A recent paper looks at a new dataset that includes stock information before the well-known CRSP data set. We now have information back before the 20th century. This was a mammoth project, but it provides us the ability to check factor analysis for a period of 150 plus years. See "The cross-section of stock returns before CRSP".
Here are some of the highlights from the data:
The beta of the CAPM does not seem to work. There is no relationship between return and risk. This data just adds another nail in the failure of the CAPM model
There does not seem to be a small cap effect. This again seems to tell us that size is not a separate factor.
Value investing does seem to work. You will be rewarded for buying cheap stocks.
The momentum also seems to work. Buy high performers and sell losers. There is strong consistency with this factor.
The low-risk effect of being compensated for holding low risk or low beta stocks also seems to hold further back in time.
Macro risks do not seem explain what is going with the risk and return trade-off with stocks. Delegated management may be an explanation for these factors, but these results do not seem to fit this story for the pre-CRSP dataset. Market crashes suggest that momentum investors are being compensate for the risk of large market down moves. Again, it does not seem as though this data fits this narrative.
Machine learning was applied to this data set, and it finds that momentum, value, and low risk are the key factors extracted from the data. These strategies may be associated with behavioral finance explanations. For example, the endowment effect can explain the value factor. Herd behavior (FOMO) can explain the momentum effect. Low risk investing is associated with the aversion to loss. Investors want to hold low beta stocks. These theories seem to offer go explanations for these older data.
There is no evidence of significant out of sample decay, so value, momentum, and low risk continue to work over the long-term.
No comments:
Post a Comment