How to Read MT5 Trading Statistics (Profit Factor, Sharpe, Drawdown)

分步说明

1. 第 1 步:Profit Factor — the single most useful metric

   Profit Factor = gross_profit / gross_loss. Ignores trade count, time, and equity progression — purely a ratio of winning gross to losing gross.

   Interpretation: • PF < 1.0 — EA loses money. Reject. • 1.0–1.3 — barely profitable. After accounting for spread/commission slippage in live, likely unprofitable. Marginal. • 1.3–1.5 — viable but tight. Live performance needs to closely match backtest to remain profitable. High risk. • 1.5–2.0 — robust. Modest live-vs-backtest divergence still leaves the EA profitable. • 2.0–3.0 — strong. Reasonable cushion against execution friction. • 3.0+ — suspicious. Either the backtest period was unusually favourable, the EA was curve-fit, or the data window was too short. Investigate.

   Profit Factor is symmetric across long/short and across symbols, so it's the right metric for comparing EAs that trade differently. Use as the primary filter after Drawdown.

2. 第 2 步:Maximum Drawdown — the loss you have to survive

   Drawdown is the peak-to-trough percentage equity decline during the backtest. MT5 reports both Balance Drawdown (closed P&L) and Equity Drawdown (mark-to-market including open positions). For EA evaluation, use Equity Drawdown — it captures the open-position pain you'd actually feel.

   Acceptable ranges by EA type: • Conservative trend-follower — 5–10% max DD • Standard EA — 10–20% max DD • Aggressive scalper — 15–25% max DD • Grid / martingale — 30%+ max DD (often blows up in live despite favourable backtest) • Retail EA above 30% DD — high blow-up risk; not suitable for live capital • Above 50% — likely blows up on a single bad month

   If your tolerance is 'never lose more than 10%', size positions so the EA's backtest DD scales to 10%. An EA with 25% backtest DD trading 1.0 lots becomes a 10% DD EA trading 0.4 lots — at the cost of 40% of the absolute profit.

   Also check the Drawdown Duration — how long the EA stayed below its previous equity peak. A 15% DD that recovers in 2 weeks is fine; a 15% DD that takes 18 months to recover is psychological torture and most traders quit before the recovery.

3. 第 3 步:Sharpe Ratio — risk-adjusted return

   Sharpe = (average return - risk-free rate) / standard deviation of returns. Higher is better. MT5 build 4000+ reports Sharpe in the Results tab; older builds require manual computation from the equity curve.

   Interpretation: • Sharpe < 0.5 — return barely beats variance. The EA's profit is mostly luck. • 0.5–1.0 — modest. Tolerable for conservative strategies. • 1.0–2.0 — good. Most professional managers target this range. • 2.0–3.0 — excellent. • 3.0+ — exceptional but rare. Verify the calculation; many short backtests inflate Sharpe.

   Caveat: Sharpe penalises upside volatility the same as downside volatility, which is mathematically wrong for trading (you want upside volatility). Sortino Ratio is the fix — only penalises downside variance. MT5 doesn't report Sortino natively; export the equity curve to Excel/Python and compute it for serious analysis.

   Also: Sharpe is only meaningful over 1+ year of data. Short backtests give Sharpe values that look impressive but reflect noise.

4. 第 4 步:Recovery Factor — return per unit of drawdown

   Recovery Factor = net_profit / max_drawdown. Conceptually: 'how many times did the EA recover from its worst drawdown'.

   Interpretation: • RF < 1.0 — the EA's worst drawdown exceeds its total profit. Unacceptable. • 1.0–2.0 — marginal. The EA spends most of its life recovering from drawdowns. • 2.0–5.0 — acceptable. Reasonable cushion. • 5.0–10.0 — strong. • 10.0+ — excellent.

   Recovery Factor is the metric that captures 'how comfortable would I be running this EA'. An EA with 50% profit and 25% drawdown (RF 2.0) and an EA with 30% profit and 5% drawdown (RF 6.0) — most traders prefer the latter despite the lower absolute profit.

   For multi-EA portfolios, use Recovery Factor as the position-sizing metric: allocate capital proportional to RF, not to absolute profit.

5. 第 5 步:Win Rate alone is misleading — pair with Avg Win/Loss

   Win Rate = winning_trades / total_trades. By itself, win rate tells you nothing about profitability — an EA with 90% win rate and 1-pip wins and 50-pip losses still loses money.

   The correct pair: Win Rate × Average Win/Loss Ratio.

   Avg Win/Loss Ratio = average_winning_trade / average_losing_trade. Sometimes called Payoff Ratio.

