Backtesting Lies: Why Paper Performance Rarely Survives Live Execution

A backtest that shows a 70% win rate and a smooth equity curve is not lying about the past. Run correctly, it is an accurate account of exactly what would have happened if those trades had been taken on that data. The lie is in the gap between that account and what actually happens when the same rules meet a live market — a gap most traders never measure, because the backtest and the live journal are kept in two different places and never put side by side.

Fills are the first thing that changes

A backtest almost always assumes an order fills at the price the signal fired on. Live, it fills at whatever price is available by the time the order reaches the market — after the tick that triggered the signal, after the platform processes it, after the broker routes it. On a slow-moving pair with tight spreads this difference is small enough to ignore. On anything volatile, or during news, or on a strategy that trades near session opens, the fill can be several pips away from the modeled price on a meaningful share of entries. A strategy with a thin edge on paper can have no edge at all once every entry is priced worse than the backtest assumed.

Spread and commission are usually modeled once, not per trade

Many backtests apply a fixed, average spread and a flat commission across the whole test. Real spread widens around news releases, thins during quiet hours, and varies by broker and by symbol. A strategy that concentrates its entries around session opens or news events — often the times a backtest looks best, because that is when the moves it is designed to catch happen — is also the strategy most exposed to the difference between average spread and actual spread at the moment it trades. The backtest charges the average. Live, you pay whatever the market is asking right then.

Backtests don't hesitate, and you do

A backtested rule fires the instant its condition is met, every time, without exception. A live trader looking at the same setup checks it twice, waits for one more candle to confirm, or skips it because the last three signals like it lost. None of that shows up in the backtest's numbers, but all of it shows up in the live results — as entries taken later than the rule specified, entries skipped that would have won, and entries taken on a whim that the rule never called for in the first place. The strategy being graded and the strategy being traded are rarely the exact same thing by the time discretion gets involved.

Overfitting looks like a great backtest and nothing else

A rule set with enough parameters — a moving average length, an RSI threshold, a specific session window, a minimum candle size — can be tuned against historical data until it produces an excellent equity curve on that exact data. The tuning is not visible in the result; a curve-fit strategy and a genuinely robust one can look identical in a backtest report. The tell is usually in how the parameters were chosen: if a setting was picked because it happened to maximize the historical result rather than because it reflects something true about how the instrument trades, the backtest is measuring how well the strategy fits the past, not how well it will handle a future the fitting process never saw.

What actually closes the gap

Not more backtesting — a live sample checked against the backtest's assumptions, specifically. That means logging, per trade, the price the signal called for against the price actually filled, and the spread the backtest assumed against the spread paid. Once that comparison exists for a real sample of trades, it tells you exactly which assumption is costing you: consistent negative slippage on entries points at execution speed or broker routing; a spread gap concentrated around specific hours points at trading through conditions the backtest never modeled; a live win rate that tracks the backtest but a live expectancy that doesn't points at fills, not the rules themselves.

What to actually track

• Modeled fill vs. actual fill — the price the signal specified against what the order actually executed at, trade by trade.
• Assumed spread vs. paid spread — especially around news windows and session opens, where the two diverge the most.
• Rule-fired vs. discretion-taken — which entries and exits followed the backtested rule exactly, and which were adjusted in the moment.
• Live expectancy vs. backtested expectancy — on a matched sample size, not a handful of trades against years of history.

A backtest tells you whether an idea has a plausible edge. Only a live journal, checked against that same backtest's assumptions, tells you whether the edge survived contact with real fills and real spread. getALPHA's journal captures execution detail on every synced trade, so that comparison is something you can actually run instead of something you assume away.