Measuring the efficient market: −$0.18 a trade over 1,162 trades

When we ran the gold breakout's post-mortem (E-05) on EURUSD for comparison, the fold tags told a story: seven of eight EURUSD folds were range regimes. We had been running a momentum strategy on the most mean-reverting major in FX — a textbook mismatch (it lost −$3,092 there; the pair's tiny cost wall halved the damage relative to gold, but negative is negative). The mirror-image hypothesis wrote itself: fade the extremes instead of chasing them. For once, an idea motivated by market structure rather than mined from the data.

By this point the program had killed five strategies, and we knew the failure mode of researchers as well as the failure modes of models: there is always one more variant, one more parameter, one more asset. So before running the test, we pre-committed, in writing: this is the final principled test. If it does not show a robust, cross-regime, out-of-sample edge, the trading-research thread closes permanently and the infrastructure pivots to publishing analytics. No renegotiation with the results afterward.

3.5 years of EURUSD 5-minute data — 254,000 bars, median bar range 2.8 pips — with the economics re-parameterized for the pair (100,000-unit contract, ~0.1-pip raw spread, $7 round-turn commission). The strategy: fade z-score extremes (|z| ≥ 2) with a 1.5 ATR target against a 2 ATR stop. And one structural addition — a calm-regime gate: our validated volatility classifier, inverted. Mean reversion works in quiet markets and gets steamrolled by breakouts, so the model that predicts expansions was used to veto trading when one was coming.

Every qualitative prediction of the mean-reversion hypothesis came true. Win rates ran 54–65% across folds — the fade was genuinely mean-reverting. The calm gate was genuinely predictive — it cut losses by 93% against the ungated version (−$2,868). Two models and a market theory, all working as designed.

Net result: −$205 over 1,162 trades. Eighteen cents of loss per trade. Three of eight folds positive. Against a pre-committed bar of robust cross-regime profit, that is not a pass. It is the most precise zero we ever measured.

This number deserves a careful reading, because it is not a failure of the model — it is a successful measurement of the market. The mean-reversion effect exists; the 54–65% win rates prove it. But an effect that every quantitative desk on earth can see, executed at retail latency and retail costs, returns precisely its residual: zero, minus a rounding error. The institutions running the same arithmetic faster and cheaper have consumed the effect down to the boundary where collecting it costs exactly what it pays. Eighteen cents per trade is what the textbook definition of an efficient market looks like when you stop reading about it and go measure it with your own money at stake.

Verdict: statistical breakeven — killed by the efficient market itself, and by a stopping rule we refused to renegotiate. The pre-commitment executed as written: the trading-research thread closed, permanently. What had been a trading project became what you are reading — the pipeline, the validated volatility model, and the cost-honest harness, pivoted from hunting an edge to publishing measurements. The full story of that decision is our methodology; the two experiments that were already in flight when the rule fired are E-07 and E-08.