1. Charles University FSV UK STAKAN III Institute of Economic Studies Faculty of Social Sciences Institute of Economic Studies Faculty of Social Sciences Jan Ámos Víšek Econometrics Tuesday, 14.00 – 15.20 Charles University Eleventh Lecture

2. ● Inspiration for Generalized Method of Moments estimation ● Examples of GMM-reformulation of problems within the regression framework ● Recalling Cragg’s improvement of the estimates of regression coefficients under heteroscedasticity ● Reformulating Cragg’s approach as GMM-estimation ● Recalling Whites results of estimating covariance matrix of the OLS-estimator of regression coefficients Schedule of today talk Prior to all: Generalized Least Squares

3. The Generalized Least Squares Let us assume that - regular, i.e. homoscedasticity is broken. - regular and symmetric and put multiplying the basic model from the left by..

4. For we have, i.e. we have reached homoscedasticity. Then. Recalling that, i. e. Generalized Least Squares The Generalized Least Squares continued

5. An example What is Generalized Method of Moments (GMM) ? Wooldridge, J. M. (2001): Applications of generalized method of moments estimation. J. of Economic Perspective, 15, no 4, 87 - 100. Inspiration taken from Transactions on the household account

6. continued An example Two orthogonality conditions What is Generalized Method of Moments (GMM) ? symmetric, positive definite which can’t be simultaneously fulfilled !

7. The empirical counterparts Various covariance structure, various distributional framework A collection of orthogonality conditions The first attempt for a general framework Generalized Method of Moments A population of r.v.’s symmetric, positive definite

8. The empirical counterparts = normal equations A collection of orthogonality conditions and the classical least squares Another example Generalized Method of Moments Let’s consider linear regression model

9. continued Another example Generalized Method of Moments The ordinary least squares symmetric, positive definite

10. Hansen, L. P. (1982): Large sample properties of generalized method of moments estimators. Econometrica, 50, no 4, 1029 - 1054. Pioneering paper Commonly accepted framework Generalized Method of Moments We would like to estimate consistently the model it seems that we have to believe that the disturbances are orthogonal to the model ! Observed data An assumed “casual” model

11. continued It is evident that it can’t be generally true we look for some instruments, being close to model, however orthogonal to disturbances ! Disturbances Instruments Orthogonality conditions Kronecker product This equality defines function Commonly accepted framework Generalized Method of Moments

12. Commonly accepted framework Generalized Method of Moments continued symmetric, positive definite Empirical counterpart of the orthogonality conditions An estimate - symmetric, positive definite

13. ● Heteroscedastic disturbances ● Instrumental variables Another examples of the problems in classical regression analysis with the straightforward GMM-reformulation ● Collinearity Discussed in details Generalized Method of Moments

14. ● Data in question represent the aggregates over some regions. ● Explanatory variables are measured with random errors. ● Models with randomly varying coefficients. ● ARCH models. ● Probit, logit or counting models. ● Limited and censored response variable. Can we meet with the heteroscedasticity frequently ? ● Error component (random effects) model. Heteroscedasticity is assumed by the character (or type) of model.

15. ● Expenditure of households. ● Demands for electricity. ● Wages of employed married women. ● Technical analysis of capital markets. Can we meet with heteroscedasticity frequently ? continued Heteroscedasticity was not assumed but “empirically found” for given data. ● Models of export, import and FDI ( for industries ).

16. Is it worthwhile to take seriously heteroscedasticity ? Let’s look e. g. for a model of the export from given country

17. Ignoring heteroscedasticity, we arrive at: B means backshift

18. Other characteristics of model White het. test = 244.066 [.000]

19. Significance of explanatory variables when White’s estimator of covariance matrix of regression coefficients was employed.

20. Reducing model according to effective significance

21. White het. test = 116.659 [.000] Other characteristics of model

22. ● - independently (non-identically) distributed r.v.’s ● ● - absolutely continuous d. f.’s,,, White, H. (1980): A heteroskedasticity-consistent covariance matrix estimator and a direct test for heteroscedasticity. Econometrica, 48, 817 - 838. Recalling White’s ideas - assumptions

23. ● ● for large T, for large T.,,, continued Recalling White’s ideas - assumptions ● No assumption on the type of distribution already in the sense of Generalized Method of Moments. Remark.

24. Recalling White’s results

25. continued

26. Recalling White’s results continued

27. Recalling White’s results continued

28. We should use has generally elements T ( T + 1 ) 2 Cragg, J. G. (1983): More efficient estimation in the presence of heteroscedasticity of unknown form. Econometrica, 51, 751 - 763. Recalling Cragg’s results

29. continued We put up with has T unknown elements, namely Even if rows are independent

30. Recalling Cragg’s results continued An artificial system of r.v.’s

31. Recalling Cragg’s results continued Should be positive definite. Nevertheless, is still unknown An improvement if

32. Recalling Cragg’s results continued Asymptotic variance Estimated asymptotic variance

33. Example – simulations Recalling Cragg’s results Model Heteroscedasticity given by 1000 repetitions T=25 Columns of matrix P

34. continued Example Recalling Cragg’s results Example – simulations Asymptotic Estimated Actual = simulated LS1.0001.0110.7641.0000.9800.701 0.4090.4780.4000.5900.7420.442 0.2780.3370.2860.4710.6290.309 0.2540.3310.2660.4450.6260.270 0.2470.3460.2470.4370.6610.230 j=1,2,3,4 j=1 j=1,2 j=1,2,3 Asymptotic Actual Estimated

35. Example Reformulating Cragg’s results as GMM estimation An artificial system of r.v.’s I.e., the error term The orthogonal conditions Recalling Cragg’s approach and their covariance matrix

36. continued Example Reformulating Cragg’s results as GMM estimation Putting Normal equations

37. What is to be learnt from this lecture for exam ? Generalized method of moments (inspiration, main idea – up to the slide 7). Cragg’s recommendation in the case of heteroscedsticity. All what you need is on http://samba.fsv.cuni.cz/~visek/