Effect of 2008 financial crisis on pairs trading

Discussion in 'Trading' started by ar1zona, Oct 9, 2011.

  1. ar1zona



    I know there are many pair traders out there and you all have your own tweaks so performance varies, but in general, do you find the financial crisis had a strong influence on your performance?

    I have been working on a pair trading model, the hypothetical equity curve from June 2000 to October 2011 is shown below.


    The two year period between the green lines are in sample which I optimized the parameters, and then tested for out sample on both directions (going back and forward). The orange line marks the approximate date of Lehman Brothers bankruptcy. I know going back too much will subject to many bias like survival bias, but I just want to get an idea of the historical equity curve.

    This post is not asking to debug my algorithm, since there's not enough details provided to do so, but to answer the original question. Below is the same equity curve but only for the last four years.


    As you can see, from the forward test results, going out of sample did not have a significant effect on the slope of equity curve as opposed to the financial crisis. Is this what everything else is observing? And do you find the time just before the crisis had a positive impact on your trading as well? Any input? Thanks.
  2. It is because you are going the reverse way. Your in sample should be before the crisis and your out of sample the crisis period. It is highly likely that if a system was optimized to produce good results through a terrible period, it will produce as good or better results through better periods.

    So I suggest you optimize your system from 2000 to 2008 and then test during the crisis period. What you have done is called data snooping.
  3. ar1zona


    I will re-run it with an in sample before the crisis and see what happens.

    Could you elaborate on the data snooping part? Suppose the time is at the end of the in sample period and I dunno anything about the future, then I can not make the assumption that my in sample period was more volatility or whatever against any future period, so I can not intentionally not perform a backtest for 2007-2009 right? Please advice.