Other Models of Labor Dynamics

Transcription

1 Other Models of Labor Dynamics Christopher Taber Department of Economics University of Wisconsin-Madison March 2, 2014

2 Outline 1 Kambourov and Manovskii 2 Neal 3 Pavan

3 Occupational Specificity of Human Capital Kambourov and Manovskii want to estimate something like the returns to tenure specification, but allow for occupation and industry specific human capital

4 They use the model log(w ijmnt ) = β 0 Emp_Ten ijt + β 1 OJ ijt + β 2 Occ_Ten imt + β 3 Ind_Ten int +Work_Exp it + θ it where Emp_Ten ijt OJ ijt Occ_Ten imt Ind_Ten int Work_Exp it Tenure at the employer Dummy for first year on the job Tenure in the current occupation Tenure in the Current Industry Total Work Experience

5 Data One hard part of this is that they need to get good data on occupation which is often measured poorly They use the PSID They make use of the PSID Retrospective Occupation-Industry Supplemental Data Files which retrospectively get better measures of occupations for the period They are going to make a distinction between 1, 2 and 3 digit occupations and industries. Lets see what that means

6

7

8 The error term in the model is quite complicated with θ it = µ i + λ ij + ξ im + v in + ε it where µ i is individual effect, λ ij is job match, ξ im is occupation match, and v in is industry match (and as usual ε it is noise) This probably means about everything is biased upward

9 They will deal with this using the Altonji/Shakotko approach That is, they will use Emp_Ten ijt Emp_Ten ij as an instrument for Emp_Ten ijt Occ_Ten ijt Occ_Ten ij as an instrument for Occ_Ten ijt Ind_Ten ijt Ind_Ten ij as an instrument for Ind_Ten ijt

10

11

12 They do a lot of other robustness checks Basic results seem robust: Occupational specific tenure is really important Firm specific tenure is not important Industry specific tenure is somewhere in between

13 Outline 1 Kambourov and Manovskii 2 Neal 3 Pavan

14 Neal 1999 Neeal distinguishes between complex job switches in which workers switch careers from simple job shifts in which workers switch firms but do not switch careers He adds uncertainty into our framework: when workers try a new job they don t know whether they will be good at it or not He develops a simple model of this and shows that the data is consistent with the basic predictions of the model: workers first shop for a career and then shop for a firm within the career

15 The key components of the model are: Career match θ distributed F(θ) Job match ξ distributed G(ξ) The key restriction of the model is that to switch careers, you must switch firms, but to switch firms, you do not have to switch careers He is abstracts from all but the most necessary components-clearly one could make this model more complicated if you want.

16 Assuming that people are paid θ + ξ and that there are no search costs in the sense that you can always find a new job of the type you want-but you don t observe the match component until you start working there You can write the Belman equation as { V (θ, ξ) = θ + ξ + βmax V (θ, ξ), } V(x, s)df(x)dg(s) V(θ, s)dg(s), where V is the value function and β is the discount factor.

17 characterized by figure 1. The variables 0 - and 4* serve as quasi-reservation values for each type of match, and based on these values, the figure is divided into three regions. Workers holding a pair (0, 4) that lies in region A choose to draw a new pair at the beginning of the next period. Workers holding (0, A) in region B keep their current career match 0 but draw a new firm match at the beginning of the next period. Workers who hold (0, 4) in region C cease searching. You can think of it in terms of the following figure Stop C Change 0 and 4 A Change 4 B FIG. 1 Given this search strategy, workers never change careers after changing firms within a given career. In this model, workers who make simple firm

18 Note that once you get to region B, you will never go back to A Once you get to C, you will stay This has the implication that as workers age, the fraction of job changes that are complex should fall Note also that if we condition on people who have never made a simple job change, the probability that the next job change will be simple does not depend on age Neal looks for these implications in the data

19 Data He uses the NLSY79 which is great for constructing data on job changes and how they vary with occupation and industry He looks at Males only The question here is what represents a career Neal assumes that a complex job change represents both an occupation change and an industry change Lets look at the first piece of evidence. Each observations is a sequence of job changes. He groups by the total number of job changes and documents the fraction consistent with the pure model (i.e. no complex changes following simple changes)

20

21 You can see that the results are not precisely the two stage model, but they are much closer than you would expect by chance Next an observation is a single job change He groups by the number of simple changes since working in the current career (and by education)

22

23 Key thing is that (for example) for high school graduates for whom this is there first firm in the career, the chances that the next switch is complex is 69% However, for those who underwent a previous job switch in this career, it is only 22% The next tables are similar, but we group by experience

24

25

26 One concern is that this could be about career specific human capital rather than about search. Neal addresses this with the following Table

27

28 While the strict version of the model is not precisely true, the data is broadly consistent with the idea.

29 Outline 1 Kambourov and Manovskii 2 Neal 3 Pavan

30 Career Choice and Wage Growth Pavan implements a structural extension of Neal s model Like Neal he uses a similar definition of Career change.

31 Data He also uses the NLSY79 Individuals born between 1957 and 1964 Representative males (with some restrictions to simplify sample) Different definitions of career (3 digit) occupation and industry change (t 1, t) to (t + 1, t + 2) occupation and industry change t to t + 1 occupation t to t + 1 industry t to t + 1

32

33 He then does something similar to what Komogorov and Manovskii do, but for career (using his four definitions of career)

34

35

36 Structural Model h e : general human capital θ c : career specific human capital ε t : firm specific human capital All three variables are initially drawn from a normal h 1 N(µ h1,σ 2 h) θ 1 N(0, σ 2 θ ) ε 1 N(0, σ 2 ε) as in Neal, you draw a new θ when you switch career and a new ε when you switch firms Don t get to see those variables unless you actually move

37 These things evolve h e =h(e, h 1 ) θ c =µ θ + α 1 (h 1 µ h1 ) + θ c 1 + u θc µ θc N(0, ηθ 2 ) ε t =µ ε + α 2 (h 1 µ h1 ) + ε t 1 + u εt µ εt N(0, ηε) 2 Wages are log(w ect ) =δ X e + h e + θ c + ε t econometrician observes wages measured with iid normal measurement error

38 Utility Let H e be hours worked K e (H e ) disutility of work Ũ = log(w e H e ) K e (H e ) =δ X e + h e + θ c + ε t + log(h e ) K e (H e ) First order condition for H e implies 1 H e =K e (H e )

39 Dynamics Every period I receive an offer in a different career and another in the same career Separate from firm exogenously at rate p f When that happens with additional probablity p c he may also have to switch career

40 Value Function This gives the value function V(h 1, e, θ, ε, X e ) =δ X e + h e + θ c + ε t + log(h e ) K e (H e ) + β(1 p f )V NS (h 1, e + 1, θ, ε, X e+1 ) + βp f p c V SC (h 1, e + 1, X e+1 ) + βp f (1 p c )V SF (h 1, e + 1, θx e+1 ) Where ( ) V NS (h 1, e + 1, θ, ε, X e+1 ) = max{e V(h 1, e + 1, θ, ε, X e ) θ, ε, ( ) E V(h 1, e + 1, θ, ε 1, X e ) θ C f, E (V(h 1, e + 1, θ 1, ε 1, X e )) C c }

41 and V SC (h 1, e + 1, X e+1 ) =E (V(h 1, e + 1, θ 1, ε 1, X e )) C c ( ) V SF (h 1, e + 1, θx e+1 ) = max{e V(h 1, e + 1, θ, ε 1, X e ) θ C f, E (V(h 1, e + 1, θ 1, ε 1, X e )) C c }

42 Estimation Pavan solves this model by Maximum Likelihood (Much much easier said than done, see his paper for details)

43

44 Wage Decompositions The model is complicated To understand it better Pavan does three different wage growth decompositions Remember that we can write log(w ect ) =δ X e + h e + θ c + ε t

45 The X s he uses don t change over time so log(w e+1c t ) log(w ect) =h e+1 h e + θ c θ c + ε t ε t =h e+1 h e + (θ c θ c ) 1 ( c = c + 1 ) + (θ c θ c ) 1 ( c = 1 ) + + (ε t ε t ) 1(t = t + 1) + (ε t ε t ) 1(t = 1) =h e+1 h e + (µ θ + α 1 (h 1 µ h1 ) + u θc ) 1 ( c = c + 1 ) + (θ 1 θ c ) 1 ( c = 1 ) + (µ ε + α 2 (h 1 µ h1 ) + u εt ) 1(t = t + 1) + (ε 1 ε t ) 1(t = 1) (there is a typo in the paper) Do this decomposition for three different cases 1 People who stay at the same job 2 People who stay in the same career 3 Everyone

46

47

48

49 Altonji and Shakatko approach This suggests there is a return to firm tenure, but Altonji and Shakotko didn t So what is the reason? Pavan simulates data from his model and repeats the procedure on the simulated data This suggests that this is not a good way to do this and we really need structural models

50