What is a backtest?

A backtest is a simulation. You take a fixed rule — entry condition, exit condition, instrument universe, position sizing — and apply it to historical market data exactly as it stood at each point in time. The output is a record of what would have happened if that rule had been followed mechanically through the chosen window, under stated assumptions about execution, costs, and liquidity.

The keyword is fixed. Rules are decided before the simulation runs and do not change in response to what the simulation produces. The moment the rule is altered because the output looked unflattering, the exercise stops being a backtest and becomes parameter mining.

Backtests are how a researcher converts an opinion into evidence. Before a backtest, a statement like 'expiry-day NIFTY behaviour favours premium sellers' is folklore. After a properly designed backtest, the same statement becomes a measured distribution of historical outcomes under disclosed assumptions — much easier to argue with, agree with, or disprove.

The point is not to find a strategy to deploy tomorrow. The point is to learn how a rule would have behaved across a long enough window to cover more than one regime, and to identify the conditions under which it broke down. Most published backtests skip this second part, which is exactly why most published backtests are not useful.

A clean backtest can tell you the distribution of historical outcomes under your assumptions: the spread of returns per observation, the magnitude and duration of drawdowns, the frequency and clustering of losses, the sensitivity of outcomes to costs, and the way performance varies by regime. These are real properties of the historical sample.

It can also tell you the structural risk profile of the rule. A short-premium structure will show a smooth equity curve with rare deep losses. A trend-following structure will show frequent small losses and rare large gains. The backtest makes the shape of the return distribution visible — and that shape is far more useful than the headline CAGR.

A backtest cannot tell you what the rule will do next week, next month, or next year. It is a measurement of the past under one set of assumptions. Markets change: weekly expiries appeared, lot sizes were revised, retail participation shifted, peak margin rules tightened, and STT was restructured. A rule that produced a clean historical curve before any of those changes is not necessarily a rule that will produce a clean curve after them.

A backtest also cannot tell you anything about a question it was not designed to answer. A backtest of expiry-day index option behaviour says nothing about single-stock futures. A backtest using daily closes says nothing about intraday execution quality.

Suppose the question is: 'how did a defined NIFTY weekly-expiry options rule behave between 2018 and 2024 under conservative cost assumptions?' The study fixes the universe (NIFTY weekly options), the rule (a specific entry and exit condition decided in advance), the execution model (signal at close, fill at next available print), the costs (brokerage, STT, GST, exchange and SEBI charges modelled separately), and the slippage (scaled by strike liquidity).

The output is a per-observation log of historical outcomes, an equity curve, a drawdown series, and a sensitivity panel that doubles the slippage assumption and re-runs. The report then states limitations explicitly: 'this study covers one regime of weekly-expiry microstructure and does not extrapolate to lot-size or STT changes after the data window'.