Dear All,
I hope you doing well.
I am currently working on so called Three-Pass Regression Filter and I would tremendously appreciate your advises.
I have a data set of 6 portfolios of companies with their return and book-to-market ratios from 1927 to 2010. The idea is that the same factor drives both, b-t-m ratios and aggregated returns. The setup is divided into 3 parts. In the first step we run time series regression for each i(portfolio) of its book to market ratio on aggregated returns. From this, we extract coefficients and use these coefficients in the second step where we run cross sectional regression for each t of book to market ratios on their coefficients estimated in the first step. In the third step we run predictive regression of realized returns on lagged factors estimated in the second step, factors in other words coefficients from the second set of regressions.
I know it kinda sounds questionable, however this was broadly cited in finance papers. The main issue is that authors do not include actual approach how it was done using software program. I have found replication of this filter, however it was done using excel which probably isn't the best software to use for this matter.
So, my approach on this is to start this filter by running rolling regression, however I get values regressed by years, example:
so it runs regression between years. Is there a some kind of option with rolling regression to run regression for every observation and not by years? Although authors do not say anything regarding this but I m not exactly sure if it's exact way to do it. Maybe you could give some advises regarding this regression approach in general or had similar experience and have suggestions? Everything will be appreciated!
Best Regards,
Marijus
I hope you doing well.
I am currently working on so called Three-Pass Regression Filter and I would tremendously appreciate your advises.
I have a data set of 6 portfolios of companies with their return and book-to-market ratios from 1927 to 2010. The idea is that the same factor drives both, b-t-m ratios and aggregated returns. The setup is divided into 3 parts. In the first step we run time series regression for each i(portfolio) of its book to market ratio on aggregated returns. From this, we extract coefficients and use these coefficients in the second step where we run cross sectional regression for each t of book to market ratios on their coefficients estimated in the first step. In the third step we run predictive regression of realized returns on lagged factors estimated in the second step, factors in other words coefficients from the second set of regressions.
I know it kinda sounds questionable, however this was broadly cited in finance papers. The main issue is that authors do not include actual approach how it was done using software program. I have found replication of this filter, however it was done using excel which probably isn't the best software to use for this matter.
So, my approach on this is to start this filter by running rolling regression, however I get values regressed by years, example:
so it runs regression between years. Is there a some kind of option with rolling regression to run regression for every observation and not by years? Although authors do not say anything regarding this but I m not exactly sure if it's exact way to do it. Maybe you could give some advises regarding this regression approach in general or had similar experience and have suggestions? Everything will be appreciated!
Best Regards,
Marijus
Comment