Testing Sampling Bias in SPXEPS Using Data to 1990

Testing Sampling Bias in SPXEPS Using Data to 1990 Header Image

Some portions this post contains premium content, which are marked, and can only be viewed by premium subscribers. To view the whole post sign up or review pricing.

When testing strategies we have to be careful from sampling bias.  So instead of using the start point of 1998 as our original data point for the trend following strategy SPXEPS, I will look at the available data from stockcharts.com dating back to 1990.  The results confirm most of the strategy already, but also highlight some new trades for the short side of SPX and buying gold during sell signals.

Approaching a strategy, in my opinion, you should be as unbiased as possible.  One trick is to cover up the ticker symbols and your conclusion should not change based on the emotion you might have in the symbol alone.  Emotion can get in the way of hard data.  One aspect of manipulating this is through sampling bias.  This can be done by selecting specific assets to test, or testing between certain strategy start and end dates to curve fit your hypothesis.

In the recent study of SPXEPS, we tested from 1998 till present to develop our indicator based on the EPS of the $SPX.  But the entire dataset from stockcharts goes back to 1990.  So I tested the data again going back to this start date.  The results do not change much but are highlighted in the following video:

1990 additional summary:

  • The strength of the long default position is increased as the study showed we rode the entire bull market of the 1990s.
  • Shorting once again produced an additional trade, which once again proved the difficulty of shorting, or reversing from long to short.
  • Going long gold also produced one trade which was a loser, but similar R values as in the 1998 study.


For premium subscribers I talk about the trades in more detail here, and what additional deductions we can make:

Premium Content

Download the spreadsheet here:

Premium Content

Final Thoughts

The inclusion of the 1990 data enhances the bias of our long thesis.  Instead of gaining only 29% the strategy gains 198% because it is able to ride the 1990s bull market where EPS also tripled over the same period.  When including this data it enhances and confirms the strength of the indicator, and confirms the viability of a potential strategy that should be forward tested.

Recent Posts

Sign up for the Newsletter

Get email alerts when a new blog is posted, or a public strategy trade is made