How long should you backtest? A practical answer
A year is the popular default for backtest length. A year is often not enough. The real question isn't time — it's how many trades you generate and how many market regimes you cover. Here's how to think about it without rules of thumb that betray you.
KEY FACTS
- A strategy that makes one trade per week over five years (260 trades) is better-validated than one that makes one trade per month over twenty years (240 trades).
- Minimum trade counts For statistical significance, most systematic strategies need 100–200 trades minimum.
- With 30 trades, even a 65% win rate could plausibly be from chance.
- With 200 trades, that same win rate is meaningful evidence of an edge.
Source: Puravida Edge, from the article below; rules data verified 29 September 2026.
Time is the wrong unit
Backtest “length” is almost always discussed in months or years. That's the wrong frame. A strategy that makes one trade per week over five years (260 trades) is better-validated than one that makes one trade per month over twenty years (240 trades). What matters is the trade count and the variety of market regimes encountered, not the calendar duration.
Minimum trade counts
For statistical significance, most systematic strategies need 100–200 trades minimum. Below that, the signal-to-noise ratio is too low to distinguish a real edge from luck. With 30 trades, even a 65% win rate could plausibly be from chance. With 200 trades, that same win rate is meaningful evidence of an edge. See risk of ruin and statistical significance for the math behind these thresholds.
Regime coverage matters more than calendar time
A backtest only validates a strategy against the regimes contained in the sample. If your year of data was all uptrend, you don't know whether the strategy survives a downtrend, a chop, or a high-volatility crash. Different markets cycle through regimes on different timescales — index futures might see all four major regime types in 2–3 years; commodity markets often need 5+ years; FX pairs vary widely.
| Regime type | Characteristics | Typical time to encounter |
|---|---|---|
| Uptrend | Sustained directional move | Months |
| Downtrend | Sustained reverse move | Months |
| Range | Bound between defined levels | Months |
| High volatility | Wide bars, frequent reversals | Weeks to months |
| Quiet | Tight ranges, low volume | Months to years |
| Crisis | Sudden gap, liquidity stress | Years apart |
Length by trade frequency
| Strategy frequency | Recommended minimum backtest | Why |
|---|---|---|
| Scalping (10+ trades/day) | 3–6 months | Generates 10,000+ trades; sample size not the constraint |
| Intraday (1–5 trades/day) | 1–2 years | 500–1,000 trades; covers 2–3 regime cycles |
| Swing (1–3 trades/week) | 3–5 years | 200–500 trades; multiple full regime cycles |
| Position (1–3 trades/month) | 7–10 years | 100–200 trades; full bull/bear cycles |
Special considerations for prop firm strategies
Prop firm evaluations typically last 30–60 days. A 1-year backtest that produces 500 trades gives roughly 40–80 trades per evaluation window — enough to see whether the strategy will reach the target within the time limit. A 1-year backtest that produces 50 trades gives 4–8 trades per evaluation, way too few to validate the strategy will reliably hit the target on time. For low-frequency prop strategies, you need 3–5+ years of backtest data. See time-to-first-payout for how this plays out in practice.
When you can't get enough data
If your strategy is so new or so low-frequency that you can't reach 100+ trades on available data, you have two honest options:
- Trade it forward in paper / sim until enough trades accumulate to validate.
- Accept that what you have is a hypothesis, not a tested strategy, and size accordingly.
Pretending a 30-trade backtest validates a strategy is how people blow up accounts. The backtest + live sample window used in the Puravida Edge methodology generates 500–2,000+ trades per portfolio depending on strategy frequency — well above the statistical significance threshold across multiple regime cycles.
Sample size is one of the inputs the backtest quality checker weighs when scoring a set of results.
How many trades should I backtest for reliability?
More than most backtests contain, and the number depends on your win rate rather than on a fixed rule of thumb. A strategy with a 55% win rate needs several hundred trades before the observed rate is a reasonable estimate of the true one. At 200 trades, a genuine 55% system can display anywhere from roughly 48% to 62% purely from sequence luck.
Two additional requirements matter more than raw count. The sample must span different market conditions — a strategy tested exclusively across a trending period has been validated for trends, not for markets. And it must include a losing period, because a backtest with no meaningful drawdown is too short to have shown you the strategy's downside rather than proof of quality.
There is a second sample nobody counts: the number of parameter combinations tested. Testing 500 variations and selecting the best performer guarantees a good-looking result whether or not an edge exists, because the best of 500 random outcomes looks impressive by construction. The count of things you tried is part of the statistics.
The practical check is sequence robustness rather than trade count alone. Take the historical trades, reshuffle their order several thousand times, and look at the distribution of outcomes — maximum drawdown, longest losing run, final equity. If most reshuffles remain acceptable, the result does not depend on the specific order history happened to produce. The Monte Carlo simulator does this, and is your backtest overfit covers the selection problem.
After backtesting and going live, how many live trades before the win rate shows?
Longer than the period most traders allow before concluding the backtest was wrong. At typical retail trade frequencies, a sample large enough to distinguish a working system from a broken one takes months, not weeks — and during that time the observed win rate will move around substantially for reasons that have nothing to do with the strategy.
Two useful reference points. Thirty live trades tell you almost nothing: a 55% system regularly produces 40% or 70% across a run of thirty. A hundred trades narrow the range but still leave room for a working system to look mediocre. The uncomfortable implication is that the period during which you cannot yet evaluate the strategy is also the period during which the temptation to modify it is strongest.
