Tuesday, April 28, 2009

Quantile Regression

Caution: If you are not an economist (or a social scientist of quantitative type), please skip this post.

Read Deaton (1997) (pages 80-83) to review the merit of running quantile regression.

1. Uncover the heteroskedasticity in the error term.

2. Figure out the shape of the conditional distribution (such as income conditional on age).

3. Obtain a more efficient estimator than OLS when the error term does not follow the normal distribution.

There is a concise survey of quantile regression in economics: Roger Koenker and Kevin F. Hallock (2001).

For an instrumental variables estimation of quantile regression, see Chernozhukov and Hansen (2005). For a method to estimate the the impact of a regressor on the unconditional distribution of the dependent variable, see Firpo, Fortin, and Lemieux (2007).

No comments: