Backtest Profitable but Losing Live: The Real Causes

"The backtest tripled the account, but running it live I'm losing." This is the most common disappointment in algorithmic trading. Most people conclude "the EA broke" or "the market changed," but the cause is usually a set of structural gaps between backtest and live trading.

There are six. Here is each one — what it is, and how to spot it before you buy. This article is a hub linking to the detailed piece on each.

Gap 1: The tick-model lie

Running the backtest on the "1-minute OHLC" model creates fills at prices that never existed. For thin-take-profit, grid and scalping EAs, this alone makes the backtest profitable and the live account red.

Fix: confirm the model is "Every tick based on real ticks." If it isn't stated, distrust it. → Detail: Why real ticks vs OHLC give opposite results

Gap 2: Spread and execution

A backtest assumes idealized execution. On a real broker, spread is wider, execution is slower, and slippage is added. The thinner the margin, the more of the profit these differences erase.

Fix: distrust backtests run on unrealistically tight spreads; demo-test on your own broker. → Detail: How spread and slippage affect EA returns · Why the same EA loses on a different broker

Gap 3: Curve fitting (over-optimization)

An EA whose parameters were fitted too closely to past data can be perfect on that history and useless on markets it has never seen. The beautiful rising equity curve may just be the fitted result.

Fix: check that it was validated with in-sample/out-of-sample splits or walk-forward. → Detail: How to avoid curve fitting

Gap 4: Numbers inflated by compounding

A net profit % like "+5,000%" is the multiplication of compounding, not skill. Live, a drawdown along the way makes compounding work against you, and you never reach the expected number.

Fix: measure skill by profit factor, not net profit %. → Detail: The compounding illusion, and what to read instead

Gap 5: Too few trades — it was luck

An EA that trades 28 times in 8 years with PF 5.5 may just have caught a few lucky big wins. Live, that "luck" doesn't continue.

Fix: confirm there are 500+ trades spanning multiple market regimes. → Detail: Why you can't trust a backtest with few trades

Gap 6: Dependence on past markets (regime dependence)

An EA that only works in a one-directional market of the last few years collapses the moment conditions change. A strong backtest can be dependent on a specific market regime.

Fix: read the year-by-year P&L. One big year and losses in the rest points to regime dependence. → Detail: We backtested 43 popular gold EAs on 11.5 years of real data

The four-step check

1. Is the model real ticks? (Gap 1)
2. Are there enough trades, across multiple regimes? (Gaps 5, 6)
3. Are you reading PF and equity DD? (Gaps 3, 4)
4. Did you demo-test on your own broker? (Gap 2)

An EA that passes these four has a small backtest-to-live gap. An EA that doesn't disclose them is much more likely to disappoint you live.

Check it yourself

Our free EA (a closed-bar Donchian breakout) is published to meet all four points: real-tick verified, 631 trades, 9.4 years, PF and equity DD disclosed, symbol specs read dynamically. Run the backtest and see what a small-gap EA looks like.

Donchian Trend Engine (free EA)

Every product is published to the same standard on the verification page.

Summary

• Backtest-profitable, live-losing is caused not by a broken EA but by six structural gaps.
• Tick model, spread/execution, curve fitting, compounding, trade count, regime dependence.
• The check: real ticks, trade count and period, PF and equity DD, demo on your own broker.
• An EA that discloses none of these is likely to disappoint live.

Related: Real ticks vs OHLC · Why the same EA loses on a different broker · The compounding illusion · Backtest sample size