The current Fixed Effects & Clustered Standard Errors chapter introduces both tools cleanly, but its running example sits in corporate finance (the investment regression), so the transition from the preceding Fama-MacBeth chapter is weaker than it could be. I'd like to add an asset pricing framing up front. Concretely: re-estimate the cross-section of returns (beta, size, book-to-market on next-month excess returns) as a single pooled panel using the same dataset as the Fama-MacBeth chapter, then show two things. First, adding time fixed effects reproduces the per-month intercepts that Fama-MacBeth's first stage estimates separately. Second, clustering by date yields inference that is close to the Newey-West standard errors we already used. This makes explicit that Fama-MacBeth is a special case of the panel toolbox, namely equal-weighted averaging of monthly slopes with cross-sectionally robust standard errors. The existing investment regression then becomes the natural contrast (firm rather than time effects), which motivates Petersen's (2008) distinction between a dominant time effect in asset pricing and a dominant firm effect in corporate finance, and explains why firm fixed effects would be inappropriate in the return regression since they remove the very cross-sectional variation we are trying to price. The net effect is a smoother bridge from the previous chapter and a clearer answer to the practical question of which dimension to fix and cluster on.
To be implemented after migration to Polars.
The current Fixed Effects & Clustered Standard Errors chapter introduces both tools cleanly, but its running example sits in corporate finance (the investment regression), so the transition from the preceding Fama-MacBeth chapter is weaker than it could be. I'd like to add an asset pricing framing up front. Concretely: re-estimate the cross-section of returns (beta, size, book-to-market on next-month excess returns) as a single pooled panel using the same dataset as the Fama-MacBeth chapter, then show two things. First, adding time fixed effects reproduces the per-month intercepts that Fama-MacBeth's first stage estimates separately. Second, clustering by date yields inference that is close to the Newey-West standard errors we already used. This makes explicit that Fama-MacBeth is a special case of the panel toolbox, namely equal-weighted averaging of monthly slopes with cross-sectionally robust standard errors. The existing investment regression then becomes the natural contrast (firm rather than time effects), which motivates Petersen's (2008) distinction between a dominant time effect in asset pricing and a dominant firm effect in corporate finance, and explains why firm fixed effects would be inappropriate in the return regression since they remove the very cross-sectional variation we are trying to price. The net effect is a smoother bridge from the previous chapter and a clearer answer to the practical question of which dimension to fix and cluster on.
To be implemented after migration to Polars.