I have found that the figures on backtests are incongruent when comparing different backtest quantities of units.
For example a backtest on 10,000 units may have no losses but a 100,000 backtest would show losses in the segment of time normally covered by a 10,000 backtest and not at the beginning of the 10,000 backtest time period which as overlap you could go some way to understanding. My logical mind says that even though I can appreciate a backtest is a rough approximation on each individual candle there should be a demarcation between each 24hr day this doesn’t seem to be whats happening when testing over a larger number of units. Maybe this has been fixed in v11, I cross my fingers and hope.
Multiple OOS backtests on a smaller amount of units may give a better representation and accuracy, just my own opinion before everyone shouts me down telling me I should be backtesting over a decade. 🙂
With 80% of JP Morgans transactions being automated trades we can’t assume that market conditions are the same 10 years ago as they are now, particularly with the amount of market manipulation I see happening on markets like the DJI. Is it crazy to think that maybe a backtest no longer than the last 2 years may give a more accurate understanding of the market conditions we are now working within ?
I’m opening this up for discussion please feel free to comment and add your views and my apologies if I’ve hijacked this thread but I do think it’s relevant.
.
.
