I am currently running some Fama French regressions for a number of specified portfolios, consisting of US stocks (for a university level course). The regressions are conducted in R, using the lm funtion.
The models look like this:
3F <- lm(ExcessReturn ~ MKT + SMB + HML, data = df) 4F <- lm(ExcessReturn ~ MKT + SMB + HML + MOM, data = df) 5F <- lm(ExcessReturn ~ MKT + SMB + HML + RMW + CMA, data = df)
The excess return is the return of a long-short portfolio, i.e., the return of Portfolio A minus return of portfolio B, minus the risk free rate. The regressions are carried out exactly like this, without anything happening in between. I am using HC4 robust standard errors (& p-values) later and override the ones given from these regressions in my output table, which decreases significance (but not noticably for the alphas).
My Problem with interpreting the results is that I do not get any significant Alphas using the 3F and 4F Models, but I do get significant Alphas for the 5F Model, which kind of confuses me, as I thought more explaning factors should decrease Alpha (especially comparing 3F and 5F).
I'd greatly appreciate any help with either troubleshooting (i.e. whether this result is technically achievable using correct methods) or, if so, an interpretation of these results. Also, do you see any potential to improve the regression?
A sincere thanks to anyone taking the time reading and thinking about this! If any more information is needed, please ask.
Edit (1): I am running time-series regressions according to the Fama French 1993 and 2015 Models as well as Carhart (1997).
(Edit (2): I am currently thinking about how I calculated the long-short portfolio's return. I think (!) I may have made an mistake subtracting the Excess Return. First indication to back that up: in Kempf & Osthoff (2007 I think), the long-short portfolio's factors are always (!) equal to the Long factor loading minus the short factor loading, which is not the case for me. It does happen though, when defining the long-short portfolio return as return(long porfolio) minus return(short portfolio). Which is the same as (return(long) - rf) - (return(short) - rf). I do think my mistake was in subtracting the risk free rate from (either of these) calculations. Is there anyone who could confirm this?)
Edit (3): While the calculation of the long-short portfolio-return was indeed problematic, my base problem still persists. For less portfolios, luckily, but I still get a significant Alpha for the 5F, but non-significant Alphas for 3F & 4F for two portfolios.