Measuring Mismatch in the U.S. Labor Market

Transcription

1 Measuring Mismatch in the U.S. Labor Market Ayşegül Şahin Federal Reserve Bank of New York Joseph Song Federal Reserve Bank of New York Giorgio Topa Federal Reserve Bank of New York, and IZA Gianluca Violante New York University, CEPR, and NBER Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 1 /44

2 Motivation Recent surge in US unemployment sharp and persistent Percent 12 Unemployment Rate Percent Source: Bureau of Labor Statistics 0 Note: Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 2 /44

3 Motivation High unemployment puzzling in light of recent rise in vacancies Vacancy Rate (JOLTS) 4.0 Beveridge Curve Vacancy Rate (JOLTS) Dec Nov Dec Feb Source: Bureau of Labor Statistics Unemployment Rate Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 3 /44

4 Potential explanations 1. Lower workers search effort (e.g., extension of UI benefits) 2. Lower employers recruiting effort (e.g., high uncertainty) 3. Higher sectoral mismatch skills/occupations/industries/locations of idle labor are poorly matched with those of job openings Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 4 /44

5 Potential explanations 1. Lower workers search effort (e.g., extension of UI benefits) 2. Lower employers recruiting effort (e.g., high uncertainty) 3. Higher sectoral mismatch skills/occupations/industries/locations of idle labor are poorly matched with those of job openings We develop a framework to measure: 1. how much of (the rise in) unempl. is due to (the rise in) mismatch 2. which dimensions of mismatch are the most important Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 4 /44

6 Methodology Economy with I distinct frictional labor markets {u i }: observed allocation {u i }: allocation selected by a planner who can freely move unemployed across markets (constrained first-best) Difference between {u i } and {u i } lower job finding rate additional (mismatch) unemployment Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 5 /44

7 Methodology Economy with I distinct frictional labor markets {u i }: observed allocation {u i }: allocation selected by a planner who can freely move unemployed across markets (constrained first-best) Difference between {u i } and {u i } lower job finding rate additional (mismatch) unemployment Same insight as misallocation literature: distance from first-best Specifically, we build on Jackman-Roper (OBES, 1987) Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 5 /44

8 What we don t do 1. We have little to say about the deep causes of mismatch: moving/retraining costs borrowing constraints information imperfections wage rigidity government policies Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 6 /44

9 What we don t do 1. We have little to say about the deep causes of mismatch: moving/retraining costs borrowing constraints information imperfections wage rigidity government policies 2. We can t tell whether mismatch is constrained efficient need a model where mismatch is an equilibrium outcome Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 6 /44

10 What we don t do 1. We have little to say about the deep causes of mismatch: moving/retraining costs borrowing constraints information imperfections wage rigidity government policies 2. We can t tell whether mismatch is constrained efficient need a model where mismatch is an equilibrium outcome 3. We abstract from the effect of mismatch on vacancy creation Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 6 /44

11 From mismatch to unemployment: two channels u = s s+f 1. More mismatch lower job finding rate f higher u Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 7 /44

12 From mismatch to unemployment: two channels u = s s+f 1. More mismatch lower job finding rate f higher u 2. Effect of higher sep. rate on u increasing in mismatch through f du ds d 2 u dsdf = = f (s+f) 2 > 0 s f (s+f) 3 < 0 since f s Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 7 /44

13 Outline of the rest of the talk 1. Environment and solution to planner s problem 2. Derivation of mismatch indexes and their interpretation 3. Explanation of counterfactuals 4. Results based on JOLTS vacancies 5. Results based on HWOL job advertisements Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 8 /44

14 1. ECONOMIC ENVIRONMENT AND PLANNER S PROBLEM Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 9 /44

15 Demographics, preferences and geography Measure one of ex-ante equal agents Individuals can be employed, unemployed, or OLF Linear utility over consumption, disutility of search effort ξ I distinct frictional labor markets (sectors) Free mobility of labor across sectors Aggregate labor force: l = I i=1 (e i +u i ) 1 Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 10 /44

16 Frictions, heterogeneity and uncertainty New production opportunities (vacancies) v i arise exogenously in each market i Labor markets are frictional: h i = Φφ i m(u i,v i ) Existing matches in sector i produce Zz i units of output Matches destroyed exogenously at common rate δ Employed workers can quit into unemployment/olf Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 11 /44

17 Timing of events 1. Exogenous states S = (Z,δ,Φ), and s = (v,φ,z) are observed. Endogenous states e = {e 1,...e I } and u also given. 2. Unemployed direct their job search towards sector i {u i } 3. Matching process h i = Φφ i m(u i,v i ) new hires 4. Production takes place in the e i +h i matches 5. Fraction δ of matches destroyed and σ i workers quit e 6. Labor force participation decision l u 7. New realizations of exogenous states Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 12 /44

18 Planner s problem V (u,e;s,s) = subject to: : I u i u i=1 max {u i,σ i,l } I Zz i (e i +h i ) ξu+βe[v (u,e ;s,s )] i=1 h i = Φφ i m(u i,v i ) e i = (1 δ)(e i +h i ) σ i I u = l i=1 e i u i [0,u],l [0,1],σ i [0,(1 δ)(e i +h i )] Γ Z,δ,Φ (Z,δ,Φ ;Z,δ,Φ), Γ v (v ;v,z,δ,φ ),Γ φ (φ ;φ),γ z (z ;z) Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 13 /44

19 Solution The FOC wrt u i yields: ( ) vi Zz i Φφ i m u u i ( ) vi +βe[v ei ( ) V u ( )](1 δ)φφ i m u u i = µ Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 14 /44

20 Solution The FOC wrt u i yields: ( ) vi Zz i Φφ i m u u i ( ) vi +βe[v ei ( ) V u ( )](1 δ)φφ i m u u i = µ The FOC wrt l is: E[V u (u,e ;φ,z,v,z,δ,φ )] = 0 The Envelope condition wrt u is: V u (u,e;φ,z,v,z,δ,φ) = µ ξ Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 14 /44

21 Solution The FOC wrt u i simplifies to: ( ) vi Zz i Φφ i m u u i ( ) +βe[v ei (u,e ;s,s vi )](1 δ)φφ i m u u i = µ Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 15 /44

22 Solution The FOC wrt u i simplifies to: ( ) vi Zz i Φφ i m u u i ( ) +βe[v ei (u,e ;s,s vi )](1 δ)φφ i m u u i = µ The Envelope condition wrt e i is: V ei (u,e;φ,z,v,z,δ,φ) = Zz i +β(1 δ)e[v ei (u,e ;φ,z,v,z,δ,φ )] Guess and verify that: V ei (u,e;φ,z,v,z,δ,φ) = z i Ψ(Z,δ,Φ) Conjecture true if: E[z i ] = ρz i Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 15 /44

23 Solution Using this result into the FOC wrt u i : ( ) vi ZΦz i φ i m u u i ( ) +β(1 δ)ρe[ψ(z,δ,φ vi )]Φz i φ i m u u i = µ Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 16 /44

24 Solution Using this result into the FOC wrt u i : ( ) vi ZΦz i φ i m u u i ( ) +β(1 δ)ρe[ψ(z,δ,φ vi )]Φz i φ i m u u i = µ which yields the generalized Jackman-Roper condition: ( ) v1 z 1 φ 1 m u u 1 =... = z i φ i m u ( vi u i ) =... = z I φ I m u ( vi u I ), Convenient static condition to manipulate into mismatch indexes" Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 16 /44

25 2. MISMATCH INDEXES Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 17 /44

26 Mismatch indexm u t At date t, {v it } and u t given, hence θ t = v t /u t given W/o heterogeneity in (z i,φ i ), optimality requires u it = 1 θ t v it Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 18 /44

27 Mismatch indexm u t At date t, {v it } and u t given, hence θ t = v t /u t given W/o heterogeneity in (z i,φ i ), optimality requires u it = 1 θ t v it Number of mismatched unemployed: u M t = 1 2 I u it u it = 1 2 i=1 I i=1 u it u t 1 θ t vit u t u t = 1 2 I i=1 u it u t v it v t u t Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 18 /44

28 Mismatch indexm u t At date t, {v it } and u t given, hence θ t = v t /u t given W/o heterogeneity in (z i,φ i ), optimality requires u it = 1 θ t v it Number of mismatched unemployed: u M t = 1 2 I u it u it = 1 2 i=1 I i=1 u it u t 1 θ t vit u t u t = 1 2 I i=1 u it u t v it v t u t Mismatch unemployment as a share of total is: M u t um t u t = 1 2 I i=1 u it u t v it v t which can be computed from observed distribution {u it,v it } Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 18 /44

29 Mismatch indexm u t (contd.) With heterogeneity in φ i and m(u it,v it ) = Φ t φ i v α it u1 α it : M u φt = 1 2 I i=1 u it u t ( ) 1 φi α v it φ t v t where φ t = [ I i=1 φ 1 α i ( vit v t ) ] α Similarly for the model with heterogeneous productivities M u zt Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 19 /44

