Tuesday, May 20, 2008

Survivorship Disturber II

In an earlier post, we discussed the requirement to determine what happened to de-listed companies when running analyses on historical stocks. If you don't do that, you will bias your results because companies that failed will not be reflected in your results.

Unfortunately, there is an inherent bias even within databases like Compustat, so even if you think you've accounted for stocks that have disappeared, you can't do anything about the stocks that never appeared in the first place.

You see, Compustat doesn't contain every stock on ever exchange. It only adds in stocks that have been around for a while. In this way, the database is biased towards companies that have not lost so much money that they disappear!

So how do you account for this? Fama and French argue in "The CAPM is Wanted: Dead or Alive" that survivorship bias in the Compustat database is statistically insignificant in their researching proving low book value to market value stocks outperform the market. Nevertheless, many a proponent of Efficient Markets Hypothesis (EMH) has argued that Fama and French results suffer from survivorship bias.

For our purposes, we have decided to use two methods to reduce the effects of this inherent bias:

1) Acquire results from specific exchanges where Compustat coverage is high (and therefore selection into Compustat is not as judgement based).
2) Acquire results from specific indices. Yes, indices only choose stocks that have performed well, but because they have a fixed number of stocks on them, there are de-listed losers as well which we can track and ensure their results are accounted for.

Enjoy the long weekend!

No comments: