ECON 5120: Panel data econometrics

Seminar 3: October 2., 2007

Problem 3.1: Error autotocorrelation and heteroskedasticity

Standard variance components model:

(0.1)

Rewriting the model in matrix form gives:

,,

are uncorrelated and have zero expectation, with covariance matrix

. Model (0.1) can thus be written compactly as:

(0.2) ,

where

(0.3)

Models with autocorrelation:

Model A1:

Assume that the AR(1) process started infinetly long back in time. Then .

We derive the variance of :

From Lillard and Willis (1978) p.989[1], we know that

The first line is the variance, the second the covariance between inventions in different periods, for the same individual. Thus, the covariance of is:

We can write the residual covariance matrix of the genuine disturbance, :

and the covariance matrix of as:

In model A1, there is homoschedasiticity and autocorrelation, but not equicorrelation unless . This can be seen from the fact that[2]:

The correlation is independent of i but varies with for .

Model A2:

We go through the same steps as for model A1, the only difference being that is now individual specific (has subscript i). The aggregate covariance matrix is thus:

In this model, is both heteroskedastic and autocorrelated. We now have:

which varies with both i and for .

Model A3:

We go through the same steps as for models A1 and A2, the only difference being that now both and are individual specific (have subscript i). The aggregate covariance matrix is thus:

In this model, as in A2, is both heteroskedastic and autocorrelated. We now have:

which varies with both i and for .

To sum up, the models A1-A3 all have autocorrelation, and A2-A3 also have heteroskedasticity. However, none of them have equicorrelation, as the correlation of varies with for .

Models with error heteroskedasticity:

Model H1:

The covariance matrix for this model is: which can be written out just as (0.3), only with individual . We thus have:

in this model is heteroskedastic, and it is equicorrelated for individual i.

Model H2:

The covariance matrix for this model is: which can be written out just as (0.3), only with individual . We thus have:

in this model is heteroskedastic, and it is equicorrelated for individual i.

Model H3:

The covariance matrix for this model is: which can be written out just as (0.3), only with individualand . We thus have:

in this model is heteroskedastic, and it is equicorrelated for individual i.

All of the models H1-H3 have heteroskedastic disturbances, ’s, which are equicorrelated for individual i.

Would you consider any of these extensions as improvements of the model? That depends on the real structure of the data. For instance, there is no point in using an autocorrelation covariance matrix for estimation unless there actually is autocovariance in the disturbances. We lose more periods when we have to account for autocorrelation.

Problem 3.2 A: Instrument variables

If we treat all variables as exogenous, we can use the one-stage within estimator. Xtreg

We assume that the model can be written

=

= ~IID()

. xtreg ln_wage age* ten not_s uni so, fe i(idcode)

Fixed-effects (within) regression Number of obs = 19007

Group variable (i): idcode Number of groups = 4134

R-sq: within = 0.1333 Obs per group: min = 1

between = 0.2375 avg = 4.6

overall = 0.2031 max = 12

F(6,14867) = 381.19

corr(u_i, Xb) = 0.2074 Prob > F = 0.0000

------

ln_wage | Coef. Std. Err. t P>|t| [95% Conf. Interval]

------+------

age | .0311984 .0033902 9.20 0.000 .0245533 .0378436

age2 | -.0003457 .0000543 -6.37 0.000 -.0004522 -.0002393

tenure | .0176205 .0008099 21.76 0.000 .0160331 .0192079

not_smsa | -.0972535 .0125377 -7.76 0.000 -.1218289 -.072678

union | .0975672 .0069844 13.97 0.000 .0838769 .1112576

south | -.0620932 .013327 -4.66 0.000 -.0882158 -.0359706

_cons | 1.091612 .0523126 20.87 0.000 .9890729 1.194151

------+------

sigma_u | .3910683

sigma_e | .25545969

rho | .70091004 (fraction of variance due to u_i)

------

F test that all u_i=0: F(4133, 14867) = 8.31 Prob > F = 0.0000

(We note that only 19007 obs are used in the regression, due to missing variables in UNION)

The F test is a test for absence of fixed effects. We can assume fixed effects.

We can make a plot to look at the residuals. First we predict Xb, We have named it “yhatt”

We then compute the residuals: y- yhatt: We have named it “res”, or we can type:

predict <varname>, ue

Either way we obtain , the combined residual.

. predict yhatt

. gen res= ln_wage- yhatt

(or predict res_ ,ue)

. hist res

. hist res, normal

. twoway (scatter res yhatt) (mspline res yhatt)

. twoway (scatter res year) (mspline res year)

This gives the plots below. By looking at the plot it seems that our model seems to fit assumptions on ~IID()

We can also predict the first differenced overall component =

, by typing:

predict res2,e

We then obtain these plots.

IV-modell 2:

xtivreg ln_wage age* not_s (tenure = south union), fe

If we believe that tenure is an endogenous variable, we can try to handle this with instruments. It is suggested that we use union and south as instruments for tenure.

We then need another specification of the model.

