monthly frequency) Interest rate: r = (Cooley, 1995) Separation rate: σ = (Shimer, 2005a) Job finding rate: λ u = 0.39 (Shimer, 2005a) Flow value of unemployment: ρ = 0.40 (Shimer, 2005b) Mm = Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 10/4

22 Some unpleasant search arithmetic Mm = = λ u r+σ + 1 λ u r+σ + ρ ρ = ρ = if we are willing to accept that ρ 0 Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 11/4

23 Interpretation: good things come to those who wait Unemployed workers search longer (and turn down jobs) if there is a large option value of waiting for better offers Option value is determined by wage dispersion Short unemployment duration, as in U.S. data, reveals that dispersion is small!! Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 12/4

24 Interpretation: good things come to those who wait Unemployed workers search longer (and turn down jobs) if there is a large option value of waiting for better offers Option value is determined by wage dispersion Short unemployment duration, as in U.S. data, reveals that dispersion is small!! How large is frictional wage dispersion in the data? Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 12/4

25 Measurement of residual wage dispersion Empirical counterpart of theory: wage observations for ex-ante similar workers searching in the same labor market Three key data issues: 1. Definition of labor market: occupation/geographical area Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 13/4

26 Measurement of residual wage dispersion Empirical counterpart of theory: wage observations for ex-ante similar workers searching in the same labor market Three key data issues: 1. Definition of labor market: occupation/geographical area 2. Control for ex-ante individual heterogeneity observable: e.g., education, experience, gender, race unobservable: e.g., innate workers characteristics Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 13/4

27 Measurement of residual wage dispersion Empirical counterpart of theory: wage observations for ex-ante similar workers searching in the same labor market Three key data issues: 1. Definition of labor market: occupation/geographical area 2. Control for ex-ante individual heterogeneity observable: e.g., education, experience, gender, race unobservable: e.g., innate workers characteristics 3. Measurement error in hourly wages Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 13/4

28 Three sources of micro data 1. 5% sample of 1990 US Census ( 4,600,000 obs.) 487 occupations, 799 geographical areas controls: gender, race, edu, exp, marital status Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 14/4

29 Three sources of micro data 1. 5% sample of 1990 US Census ( 4,600,000 obs.) 487 occupations, 799 geographical areas controls: gender, race, edu, exp, marital status 2. Occupational Employment Statistics ( 1,200,000 establishments) 637 occupations, 337 metropolitan areas no controls available, but no reporting errors Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 14/4

30 Three sources of micro data 1. 5% sample of 1990 US Census ( 4,600,000 obs.) 487 occupations, 799 geographical areas controls: gender, race, edu, exp, marital status 2. Occupational Employment Statistics ( 1,200,000 establishments) 637 occupations, 337 metropolitan areas no controls available, but no reporting errors 3. PSID ( 80,000 obs.) controls: gender, race, edu, exp, marital status, region, occ, occ region panel data control for fixed individual effects Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 14/4

31 Empirical Findings Table 1: Dispersion measures for hourly wage from various data sources Min. obs. Number of Ratio of mean wage to CV per cell labor mkts min. 1st pct. 5th pct. 10th pct. Panel Study of Income Dynamics nd stage residuals Occupational Employment Statistics 2000 Occ./Geog. Area 106, % sample of1990 US Census Occ./Geog. Area (N 50) 13, Fulltime/Fullyear (N 50) 7, Unskilled Occ. (N 50) 1, Wetarget =1 70and =0 30 Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 15/4

32 Distribution of Mp5 across US labor markets (Census) Density ,246 cells Median Mm= Mean / 5th percentile Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 16/4

33 Matched employer-employee data (Denmark) Private Firm Hourly Wage Dispersion Measures Danish IDA Data, Occupation # of firms 1pct 5pct 10pct Full Sample 113, Management 49, Salaried 57, Skilled 44, Unskilled 70, Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 17/4

