Seasonal adjustment and time fixed effects

Question

Apologies if this seems mundane.

I have monthly data on 65 regions and I want to seasonally adjust it using seasonal dummies in Stata. I have two questions:

1. If I plan on analyzing the data using xtreg, fe and controlling for time fixed effects, how should this be done for monthly data? Is including i.monthly_date sufficient to control for seasonality in xtreg, fe?
2. If I intend to seasonally adjust the series before running regressions, should I use the predicted residuals or fitted values as the adjusted series? After running reg x i.monthly_date, should I use predict adj_series, residuals or predict adj_series, xb to get the seasonally adjusted values? Additionally, is the regression reg x i.monthly_date enough to adjust for seasonality?

I will appreciate any guidance.

Answer 1

If I plan on analyzing the data using xtreg, fe and controlling for time fixed effects, how should this be done for monthly data? Is including i.monthly_date sufficient to control for seasonality in xtreg, fe

In principle such monthly fixed effects would absorb the seasonal effects. However, there are other things to consider. These monthly fixed effects will cost you a lot of degrees of freedom. Depending on your dataset size and other characteristics, and other research details there could be better ways to control for seasonality. It’s difficult to give more specific advice without seeing the data.

If I intend to seasonally adjust the series before running regressions, should I use the predicted residuals or fitted values as the adjusted series? After running reg x i.monthly_date, should I use predict adj_series, residuals or predict adj_series, xb to get the seasonally adjusted values? Additionally, is the regression reg x i.monthly_date enough to adjust for seasonality?

If you are seasonally adjusting the series using some model of course you should use the fitted values from that model. I am not sure what your rationale for using residuals would be.

In principle reg y x i.monthly_date is sufficient to control for seasonal effects.