1) = = ~IID()

X is still a vector of exogenous variables; Y is a vector of observations of endogenous variables, that are allowed to correlate with. N is the number of observations, and n is the number of girls.(groups)

We then use xtivreg which is a twostage estimator.

First we estimate

It is important to construct instruments that are strongly correlated with the endogenous variable, but not . We find that the correlation between and not_smsa is -0.1451 and between and union is 0,0792. The correlations between the endogenous variable and the instruments are -0.0266 and -0.1321. So they are not very good instruments.

. corr tenure south union

(obs=19007)

| tenure south union

------+------

tenure | 1.0000

south | -0.0266 1.0000

union | 0.1600 -0.1321 1.0000

. corr res3 not_smsa union

(obs=19007)

| res3 not_smsa union

------+------

res3 | 1.0000

not_smsa | -0.1451 1.0000

union | 0.0792 -0.0693 1.0000

IV-modell 3:

xtivreg ln_wage age* not_s (tenure = south union), fe

Fixed-effects (within) IV regression Number of obs = 19007

Group variable: idcode Number of groups = 4134

R-sq: within = . Obs per group: min = 1

between = 0.1304 avg = 4.6

overall = 0.0897 max = 12

F(4138,14869) = 74.14

corr(u_i, Xb) = -0.6843 Prob > F = 0.0000

------

ln_wage | Coef. Std. Err. t P>|t| [95% Conf. Interval]

------+------

tenure | .2403531 .0373419 6.44 0.000 .1671583 .3135478

age | .0118437 .0090032 1.32 0.188 -.0058037 .0294912

age2 | -.0012145 .0001968 -6.17 0.000 -.0016003 -.0008286

not_smsa | -.0167178 .0339236 -0.49 0.622 -.0832123 .0497767

_cons | 1.678287 .1626657 10.32 0.000 1.359442 1.997132

------+------

sigma_u | .70661941

sigma_e | .63029359

rho | .55690561 (fraction of variance due to u_i)

------

F test that all u_i=0: F(4133,14869) = 1.44 Prob > F = 0.0000

------

Instrumented: tenure

Instruments: age age2 not_smsa south union

. correlate, _coef

| tenure age age2 not_smsa _cons

------+------

tenure | 1.0000

age | -0.3709 1.0000

age2 | -0.7337 -0.3543 1.0000

not_smsa | 0.4131 -0.1526 -0.3054 1.0000

_cons | 0.6172 -0.9515 0.0637 0.2079 1.0000

Note: corr(u_i, Xb) = -0.6843 is high, the model seems to be even worse than before

(note: this does not happen if we use another instrument than union, for instance hours)

Looking at the residuals we see that they do not seem to fit assumptions on IID (0,σ2). Our model specification with instrument variables does not improve our estimation.

By using tenure as a endogenous variable, using south and union as instruments, we find that age and not_smsa are no longer significant. If we believe for instance from other studies that these should be significant, we should use a different model specification.

IV-modell 4:

We are asked to use a between estimation.

After passing 1) trough the between estimator we are left with

Where for

We similarly define as the matrix of instruments after they have passed trough the between transformation. These instruments are used to correct the biases on the coefficients. We do not succeed, we find that sd(u_i + avg(e_i.))= 0,4445007. The residual plots also show that the coefficients are not constant for different values of X.

xtivreg ln_wage age* not_smsa (tenure= union south), be i(idcode)

Between-effects IV regression: Number of obs = 19007

Group variable: idcode Number of groups = 4134

R-sq: within = 0.0881 Obs per group: min = 1

between = . avg = 4.6

overall = 0.1483 max = 12

chi2(4) = 512.38

sd(u_i + avg(e_i.))= .4445007 Prob > chi2 = 0.0000

------

ln_wage | Coef. Std. Err. z P>|z| [95% Conf. Interval]

------+------

tenure | .1349486 .0102693 13.14 0.000 .1148212 .155076

age | .0424055 .0135488 3.13 0.002 .0158503 .0689607

age2 | -.0008035 .0002143 -3.75 0.000 -.0012235 -.0003835

not_smsa | -.2619405 .0162784 -16.09 0.000 -.2938457 -.2300354

_cons | .8430686 .2029267 4.15 0.000 .4453395 1.240798

------+------

Instrumented: tenure

Instruments: age age2 not_smsa union south

. correlate, _coef

| tenure age age2 not_smsa _cons

------+------

tenure | 1.0000

age | -0.1843 1.0000

age2 | 0.0832 -0.9898 1.0000

not_smsa | 0.0096 -0.0202 0.0156 1.0000

_cons | 0.1310 -0.9921 0.9767 0.0016 1.0000

IV-modell 5:

If we believe, or are willing to assume, that all’s are uncorrelated with the other covariates, we can fit the random-effects model. There are two variance components to estimate, the variance of and . Since the variance components are unknown, the consistent estimates are required to implement feasible GLS.

A consistent estimator is obtained by

The residuals in estimating are first obtained form OLS regression.

The estimates and their standard errors are calculated using .

(note: We are not quite sure about this, and hope that this can be commented on at the seminar.)

xtivreg ln_wage age* not_s (tenure = south union), re i(idcode)

. xtivreg ln_wage age* not_s (tenure = south union), re i(idcode)

G2SLS random-effects IV regression Number of obs = 19007

Group variable: idcode Number of groups = 4134

R-sq: within = 0.0620 Obs per group: min = 1

between = 0.1745 avg = 4.6

overall = 0.1206 max = 12

Wald chi2(4) = 941.52

corr(u_i, X) = 0 (assumed) Prob > chi2 = 0.0000

------

ln_wage | Coef. Std. Err. z P>|z| [95% Conf. Interval]

------+------

tenure | .1772948 .0111724 15.87 0.000 .1553972 .1991924

age | .0191674 .0066388 2.89 0.004 .0061555 .0321792

age2 | -.0008496 .0001057 -8.04 0.000 -.0010567 -.0006425

not_smsa | -.2119932 .0130456 -16.25 0.000 -.2375622 -.1864243

_cons | 1.42761 .1037797 13.76 0.000 1.224205 1.631014

------+------

sigma_u | .33156584

sigma_e | .63029359

rho | .21674808 (fraction of variance due to u_i)

------

Instrumented: tenure

Instruments: age age2 not_smsa south union

. correlate, _coef

| tenure age age2 not_smsa _cons

------+------

tenure | 1.0000

age | -0.2370 1.0000

age2 | -0.2199 -0.8895 1.0000

not_smsa | 0.0847 -0.0247 -0.0171 1.0000

_cons | 0.3147 -0.9874 0.8287 -0.0007 1.0000

3.2.B

The STATA output from running the regressions can be found on the pages below. We have given the models numbers, and comment on all the models first.

Model 3.2.B1. xtabond n w k ys yr1980-yr1984, lags(1)

Model 3.2.B2 xtabond n w k ys yr1980-yr1984, lags(1) robust

Model 3.2.-B3 xtabond n w k ys yr1980-yr1984, lags(1) twostep

Model 3.2.-B4 xtabond n w k ys yr1980-yr1984, lags(2)

Model 3.2.-B5 xtabond n l(0/1).w l(0/1).k l(0/1).ys yr1980-yr1984, lags(1)

Model 3.2.-B6 xtabond n l(0/1).w l(0/1).k l(0/1).ys yr1980-yr1984, lags(1) robust

Model 3.2.-B7 xtabond n l(0/1).w l(0/1).k l(0/1).ys yr1980-yr1984, lags(1) twostep

Model 3.2.-B8 xtabond n l(0/2).w l(0/2).k l(0/2).ys yr1980-yr1984, lags(2)

Model 3.2.-B9 xtabond n l(0/2).w l(0/2).k l(0/2).ys yr1980-yr1984, lags(2) robust

Model 3.2.-B10 xtabond n l(0/2).w l(0/2).k l(0/2).ys yr1980-yr1984, lags(2) twostep

Dynamic panel data models allow past realisations of the dependent variable to affect its current level.

1)

~IID(o,)

and are assumed to be independent for each i over all t.

is a vector of parameters to be estimated

Var(

Arellano and Bond derive a generalized method-of-moments estimator for , using lagged levels of the dependent variable as instruments.

This method assumes that there is no second-order autocorrelation in the .

xtabond includes the test for autocorrelation and the Sargan test of over-identifying restrictions for this model.

We do not know the AR structure but it can be different for each individual () and the variances () may be different for different individuals. (se Model A1-A3)

First differencing of the equation removes theand produces an equation that can be estimated using instrumental variables.

In all the models we use the lagged dependent variable as an instrument variable. We have then lost the three first observations, to lags and differencing. Since contains only strictly exogenous covariates, will serve as its own instrument. The instrument matrix has one row for each time period we are instrumenting.

The difficult part is to define and implement this kind of instrument matrix for each i. We have tried different methods of this in model B1-B10, without much success. It might be that we have omitted variables, and that our attempts will be no use with this model.

We have an unbalanced panel. This makes the algebra more difficult as we can not use kroneker products. But stata handles this. Missing observations are handled by dropping the rows for which there are no data, and filling inn zeroes in columns where missing data would be required.

It=Index set of individuals which are observed in period t . t =1,….T

Pi=Index set of periods where individual i is observed i =1,…N

T the number of periods when at least one individual is observed.

N is the number of individuals which are observed at least one period.

Vi define D as a (NxT) matrix whose element (i,t) is

Dit =

In model Model 3.2.B1 the genuine disturbance follows an AR(1) prosess.

Sargan test of over-identifying restrictions is rejected. Possibly due to heteroskedasticity. The presence of second order autocorrelation would imply that the estimates are inconsistent.

Model 3.2.B2 is similar to B1 but we now have computed robust standard errors, taken into account that we suspect heteroskedasticity .We see that the coefficients are the same, as they should be, and the (robust) standard errors are larger. But we still suspect that the estimates are inconsistent, because of the presence of second order autocorr.

Model 3.2.B3.

Areallo Bond recommends one step, but we see that Sagran test is not rejected and the autocorrelation test says there is no first order autocorrelation. But the estimates may still be inconsistent, because of the presence of second order autocorrelation.

We also note that several of the coeff. have changed, one has even switched sign.

Model 3.2.-B4

We use two lags of the dependent variable, but it is not significant for lag 2. The other results do not differ much.

Model 3.2.-B5 and B6

Here we use both the 1 difference and the lagged variable. The results do not differ much.

Model 3.2.-B7-10

We now use two lags of the dependent variable. But the estimates may still be inconsistent, because of the presence of second order autocorrelation.

xtabond n w k ys yr1980-yr1984, lags(1)

Arellano-Bond dynamic panel-data estimation Number of obs = 751

Group variable (i): id Number of groups = 140

Wald chi2(9) = 645.91

Time variable (t): year Obs per group: min = 5

avg = 5.364286

max = 7

One-step results

------

D.n | Coef. Std. Err. z P>|z| [95% Conf. Interval]

------+------

n |

LD. | .3566042 .0761519 4.68 0.000 .2073492 .5058592

w |

D1. | -.5114253 .0526485 -9.71 0.000 -.6146145 -.4082361

k |

D1. | .3086461 .0282417 10.93 0.000 .2532934 .3639988

ys |

D1. | .5032803 .0958316 5.25 0.000 .3154537 .6911069

yr1980 |

D1. | .0195602 .0143097 1.37 0.172 -.0084863 .0476067

yr1981 |

D1. | .0205486 .0226508 0.91 0.364 -.0238461 .0649432

yr1982 |

D1. | .0432438 .0296175 1.46 0.144 -.0148054 .1012929

yr1983 |

D1. | .0742359 .0370875 2.00 0.045 .0015457 .1469261

yr1984 |

D1. | .0918581 .0444807 2.07 0.039 .0046775 .1790387

_cons | -.0148382 .0056797 -2.61 0.009 -.0259702 -.0037062

------

Sargan test of over-identifying restrictions:

chi2(27) = 83.97 Prob > chi2 = 0.0000

Arellano-Bond test that average autocovariance in residuals of order 1 is 0:

H0: no autocorrelation z = -2.79 Pr > z = 0.0052

Arellano-Bond test that average autocovariance in residuals of order 2 is 0:

H0: no autocorrelation z = -0.72 Pr > z = 0.4745

Model 3.2.B2 xtabond n w k ys yr1980-yr1984, lags(1) robust

Arellano-Bond dynamic panel-data estimation Number of obs = 751

Group variable (i): id Number of groups = 140

Wald chi2(9) = 433.33

Time variable (t): year Obs per group: min = 5

avg = 5.364286

max = 7

One-step results

------

| Robust

D.n | Coef. Std. Err. z P>|z| [95% Conf. Interval]

------+------

n |

LD. | .3566042 .1371188 2.60 0.009 .0878562 .6253522

w |

D1. | -.5114253 .1701517 -3.01 0.003 -.8449164 -.1779342

k |

D1. | .3086461 .0534522 5.77 0.000 .2038817 .4134105

ys |

D1. | .5032803 .1513647 3.32 0.001 .2066109 .7999496

yr1980 |

D1. | .0195602 .013986 1.40 0.162 -.0078518 .0469722

yr1981 |

D1. | .0205486 .0303305 0.68 0.498 -.038898 .0799952

yr1982 |

D1. | .0432438 .0395268 1.09 0.274 -.0342273 .1207148

yr1983 |

D1. | .0742359 .0459919 1.61 0.107 -.0159065 .1643784

yr1984 |

D1. | .0918581 .0573505 1.60 0.109 -.0205468 .204263

_cons | -.0148382 .0061046 -2.43 0.015 -.0268031 -.0028734

------

Arellano-Bond test that average autocovariance in residuals of order 1 is 0:

H0: no autocorrelation z = -2.22 Pr > z = 0.0263

Arellano-Bond test that average autocovariance in residuals of order 2 is 0:

H0: no autocorrelation z = -0.61 Pr > z = 0.5443

Model 3.2.-B3 xtabond n w k ys yr1980-yr1984, lags(1) twostep

Arellano-Bond dynamic panel-data estimation Number of obs = 751

Group variable (i): id Number of groups = 140

Wald chi2(9) = 618.30

Time variable (t): year Obs per group: min = 5

avg = 5.364286

max = 7

Two-step results

------

D.n | Coef. Std. Err. z P>|z| [95% Conf. Interval]