34 How far is the model from the data? 1 Net value of leisure as a fraction of w (ρ) Reasonable pairs Pairs of (ρ,r) consistent with Mm= Monthly interest rate (r) Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 18/4

35 Why is ρ = 3 implausible? In order to avoid a week of unemployment, a worker would be willing to: work for free for a week pay $1,500 (3 times the average weekly salary) and, at the end of the week, draw a wage offer from the same distribution he would face if unemployed Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 19/4

36 Resolving the data-model discrepancy 1. Nothing wrong with the model frictional wage dispersion is negligible compared to unobserved workers heterogeneity Bad news for search theory? Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 20/4

37 Resolving the data-model discrepancy 1. Nothing wrong with the model frictional wage dispersion is negligible compared to unobserved workers heterogeneity Bad news for search theory? 2. Nothing wrong with the model + workers truly hate unemployment frictional wage dispersion is large Bad news for the rest of macro-labor? Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 20/4

38 Resolving the data-model discrepancy 1. Nothing wrong with the model frictional wage dispersion is negligible compared to unobserved workers heterogeneity Bad news for search theory? 2. Nothing wrong with the model + workers truly hate unemployment frictional wage dispersion is large Bad news for the rest of macro-labor? 3. Model fails quantitatively it needs to be augmented We take this route here, but first... Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 20/4

39 In defense on the textbook model 1. "Model designed to study unemployment, not wage dispersion" Given M m = 1.7, model predicts expected unemployment duration of 91 months!! Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 21/4

40 In defense on the textbook model 1. "Model designed to study unemployment, not wage dispersion" Given M m = 1.7, model predicts expected unemployment duration of 91 months!! 2. "The model will perform better on European data" In steady-state: λ uu = σ(1 u) Mm = λ u r+σ + 1 S.S. λ u r+σ + ρ = σ 1 u r+σ σ 1 u r+σ u u + 1 r small + ρ 1 u u + 1 ρ>0 1 u u + ρ < 1 1 u Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 21/4

41 In defense on the textbook model 1. "Model designed to study unemployment, not wage dispersion" Given M m = 1.7, model predicts expected unemployment duration of 91 months!! 2. "The model will perform better on European data" In steady-state: λ uu = σ(1 u) Mm = λ u r+σ + 1 S.S. λ u r+σ + ρ = σ 1 u r+σ σ 1 u r+σ u u + 1 r small + ρ 1 u u + 1 ρ>0 1 u u + ρ < 1 1 u 3. "Mm ratio is an unconventional measure of dispersion" Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 21/4

42 Using the Mm tool in richer search models 1. Risk-aversion 2. Volatile wages during employment 3. On-the-job search The critical rule of the game is not adding unobserved heterogeneity Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 22/4

43 Risk Aversion Let preferences be u(c) = c1 θ 1 θ, and let c = w Upper bound for effects of risk-aversion: no self-insurance Second-order Taylor expansion of reservation wage equation: Mm ( θ (θ 1) CV (w)2) λ u r+σ + ρ1 θ λ u r+σ θ 1 Calibration: only new number we need is CV (w) = 0.30 Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 23/4

44 Numerical results for the model with risk-aversion 0.7 Net value of leisure as a fraction of w (ρ) Pairs (ρ,θ) consistent with Mm=1.70 θ= Relative risk aversion (θ) Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 24/4

45 Wage shocks during employment (M-P, 1994) Wages fluctuate randomly along the employment spell At rate δ, employed workers draw new wage from F(w) Endogenous separations at rate σ = δf(w ) Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 25/4

46 Wage shocks during employment (M-P, 1994) Wages fluctuate randomly along the employment spell At rate δ, employed workers draw new wage from F(w) Endogenous separations at rate σ = δf(w ) Solve the model and obtain... Mm = λ u δ+σ r+δ + 1 λ u δ+σ r+δ + ρ As δ λ u, Mm 1/ρ 1 δ: autocorrelation coefficient of the wage process Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 25/4

