/ Check 10 · Forward record

How long should you forward test a strategy?

Long enough to do two jobs: show that the live trades look like the backtest, and, much later, show an edge on their own. The first can take a few months. The second, for a thin edge, can take years.

The short answer

A forward record is the only part of a strategy's history that could not have been fitted to. Every trade in it was decided before the price that settled it existed. A backtest, however careful, was run by someone who could see the whole window. So the question is not how many weeks a forward test should last. It is how many trades it needs before it can say something, and that depends on the strategy's edge and how often it trades.

What counts as a forward record

  • Frozen parameters. The settings are written down, with a date, before the first forward trade. A record whose settings changed partway through is several short records, not one long one.
  • A dated ledger. Each entry and exit is logged when it happens, with its size and its costs. A list put together afterwards from the chart is a backtest of a later window.
  • Paper or live, on the stated venue. Paper trading on the venue the backtest names is a forward record. Re-running the backtest on newer bars is not, if the parameters were tuned on anything you have already seen.

If none of the three exist, the result is a simulation, full stop. That is not an insult to the strategy; it is what the word means.

A worked example: same backtest, two forward records

A made-up strategy's backtest shows 110 wins in 200 trades, a 55% win rate, with wins and losses about the same size. Break-even at that payoff is 50%. The 95% interval on the backtest is 48.1% to 61.7%, which already reaches below break-even: two hundred trades have not settled whether this edge is real.

The parameters are frozen and it trades forward at about 8 trades a month. After five months the ledger holds 40 trades. Two ways that can go:

  • Record A: 22 wins in 40. That is 55%, with a 95% interval of 39.8% to 69.3%. The backtest's 55% sits comfortably inside it. The forward record agrees with the backtest, and it says nothing yet about an edge, because the interval also contains every losing win rate down to 39.8%.
  • Record B: 14 wins in 40. That is 35%, with a 95% interval of 22.1% to 50.5%. The backtest's 55% sits above the whole range. Forty trades are enough to say this record and that backtest are unlikely to describe the same strategy. The usual reason is that the backtest was fitted to its window.

That is the first job a forward record does, and it can do it early: a record that contradicts its backtest shows up in a few dozen trades.

The second job takes much longer

For the forward record to show an edge on its own, its interval has to clear break-even. At a 55% win rate and equal-sized wins and losses, that takes about 385 trades. At 8 a month that is about 48 months, four years. If the same strategy won 60% of the time, it would take about 97 trades, roughly a year.

Try your own numbers in the trades-needed calculator, and put an interval around the record you already have with the win-rate interval calculator. Divide the trades needed by the trades a month the strategy makes and you have the honest length of the forward test.

How to run the check on any report

1. Ask whether a forward record exists. Paper or live, frozen parameters, dated ledger. If the answer is no, everything else in the report is a simulation.

2. Find the freeze date. Every trade before it is backtest, however it is labelled. Only trades after it count.

3. Compare the two. Put an interval around the forward win rate and see whether the backtest's rate sits inside it. Outside is a warning even on a small sample.

4. Count what is left. Work out how many forward trades the edge needs, and how many months that is at the strategy's pace. A record shorter than that is a consistency check, not proof.

Why this is the last check

The other nine can all be run on a backtest alone, and a strategy that fails any of them does not need a forward test. A strategy that passes all nine has earned one. It has not earned anything more until the dated record exists.

This is check 10 of 10

The one-page Backtest Verification Checklist, free. Confirm the email and the PDF opens straight away; one worked check each Tuesday after that.

How these numbers were worked out

Every figure is a made-up strategy, not a result of any system. Intervals are Wilson 95% score intervals (z = 1.96) on the stated wins and trades: 110 of 200 gives 48.1% to 61.7%, 22 of 40 gives 39.8% to 69.3%, 14 of 40 gives 22.1% to 50.5%. Break-even with equal-sized wins and losses is 1 ÷ (1 + 1) = 50%. “Trades needed” is the smallest count at which the lower end of the interval sits above break-even, assuming the win rate holds and trades are independent, the same search the calculator runs: 385 at 55%, 97 at 60%. Months are trades divided by 8 a month: 40 ÷ 8 = 5; 385 ÷ 8 = 48.1; 97 ÷ 8 = 12.1. Nothing here is investment advice.

Historical simulations are not live results. Nothing on this site is investment advice.