How to read a backtest report
TRADINGPROFITTEVOLE · FREE GUIDE 05/05
How to read a backtest report
Profit, drawdown, recovery, risk and data quality: a practical guide to interpreting EA Lab reports critically.
Whenever I open a MetaTrader report, the first temptation is to look at final profit. I know the appeal of a rising curve. But how much risk and time did that result require? Here I explain the terms used in my EA Lab reviews with deliberately simple examples.
1. Test conditions come before results
I begin with the platform (MT4 or MT5), the EA version, symbol, exact testing dates, initial deposit, account currency, leverage, volume, price model and broker. Without them, two identical profits can tell entirely different stories.
Fixed volume of 0.01 lots. Where possible, I use a small fixed lot and disable dynamic money management to make comparisons between datasets and robots clearer. But 0.01 lots on XAUUSD, EURUSD or a stock index do not imply identical risk: tick values, contracts, volatility and stops differ.
Fixed and floating spreads. Spread is the difference between ask and bid. In a Tickstory test I may fix it at, say, 25 points for a reproducible entry cost, while an IC Markets broker-history test uses variable market spreads. Commissions, slippage and swaps may also change the outcome.
Historical data quality. A “99%” quality figure in a tester describes the simulation’s reported data quality; it is not an independent certification of an Expert Advisor. I want to know the source, completeness, gaps and symbol specifications.
2. Net profit and average results
Net profit is the final trading result after commissions and swaps included in the report. If an account grows from €1,000 to €1,400, the net profit is €400 (+40% of starting capital), assuming no deposits or withdrawals.
Monthly and yearly profit. In EA Lab I allocate transactions to their closing calendar month or year. The displayed period percentage is always profit for that period / starting deposit × 100. A €50 gain with €1,000 initial capital is reported as +5% even if the balance later grew to €3,000: this is not the percentage return on that month’s actual balance.
Average monthly and annual result. I divide total net profit by observed months and multiply the monthly average by 12 for an annual equivalent. €1,200 over 24 months equals €50 a month or €600 a year, namely +5% and +60% relative to an initial €1,000. These are neither compounded returns nor future forecasts; incomplete years must be identified.
3. Profit factor, winners and losers
Profit factor (PF) equals gross profit divided by the absolute value of gross loss. €3,000 of gains against €1,500 of losses gives PF = 2.00. PF above 1 describes a positive gross outcome in that sample; it does not settle questions of trade count, consistency or execution quality.
Win rate equals winning trades divided by closed trades, multiplied by 100. Winning 85 of 100 trades is an 85% win rate, but a strategy can still lose money when its losing trades are much larger than its winners. Average winning and losing trades expose this asymmetry.
Trade count and holding time. These reveal how much evidence we have and how long capital may be tied up. A position held for 20 days faces different swap, rollover and market-event exposure from a 20-minute trade.
4. Balance, equity and three drawdown measures
Balance reflects completed trades. Equity includes unrealised gains and losses from open positions. A rising balance curve can coexist with a much lower intratrade equity level.
Absolute drawdown in the tester is the shortfall from starting capital to the lowest level below that starting point. Maximum monetary drawdown is the largest currency decline from one peak to a subsequent trough; the percentage shown beside it refers to that episode. Relative drawdown is the largest percentage decline from its own peak, which may be a different event.
Example: a decline from €2,000 to €1,800 is €200 or 10%, while a fall from €5,000 to €4,600 is €400 or 8%. Maximum drawdown in euros is €400, but maximum relative drawdown is 10%. They do not describe the same event.
5. Time spent in drawdown and recovery
Drawdown is not only about how far an equity or balance curve falls; duration matters too. I define total drawdown duration as the time from a previous peak until the first recovery of that peak. Recovery time from the trough is the period between the lowest point and the recovery of the prior peak.
Example: balance €3,000 on 1 January, €2,800 on 20 January, and back to €3,000 on 10 February. That is a €200 drawdown (6.67%), lasting approximately 40 days from the peak and around 21 days from the trough. An episode not yet recovered must be marked as open; it is not a zero-day recovery.
To prevent thousands of tiny fluctuations distorting average duration, I also look at declines amounting to at least 50% of the maximum balance drawdown and measure durations within that subset. Deal history may support balance-based timing. Intratrade equity timing requires sufficiently detailed equity observations, which are not always available in MetaTrader HTML reports.
6. Recovery factor, return/risk and Sharpe
Recovery factor in MetaTrader compares net profit with maximum monetary drawdown: net profit / maximum monetary DD. A €1,000 gain with a €250 maximum drawdown gives a factor of 4.00. It is useful for describing historical reward relative to temporary loss, but does not measure the time required to recover or the worst possible future loss.
Sharpe ratio relates returns to their variability. Larger values imply greater consistency under the calculation used, but MT5’s reported Sharpe depends on the tester methodology and return series. I do not automatically equate it with values calculated by unrelated sources or traditional funds.
Single-trade reward/risk is separate: a potential €100 target against a €50 stop has a ratio of 2:1. That is not the recovery factor of an entire backtest.
7. Distribution by time and MFE/MAE
Hourly, daily and monthly distributions help reveal when an EA trades and whether its historical results are widespread or concentrated in a handful of periods. A historically profitable hour is not automatically the “best” hour for future trading; it may simply reflect past conditions.
MFE (Maximum Favorable Excursion) is the largest favourable move while a position was open; MAE (Maximum Adverse Excursion) is the largest adverse move. A trade that closes at +€10 after reaching +€50 and −€20 has MFE = €50 and MAE = −€20. These measures help analyse stops and exits, but do not prove those hypothetical values were executable.
8. Comparability, robustness and real-world risk
My first filter is a double backtest: the same EA, input set and fixed lot size, tested once on broker history with floating spread and once on documented tick data, potentially with a fixed spread. Unexplained material differences are a warning sign; even agreement between two tests remains historical evidence, not a prediction.
Next I look for an untouched out-of-sample period, higher-cost stress tests (spread and slippage) and reasonable parameter stability. When these checks have not been performed, I say so.
When comparing historical monetary drawdown with the capital required, I keep lot size constant. For example, €240 of drawdown represents 24% of €1,000, 8% of €3,000 and 4.8% of €5,000. This is arithmetic, not a forecast or guarantee. Margin, liquidity and live execution may change the result substantially.
“I would rather study an older Expert Advisor that still works under present conditions than trust a newly published one purely because its historical curve looks perfect.”
This principle comes from observing many attractive backtests. Far fewer systems withstand comparison across datasets; fewer still justify a cautious demo and forward-test period, and only some merit a carefully limited live trial. A product’s age proves nothing on its own—what matters is coherent behaviour under new conditions.
A practical comparison: Gold Beaver v5.57
To see how I apply these criteria to historical reports, visit the EA Lab archive in English. The detailed Gold Beaver v5.57 case study is currently available in Italian.
For education only. Trading may result in losses, potentially greater than the losses observed in historical simulations. This is not personalised investment advice, and each trader remains responsible for their own decisions.