47 Numerical results for the model with stochastic wages 0 Net value of leisure as a fraction of w (ρ) Pairs of (1 δ,ρ) consistent with Mm=1.70 Reasonable pairs Annual autocorrelation coefficient of wage process (1 δ) Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 26/4

48 On-the-job search (Burdett, 1978) Workers draw wage offers from F(w) at rate λ u if unemployed, and at rate λ w if employed F(w) can be any wage offer distribution When employed, accept offer w if w > w: F(w) G(w) Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 27/4

49 On-the-job search (Burdett, 1978) Workers draw wage offers from F(w) at rate λ u if unemployed, and at rate λ w if employed F(w) can be any wage offer distribution When employed, accept offer w if w > w: F(w) G(w) Solve the model for the realized steady-state wage distribution G(w) and, for r small, obtain... Mm λ u λ w r+σ+λ w + 1 λ u λ w r+σ+λ w + ρ As λ w λ u, Mm 1/ρ Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 27/4

50 Restricting the value of λ w Average separation rate χ in the model is: χ = σ + λ w [1 F (w)]dg (w) = σ (λ w + σ) log w λ w ( σ+λw σ ) BLS (JOLTS): monthly separation rate χ = 0.04 Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 28/4

51 Restricting the value of λ w Average separation rate χ in the model is: χ = σ + λ w [1 F (w)]dg (w) = σ (λ w + σ) log w λ w ( σ+λw σ ) BLS (JOLTS): monthly separation rate χ = 0.04 Set λ w to be consistent with labor mobility facts and explore implications for Mm ratio Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 28/4

52 Numerical results for the model with on the job search 1 Net value of leisure as a fraction of w (ρ) Reasonable pairs Pairs of (ρ,λ w ) consistent with Mm= Monthly offer arrival rate on the job ( λ ) w Monthly separation rate JOLTS estimate of average separation rate Monthly offer arrival rate on the job ( λ ) w Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 29/4

53 Numerical results for the model with on the job search 1 0 Net value of leisure as a fraction of w (ρ) Iso Mm curves Mm=1.7 Mm=1.2 Mm= Monthly offer arrival rate on the job ( λ w ) Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 30/4

54 Endogenous search effort (CLMNW, 2005) Unemployed and employed workers choose search effort, i.e., contact rate λ Every worker faces the same search effort cost function c(λ), with c > 0, c > 0 Reservation wage: w = b Mm = 1 ρ = 2.5 Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 31/4

55 Endogenous search effort (CLMNW, 2005) Unemployed and employed workers choose search effort, i.e., contact rate λ Every worker faces the same search effort cost function c(λ), with c > 0, c > 0 Reservation wage: w = b Mm = 1 ρ = 2.5 Disutility of unemployment is now: b c(λ ) Assume c(λ) = λ γ, with γ = 2. Then, for the unemployed worker: ρ = b c(λ ) w ( 0.5, 0.3)...still negative! Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 31/4

56 A journey through the empirical search literature Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 32/4

57 A journey through the empirical search literature Either accept implausible parameter estimates... Postel-Vinay and Robin (ECA, 2002): for unskilled occupations, r = 57% per year Flinn (ECA, 2006): for realistic values of r, ρ = 4 Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 32/4

58 A journey through the empirical search literature Either accept implausible parameter estimates... Postel-Vinay and Robin (ECA, 2002): for unskilled occupations, r = 57% per year Flinn (ECA, 2006): for realistic values of r, ρ = 4...or need huge measurement error... Eckstein and Wolpin (JPE, 1990) Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 32/4

