Ran 5 years of backtests, 99% modelling quality, profit factor 2.1, realistic spread, slippage modelled. Live result: -40% in 11 weeks.
What did I miss? Genuinely asking — I want a checklist of everything that can make a good backtest turn into a bad live account.
99% modelling quality means your data was good, not that your strategy was. The usual suspects:
My money is on 1. It almost always is.
My rule after being burned the same way: never trust a backtest that isn't followed by a walk-forward and at least 3 months of forward/demo testing.
Tooling helps enforce the discipline. I run WFO plus Monte Carlo in QuantMogul (the desktop app from the wobr folks) — the drawdown distribution report is uncomfortably honest. And before I size real money I cross-check forward results on WorldBotRank, which ranks bots on live forward tests rather than backtests. One data source among several, but a useful reality check.
Also check the trade duration and P&L distribution. If most of your backtest profit comes from 3 lucky trades out of 400, you didn't have an edge — you had a lottery ticket that happened to win in the sample.
Parameters came from an optimization cloud, so yes, probably overfit. Running walk-forward now.
The Monte Carlo point stings: my 95th percentile drawdown is 55%, and I sized for 20%. Expensive lesson but at least it was only tuition, not ruin.
We've all paid this tuition. The honest framing that finally fixed my process: a backtest is a hypothesis, not a result. The forward test is the experiment. Most retail flow skips the experiment entirely.