------+------

n |

LD. | .2651432 .0559895 4.74 0.000 .1554058 .3748807

w |

D1. | -.4103142 .0430384 -9.53 0.000 -.494668 -.3259604

k |

D1. | .2563969 .0351334 7.30 0.000 .1875368 .3252571

ys |

D1. | .5436233 .0916815 5.93 0.000 .3639307 .7233158

yr1980 |

D1. | .0203073 .0099064 2.05 0.040 .0008911 .0397236

yr1981 |

D1. | .003441 .0192103 0.18 0.858 -.0342105 .0410925

yr1982 |

D1. | .0051906 .0264286 0.20 0.844 -.0466085 .0569898

yr1983 |

D1. | .0205109 .0318032 0.64 0.519 -.0418222 .082844

yr1984 |

D1. | .0190986 .0360729 0.53 0.596 -.0516029 .0898001

_cons | -.0119679 .004482 -2.67 0.008 -.0207523 -.0031834

------

Warning: Arellano and Bond recommend using one-step results for

inference on coefficients

Sargan test of over-identifying restrictions:

chi2(27) = 32.22 Prob > chi2 = 0.2242

Arellano-Bond test that average autocovariance in residuals of order 1 is 0:

H0: no autocorrelation z = -1.24 Pr > z = 0.2165

Arellano-Bond test that average autocovariance in residuals of order 2 is 0:

H0: no autocorrelation z = -0.32 Pr > z = 0.7473

Model 3.2.-B4 xtabond n w k ys yr1980-yr1984, lags(2)

Arellano-Bond dynamic panel-data estimation Number of obs = 611

Group variable (i): id Number of groups = 140

Wald chi2(10) = 429.41

Time variable (t): year Obs per group: min = 4

avg = 4.364286

max = 6

One-step results

------

D.n | Coef. Std. Err. z P>|z| [95% Conf. Interval]

------+------

n |

LD. | .3809966 .0913604 4.17 0.000 .2019335 .5600597

L2D. | -.0314535 .0372183 -0.85 0.398 -.1044 .041493

w |

D1. | -.5582806 .0595507 -9.37 0.000 -.674998 -.4415633

k |

D1. | .3604439 .0334723 10.77 0.000 .2948394 .4260483

ys |

D1. | .506865 .1101652 4.60 0.000 .2909451 .7227848

yr1980 |

D1. | .0058845 .0194738 0.30 0.763 -.0322833 .0440524

yr1981 |

D1. | -.0010127 .032771 -0.03 0.975 -.0652427 .0632172

yr1982 |

D1. | .0158584 .0452833 0.35 0.726 -.0728953 .1046121

yr1983 |

D1. | .0370505 .0581743 0.64 0.524 -.0769689 .15107

yr1984 |

D1. | .0427605 .071393 0.60 0.549 -.0971672 .1826881

_cons | .0009947 .0124716 0.08 0.936 -.0234491 .0254385

------

Sargan test of over-identifying restrictions:

chi2(25) = 74.97 Prob > chi2 = 0.0000

Arellano-Bond test that average autocovariance in residuals of order 1 is 0:

H0: no autocorrelation z = -3.13 Pr > z = 0.0017

Arellano-Bond test that average autocovariance in residuals of order 2 is 0:

H0: no autocorrelation z = -0.39 Pr > z = 0.6973

Model 3.2.-B5 xtabond n l(0/1).w l(0/1).k l(0/1).ys yr1980-yr1984, lags(1)

Arellano-Bond dynamic panel-data estimation Number of obs = 751

Group variable (i): id Number of groups = 140

Wald chi2(12) = 813.95

Time variable (t): year Obs per group: min = 5

avg = 5.364286

max = 7

One-step results

------

D.n | Coef. Std. Err. z P>|z| [95% Conf. Interval]

------+------

n |

LD. | .5630709 .1094238 5.15 0.000 .3486042 .7775376

w |

D1. | -.5534161 .0561889 -9.85 0.000 -.6635442 -.4432879

LD. | .3098602 .0751487 4.12 0.000 .1625714 .4571489

k |

D1. | .3063797 .0297639 10.29 0.000 .2480435 .3647159

LD. | -.0522857 .0503968 -1.04 0.300 -.1510616 .0464901

ys |

D1. | .6228309 .1195031 5.21 0.000 .3886092 .8570526

LD. | -.597117 .143242 -4.17 0.000 -.8778661 -.3163679

yr1980 |

D1. | .0044292 .0156642 0.28 0.777 -.026272 .0351304

yr1981 |

D1. | -.0377724 .0244028 -1.55 0.122 -.085601 .0100561

yr1982 |

D1. | -.0710787 .032883 -2.16 0.031 -.1355282 -.0066292

yr1983 |

D1. | -.0812401 .0425751 -1.91 0.056 -.1646857 .0022055

yr1984 |

D1. | -.080054 .0513176 -1.56 0.119 -.1806347 .0205267

_cons | .0050601 .0078047 0.65 0.517 -.0102369 .020357

------

Sargan test of over-identifying restrictions:

chi2(27) = 77.00 Prob > chi2 = 0.0000

Arellano-Bond test that average autocovariance in residuals of order 1 is 0:

H0: no autocorrelation z = -3.39 Pr > z = 0.0007

Arellano-Bond test that average autocovariance in residuals of order 2 is 0:

H0: no autocorrelation z = -1.23 Pr > z = 0.2203

.