30 Mismatch indexm u t (contd.) With heterogeneity in φ i and m(u it,v it ) = Φ t φ i v α it u1 α it : M u φt = 1 2 I i=1 u it u t ( ) 1 φi α v it φ t v t where φ t = [ I i=1 φ 1 α i ( vit v t ) ] α Similarly for the model with heterogeneous productivities M u zt M u t : fraction of unemployed searching in the wrong sector Hence, index of misallocation of unemployed workers Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 19 /44

31 Mismatch indexm h t Assume Cobb-Douglas matching function: h it = Φ t v α it u1 α it Summing across sectors, aggregate hires equal: h t = Φ t v α t u 1 α t [ I i=1 ( vit v t ) α ( uit u t ) 1 α ] and optimal aggregate hires are h t = Φ t v α t u 1 α t Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 20 /44

32 Mismatch indexm h t Assume Cobb-Douglas matching function: h it = Φ t v α it u1 α it Summing across sectors, aggregate hires equal: h t = Φ t v α t u 1 α t [ I i=1 ( vit v t ) α ( uit u t ) 1 α ] and optimal aggregate hires are h t = Φ t v α t u 1 α t Alternative mismatch index: M h t h t h t h t = 1 I ( vit ) α ( uit v i=1 t u t ) 1 α measures the fraction of hires lost because of misallocation Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 20 /44

33 3. COUNTERFACTUALS Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 21 /44

34 Explaining the shift in the Beveridge curve Aggregate matching function: h t = ( 1 M h t ) Φt v α t u 1 α t Take logs: logh t = log [( ] 1 M h t ) Φt }{{} Aggr. matching efficiencya t +αlogv t +(1 α)logu t Estimate {A t } residually Given our estimate of {1 M h t}, we can measure how much of the observed shift in aggr. efficiency is due to increased mismatch Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 22 /44

35 Counterfactual unemployment dynamics Observed unemployment dynamics u t+1 = u t +s t (1 u t ) f t u t Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 23 /44

36 Counterfactual unemployment dynamics Observed unemployment dynamics u t+1 = u t +s t (1 u t ) f t u t Aggregate job finding rate: 1. observed: f t = (1 M h t) Φ t ( ) α v t u t 2. no mismatch: f t = Φ t ( vt u t ) α = f t (1 M h t ) ( ut u t ) α Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 23 /44

37 Counterfactual unemployment dynamics Observed unemployment dynamics u t+1 = u t +s t (1 u t ) f t u t Aggregate job finding rate: 1. observed: f t = (1 M h t) Φ t ( ) α v t u t 2. no mismatch: f t = Φ t ( vt u t ) α = f t (1 M h t ) ( ut u t ) α Counterfactual unemployment dynamics in absence of mismatch: u t+1 = u t +s t (1 u t) f t u t u u : how much of the observed rise in unemployment is due to increased mismatch Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 23 /44

38 4. ANALYSIS BASED ON JOLTS Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 24 /44

39 Sources of data Vacancies: JOLTS 2000: :2 Disaggregation: 16 industries in the private sector + government, and 4 Census regions Unemployment: Monthly CPS Information on industry and occup. of last employment only Productivity: Average hourly earnings by industry (CES) Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 25 /44

40 Matching function specification For 2-digit industries, we estimate CES matching function: ln ( hit u it ) = logφ t +logφ i + 1σ [ ( ) σ log vit α +(1 α)] u it ˆσ % Conf. Int. [ 0.267, 0.081] Recall: σ (,1), with σ = 0 for Cobb-Douglas Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 26 /44

41 Matching function specification For 2-digit industries, we estimate CES matching function: ln ( hit u it ) = logφ t +logφ i + 1σ [ ( ) σ log vit α +(1 α)] u it ˆσ % Conf. Int. [ 0.267, 0.081] Recall: σ (,1), with σ = 0 for Cobb-Douglas When restricting to Cobb-Douglas: we estimate ˆα = 0.60 and ˆφ i for each industry Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 26 /44

42 Labor demand shifts across industries Vacancy Share 0.1 Construction Manufacturing Durables Finance Health Date Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 27 /44

43 Correlation between(u, v) shares across industries Correlation Coefficient Date Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 28 /44

44 Mismatch indexm u t (JOLTS) M u M u φ M u z M u x 0.35 Index Date After the recession: additional 5% of unemployed misallocated Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 29 /44

45 Mismatch indexm h t (JOLTS) 0.08 M h M h φ M h z M h x 0.06 Index Date After the recession: additional 2% of monthly hires lost bc of mismatch Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 30 /44

46 Accounting for shift in aggregate matching function logh t = log [( 1 M h t) φxt Φ t ] +αlogvt +(1 α)logu t Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 31 /44

