When selecting or evaluating an Expert Advisor (EA), Profit Factor (PF) is one of the first metrics to examine. Yet the rule of thumb that “PF 1.5 or above means pass” has taken on a life of its own, and it is not uncommon to see EAs adopted without a clear understanding of the assumptions behind that figure or its inherent limitations. This article defines PF, explains how it is calculated, examines the rationale behind the benchmarks used in EA evaluation, and clarifies the fundamental limitations of the metric.
What Is Profit Factor: Definition and Calculation
Profit Factor (PF) is a straightforward metric that indicates the profitability of an EA or trading strategy. The formula is as follows.
PF = Gross Profit / Gross Loss (absolute value)
For example, if total profit during a backtest period is 1,200,000 yen and total loss is 800,000 yen, then PF = 1,200,000 / 800,000 = 1.50. A PF below 1.0 means losses exceed profits and, in theory, the account balance will continue to shrink under continued operation. A PF above 1.0 means profits exceed losses by that ratio.
An important point: PF is a metric that is independent of win rate. Even with a 30% win rate, if the average profit is more than four times the average loss, PF will exceed 1.2. Conversely, even with an 80% win rate, a few large losses can push PF below 1.0. PF is simply the ratio of total profit to total loss; it contains no information about the distribution or continuity of individual trades. For a foundational guide to reading PF, see also PF Indicator Benchmarks: A Guide from the Basics.
EA Evaluation Benchmarks and Practical Calculation Examples
Below is a summary of common PF benchmarks used when evaluating EAs.
| PF Level | Evaluation Benchmark | Notes |
|---|---|---|
| Below 1.0 | Below breakeven | Not suitable for continued operation |
| 1.0 – 1.3 | Marginal level | Easily undermined by spread and commission changes |
| 1.3 – 1.5 | Minimum practical level | Trade count and verification period must be scrutinized |
| 1.5 – 2.0 | Practical benchmark | Must be evaluated alongside MDD and trade count |
| 2.0 and above | High profitability or possible overfitting | Suspect over-optimization (curve fitting) |
Consider a concrete example. Suppose EA-A shows a backtest gross profit of 300,000 yen and gross loss of 150,000 yen (PF = 2.00), while EA-B shows a gross profit of 200,000 yen and gross loss of 160,000 yen (PF = 1.25). On numbers alone, EA-A appears overwhelmingly superior. But if EA-A had only 12 trades and EA-B had 380 trades, the statistical credibility is reversed. With only 12 trades, it is impossible to rule out the possibility that one or two lucky wins simply inflated the PF.
In gold EA verification, high volatility has a strong influence on PF. For an evaluation perspective that accounts for currency-pair-specific characteristics, see also Key Metrics to Check in Gold EA Verification.
Common Pitfalls for Beginners
- Trusting PF with an insufficient number of trades
PF becomes statistically noisier the fewer trades there are. As a general rule, fewer than 200 to 300 trades means the figure should be considered unreliable. Even if a short backtest produces a PF above 2.0, the number may be inflated by statistical noise. - Backtest period biased toward a one-directional market
For example, a long-biased EA tested only during the gold surge of 2020-2021 will show a PF tailored to that specific period. A proper verification crosses multiple market environments — trending, ranging, high-volatility, and low-volatility — using five or more years of data. - Judging solely by PF while ignoring MDD (maximum drawdown)
Even with a PF of 1.8, an EA whose maximum drawdown reaches 60% of the account balance carries a real risk of running out of capital in live operation. PF and MDD must always be reviewed together and assessed compositely from a risk-efficiency perspective. - Underestimating the gap between backtest and live PF
Spread, slippage, and fill latency are often set to ideal values in backtests. In live operation, a PF drop of 0.2 to 0.5 points is not uncommon. It is essential to compare the PF of a forward test (an operational record on a live price feed) against the backtest PF. - Failing to detect PF inflation in martingale-type EAs
Martingale and grid EAs are structurally designed to defer stop-losses, which means unrealized losses during a backtest tend not to be recorded as realized losses, making the apparent PF artificially high. For the structural background behind this, see Why Martingale EAs Experience Deep Drawdowns.
FX AI Lab’s Assessment
At our lab, PF is not treated as a standalone pass/fail criterion when evaluating EAs. Our current evaluation standard requires the simultaneous satisfaction of “PF 1.3 or above, MDD below 30%, and a trade count of 200 or more” as a minimum threshold. We also verify whether the PF during the forward-test period maintains at least 80% of the backtest PF. Reports based on these criteria for our in-house EAs under verification and for copy trade configurations are published progressively in our Verification Report Library. Feel free to contact us for participation in copy trading or for further details.
Related Links
This article is provided for informational purposes only and does not constitute a recommendation to buy or sell any specific EA or trading method. FX trading involves leverage and may result in losses exceeding your principal. All final investment decisions are made at your own risk. For details, please review our Risk Disclosure.