The Anatomy of a Backtest

intermediate8 min read

Universe, signals, entries, exits, costs — the moving parts every honest backtest must define.

Every honest backtest is built from the same handful of clearly-defined parts. If any one is vague, the whole result is suspect. Knowing the anatomy lets you both build a sound test and audit anyone’s claimed backtest.

Universe — which instruments you trade (e.g. Nifty 500 stocks), defined to avoid survivorship bias (include delisted names).
Signals — the precise conditions that identify a candidate trade (e.g. price > 200-day MA and momentum in top decile).
Entry rules — exactly when and at what price you get in (and crucially, using only data available at that moment).
Exit rules — targets, stops, time exits, or signal reversal; exits drive most of the result.
Position sizing — how much capital per trade (equal weight, risk-based, etc.).
Costs — brokerage, taxes and slippage applied to every fill (covered next).

The discipline that makes a backtest honest is point-in-time correctness: at every moment in the simulation, you may use only the information that actually existed then. This sounds obvious but is violated constantly — using a stock’s current index membership for past dates, today’s restated earnings for old trades, or a signal computed from the day’s close to enter at that same close. Each tiny leak of future knowledge inflates results and is invisible in the equity curve. So the anatomy isn’t just a checklist of parts — it’s a discipline of asking, for every single rule: “could I actually have known this at the time?” Define each component precisely and point-in-time, or your backtest is quietly cheating.

ExampleA backtest buys “the Nifty 50 stocks” using today’s Nifty 50 list for trades in 2015 — but that list has changed; you’re unknowingly buying the winners that got promoted into the index later (survivorship + look-ahead). The fix: use the historical index membership as it existed on each date. Same idea, point-in-time correct.

Key takeawayA sound backtest defines universe, signals, entries, exits, sizing and costs — each precisely and point-in-time correct (using only data available at the moment). The discipline behind the anatomy is asking, for every rule, “could I have known this then?” — get that wrong and the backtest cheats invisibly.

FAQs

Which component matters most?

Exits and costs are the most underrated — exits drive a large share of the result, and ignored costs can turn a “winning” backtest into a real-world loser. But *point-in-time correctness* across all components is the make-or-break: a single look-ahead leak (often in the universe or signal) can fabricate an edge that doesn’t exist.