Lecture VIII. Income Process: Facts, Estimation, and Discretization

Transcription

1 Lecture VIII Income Process: Facts, Estimation, and Discretization Gianluca Violante New York University Quantitative Macroeconomics G. Violante, Income Process p. 1 /26

2 Estimation of income process Competitive model: unique market price per efficiency unit wage distribution reflects heterogeneity in efficiency units: Y iat = P t exp (Ỹiat ) Compute residuals y iat from regression in logs logy iat = D t +β X iat +y iat time dummies D t = [logp 1,logP 2,...], and X iat are fixed observables that capture predictable part of Ỹ iat : age, edu, race Gender: be aware of of selection. Typically use data on men. Important to distinguish between wages (endogenous labor supply), individual earnings (two-earner hh, inelastic labor supply), hh earnings (one earner hh, inelastic labor supply), hh income (endowment economy) G. Violante, Income Process p. 2 /26

3 Step 1: choose statistical model Assume process is stationary: when you compute moment conditions, do it t by t, but then average across t. Identification/estimation in nonstationary process is more complicated, see Heathcote-Storesletten-Violante (2010). We choose following specification for log earnings process: Why persistent + transitory? y ia = z ia +ε ia z ia = ρz i,a 1 +u ia u ia iid (0,σ u ) z i0 iid (0,σ z0 ) ε ia iid (0,σ ε ) G. Violante, Income Process p. 3 /26

4 0.25 NLSY 79: Covariance of wages for college educated males 0.2 COV (age 22, a) COV( age 23,a) COV( age 24, a) Covariance Age (a) G. Violante, Income Process p. 4 /26

5 Identification in levels y ia z ia = z ia +ε ia = ρz i,a 1 +u ia var(y i0 ) = σ z0 +σ ε for a = 0 var(y ia ) = var(z ia )+σ ε for a > 0 var(z ia ) = ρ 2 var(z i,a 1 )+σ u cov(y ia,y i,a j ) = cov(z ia,z i,a j ) for j > 0 cov(z ia,z i,a j ) = ρ j var(z i,a j ) Identification of ρ from the slope of the covariance at lags > 0: cov(y ia,y i,a 2 ) cov(y i,a 1,y i,a 2 ) = ρ2 var(z i,a 2 ) ρvar(z i,a 2 ) = ρ G. Violante, Income Process p. 5 /26

6 Identification in levels Identification of σ ε from the difference between variance and covariance var(y i,a 1 ) ρ 1 cov(y ia,y i,a 1 ) = var(z i,a 1 )+σ ε var(z i,a 1 ) Given σ ε, obtain σ z0 residually from var(y i,0 ) Finally, identification of σ u : var(y i,a 1 ) cov(y ia,y i,a 2 ) σ ε = ρ 2 var(z i,a 2 )+σ u +σ ε ρ 2 var(z i,a 2 ) σ ε = σ u Identification achieved with two lags: (a 2,a 1,a). Typically, with long panels (T ) model is largely overidentified, and parameters tightly estimated because N is large (thousands of observations) but if nonstationary... G. Violante, Income Process p. 6 /26

7 Identification in first differences (for ρ = 1) y ia = u ia +ε ia ε i,a 1 y i,a 1 = u i,a 1 +ε i,a 1 ε i,a 2 var( y ia ) = σ u +2σ ε cov( y ia, y i,a 1 ) = σ ε Of course, you can use moments in first differences as additional moments, together with those in level, to estimate the parameters of AR(1) Not-so-much-fun fact: if you estimate a permanent-transitory process in levels and first-differences, you obtain very different results. Why? G. Violante, Income Process p. 7 /26

8 Heterogeneous Income Profiles A common alternative specification to the persistent-transitory representation of idiosyncratic income risk is: y ia = α i +β i a+z ia z ia (α i,β i ) = ρz i,a 1 +u ia u ia iid (0,σ u ) iid (0, Σ) (α i,β i ) are, respectively, constant and slope of earning profile, heterogeneous across individuals Typically, z it estimated to be a lot less persistent HIP difficult to distinguish, empirically, from persistent-transitory G. Violante, Income Process p. 8 /26

9 Step 2: Estimation You have an unbalanced panel of individuals aged a = 1,...,A. For every individual i = 1,...,I define a (A 1) vector d i = d i1... d ia where d ia {0,1} is an indicator variable for whether individual i is present at age a in the sample. Analogously to d ia define an (A 1) vector y i = y i1... y ia since the panel is unbalanced, must set missing elements to zero G. Violante, Income Process p. 9 /26