Model 3.2.-B6 xtabond n l(0/1).w l(0/1).k l(0/1).ys yr1980-yr1984, lags(1) robust

Arellano-Bond dynamic panel-data estimation Number of obs = 751

Group variable (i): id Number of groups = 140

Wald chi2(12) = 624.34

Time variable (t): year Obs per group: min = 5

avg = 5.364286

max = 7

One-step results

------

| Robust

D.n | Coef. Std. Err. z P>|z| [95% Conf. Interval]

------+------

n |

LD. | .5630709 .1197828 4.70 0.000 .3283009 .7978409

w |

D1. | -.5534161 .1772358 -3.12 0.002 -.9007918 -.2060403

LD. | .3098602 .1202698 2.58 0.010 .0741357 .5455846

k |

D1. | .3063797 .0547069 5.60 0.000 .1991562 .4136032

LD. | -.0522857 .0679217 -0.77 0.441 -.1854099 .0808384

ys |

D1. | .6228309 .1694083 3.68 0.000 .2907968 .954865

LD. | -.597117 .1872489 -3.19 0.001 -.9641182 -.2301159

yr1980 |

D1. | .0044292 .0144535 0.31 0.759 -.023899 .0327575

yr1981 |

D1. | -.0377724 .0260604 -1.45 0.147 -.0888499 .013305

yr1982 |

D1. | -.0710787 .0357855 -1.99 0.047 -.141217 -.0009405

yr1983 |

D1. | -.0812401 .0470945 -1.73 0.085 -.1735437 .0110635

yr1984 |

D1. | -.080054 .0564116 -1.42 0.156 -.1906187 .0305108

_cons | .0050601 .0094224 0.54 0.591 -.0134075 .0235276

------

Arellano-Bond test that average autocovariance in residuals of order 1 is 0:

H0: no autocorrelation z = -3.23 Pr > z = 0.0012

Arellano-Bond test that average autocovariance in residuals of order 2 is 0:

H0: no autocorrelation z = -1.25 Pr > z = 0.2099

Model 3.2.-B7 xtabond n l(0/1).w l(0/1).k l(0/1).ys yr1980-yr1984, lags(1) twostep

Arellano-Bond dynamic panel-data estimation Number of obs = 751

Group variable (i): id Number of groups = 140

Wald chi2(12) = 1060.48

Time variable (t): year Obs per group: min = 5

avg = 5.364286

max = 7

Two-step results

------

D.n | Coef. Std. Err. z P>|z| [95% Conf. Interval]

------+------

n |

LD. | .4670255 .0745046 6.27 0.000 .3209992 .6130518

w |

D1. | -.4870518 .0470883 -10.34 0.000 -.5793432 -.3947605

LD. | .2396211 .0611649 3.92 0.000 .1197401 .3595021

k |

D1. | .2229986 .0403502 5.53 0.000 .1439136 .3020836

LD. | .0524942 .0500748 1.05 0.294 -.0456507 .1506391

ys |

D1. | .600489 .1031834 5.82 0.000 .3982532 .8027247

LD. | -.4223655 .1189128 -3.55 0.000 -.6554304 -.1893007

yr1980 |

D1. | .0028122 .0107633 0.26 0.794 -.0182834 .0239079

yr1981 |

D1. | -.0430203 .0201913 -2.13 0.033 -.0825945 -.0034462

yr1982 |

D1. | -.0651432 .0270214 -2.41 0.016 -.1181041 -.0121823

yr1983 |

D1. | -.0671289 .0310614 -2.16 0.031 -.1280081 -.0062497

yr1984 |

D1. | -.0738373 .0354618 -2.08 0.037 -.1433411 -.0043335

_cons | .0007818 .0053805 0.15 0.884 -.0097637 .0113274

------

Warning: Arellano and Bond recommend using one-step results for

inference on coefficients

Sargan test of over-identifying restrictions:

chi2(27) = 37.14 Prob > chi2 = 0.0925

Arellano-Bond test that average autocovariance in residuals of order 1 is 0:

H0: no autocorrelation z = -2.55 Pr > z = 0.0108

Arellano-Bond test that average autocovariance in residuals of order 2 is 0:

H0: no autocorrelation z = -1.02 Pr > z = 0.3076

Model 3.2.-B8 xtabond n l(0/2).w l(0/2).k l(0/2).ys yr1980-yr1984, lags(2)

Arellano-Bond dynamic panel-data estimation Number of obs = 611

Group variable (i): id Number of groups = 140

Wald chi2(16) = 549.88

Time variable (t): year Obs per group: min = 4

avg = 4.364286

max = 6

One-step results

------

D.n | Coef. Std. Err. z P>|z| [95% Conf. Interval]

