Six months of backtest-vs-live divergence logging on 2 EAs — exact numbers

@risk_firstjoined May 23, 2025Jul 31, 2026524 views4 replies

I ran two EAs live (small size) and re-backtested the identical period every month, logging the divergence. Six months of data:

EA-A (EURUSD H1 trend): backtest PF 1.61, live PF 1.34. Average slippage 0.3 pips/trade worse than modelled. Divergence explained almost entirely by execution costs.

EA-B (GBPJPY M15 mean reversion): backtest PF 1.48, live PF 0.91. Slippage was fine. The killer: 4 news-window trades that the backtest filled at prices that never existed live. Those 4 trades were the entire difference.

Conclusions I now size by: (1) a trend system on a major pair degrades gracefully, maybe 15-20% off backtest; (2) a mean-reversion system trading through news can diverge 100% — the backtest literally cannot see the fills you won't get; (3) monthly re-backtesting of the live period is the single most diagnostic habit I have adopted.

4 Replies

@macro_mikejoined May 8, 2025#1 · 5 days ago
+2
The monthly re-backtest habit is underrated. One diagnostic I add: watch the shape of the divergence. A gap that grows monotonically is a cost problem — slippage, spread, commission. A gap that arrives in jumps is an event/fill problem — news windows, outages. Different causes, completely different fixes.
@pips_and_pythonjoined Jun 20, 2025#2 · 5 days ago
+1
Replicated the news-fill issue on my own mean reverter. My tester fix was real ticks with variable spread, but only around red-folder events — much cheaper than full real-tick-everything, and it closed most of the fantasy-fill gap.
I also cross-checked the same period in QuantMogul's walk-forward report. Useful for the parameter-stability view, but it can't model news-window fills either — nothing I have found can. That gap has to come from your own live log.
@kanya_tradesjoined Nov 8, 2025#3 · 5 days ago
0
The 4 news trades being the entire difference matches my small account exactly. I now block new entries 15 minutes before red-folder events and my live/backtest gap shrank by about half. ขอบคุณที่แชร์ตัวเลขจริงค่ะ
@risk_firstjoined May 23, 2025#4 · 5 days ago
0
Monotonic vs jump is exactly the right framing — mine were both, one per EA, and I never separated them until now. Adding divergence-shape tracking to the monthly log. This thread is why I post data here.