Carhart 4 factor model: Fixed-effects panel
Posted: Sun Jul 13, 2014 2:07 am
Hi everyone,
I am in the process of analysing mutual fund returns using the Carhart 4-factor model, which is an extension of the Fama & French 3-factor model.
I have an unbalanced panel with 186 cross sections and 282 periods (monthly data). Thus, 52,452 observations.
For my model I ran a cross-section fixed-effects LS regression on my dependent variable gf_eu. My explanatory variables are mkt_rf_eu, wml_eu, hml_eu and smb_eu. The results can be found in the uploaded EViews file. The equation results are saved under green_eu.
In this context, I encounter the following problem. My R squared is very low (between 0.08-0.2, depending on the adjustments I made) while I have for several variables very high t-statistics and thus 3 being significant at the 1% level (including the coefficient). This seems very suspicious to me, not least because prior studies all found R squared between 0.7 to 0.9.
I performed the Hausman test to check if I use the right model specifications, performed unit root tests, cointegration etc. I cannot find my mistake. However there must be one as I got the data from the renowned Kenneth French data library and Datastream and it thus should be correct as it is used for studies all over the world. Thus, my R squared should at least be similar to previous studies.
I am kind of stuck on this issue, but I need to solve it over the week to come and would appreciate any hint or help.
I attached you both a simplified version of my excel sheet used to generate the final dataset and a copy of my EViews model.
Thank you very much.
Kind regards,
Tom
P.S. I am using EViews 8
I am in the process of analysing mutual fund returns using the Carhart 4-factor model, which is an extension of the Fama & French 3-factor model.
I have an unbalanced panel with 186 cross sections and 282 periods (monthly data). Thus, 52,452 observations.
For my model I ran a cross-section fixed-effects LS regression on my dependent variable gf_eu. My explanatory variables are mkt_rf_eu, wml_eu, hml_eu and smb_eu. The results can be found in the uploaded EViews file. The equation results are saved under green_eu.
In this context, I encounter the following problem. My R squared is very low (between 0.08-0.2, depending on the adjustments I made) while I have for several variables very high t-statistics and thus 3 being significant at the 1% level (including the coefficient). This seems very suspicious to me, not least because prior studies all found R squared between 0.7 to 0.9.
I performed the Hausman test to check if I use the right model specifications, performed unit root tests, cointegration etc. I cannot find my mistake. However there must be one as I got the data from the renowned Kenneth French data library and Datastream and it thus should be correct as it is used for studies all over the world. Thus, my R squared should at least be similar to previous studies.
I am kind of stuck on this issue, but I need to solve it over the week to come and would appreciate any hint or help.
I attached you both a simplified version of my excel sheet used to generate the final dataset and a copy of my EViews model.
Thank you very much.
Kind regards,
Tom
P.S. I am using EViews 8