IIRC optimising the start time suggested that UTC (i.e. no offset) start time was the most robust, so I will leave it like that on the Zorro 1.32 version.
Incidentally I backtested the original unoptimised script and in 2015 so far it has performed really well. Considering this would be more 'out of sample' data it adds evidence for the hypothesis that is a good system for decent returns whilst keeping capital deployed for regular but short time periods.
BackTest ONS Portfolio portfolio
Simulated account AssetsFix.dta Bar period 24 hours (avg 2028 min) Test period 02.01.2015-02.12.2015 (238 bars) Lookback period 40 bars (8 weeks) Monte Carlo cycles 200 Assumed slippage 10.0 sec Capital invested 50000$
Gross win/loss 40797$ / -12491$ (+2436p) Average profit 30955$/year, 2580$/month, 119$/day Max drawdown -2461$ 9% (MAE -2461$ 9%) Total down time 59% (TAE 8%) Max down time 5 weeks from Apr 2015 Max open margin 7676$ Max open risk 7301$ Trade volume 9002053$ (9844311$/year) Transaction costs -724$ spr, 0.00$ slp, 0.00$ rol, -325$ com Capital required 12133$
Number of trades 81 (89/year, 2/week, 1/day) Percent winning 62% Max win/loss 6338$ / -2011$ Avg trade profit 349$ 30.1p (+70.2p / -34.7p) Avg trade slippage 0.00$ 0.0p (+0.0p / -0.0p) Avg trade bars 1 (+1 / -1) Max trade bars 1 (24 hours) Time in market 34% Max open trades 5 Max loss streak 4 (uncorrelated 5)
Annual growth rate 63% Profit factor 3.27 (PRR 2.38) Sharpe ratio 2.43 Kelly criterion 9.55 R2 coefficient 0.851 Ulcer index 4.1% Prediction error 64%
