Aug

3

I was having a discussion of survivorship bias the other day which seems worth sharing in case others could expand:

An example of survivorship bias is when you look at an index composition today,
at this point in time. The stocks you see today are "survivors" and if
you then go and do some analysis on those stocks as, say, representing
the small-cap or mid-cap universe, you're ignoring the stocks that
crashed, i.e., you are showing a bias towards survivors.

An example would be: You want to study smaller-cap stocks over time,
so you get a list of the current components of the Russell 2000 and then
look up their individual price histories for the last ten years to
study volatility or return characteristics - you would be getting a
heavy dose of survivorship bias. To avoid the bias, you would have to
get a list of the Russell 2000 components from 2005 and work forward
from there.

I think you can generalize survivorship bias, for a study over some
particular time period, as using a selector or filter from the end of
the time period. Indices are selectors/filters over time, so you have to
use the version that existed at the beginning of the time period. In
that case the study can become a test of the selector/filter as a
predictor.

Steve Ellison explains his solution:

I have maintained a database of what the 2000 or so stocks included in Value Line were since 2005, including the dates each stock was added or dropped. My effort has flagged a bit recently because I seldom trade individual stocks.

Stefan Martinek writes:

I think indexes are the best to avoid this bias if you trade indexes directly. If you research individual stocks you have to deal with this bias and use some good DBs not indexes.


Comments

Name

Email

Website

Speak your mind

Archives

Resources & Links

Search