Sin stocks

[tomz23]

Patrick Schwarz writes:

„In your code on GitHub (line 20 in Signalhorn.do) and also in the online appendix (page 25) you write that you assume as in the original paper that once a stock is classified as Sin, the entire past and the entire future is also considered Sin.

I read the Hong Paper as saying that there is no forward classification (forward filling) and only a look back (backward filling) for periods before 1985 due to lack of data.

The implementation with forward and backward filling contains thus nevertheless unnecessarily much noise, which seems unnecessary to me at least in relation to the forward filling. Also noteworthy is that Hong 2009 use the Fama French industry classification which is primarily based on historical COMPUSTAT SIC codes and it seems to me that your code is using historical CRSP SIC codes, which often times are different from the COMPUSTAT SIC codes.

[tomz23]

Update here: https://github.com/OpenSourceAP/CrossSection/commit/53485730065edc6a6fa5caa573a3d261c15cae16

• This update is a complete rewrite of sinAlgo.
• It follows the OP (relatively) closely and fixes the forward-backward-fill. It also makes sure to not classify tobacco stocks as sin stocks before 1965.
• I continued to use CRSP SIC codes because CS codes are not available before 1950 and the tests in table 4B have data back until 1926.
• sinAlgo replicates the stats in table 4A and B quite well, with similar monthly returns (21bps and 24bps) and slightly lower t-stats. Our data are a bit different because our segment data start in 1976.