47 Accounting for shift in aggregate matching function logh t = log [( 1 M h t) φxt Φ t ] +αlogvt +(1 α)logu t Shift in the Beveridge Curve Data Counterfactual log(1 M 0.25 h ) Counterfactual log(1 M h x ) log(φ x ) Date Industry mismatch explains a tiny fraction of the observed shift Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 31 /44

48 Accounting for the rise in US unemployment U.S. Data Counterfactual M h 0.09 Unemployment Rate Date At most 0.7 pct points of rise in u explained by industry mismatch Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 32 /44

49 Geographical mismatch (4 Census regions) 0.3 M u M u φ Index Date Geographical mismatch shows no significant trend Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 33 /44

50 5. ANALYSIS BASED ON HWOL Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 34 /44

51 The HWOL data: July HWOL program is targeted to cover the full universe of all online advertised vacancies which are posted directly on internet job boards or through newspaper online ads Four million ads per month (four thousand in JOLTS) Unduplication algorithm to identify ads posted on multiple boards Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 35 /44

52 The HWOL data: July HWOL program is targeted to cover the full universe of all online advertised vacancies which are posted directly on internet job boards or through newspaper online ads Four million ads per month (four thousand in JOLTS) Unduplication algorithm to identify ads posted on multiple boards Info by ad: Job board, Full/Part time, Location (county), SOC (6-digit), Education level, NAICS (6-digit), Salary (where available) Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 35 /44

53 The HWOL data: July HWOL program is targeted to cover the full universe of all online advertised vacancies which are posted directly on internet job boards or through newspaper online ads Four million ads per month (four thousand in JOLTS) Unduplication algorithm to identify ads posted on multiple boards Info by ad: Job board, Full/Part time, Location (county), SOC (6-digit), Education level, NAICS (6-digit), Salary (where available) Two major measurement issues: 1. Upward trend in the use of online advertisement 2. Number of vacancies in each ad Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 35 /44

54 JOLTS-HWOL comparison by Census region 1000 JOLTS HWOL 1600 JOLTS HWOL Vacancies (In Thousands) Vacancies (In Thousands) Date Date 1000 JOLTS HWOL 2000 JOLTS HWOL Vacancies (In Thousands) Vacancies (In Thousands) Date Date Correlation between aggregate time series is 0.91 Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 36 /44

55 Labor demand shifts across occupations Business and Financial Operations (Left Axis) Computer and Math (Left Axis) Healthcare Practitioner (Left Axis) Construction and Extraction (Right Axis) Production (Right Axis) Vacancy Share Vacancy Share Date Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 37 /44

56 Correlation between(u, v) shares across occupations Correlation Coefficient Date Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 38 /44

57 Mismatch indexm h t (HWOL 2 digit occ.) 0.16 M h 0.14 Index Date After the recession: additional 3% of monthly hires lost bc of mismatch Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 39 /44

58 Accounting for the rise in US unemployment 2 Digit SOC (CF_HWOL_soc.eps) U.S. Data Counterfactual M h 0.09 Unemployment Rate Date At most 1.3 pct points of rise in u explained by occupational mismatch Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 40 /44

59 Vacancy and unemployment shares by state California (Right Axis) Florida (Left Axis) New York (Left Axis) Ohio (Left Axis) Texas (Left Axis) Vacancy Share Vacancy Share Unemployment Share Unemployment Share Date 0.04 California (Right Axis) Florida (Left Axis) New York (Left Axis) Ohio (Left Axis) Texas (Left Axis) Date 0.13 Significant shifts for some big states, but small or no shifts for all others Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 41 /44

60 Geographical mismatch (50 states) 0.08 M h 0.06 Index Date Geographical mismatch across states shows a slight decline Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 42 /44

61 Accounting for the rise in US unemployment State (CF_HWOL_State.eps) U.S. Data Counterfactual M h 0.09 Unemployment Rate Date Role of geographical mismatch appears irrelevant Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 43 /44

62 Conclusions Building on Jackman-Roper (1987), we develop an approach to measure mismatch unemployment in the labor market Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 44 /44

63 Conclusions Building on Jackman-Roper (1987), we develop an approach to measure mismatch unemployment in the labor market Main findings: 1/4 to 1/7 of observed rise in unemployment due to mismatch Misallocation by industry/occupation, but not by geography Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 44 /44

64 Conclusions Building on Jackman-Roper (1987), we develop an approach to measure mismatch unemployment in the labor market Main findings: 1/4 to 1/7 of observed rise in unemployment due to mismatch Misallocation by industry/occupation, but not by geography Future work: Correction for industries/occupation of unemployed Mismatch indexes by education level Access to UI records for selected states Şahin-Song-Topa-Violante, Measuring Mismatch in the U.S. Labor Market p. 44 /44