Econometrics: weeks 7-10 (March 26th to April 22th)

In weeks 5-6 we:

• Explained Lesson 3.2 (Inference)
• Continued the work with Many least squares calculations done (almost) by hand and Weight vs. age for a sample of children
• In the March 21th seminar, we concluded the work with the CEO compensation dataset

The second and last work session with the CEO compensation data started by

• Deleting some outliers. We found that the model fit improved.
• Fitting a model for the log of salaries. Natural logs of non-negative economic data are typically closer to the normal model.

When we tried to discern which model (for salaries or log-salaries) was better, we discovered that:

You should not compare apples and oranges

...meaning that R-squared values and information criteria are useful to compare alternative models only if the endogenous variable in all of them is the same. This does not happen in this case.

When comparing models for variables with different transformations one may use other criteria, such as chosing:

• the model that is closer to fulfilling the regression model hypotheses or, alternatively,
• the one that provides the best out-of-sample forecasts

Using the first criterion, the log-transformed model provides residuals are closer to normality. This is specially important when you want to do test hypotheses because standard testing results depend critically on normality.

The goals for weeks 7-10 are:

• Explaining Lessons 4.1 (Discrete and semi-continuous regressors. Polynomial terms) and 4.2 (Collinearity).
• Concluding the revision of the practical materials Many least squares calculations done (almost) by hand
• In the April 18th seminar, we will revise and make further progress with the homework defined below

Activities:

Download again Lesson 3.2 (Inference), as I corrected some errata and added a short section about out-of-sample forecasting

Graded personal homework

Send an e-mail to this address (last day for delivery: April, 17th). This message should:

• Describe any doubts 
• Describe broadly the results obtained when: (a) solving this exercise sheet, and (b) performing a forecasting exercise with the Beauty dataset that is described here

This e-mail will be assessed as part of the "active participation" item (10% of final grade). To facilitate processing the e-mails it is important to comply with the deadline and follow the delivery procedure (please use this email address). I would also help if all the text is included in the body of the message body, avoiding attachments whenever possible.