------+------

n |

LD. | .7590458 .1534595 4.95 0.000 .4582706 1.059821

L2D. | -.1182499 .0491858 -2.40 0.016 -.2146523 -.0218474

w |

D1. | -.6264705 .0683354 -9.17 0.000 -.7604053 -.4925357

LD. | .4450418 .1093473 4.07 0.000 .2307251 .6593584

L2D. | -.1459958 .0759505 -1.92 0.055 -.294856 .0028644

k |

D1. | .3552865 .0379609 9.36 0.000 .2808846 .4296884

LD. | -.0810551 .0601376 -1.35 0.178 -.1989226 .0368124

L2D. | -.0184798 .0422759 -0.44 0.662 -.101339 .0643794

ys |

D1. | .6353047 .1386783 4.58 0.000 .3635001 .9071092

LD. | -.8009587 .1938173 -4.13 0.000 -1.180834 -.4210837

L2D. | .2040576 .1563103 1.31 0.192 -.102305 .5104202

yr1980 |

D1. | .0108957 .0221529 0.49 0.623 -.0325231 .0543146

yr1981 |

D1. | -.0227497 .0370657 -0.61 0.539 -.0953972 .0498978

yr1982 |

D1. | -.0338001 .0509725 -0.66 0.507 -.1337044 .0661041

yr1983 |

D1. | -.0194175 .0673381 -0.29 0.773 -.1513978 .1125628

yr1984 |

D1. | -.0011615 .084187 -0.01 0.989 -.166165 .1638419

_cons | -.0004955 .0150878 -0.03 0.974 -.0300669 .029076

------

Sargan test of over-identifying restrictions:

chi2(25) = 59.25 Prob > chi2 = 0.0001

Arellano-Bond test that average autocovariance in residuals of order 1 is 0:

H0: no autocorrelation z = -4.26 Pr > z = 0.0000

Arellano-Bond test that average autocovariance in residuals of order 2 is 0:

H0: no autocorrelation z = -0.11 Pr > z = 0.9096

Model 3.2.-B9 xtabond n l(0/2).w l(0/2).k l(0/2).ys yr1980-yr1984, lags(2) robust

Arellano-Bond dynamic panel-data estimation Number of obs = 611

Group variable (i): id Number of groups = 140

Wald chi2(16) = 647.69

Time variable (t): year Obs per group: min = 4

avg = 4.364286

max = 6

One-step results

------

| Robust

D.n | Coef. Std. Err. z P>|z| [95% Conf. Interval]

------+------

n |

LD. | .7590458 .1341298 5.66 0.000 .4961561 1.021935

L2D. | -.1182499 .0457147 -2.59 0.010 -.2078491 -.0286506

w |

D1. | -.6264705 .1906682 -3.29 0.001 -1.000173 -.2527678

LD. | .4450418 .1795079 2.48 0.013 .0932128 .7968707

L2D. | -.1459958 .0873153 -1.67 0.095 -.3171306 .0251389

k |

D1. | .3552865 .0601116 5.91 0.000 .2374698 .4731031

LD. | -.0810551 .0744821 -1.09 0.276 -.2270374 .0649272

L2D. | -.0184798 .032538 -0.57 0.570 -.0822531 .0452934

ys |

D1. | .6353047 .1773702 3.58 0.000 .2876654 .9829439

LD. | -.8009587 .262686 -3.05 0.002 -1.315814 -.2861035

L2D. | .2040576 .1642452 1.24 0.214 -.117857 .5259722

yr1980 |

D1. | .0108957 .0175574 0.62 0.535 -.0235161 .0453075

yr1981 |

D1. | -.0227497 .0312617 -0.73 0.467 -.0840216 .0385222

yr1982 |

D1. | -.0338001 .041608 -0.81 0.417 -.1153503 .0477501

yr1983 |

D1. | -.0194175 .0558735 -0.35 0.728 -.1289274 .0900925

yr1984 |

D1. | -.0011615 .073711 -0.02 0.987 -.1456325 .1433095

_cons | -.0004955 .0126406 -0.04 0.969 -.0252707 .0242797

------

Arellano-Bond test that average autocovariance in residuals of order 1 is 0:

H0: no autocorrelation z = -3.95 Pr > z = 0.0001

Arellano-Bond test that average autocovariance in residuals of order 2 is 0:

H0: no autocorrelation z = -0.10 Pr > z = 0.9206

Model 3.2.-B10 xtabond n l(0/2).w l(0/2).k l(0/2).ys yr1980-yr1984, lags(2) twostep

Arellano-Bond dynamic panel-data estimation Number of obs = 611

Group variable (i): id Number of groups = 140

Wald chi2(16) = 1059.42

Time variable (t): year Obs per group: min = 4

avg = 4.364286

max = 6

Two-step results

------

D.n | Coef. Std. Err. z P>|z| [95% Conf. Interval]

