Sie sind auf Seite 1von 2

Final Exam, Econometrics Ph.D.

, January 2013
Directions 1. Provide an answer to each of the following problems. Include in your answer both Stata output from the estimation of the models and the commands you submitted to Stata. Do not submit output containing results that are not relevant to your answers. 2. Your answers to all three problems should be combined and submitted as a single Microsoft Word or Adobe Acrobat file. This file should be sent electronically as an email attachment to miroslav.verbic@ef.uni-lj.si by midnight, Thursday, 31 January 2013. 3. You may consult your lecture notes, textbooks, and Stata manuals as you work on this examination. However, you should not seek help from professors or other students. The exam is individual; one student per exam paper. 4. Please, read each question carefully. Be sure that you do everything as asked. Exam Questions 1. Stata data file qmacro.dta contains quarterly data on macroeconomic time series for the period 1950Q1 2000Q4. a) Define the quarterly time dimension appropriately, i.e. create a time variable. Then estimate the following classical linear regression model: m1t = 1 + 2 realgdpt + 3tbilratet + 4 cpit + ut . Interpret all the regression coefficients. Do the estimates seem reasonable? b) Perform a logarithmic transformation of the variables and estimate the model above with the transformed variables:
log ( m1t ) = 1 + 2 log ( realgdpt ) + 3 log ( tbilrate )t + 4 log ( cpit ) + ut .

Again, interpret all the regression coefficients. c) Test formally for normality of the residuals in the initial model (exercise a) and in the transformed model (exercise b). Write down the procedures and the results. Compare the results and explain your findings. d) Test formally in the transformed model (exercise b) that 4 = 2 2 . Write down the procedures and the results. Perform this test both manually (on paper based on the estimated coefficients and variance-covariance matrix) and automatically with the use of appropriate testing commands in Stata. e) Based on a general specification of a regression model with an intercept and two numeric explanatory variables, show theoretically that it holds t et x jt = 0 , where ei are the residuals of the model. In addition, show that it holds there is an intercept in the model. 1

t t

= 0 if and only if

f) Create a time series plot of tbilrate. Does it appear to be stationary? Explain. Check for stationarity using an appropriate test; write down the null and alternative hypothesis, the test statistic and the results. How many lags should you use? g) If necessary, make the series tbilrate stationary by using first differences. Does it appear to be stationary now; conditionally and unconditionally? Test for stationarity again. 2. Using the Stata data file ngrades.dta, estimate the effect that gpa, tuce and psi have on the probability that an individual passes an end-of-term examination in economics (grade = 1 if passes, 0 otherwise). To estimate the effect of gpa, tuce and psi on grade, we have to use a probit model, because the dependent variable is a binary variable. a) Are the explanatory variables in the model jointly significant at the 5% level? If not, what is their level of joint significance? b) What is the interpretation of the coefficient on tuce? Calculate and interpret its marginal effect as well. c) Produce a cross-tabulation of actual with predicted values for grade. How many of the observations in which grade = 1 are correctly predicted by your model? How many of the observations in which grade = 0 are correctly predicted? d) You just predicted that grade = 1 whenever the predicted P(grade = 1) is greater than 0.5; otherwise grade = 0. Starting from your predicted values, improve this procedure (find a new threshold) to obtain a more accurate measure of the predictive power of this model. Explain your procedure and produce a new cross-tabulation. How many of the observations in which grade = 0 are correctly predicted by your new method? e) Based on the general specification of a probit model and its likelihood function, show theoretically that the value of the log likelihood of a probit model is always negative. 3. A friend of yours is estimating a regression model with panel data. The data are contained in the Stata data file firm_data.dta. The friend is estimating the effect of x and z on y. There are 20 firms in the data identified by the variable firmid. Each firm is observed five times. Your friend wants to estimate a random effects model, but he is worried that this choice may be inappropriate. He has heard that a Hausman test can be used to test the random effects model if he estimates both fixed and random effects models. However, when he runs the following commands in Stata, he notices that apparently the variable z has been dropped from the model by Stata.
iis firmid xtreg y x z, fe

Seeing this, your friend thinks that he should drop variable z from the model altogether and base his Hausman test on a regression of y on x alone. a) Explain to your friend why this is not a good idea. b) Show him how to estimate his model and how to conduct a Hausman test properly; provide commands and output. Explain the obtained Hausman statistic value. c) Answer the question, should your friend use a random or fixed effects model, explaining why.

Das könnte Ihnen auch gefallen