In pairs trading, you have to choose a window size for doing the regression. In fact, this is probably the only parameter that's needed. But the performance of pairs trading is very sensitive to this parameter, and in fact, the PNL can vary from very negative to very positive, depending on the value of this parameter. The sensitivity curve can be very peaked, with good Sharpe ratios on only few window size values, and with mediocre (around 0) Sharpe ratios on a lot window size values and with negative Sharpe ratios on also a lot window size values. Any thoughts? Thanks a lot!
This is something I find myself pondering more and more. You'll find that there is a wide range of window sizes used in academic papers, so there's little help in finding a definitive answer from published work. Unfortunately, I haven't found the time to do a full study on my own. I've had a lot of success with a 130 day length, but I'm also a relatively discretionary pairs trader so if something looks wrong I'll do several other regressions to further confirm the entry. What have you experimented with thus far?
just thought of this - what if you measure the variance for multiple look back periods while looking for an abnormally large deviation in that value from one time frame to the next longer frame. for example - if variance typically increases linearly with a time frame increase but suddenly it breaks that relationship by a certain standard deviation than this becomes your cut off point for the look back period. I'm testing this idea today and this weekend. I'll let you know what I find.
I take that back - I am looking for the variance to be consistent and as soon as it breaks from that consistency than I am cutting the time frame. So far the results have been pretty close to my typical look back period.