10 Step 2: Estimation The covariance of earnings is then a (A A) symmetric matrix computed as C = Y D This is an element by element division, where Y and D are (A A) matrices given by Y = I y i y i D = i=1 I d i d i i=1 Each element of y i y i Each element of d i d i is product of earnings at ages (p,q) for i is 1 only if i is present at both ages (p,q). Each entry of C is the cross-sectional covariance of earnings at ages (p,q) G. Violante, Income Process p. 10 /26

11 Step 2: Estimation Take the upper triangular portion of C and vectorize it into an (A(A+1)/2,1) vector. m =vech ( C UT) Let f (Θ) be the comforming (A(A+1)/2,1) vector of empirical moments. Estimation based on a minimum distance estimator that minimizes the distance btw the model covariance matrix and the empirical covariance matrix (Chamberlain, 1984) family of GMM. The estimator that solves the following minimization problem min Θ [m f (θ)] Ω[m f (θ)] where Ω is a weighting matrix and θ is a (n,1) vector of parameters. G. Violante, Income Process p. 11 /26

12 Weighting matrix Define, comformably with m, the vector m i that contains the distinct elements of the cross-product matrix y i y i. Similarly, define the vector s = vech ( D UT). Chamberlain proved that m has asymptotic variance V = I i=1 (m m i)(m m i ) ss where V is a (A(A+1)/2,A(A+1)/2) matrix Chamberlain shows that the optimal weighting matrix Ω is V 1 Altonji and Segal (1996) find, by MC simulations, that there is substantial small sample bias in the estimates of θ and recommend using (i) Ω = I (equally weighted estimator) or Ω = diag ( V 1) (diagonally weighted estimator) G. Violante, Income Process p. 12 /26

13 Step 3: S.E. and Testing Standard errors of θ are then computed as (G ΩG) 1 G ΩVΩG(G ΩG) 1 where G is the (A(A+1)/2,n) gradient matrix f (θ)/ θ evaluated at θ. Finally, the chi-squared goodness of fit statistic is [m f (θ )]R [m f (θ )] χ 2 q where R is the generalized inverse of R = WVW where W = I G(G ΩG) 1 G Ω, and q = A(A+1)/2 n, are the the degrees of freedom. G. Violante, Income Process p. 13 /26

14 Higher moments of income distribution Three approaches to think about skewness and excess kurtosis with a continuous process: 1. Assume innovations drawn from skewed and leptokurtic distributions and add higher moments to estimation 2. Clark (ECMA, 1973): subordinated stochastic processes Y(T(t)), where T(t) is the directing process that sets number of times Y changes in period t Both Y(t) and T(t) have independent increments If Y is normal and T lognormal, Y(T(t)) displays kurtosis. 3. Kou (Management Science, 2002): combination of diffusion plus jump process, where jump is drawn from an asymmetric double-exponential distribution G. Violante, Income Process p. 14 /26

15 Step 4: Discretization Two dominant methodologies: 1. Tauchen method (Tauchen EL, 1986) 2. Rowenhorst method (article in Frontiers of business cycle, (Tom Cooley, editor, 1995) Rowenhorst method better as ρ 1, plus one can derive first four moments in closed form Two approaches to derive moments in closed form: Kopecki-Suen (RED, 2010) and Lkhagvasuren (2012) Multivariate version: Terry-Knotek (EL, 2011), Galindev-Lkhagvasuren (JEDC, 2010) G. Violante, Income Process p. 15 /26

16 Tauchen vs Rowenhorst (9 points,σ y = 1) Continuous, ρ = 0.99 Continuous, ρ = Tauchen, ρ = Tauchen, ρ = Rouwenhorst, ρ = Rouwenhorst, ρ = G. Violante, Income Process p. 16 /26

17 Rowenhorst method to discretize AR(1) Consider the following two-state Markov chain x {0,1} where Pr(x = 0 x = 0) = p Pr(x = 0 x = 1) = 1 q Pr(x = 1 x = 0) = 1 p Pr(x = 1 x = 1) = q Compute the invariant distribution [ 1 α α ] [ p 1 p 1 q q ] = [ 1 α α ] A system of 2 equations in 2 unknowns with with solution α Pr(x = 1) = 1 p 2 p q G. Violante, Income Process p. 17 /26

18 Moments in closed form Bernoulli invariant distr. we can compute moments analytically: E(x) = α = 1 p 2 p q Var(x) = α(1 α) = (1 p)(1 q) (2 p q) 2 Skew(x) = 1 2α α(1 α) = p q (1 p)(1 q) Ekurt(x) = α(1 α) = 2+ (p q) 2 (1 p)(1 q) Corr(x,x ) = p+q 1 G. Violante, Income Process p. 18 /26