------+------

n |

LD. | .7219585 .0872442 8.28 0.000 .550963 .892954

L2D. | -.0968684 .0277448 -3.49 0.000 -.1512472 -.0424896

w |

D1. | -.5542483 .0568186 -9.75 0.000 -.6656107 -.4428859

LD. | .4028884 .0935179 4.31 0.000 .2195968 .5861801

L2D. | -.1332653 .053101 -2.51 0.012 -.2373413 -.0291892

k |

D1. | .2791604 .0455152 6.13 0.000 .1899522 .3683685

LD. | -.0196619 .0552974 -0.36 0.722 -.1280427 .0887189

L2D. | -.0470922 .0263051 -1.79 0.073 -.0986492 .0044649

ys |

D1. | .5826981 .1170406 4.98 0.000 .3533028 .8120935

LD. | -.6633483 .143779 -4.61 0.000 -.94515 -.3815466

L2D. | .2129541 .119221 1.79 0.074 -.0207148 .4466229

yr1980 |

D1. | .004616 .0132031 0.35 0.727 -.0212616 .0304936

yr1981 |

D1. | -.0434272 .0247054 -1.76 0.079 -.091849 .0049945

yr1982 |

D1. | -.0524147 .0323349 -1.62 0.105 -.1157899 .0109604

yr1983 |

D1. | -.0320466 .0419099 -0.76 0.444 -.1141886 .0500954

yr1984 |

D1. | -.0347231 .0531522 -0.65 0.514 -.1388994 .0694532

_cons | .0022866 .0090722 0.25 0.801 -.0154946 .0200678

------

Warning: Arellano and Bond recommend using one-step results for

inference on coefficients

Sargan test of over-identifying restrictions:

chi2(25) = 31.68 Prob > chi2 = 0.1673

Arellano-Bond test that average autocovariance in residuals of order 1 is 0:

H0: no autocorrelation z = -3.48 Pr > z = 0.0005

Arellano-Bond test that average autocovariance in residuals of order 2 is 0:

H0: no autocorrelation z = -0.25 Pr > z = 0.8048

Looking at the residuals=predicted_y - y, it does not seem like we can assume standard assumptions of normality and constant variance.

. correlate, _coef

| LD. L2D. D. LD. L2D. D. LD. L2D.

| n n w w w k k k

------+------

n |

LD. | 1.0000

L2D. | -0.4966 1.0000

w |

D1. | 0.0185 -0.2676 1.0000

LD. | 0.4882 -0.0882 -0.7629 1.0000

L2D. | -0.1588 -0.0786 0.6073 -0.5442 1.0000

k |

D1. | -0.0416 -0.0503 -0.0785 0.1200 0.0040 1.0000

LD. | -0.5603 0.2779 0.0322 -0.3463 0.0884 -0.6392 1.0000

L2D. | -0.2603 -0.2597 0.0791 -0.2135 -0.0362 0.2655 -0.1719 1.0000

ys |

D1. | 0.0966 -0.0180 -0.5238 0.5203 -0.1663 0.0837 -0.1507 -0.0715

LD. | -0.3557 -0.0026 0.7719 -0.8535 0.5145 -0.0846 0.2683 0.2198

L2D. | 0.1330 0.0324 -0.5510 0.5372 -0.5240 0.0052 -0.1195 -0.2073

yr1980 |

D1. | -0.0800 0.2176 -0.3169 0.1438 -0.3610 0.0955 -0.0237 0.0962

yr1981 |

D1. | -0.0505 0.1136 -0.3418 0.2121 -0.4099 0.1523 -0.0725 0.1953

yr1982 |

D1. | -0.0930 0.1934 -0.2635 0.0739 -0.4432 0.0434 0.0826 0.0797

yr1983 |

D1. | -0.0177 0.1196 -0.1840 -0.0008 -0.4918 -0.0791 0.0910 0.0821

yr1984 |

D1. | -0.0732 0.1882 -0.3305 0.0932 -0.5857 -0.0790 0.1422 0.0951

_cons | 0.1179 -0.2728 0.2353 -0.0558 0.5535 0.2314 -0.2538 -0.0425

| D. LD. L2D. D. D. D. D. D.

| ys ys ys yr1980 yr1981 yr1982 yr1983 yr1984

------+------

ys |

D1. | 1.0000

LD. | -0.6909 1.0000

L2D. | 0.1428 -0.6316 1.0000

yr1980 |

D1. | 0.3594 -0.2632 0.1602 1.0000

yr1981 |

D1. | 0.3682 -0.1718 0.0343 0.8633 1.0000

yr1982 |

D1. | 0.0774 -0.0219 0.1796 0.8038 0.8670 1.0000

yr1983 |

D1. | -0.1406 0.0445 0.2605 0.6932 0.7232 0.9192 1.0000

yr1984 |

D1. | -0.1235 -0.0656 0.3313 0.6807 0.6959 0.8886 0.9577 1.0000

_cons | 0.2124 0.0330 -0.2920 -0.6376 -0.6224 -0.8055 -0.8860 -0.9200

|

| _cons

------+------

_cons | 1.0000.

Side 1 av 21

[1] Lillard, Lee A. and Robert J. Willis (1978). “Dynamic Aspects of Earning Mobility”. Econometrica, Vol. 46, No. 5, pp. 985-1012.

[2] Note: