as a risk premia harvestoor, we all like our backtests.

(yes if it's backtesteable , probably ain't edge bruv.)

however, most of the backtests you see around are wrong, clearly very wrong.

I've spent the last year+ designing mine, to a point where I think it pretty much comes very close to reality.

but the real test of this is once it's live. how close to reality does your realized performance post fees, costs, impact, slippage, funding, etc, match the backtest engine you have?

well, I do this by tracking the realized performance vs the simulated performance on a day to day.

example, this new model I launched earlier this month, is matching the backtesting engine quite well. (ignore last data point, today's simulation is only updated at EOD.)

if it wasn't, well, probably the backtestoor is worthless.