How to Backtest

Understand, run and read EDGEwise backtests — including the Deep Backtest — in 10 minutes.

1 · What a backtest is (and why it matters)

A backtest replays the strategy over real historical prices, bar by bar, as if it had been trading live: when a signal fires, it "enters" on the next bar's open, places the stop and targets, and then watches the following bars to see what would have happened. It answers one question honestly: does this setup have an edge on this asset and timeframe, or not? A signal service without backtests is opinion; with them, it's evidence.

2 · How the engine tests — exactly like it scores live trades

Since the July 2026 methodology update, one rule governs the backtest, the live record and the quality gate alike — the TP1 standard:

Stop reached first → −1R (one full risk unit lost).
TP1 reached first → +1R (the recorded win).

TP2 and TP3 are published as extended targets for traders who choose to run part of a position further — they are never counted in any performance number. Inside any single bar the stop is always checked before the target (the conservative assumption), and realistic costs — spread, commission, slippage — are subtracted from every trade. One rule, everywhere, verifiable by anyone.

3 · Two views, one truth

🧠 Auto-learned best settings — the walk-forward optimiser tries many parameter combinations per timeframe and keeps the best, then validates it on data it never trained on (the OOS column). These are the settings that gate your live signals.
Run all timeframes — replays those same settings on every timeframe so you can compare the edge at a glance.

4 · How to read each number

Value	Meaning	What "good" looks like
Trades (count)	How many simulated trades the test produced.	20+ before trusting anything; under 20 is too small to judge.
Win rate %	Share of trades that reached TP1 before the stop.	Above 50% at 1:1 is a real edge; the higher the better.
Profit Factor (PF)	Total won ÷ total lost. The single most important number.	>1 profitable · 1.3+ real edge · 2+ excellent · below 1 losing setup.
Expectancy	Average R earned per trade, after costs.	Positive; 0.15R+ per trade is respectable.
OOS PF	Profit factor on the final ~30% of data the optimiser never saw — the anti-curve-fit check.	Should stay ≥1.0. High PF with weak OOS = settings memorised the past.
Grade	AB proven edge · C thin · D no edge · — too few trades yet.	Only A/B streams gate live signals (with "Suppress C/D" on).
Max drawdown	Worst peak-to-valley slide of the equity curve, in R and money.	Smaller is better; judge it against total return.

5 · The Deep Backtest — the institutional view

The Deep Backtest runs the same engine but reports like a fund would:

💰 Money model

Starts from your chosen capital and risks a fixed % of current equity per trade (compounding). The $ equity curve, drawdown and CAGR come from this — what a real account would have experienced, costs included.

📉 Benchmark

Compares the strategy against simply buying and holding the same asset over the same window. A strategy should beat — or strongly de-risk — buy & hold to justify itself.

🎲 Monte Carlo (400 resamples)

The order wins and losses arrive in is luck. Monte Carlo reshuffles your trade list 400 times and shows the range of outcomes: the 50% band (half of all reshuffles land here), the 90% band (almost all of them), the median path, and the gold dashed line = what actually happened. Read three things:

PF 90% range — if even the low end stays above 1.0, the edge survives bad luck.
Worst-case drawdown — size your risk % so you could stomach that number, not the average.
% chance a run ends negative — under ~15% is the comfort zone.

📒 Trade log + Export CSV

Every simulated trade: entry time, direction, in/out prices, R result, money P&L, running equity, bars held. Export the CSV and audit any trade yourself — nothing is hidden.

6 · Using it for decisions

Before trusting an asset/timeframe: check Grade + OOS PF. A/B with OOS ≥1 has earned attention.
Before sizing up: respect the Monte Carlo worst-case drawdown, not the pretty equity curve.
When two timeframes disagree: trade the one with the better PF and enough trades — never 6 lucky trades.
High PF but weak OOS: walk away — that's a memorised past, not an edge.

The golden rule: a backtest never predicts the future — it filters out setups that never worked in the past. EDGEwise runs this continuously in the background and blocks signals whose edge isn't proven (the "edge" quality gate). The live forward record then verifies the survivors in real time — that's the incubation → graduation pipeline.

Honesty note: backtests use conservative assumptions (stop before target inside a bar, costs subtracted), but real slippage in fast markets can exceed the model. That's why streams are graded on the live forward record before ever being published to subscribers.

EDGEwise — Trade with Confidence · Educational material, not financial advice · trading involves risk of loss.