Guide
Why your backtest loses money live: 5 common causes
You backtested it for months. Sharpe over 2, a clean equity curve, drawdowns you could live with. You went live with real money — and it bled. Not a black-swan crash, just a slow, steady loss that never matched what the backtest promised. This is one of the most common experiences in trading, and it almost always comes down to a handful of specific, findable gaps between the backtest and reality.
Quick answer
Backtests usually lose their edge live because of costs that were underestimated, fills that were more favourable than reality, a strategy that was tuned to fit the past rather than to persist, or market conditions that simply changed. The fix isn't a better strategy — it's a backtest that's honest about these gaps before you trade.
The backtest-to-live gap has a few specific causes
It rarely comes down to "the market is unpredictable" — that's a cop-out. In almost every case, one or more of the following was quietly wrong in the backtest, and reality corrected it the moment real money was involved.
1. Your fills were better in the backtest than they'll ever be live
A backtest often assumes you buy and sell at the exact price on the chart. Live, you don't. There's the bid-ask spread, the market moves between your signal and your order reaching the exchange, and on anything but the most liquid instruments, your own order can move the price against you. This difference is called slippage, and on a strategy that trades often, it compounds fast.
The fix: charge every backtested trade a realistic slippage estimate — not zero, and not a token symbolic amount. If your edge disappears once you do this, it was never large enough to survive contact with a real order book.
2. Commission and borrow costs were left out or lowballed
Commission is easy to forget when you're testing an idea, and borrow cost (the fee to hold a short position) is easy to forget entirely. Both are small per trade and easy to dismiss — until you're making fifty trades a month and the fees quietly outweigh the average profit per trade.
The fix: model your actual broker's commission schedule and, for shorts, actual borrow rates. A strategy needs to clear these costs on every single trade, not just on average.
3. The strategy was tuned to the past, not built for the future
If you adjusted parameters until the backtest looked good, you may have fit the strategy to the specific noise of that historical period rather than to a repeatable pattern. This is overfitting, and it's the single biggest reason a backtest and live results diverge. The strategy wasn't wrong about the past. It just never described the future.
The fix: always validate on out-of-sample data the strategy was never tuned against, and treat a big gap between in-sample and out-of-sample performance as a warning, not a detail.
4. Market conditions genuinely changed
Sometimes the backtest was honest and the strategy was real — it just relied on a condition that stopped holding. A mean-reversion strategy built during a range-bound period can struggle once the market starts trending. A volatility strategy tuned on a calm regime can misfire once volatility regimes shift. This isn't a flaw in the backtest; it's a reminder that no historical edge is guaranteed to persist.
The fix: understand why your strategy is supposed to work, not just that it did. A strategy with a clear, sensible reason for its edge is easier to judge when conditions shift than one that's just "what the optimiser found."
5. A single lucky run was mistaken for an edge
One equity curve is one path through history. If most of your strategy's return came from a handful of exceptional trades, or would evaporate under a different ordering of the same trades, you may have been looking at a lucky outcome dressed up as a strategy. Resampling the trade order thousands of times reveals whether the result is typical or a rare, lucky tail.
How to close the gap before you trade real money
None of these five causes are exotic — they're the standard, well-known ways backtests mislead people, and each one is checkable before you risk a cent. The discipline is simple to describe and easy to skip under time pressure: model real costs, validate out-of-sample, understand your edge's reason for existing, and stress-test the result before you trust it.
Or let the engine check all five automatically
The Honest Backtest Engine charges realistic costs on every trade, headlines out-of-sample results by default, and stress-tests every run with Monte-Carlo resampling — then scores how believable the edge is before you ever risk real money on it.
See how it worksFrequently asked questions
Why does my backtest look profitable but lose money in live trading?
Usually because of underestimated slippage and commission, a strategy tuned to fit past noise rather than a repeatable pattern, or a change in market conditions the strategy depended on. Each of these inflates backtested results in ways that don't survive contact with live trading.
Is slippage really that significant?
On a strategy that trades frequently, yes. Even a small per-trade slippage assumption compounds across dozens or hundreds of trades and can turn a backtested profit into a real loss.
How do I know if my strategy was overfit rather than just unlucky live?
Check whether it was validated on out-of-sample data it was never tuned against, and whether the result survives resampling the trade order. If the edge only exists on the exact history it was built on, it was likely overfit rather than unlucky.
Can a real edge still fail live?
Yes, market conditions change, and a strategy built for one regime can underperform in another. This differs from overfitting: the edge was real but relied on a condition that stopped holding. Understanding why a strategy is supposed to work helps you judge this.