LEARN · EN · easyquanttrading.com
Honest backtesting, not pretty curves
Most backtests lie politely. They leak future data, ignore costs, or show you only the lucky runs. This guide explains what honest backtesting actually requires — look-ahead-free bars, transparent costs, out-of-sample validation — and how to audit any backtest (including ours) for honesty.
- Look-ahead-free engine: bars are computed in order, never with future data
- Cost profiles: spread, swap and slippage in the model, conservative by default
- Red/green evidence — untested stages are never painted green
- Public methodology: every assumption written down, including our own limits
What 'honest backtesting' actually means
Honest backtesting is a set of constraints, not a feature badge. Three constraints matter above all: the engine must never use information that was not available at the time (no look-ahead), the cost model must reflect real trading (spread, swap, slippage), and the results you are shown must survive testing on data the optimizer never saw.
A platform is honest if it makes these constraints visible and lets you fail at them. A platform is dishonest if it hides them behind a pretty curve — whether out of marketing or incompetence.
The silent killers of backtest honesty
- Look-ahead bias: using today's close to decide today's entry, or re-balancing on data that would not have existed yet. The most common way to fake a curve.
- Missing costs: no spread, no swap, no slippage. A 2-pip edge dies on a 1-pip spread — the model must include it in every trade.
- Data snooping: trying 500 indicator combinations and showing only the best. Selection bias is invisible in a single chart.
- Survivorship: a strategy library that only shows winners. The losers you never see are the honest part of the story.
- Curve cosmetics: filling missing metrics with zeros instead of leaving them null, or drawing equity lines through data gaps.
How to audit any backtest in 5 minutes
- Ask where the data came from and whether the bars are raw or cleaned — and whether the start date was chosen after the fact.
- Check that every entry used only data available at that bar: no future closes, no next-day rebalancing.
- Look for a cost model: spread, swap and slippage per trade. If there is no cost line, the backtest is a fantasy.
- Ask for out-of-sample results: a walk-forward chain or a held-out period. In-sample curves prove nothing.
- Count the trials: how many parameter sets and strategies were tried to produce this one? Nobody counted means selection bias was free.
What honest platforms do differently
The difference is structural, not cosmetic. An honest platform fixes the testing protocol so users cannot quietly skip stages: data is processed in strict bar order, costs are read from the same configuration the strategy trades against, and validation results (walk-forward, Monte Carlo, DSR/PBO) are attached to every strategy card. Missing data stays null; untested stages stay grey.
EasyQuant is built this way: a single authoritative look-ahead-free engine, conservative cost defaults, a fixed eight-stage pipeline, and a public methodology page that lists our own limitations. We do not promise returns — we promise inspectability.
FAQ
- What is a look-ahead-free backtest?
- A backtest that computes each bar strictly in chronological order, using only information available at that moment. No future closes, no next-day decisions — the architecture of the engine enforces it.
- Is a high in-sample Sharpe deployable?
- No. Deployable requires the robustness pipeline (signal health, defense, walk-forward, Monte Carlo) to pass in fixed order. In-sample beauty is only a research filter.
- Why do backtests fail in live trading?
- Usually three reasons: costs were ignored or underestimated, the strategy was curve-fit to history, or execution assumptions (fills, slippage) did not match reality. Honest backtesting addresses all three before money is at risk.
- Do you guarantee profits?
- Never. We sell research discipline and glass-box evidence, not return promises. Historical results do not guarantee future performance.
- How do I know EasyQuant's backtests are honest?
- The methodology page documents every assumption including our known limitations, the engine is bar-order strict, costs are conservative by default, and strategy cards show validation results — including failures.
More guides
Not investment advice. Historical results do not guarantee future performance. EasyQuant is a research factory — you execute on accounts you control.