Backtesting and forward testing answer two different questions, and confusing them is one of the more expensive mistakes in systematic trading. A backtest asks: did this strategy work in the past? A forward test asks: does it still work now, on live prices, with all the frictions the historical data smoothed over? You need both, in that order, and skipping the second is how strategies that look brilliant on paper quietly bleed funded accounts.
What each test can and cannot prove
Backtesting: breadth, cheaply
A backtest's superpower is coverage. In minutes you can run a rule set across years of candles and thousands of trades, spanning trending markets, ranging markets, panics and dead summers. No forward test will ever give you that breadth: to see your strategy face five different regimes live, you would have to trade it for five years first. The backtest is where you learn the strategy's character: its maximum drawdown, its losing streaks, which sessions carry the edge, what a normal bad month looks like.
What the backtest cannot prove is that any of it survives contact with the present. Its weaknesses are structural: it can be overfitted to the very data that produced the flattering curve, it models spread and execution rather than experiencing them, and it cannot know whether the market condition that fed the edge still exists. Our backtesting guide covers the specific ways the numbers lie and how to test honestly.
Forward testing: truth, slowly
A forward test (paper trading a fixed rule set on live prices) has exactly the opposite profile. It is slow and narrow: weeks of waiting to collect a few dozen trades from whatever regime the market happens to serve. But every one of those trades is honest. The spread was the real spread at the moment of entry. The breakout either triggered on the live feed or it did not. Nothing was smoothed, interpolated or assumed.
Most importantly, a forward test is immune to overfitting by construction: the strategy cannot have memorised data that did not exist when its rules were frozen. That makes it the closest thing trading has to an out-of-sample exam that keeps renewing itself.
What forward testing catches that backtests miss
- Overfitting, exposed. The strategy that returned 40% a year in testing and treads water live is the classic signature of curve-fitting. Better to learn that on paper than on a funded account.
- Live spread and slippage. Historical spread assumptions meet the real thing, including the session-open widening that hits breakout entries precisely when they fire.
- Execution timing. The gap between "the candle closed beyond the level" and "the order filled" is real and only measurable live.
- Regime drift. If the market has changed since your test window, forward results diverge from backtest expectations early, which is exactly the warning you want.
- Your own behaviour. Watching a live losing streak, even on paper, is a rehearsal for the discipline the real account will demand.
How long should a forward test run?
Long enough to be evidence rather than anecdote. A useful rule of thumb is to keep going until the forward test has produced enough trades to compare meaningfully against the backtest's per-month behaviour, typically several weeks to a couple of months for an intraday strategy. The comparison is the point: you are not asking "did it make money this month?" but "is it behaving like the backtest said it would?" Win rate in the same neighbourhood, losses of the expected size, drawdown inside the historical envelope. A forward test that loses money in a way the backtest predicted is a pass; one that wins in a way the backtest cannot explain deserves suspicion too.
The promotion path
Put together, the two tests form a pipeline that filters ideas by increasing cost of failure:
- Backtest over years of real candles. Kill the failures cheaply. Learn the survivor's drawdown and streaks.
- Forward test the survivor on live prices with paper money until its live behaviour matches its tested character.
- Go live small, or take the evaluation, with sizing derived from tested drawdown, not from the profit target (the arithmetic is in the FTMO evaluation guide).
- Keep the forward test running as the live benchmark: if live results ever drift from the paper twin, something has changed and the strategy has earned a review, not blind loyalty.
One practical warning from the engineering side: the pipeline only works if all three stages run the same rules on the same engine. If the backtest, the paper test and the live execution each reimplement the strategy slightly differently, disagreements between them tell you nothing, because they may just be implementation gaps. This is a deliberate design decision in Forge Strategy Lab: one engine evaluates the strategy in backtests, forward tests and live execution, so the numbers you proved really are the numbers you trade, and a forward test is one click on the backtest you already ran.
This guide is education, not financial advice. Trading involves risk and past performance, tested or live, never guarantees future results. Make your own decisions and never risk money you cannot afford to lose.