1. Collect data for a country’s GDP on a quarterly frequency. Do not use too few observations (at least 30 observations).
2. Run a regression using some statistics or econometrics software (such as SPSS or EViews) to find an estimate for the GDP growth rate over the whole time period without adjusting for seasonal effects. Indicate what that estimate is.
3. Plot the residuals from step 2 against time and comment on whether there is a seasonal pattern.
4. Redo step 2 with the inclusion of dummy variables as explanatory variables to adjust for seasonal effects.
5. Comment on whether the regression in step 4 indicates the growth rate of GDP is significantly different from zero.
6. Use an F-test to determine whether you can reject the hypothesis that all the coefficients for the dummy variables in step 4’s regression are equal to zero. Note that this test does not use the automatic F statistic from step 4’s regression, since that F statistic is used to test whether all slope coefficients in the regression in step 4 are equal to zero.
7. Find two more variables for the same country on a quarterly basis that are always positive. You should collect only variables you think could affect GDP. Explain how you think these variables should affect GDP.
8. Create three new variables which are simply the percent change in GDP and the percent change of each of the two variables from step 7. Note: you can for example create the percent-change-in-GDP variable (call it PCGDP) out of your GDP variable (called GDP). In EViews you would use the command GENR PCGDP=(GDP-GDP(-1))/GDP(-1), where GDP(-1) represents the previous period’s GDP. In other statistical software, you may need to calculate it more manually, perhaps creating another variable showing the previous quarter’s GDP and then creating the PCGDP variable based on the GDP and that new variable.
9. Run a regression of the percent change in GDP against the other percent-change variables from step 8, including an intercept. Interpret the estimates, the t-statistics and the F-statistic with that regression.
10. Plot the residuals from step 9 against time and comment on whether the plot indicates any autocorrelation or heteroscedasticity. State test hypotheses for normality (e.g. Jarque-Bera), heteroscedasticity (e.g. White’s test), autocorrelation (e.g. Breusch-Godfrey’s test) and RESET and run these significance tests. Explain and draw conclusions. If you have not found that the residuals are free from most of the above problems, there is a high risk that the estimated model is not adequate. In some cases it may be a good idea to reformulate (respecify) and reestimate this model if you experience these problems. However, if just a couple of the above tests are significant – it is ok (and you do not need to respecify it). Also if you have tried two models, and both of these give significant results in these diagnostic tests of the residuals (for e.g. heteroscedasticity, autocorrelation, RESET etc), then it is ok and we accept these problems. But try at least two models before you give up (or of course only one model if the residuals are white noise already at the first attempt… e.g. no autocorrelation, no heteroscedasticity etc. in the residuals).
11. Explain what the Durbin-Watson statistic from your step 9 regression indicates.

1. Collect data for a country’s GDP on a quarterly frequency. Do no...