I am analysing ex-post a financial strategy based on European mutual funds (from 2000 to 2016). Briefly, the funds are distributed each month in a cluster (dynamic clustering method) and each month, I compare the monthly cluster to the reference cluster of the fund. In case of disconnection at least 3 periods in a row, I take a position: long if the fund underperforms the reference cluster and short if it outperforms (I compare the monthly return of the fund and the average return of the funds belonging to the reference cluster at the same period). Positions then close when monthly_cluster = cluster_ref. The idea is that the fund will return to its reference cluster afterwards. For example, a fund that underperforms would show positive future returns to return to its cluster. So I have each month a new portfolio with the short and long returns of the funds in disconnection, based on what is observed in the previous period.
In order to analyse the performance of the strategy, I regressed according to a 4 factor model:
But I am a bit surprised by the results: I was expecting a negative alpha (the strategy doesn't really seem to be profitable considering the returns) but I am especially surprised by the R-squared, which is extremely low...
Does anyone have any idea what could go wrong? Maybe add other variables? I'm a little confused.