← Blog

Why your options backtest says the strategy works

  1. Defect one: the win rate is an input
  2. Defect two: every day produces a trade
  3. Defect three: the price path cannot produce a loss
  4. The multiplier on all three: fill assumptions
  5. What a backtest is still good for
  6. How we scope ours

Most options backtests are not wrong by a small margin. They are wrong in a way that converts a losing configuration into a winning one and prints the result with two decimal places. Three defects account for the majority of it, and all three are easy to check for before you trust a single number.

Defect one: the win rate is an input

The most common structural flaw is a harness where the hit rate is a setting. Pass it a directional accuracy, and the report hands that figure back as though it were a finding. It is not a backtest at that point, it is a calculator with extra steps.

Direction accuracy has to come from somewhere, so having it as an input is not automatically wrong. What is wrong is reporting a win rate as a result when it was supplied as an assumption. Those are different numbers: getting the direction right and making money are separate events, because a correct call that never gives you a usable entry is not a win.

How to check. Search the settings and the source for a win rate, hit rate or accuracy parameter. If one exists and the output reports the same quantity, the report is a mirror.

Defect two: every day produces a trade

Real entry logic refuses most days. The setup never confirms, the spread is too wide, no contract passes the screen, liquidity never prints. A harness that quietly converts every signal into a filled position is describing a system nobody is running, and it inflates every aggregate it reports.

This one is insidious because the output looks richer. More trades, smoother curve, tighter confidence intervals — all of it manufactured by removing the filter that was the point of the strategy.

How to check. Compare signals generated against positions taken. If those two numbers are close, the harness is not modelling refusal. A well-filtered short-dated system declines a large share of sessions outright.

Defect three: the price path cannot produce a loss

A synthetic price generator calibrated to an average move will rarely manufacture the violent one that takes a position out. The distribution comes back with its left tail quietly missing, and the harness reports a strategy that almost never loses because the model almost never generated the day on which it would.

How to check. Count the exit reasons. If stop-outs are near zero, the question is not whether the strategy is good. It is whether the generator is capable of producing the trade that fails.

The multiplier on all three: fill assumptions

Each defect above is amplified by how the harness decides what price you got. Mid-price fills hand you the entire bid-ask spread on every round trip, which on a short-dated contract is a meaningful share of the premium.

The subtler version is asymmetry. A profit target is a resting limit order and fills exactly at its price. A stop is a market exit and fills at the bid, below where it triggered. Price both at the midpoint and the harness is honest about wins and optimistic about losses — a bias that points one way and never shows up as an obvious error.

What a backtest is still good for

Plenty, once you stop asking it the wrong question.

It is excellent at structure: whether the logic behaves consistently, whether it responds to the parameters it should respond to, whether the machinery deadlocks or leaves orders working. Those answers are robust even when the absolute returns are not.

It is poor at exposure, and exposure is what decides whether you survive a bad week. How much of the account a single loss costs is arithmetic. It needs no simulator, it cannot be flattered by a price model, and it is the number most traders never actually work out.

How we scope ours

The simulator shipped with Options Sniper drives the real engine — the same entry logic, the same limit ladder, the same position manager the live app uses — over thousands of synthetic opening ranges. It is deliberately scoped to mechanics rather than edge, and the tool states that in its own output rather than letting you assume otherwise.

The app is free, runs paper or live, and goes open source on 30 November 2026, so none of this has to be taken on trust.

Common questions

Why do options backtests overstate returns so badly?

Usually a combination of three things: a hit rate supplied as a setting and reported as a finding, a model where every signal becomes a filled position, and a price generator that rarely produces the violent move that stops a trade out. Fill assumptions then multiply whatever those three produce.

How can I tell if a backtest models losing trades properly?

Count the exit reasons in the output. If stop-outs are close to zero, the useful question is whether the price generator can produce a losing day at all, rather than whether the strategy is good.

Is a synthetic price model useless for options backtesting?

No, but it answers a narrower question than most people assume. Synthetic paths are reasonable for testing whether execution machinery behaves consistently. They cannot establish that an edge exists, because the path was assumed rather than observed.

What should decide my position size, if not the backtest?

Arithmetic. What a single loss costs as a percentage of the account, and what two or three consecutive losses would do to it. That needs no simulator and cannot be flattered by a price model.

Run it yourself

Options Sniper is a free desktop terminal for 0‑5 DTE options. It runs paper or live and ships with a simulator that drives the same code the live app uses. Open source on 30 November 2026.

Download Options Sniper

Related reading: Six diagnostics for a flattering backtest · Options backtesting software compared · All automated options guides

Trade with the desk

Investors University is capped at 40 students, and 11 of the 50 N‑T PRO licences are left. Leave an address and you will hear when a place opens.

One message when something opens. Nothing else, and never shared.

See what N‑T PRO is · Book fifteen minutes