I am only Breakeven last 3 months doing that. Wanna ask others how they are doing recently ? Any suggestions for tweaks ?
1) Read Lütkepohl 's "New Introduction to Multiple Time Series Analysis," especially the chapter on VECM. Know that chapter backwards and forwards. Actually read anything you can find by that author. Also you will find "Regression Diagnostics" but Besley, Kuhn, and Welch extremely useful. 2) Model cointegrating relationships in the frequency domain as well as the time domain. You will want to sample your data at 10 times (or more) the rate of your likely decision frequency. 3) Model the relationships as time-varying cointegration -- start with two states (cointegrated and not) and go from there. Similar in spirit to threshold cointegration, but more sophisticated. Only fit the model parameters when the data series are in a cointegrated state -- nearly integrated processes (via unit root tests) imply that cointegration is a necessary condition both for consistent estimation of the parameters of the trading model and compatibility between the model and the data. 4) Model cointegration in variables other than delta-1... Use Box-Tiao, Bewley... ... I have a lot of other suggestions, but these will do for a start.
Info for beginners (Pairs Trading, Correlation, Cointegration, z-scores, Dickey-Fuller test, etc.): https://blog.quantinsti.com/pairs-trading-basics/
%% Assume end of 3rd month JAN is going about the same LOL. Maybe, with 2.5 days left. But SPY + the other 4 had a good 3 month run. NOT really pairs, with me . [ Commissions + bid\ask spread make not much difference the way i trade them] But looking @ a huge number of time frames, on one of those , 3 months is the worst; other 4 maybe OK,3 months. 1st week + 1st 5 days JAN never mean anywhere near as much forecast to me as month of JAN, but with 2.5 days left maybe close enough......... Even with some seasonals good JAN help; TrAndy that good study you posted a had a 10 years section of JAN forecast=sub-average performance. So 10 years of sub -average of 50 years = 40 years worthwhile. And 80 or 90 years of 100 years = good enough for me[hindsight]; + my metals business is not correlated or connected much to any of that.