19 Generalization to n-state process Consider the auxiliary process X n = n 1 i=1 x i where each x is an independent 2-state Markov chain Therefore X n {0,1,...,k,...,n 1}. Define the process: y = n 1 i=1 (a+bx i ) and choose a and b so that y takes values in [m,m+ ] where m is the mean and 2 the range of the state space. G. Violante, Income Process p. 19 /26

20 Generalization to n-state process By setting y min = m and y max = m+, it is easy to see that a = m n 1 2 b = n 1 It follows that the kth point of the grid (0 < k < n 1) is: y k = m + 2 n 1 k We have the grid for y as a function of three parameters (n,m, ) G. Violante, Income Process p. 20 /26

21 Moments in closed form for n-state process Then, it is easy to see that: E(y) = (n 1)[a+bE(x)] = m +2 1 p 2 p q Var(y) = (n 1)b 2 Var(x) = 4 2 n 1 Corr ( y,y ) = Corr ( x,x ) = p+q 1 (1 p)(1 q) (2 p q) 2 = m+ q p 2 p q Note: don t need to set q = p for the process to have mean zero. Less easy to show that: Skew(y) = Skew(x) (n 1) = p q (n 1)(1 p)(1 q) Ekurt(y) = Ekurt(x) n 1 = 2 n 1 + (p q) 2 (n 1)(1 p)(1 q) therefore, important constraint: Ekurt(y) = 2/(n 1)+Skew(y) 2 G. Violante, Income Process p. 21 /26

22 Exact moment matching 5 moments {E(y),Var(y),Corr(y,y ),Skew(y),Ekurt(y)} and 5 parameters {m,,n,p,q} Moment matching: (p,q,n) Corr(y,y ),Skew(y),Ekurt(y) (m, ) E(y),Var(y) In the data (SCF log household earning residuals, ages 22-60): E(y) = 0,Var(y) = 0.8,Corr(y,y ) = 0.95,Skew(y) = 1.1,Ekurt = 4 so one must give up either on the skewness or on the kurtosis... Homework: estimate parameters to fit those moments. Simulate process. Can you replicate the var, skewness and kurtosis of the one year changes? Can you derive the moments in first differences in closed form like the ones in level? G. Violante, Income Process p. 22 /26

23 Stationarizing the stochastic growth model The representative household solves: max {I t,c t,n t } E 0 t=0 β t [C φ t (1 N t ) 1 φ] 1 σ 1 σ s.t. C t +I t = (1 τ t )W t N t +r t K t +Φ t K t+1 = (1 δ)k t +I t Z t = G t z t, z t stationary, G = 1+g The representative firm solves (F is CRS): max {N t,k t } F (K t,z t N t ) W t N t r t K t The government balances its budget: Φ t = τ t W t N t G. Violante, Income Process p. 23 /26

24 Stationarizing the stochastic growth model Define a blue new set of stationary variables: c t C t /G t,k t K t /G t,i t I t /G t,φ t Φ t /G t,w t W t /G t,z t Z t /G t Rewrite the household problem as: max {i t,c t,n t } E 0 t=0 c t +i t β t [ c φ t (1 N t ) 1 φ] 1 σ s.t. 1 σ = (1 τ t )w t N t +r t k t +φ t k t+1 G = (1 δ)k t +i t where β t = (βg φ(1 σ)) t G. Violante, Income Process p. 24 /26

25 Stationarizing the stochastic growth model By CRS, the firm problems becomes max N t,k t F (k t,z t N t ) w t N t r t k t The government budget constarint becomes φ t = τ t w t N t Thus all the stationarized variables grow at rate (G 1) Labor (btw 0 and 1) and interest rate (Y/K) do not grow The model has a balanced growth path Key: preferences s.t. inc. effect = subst. effect on labor supply G. Violante, Income Process p. 25 /26

26 Stationarized Euler equation c φ(1 σ) 1 t (1 N t ) (1 φ)(1 σ) = λ t β t λ t G = β t+1 λ t+1 [1+F k (k t+1,z t+1 N t+1 ) δ] which implies: β t G φ(1 σ)t+1 λ t = β t+1 G φ(1 σ)(t+1) λ t+1 (1+r t+1 δ) G 1 φ(1 σ) ( ct+1 c t ) 1 φ(1 σ) ( 1 Nt 1 N t+1 ) (1 φ)(1 σ) = β[1+f k (k t+1,z t+1n t+1) δ] We could have written the FOC from the model with trends and stationarized it (note that F k is homogenous of degree zero). What is preferable? It depends how you solve the model! G. Violante, Income Process p. 26 /26