59 A journey through the empirical search literature Either accept implausible parameter estimates... Postel-Vinay and Robin (ECA, 2002): for unskilled occupations, r = 57% per year Flinn (ECA, 2006): for realistic values of r, ρ = 4...or need huge measurement error... Eckstein and Wolpin (JPE, 1990)...or need large unobserved heterogeneity Eckstein and Wolpin (REStud, 1995), Van den Berg and Ridder (ECA, 1998) Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 32/4

60 Cross-section vs. time-series Shimer-Hall: search model does not generate enough business-cycle volatility in vacancies and unemployment Hagedorn-Manovskii: we can save the standard search model as long as ρ 1 Our paper: if ρ 1, no hope to get any frictional wage dispersion from search models Tension between time-series and cross-sectional implications of search model Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 33/4

61 Concluding remarks Mm ratio: new, powerful tool to analyze wage dispersion in a large class of search models three possible conclusions: 1. Data still contain unobserved worker heterogeneity Frictional wage dispersion is small in both data and model Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 34/4

62 Concluding remarks Mm ratio: new, powerful tool to analyze wage dispersion in a large class of search models three possible conclusions: 1. Data still contain unobserved worker heterogeneity Frictional wage dispersion is small in both data and model 2. Data OK, but workers hate unemployment ρ < 0 implausible, plus watch for business-cycle implications! Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 34/4

63 Concluding remarks Mm ratio: new, powerful tool to analyze wage dispersion in a large class of search models three possible conclusions: 1. Data still contain unobserved worker heterogeneity Frictional wage dispersion is small in both data and model 2. Data OK, but workers hate unemployment ρ < 0 implausible, plus watch for business-cycle implications! 3. Data OK, and ρ is moderately high Basic search model fails, latest OJS models more promising Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 34/4

64 How far is the model from the data? 1 Net value of leisure as a fraction of w (ρ) Iso Mm curves Mm=1.7 Mm=1.2 Mm= Monthly interest rate (r) Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 35/4

65 Distribution of individual Mm (PSID) Density ,572 observations Median Mm= Mean / min Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 36/4

66 Relation between Mm and CV Assume Gamma distribution g(w; w, β, γ) = ( w w β ) γ 1 exp( w w β ) βγ(γ) Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 37/4

67 Relation between Mm and CV Assume Gamma distribution g(w; w, β, γ) = ( w w β ) γ 1 exp( w w β ) βγ(γ) Then, we can show that: CV (w) = 1 γ Mm(w) 1 Mm(w) Data: Mm(w) = 1.70 and CV (w) = 0.30 ˆγ = 1.88 Search model: Mm = CV = Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 37/4

68 Numerical results for the model with risk-aversion 0.9 Net value of leisure as a fraction of w (ρ) Iso Mm curves Mm=1.7 Mm=1.2 Mm= Relative risk aversion (θ) Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 38/4

69 Numerical results for the model with stochastic wages 0 2 Net value of leisure as a fraction of w (ρ) Mm=1.7 Mm=1.2 Mm=2.2 Iso Mm curves Annual autocorrelation coefficient of wage process (1 δ) Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 39/4

70 Reallocation shocks (Nagypal, 2005) Baseline OJS model, plus... At rate φ, employed workers make a wage draw from F(w) which either they accept, or they separate Mm λ u λ w φ r+σ+λ w +φ + 1 λ u λ w φ r+σ+λ w +φ + ρ Restrict (λ w, φ) to match: (i) Mm = 1.70 and (ii) sep. rate of 4% Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 40/4

71 Reallocation shocks (Nagypal, 2005) Baseline OJS model, plus... At rate φ, employed workers make a wage draw from F(w) which either they accept, or they separate Mm λ u λ w φ r+σ+λ w +φ + 1 λ u λ w φ r+σ+λ w +φ + ρ Restrict (λ w, φ) to match: (i) Mm = 1.70 and (ii) sep. rate of 4% Monthly rate at which workers are subject to a wage cut is: κ = φ F (w)dg (w) = Hornstein-Krusell-Violante, Frictional Wage Dispersion p. 40/4