What is measurable earlier than win rate is the gap between backtest and live execution. Compare fill prices to backtest assumptions, count trades the backtest took that you skipped live, and check whether costs match what was modeled. Those diverge immediately and visibly, and they explain more early underperformance than edge decay does — see why your Pine backtest will not match live.
While the sample builds, the correct response is to trade the plan at a size you calculated before starting, and to change nothing. Modifying during the accumulation period restarts it.
Why backtests and live results differ: the full map →
Is twelve months of backtest enough
Section merged from the former standalone article on 8 October 2026; figures as verified at that date.
Whether a year is enough depends on what is being asked of it. Twelve months of a high-frequency intraday strategy can contain a thousand trades; the same window on a swing system might hold forty. The unit of evidence is the trade, not the calendar, the statistics in how long to backtest and the sample math in the sample size problem.
What a year can prove
With hundreds of trades, twelve months can establish that the rule set has positive expectancy after realistic costs, characterize the win-rate and profit-factor personality per the WR vs PF data, and surface the typical losing-streak shape. For an intraday system with a defined session, that is genuinely useful evidence.
What it structurally cannot
One year contains one of each season and at most one regime change. A strategy born in a trending year has never met a chop year; the breakdown only arrives live, the decay criteria in when to retire a strategy. And a single historical path says nothing about sequence risk: the same trades in a crueler order produce a deeper drawdown, which is exactly what a trailing floor punishes.
Stretching the year honestly
Resampling is how a year of trades becomes a distribution instead of an anecdote: Monte Carlo reshuffles the sequence thousands of times and reveals the worst stretches the calendar happened not to deal, the method in Monte Carlo for sizing and the comparison against rolling validation in walk-forward vs Monte Carlo. That is why the published figures here pair a backtest + live sample with 1,500 resampled paths over a three-year horizon: the year supplies the trade distribution, the resampling supplies the stress.
The rule of thumb
Twelve months is enough to qualify an intraday strategy for paper deployment when the trade count is in the hundreds, the costs are honest, and the sequence risk has been resampled rather than assumed. It is never enough on its own to size against a hard floor, and it is nowhere near enough for a low-frequency swing system, where the calendar must do what the trade count cannot.
If you are judging whether a given sample is enough, the backtest quality checker scores it against the other quality signals at the same time.
Why backtests and live results differ: the full map →
Related
- Why your Pine Script backtest won't match live trading
- Choosing strategy parameters without overfitting
- Why your edge only shows up over a large sample of trades
The method, applied to the whole catalogue.
Every strategy and portfolio with modeled and live shown separately and the forecast behind each one. One PDF, then our letters with what the systems did.
By email, from Puerto Viejo, Costa Rica. Unsubscribe anytime.
Related
- How to verify a trading backtest is real
- ICT methodology meets Monte Carlo
- How to know if your trading backtest is overfit
FAQ
Is one year of data enough to backtest a trading strategy?
It depends on trade count. A high-frequency strategy generating 1,000+ trades in one year is well-validated; a low-frequency strategy generating 30 trades in one year is not. The minimum useful sample is roughly 100–200 trades across multiple regime types.
How many trades do I need to validate a trading strategy?
Most systematic strategies need 100–200 trades minimum for statistical significance. Below that, you can't reliably distinguish a real edge from random luck. For high-confidence validation, 500+ trades across multiple market regimes is the practical standard.
Can I backtest on demo or simulation data?
Yes — demo / sim is a legitimate way to accumulate trade history when historical data isn't enough. The key difference vs live: execution costs and slippage are usually unrealistically favorable in sim. Apply realistic commissions and slippage assumptions when interpreting sim results.
Not financial advice. Performance figures referenced are hypothetical, modeled outputs (1,500-path Monte Carlo on a backtest + live sample). Past performance does not guarantee future results. Tool names are referenced for education; verify current features and prop-firm rules directly.
How many trades should I backtest for reliability?
It depends on win rate, but several hundred is a realistic floor — at 200 trades a genuine 55% system can display 48% to 62% from sequence luck alone. The sample must also span different market conditions and include a meaningful drawdown. Sequence reshuffling matters more than raw trade count for testing robustness.
How many live trades before a win rate becomes meaningful?
Substantially more than thirty. A genuine 55% system regularly shows 40% or 70% across thirty trades, and a hundred trades still leave room for a working system to look mediocre. Execution differences — fills, skipped trades, costs versus modeled — diverge from the backtest much earlier and explain more early underperformance than edge decay.
How many trades does a valid backtest need?
Enough that the win rate and streak profile stabilize, typically several hundred. A year of an intraday system can supply that; a year of a swing system usually cannot, regardless of how clean the curve looks.
Can I trust a one-year backtest for a prop account?
Only after resampling. The single historical sequence understates drawdown risk, and a trailing floor punishes exactly the sequences the calendar did not happen to deal. Monte Carlo over the year's trades is the minimum honest stress.
Why pair a backtest + live sample with a 3-year Monte Carlo horizon?
The empirical year supplies a realistic trade distribution including costs; the resampled multi-year horizon reveals streaks and drawdowns beyond what one calendar path contained. Each covers the other's blind spot.