3 ms·
Yup. What you really want to know is "how problematic is the noise in this data"? One way to answer it is to create your own FF sorts [1] and regress your Yah
by rmrfstar 6y ago
Yup.
What you really want to know is "how problematic is the noise in this data"?
One way to answer it is to create your own FF sorts [1] and regress your Yahoo-derived factor returns against Ken French's (which come from CRSP/Compustat). If the intercepts are small/statistically insignificant and the R2's reasonable, then you're all set.
The big problem you will hit is survivorship bias in Yahoo!. My own research suggests that the quality is perfectly acceptable, provided you back-fill an unbiased universe (e.g. Russell 3K) from another source.
I'm actually surprised you didn't outperform the market with the survivorship bias. Back test must have covered a very small time period.
[1] https://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html https://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data...
- pmiller2 6y agoWhat exactly do you mean by "back-fill an unbiased universe (e.g. Russell 3K)"?
- rmrfstar 6y agoYahoo! doesn't have data for delisted or bankrupt companies, a "survivorship bias." You need a set of companies that lacks this bias, and the set of Russell 3000 index constituents is a good set to use. Some of the older Russel 3000 companies will be missing from Yahoo, because they are delisted or bankrupt. You need to find a secondary source for them, or "back-fill" your securities master.