SD1H4H — Euro / US Dollar (EUR/USD) Multi-TF Structure Backtest Original Model

Corrected, ex-ante methodology. Every metric below is a TRAILING window (last 20 trading days / 480 1H bars / 120 4H bars) ending at each trade's own entry time, using only data that would actually have been known at that moment — no lookahead. Each trade is one data point (not each month), giving far more statistical power. Trades are split chronologically: the first 70% (train) is used only to pick a median threshold and a direction; the last 30% (test) is untouched until that frozen rule is checked against it. An earlier version of this page aggregated metrics over whole calendar months, which leaks future-within-the-month information into the metric — see the superseded section at the bottom for why that mattered.

Bottom line: none of the six metrics tested reach statistical significance (p < 0.05) against individual trade outcomes for either strategy — correlations are all near zero (|ρ| < 0.1). Where a train/test split shows the same direction in both halves ('held up'), that is expected to happen by chance a meaningful fraction of the time when the underlying effect is this weak and six metrics are tested per strategy — it should not be read as a discovered filter. As of this data, none of these price/structure regime measures give an actionable, statistically defensible rule for skipping trades ahead of time.

This section has a lookahead-bias flaw and is kept only for transparency — do not treat it as a trading rule. Each metric here was computed over the ENTIRE calendar month a trade closed in, including days after that trade. A month's full ATR% or trend-efficiency ratio isn't knowable until the month is over, so the strong-looking correlations below describe what characterized good/bad months in hindsight, not something that could have been acted on in real time. The corrected, trade-level, ex-ante version above supersedes this.