A backtest is not a track record

A track record is a list of trades that were actually placed, filled, and settled, with real money, in real time, by someone who did not know what came next. A backtest is a simulation that assumes a set of rules was followed perfectly over data that has already happened. They can produce identical-looking equity curves and they are not the same kind of evidence.

The gap, item by item

Each row is a way for the backtest to be better than what you will experience. Some of them, like costs and fills, can be modelled carefully and Module 4 shows how. Others, like the effect of a 12R drawdown on your willingness to take the next signal, cannot be simulated at all.

Why the backtest is still worth doing

If a backtest is so much weaker than a track record, why bother? Because a track record is expensive. Building one takes months or years and it costs real money while the strategy is unproven. A backtest costs a few evenings. It cannot prove a strategy works, but it can prove, cheaply, that many strategies do not. Its power is in rejection.

Think of the pipeline as a set of filters, each more expensive than the last:

1. Hypothesis: does the idea have a mechanism? (minutes)
2. Backtest: does it survive honest data and costs? (evenings)
3. Out-of-sample: does it survive data it never saw? (one run)
4. Paper or tiny-size live trading: does it survive real fills and a real person? (months)
5. Full-size live: the track record. (years)

Most ideas should die at steps 1 to 3. That is the point. Every idea that dies there is one you did not pay to learn about live.

Key idea: A backtest is evidence that a strategy might work, and strong evidence when it does not. Only a live track record is evidence that it does work, and even then only for the period it covers.

Selection: the quiet problem

Suppose ten traders each backtest a different idea, and each idea has no real edge. By chance, a couple of them will produce a good-looking equity curve. Those two will post it. The eight who got flat or negative results will not. You see two excellent backtests and no failures, and conclude that backtesting reliably finds edges.

The same thing happens inside one person's head. You test fifteen variations, twelve are poor, and you remember the three good ones as "my research". Module 6 puts numbers on this. For now, the habit is to count. How many tests did you run to find this one? If the answer is "I don't know", the result is worth less than it looks.

Backtest claims in the wild

When someone shows you a backtest, the useful questions are:

• What was the hypothesis, and when was it written down?
• How much data was in-sample, and how many out-of-sample passes were made?
• What cost and slippage assumptions were used?
• How many variations were tried before this one?
• Has any of it been traded live, at what size, for how long?

A vendor or a forum poster who cannot answer these is showing you a picture, not evidence. A vendor who answers them honestly is rare and worth listening to, even if the numbers are less exciting.

Try it: Find a published strategy backtest, in a forum post, a blog, or a platform's public scripts. Score it against the five questions above. Most will score zero or one. That is not a reason to be cynical about backtesting; it is a reason to do yours differently.

Recap

• A backtest simulates rules over known data; a track record is real trades in real time.
• The backtest is optimistic in fills, costs, discipline and selection.
• Its value is cheap rejection: most ideas should fail before they cost money.
• Count how many tests you ran to get the result; unseen failures inflate the ones you see.
• Ask any backtest the five questions: hypothesis, split, costs, variations, live history.