   Breakeven combinations: • 30% win × 2.5 R:R (Avg Win/Loss) = profitable • 50% win × 1.0 R:R = breakeven • 70% win × 0.5 R:R = profitable • 80% win × 0.3 R:R = barely breakeven

   Trend-following EAs typically: 30–40% win × 2.5–4.0 R:R. Big winners, frequent small losers. Scalping EAs typically: 60–80% win × 0.5–1.0 R:R. Many small winners, occasional large losers. Both profiles can be profitable; neither is inherently better.

   Treat Win Rate as a personality trait of the EA, not a quality metric. Some traders psychologically need high win rate (scalpers); others tolerate low win rate (trend followers).

6. 第 6 步:Expected Payoff — average $ per trade

   Expected Payoff = net_profit / total_trades. The average $ profit (or loss) per trade across the entire backtest.

   Use this to estimate live profitability: if Expected Payoff is $4.50 and your EA averages 200 trades per month, you can expect roughly $900/month gross profit in live (assuming live execution matches backtest, which usually it doesn't perfectly).

   More importantly, compare Expected Payoff to spread + commission cost per round-turn trade. If Expected Payoff is $5 but your broker charges $4 in commission and 1 pip spread (≈$10 on a 0.1-lot trade), the EA is unprofitable after costs even though the backtest shows positive expected payoff. MT5 backtests can include broker costs if you configured them correctly (see backtest-ea-mt5 guide); always verify.

   Negative Expected Payoff with positive Net Profit is impossible mathematically — they have the same sign by definition. If you see this in a report, you're looking at two different backtests.

7. 第 7 步:Total Trades — statistical credibility floor

   Total Trades determines the statistical reliability of every other metric. Rules of thumb:

   • < 30 trades — meaningless. Win Rate could be 50% true and 80% in the backtest by pure chance. • 30–100 — directional but not reliable. Use for early validation only. • 100–500 — credible. Most metrics are within ±20% of their true value. • 500–2000 — robust. Metrics close to true values. • 2000+ — bulletproof. Any divergence between backtest and live now strongly suggests an execution or environmental difference, not statistical noise.

   For a serious live-money decision, target 500+ trades over 3+ years. If your EA's natural trade frequency is too low to hit 500 trades in 3 years (e.g. quarterly swing strategy), use a longer backtest period or accept that you have lower statistical confidence.

   This is also why Strategy Tester's 'optimization on 6 months' for a low-frequency EA is dangerous: 30–50 trades in 6 months produces metrics that are essentially noise.

8. 第 8 步:Always visually inspect the equity curve

   Summary metrics can hide bad behaviour. The Strategy Tester's Graph tab shows the equity curve over time — look at it. Things to spot:

   • Stair-step pattern with sudden drops — the EA has cliff-risk trades (martingale style: many small wins, occasional huge loss). Even if metrics look OK, the cliff trades will eventually take an outsized hit live.

   • Equity flat-line for months — the EA stopped trading for a regime it doesn't handle (e.g. low volatility). Live, this is fine, but you need to budget for these dead periods.

   • Equity rapidly climbs then crashes — late in the test period the EA's edge degraded. Look for what changed: structural market shift, EA logic flaw, broker condition change.

   • Equity grows linearly with low volatility — likely a healthy EA. Confirm with metrics.

   • Equity grows exponentially — usually a sizing artifact (compounding aggressive risk). Re-run with fixed-lot sizing to see the un-compounded version.

   An equity curve a trader would feel comfortable holding through is the real validation. If the chart makes you nervous looking at it, that's signal even when the metrics check out.

需要避免的常见错误

• ✗ Treating Net Profit as the primary metric解决方法: Net Profit scales with starting balance and lot size. Use Profit Factor and Recovery Factor as primary, Net Profit as a sanity check.
• ✗ Comparing two EAs with different starting balances by absolute profit解决方法: Convert to percentage. $5000 profit on $10k starting = 50%; on $100k = 5%. The metric is return %, not dollars.
• ✗ Trusting metrics computed on < 100 trades解决方法: Statistical noise floor. Either backtest longer or accept lower confidence.
• ✗ Optimizing for Win Rate without checking Avg Win/Loss解决方法: A 90% win rate with 10:1 negative skew (every loser is 10× a winner) loses money. Use Profit Factor instead.
• ✗ Ignoring Drawdown Duration解决方法: Max DD says how deep; Duration says how long. A 15% DD that lasts 18 months is much harder to hold through than a 25% DD that recovers in 2 months.
• ✗ Believing Sharpe > 5解决方法: Sharpe > 3 on retail forex EAs is almost always backtest artifact (short period, low trade count). Re-test with longer history before believing.

常见问题