How to Read and Research an MT5 Backtest Report: A Field Guide for EA Buyers and Testers

Read it like an auditor, not a fan

When you open a report, your brain looks for the answer to one question: did it make money? That question is the least useful one. Almost any strategy can be made to make money in a backtest, because you can choose the symbol, the period, the inputs and the data. A better question is: under which conditions would this result stop being true?

An auditor reads a set of accounts by looking for what does not add up. You will do the same with a Strategy Tester report. You will check that the setup is honest, that the risk is survivable, that the edge comes from many trades and many months rather than a few lucky ones, and that the result does not collapse when you nudge a parameter or increase costs. If the strategy survives all of that, it has earned a place on a demo account. It has not earned your trust yet.

5 steps

Setup, risk, trade quality, robustness, forward check

1 rule

Never judge a strategy by net profit alone

0 proof

A backtest cannot guarantee future performance

This guide uses the report produced by the MetaTrader 5 Strategy Tester, but the logic applies to reports from other platforms, and to the PDF or HTML reports that EA sellers publish. If you have not yet run a test of your own, start with our step-by-step backtesting guide, then come back here to interpret the result.

The anatomy of an MT5 report

The report has a fixed structure. Knowing where each block lives saves time and keeps you from jumping straight to the headline figure. The diagram below shows the five regions you will read, in order.

In the tester itself, the results appear across several tabs at the bottom: the summary of results, a graph, the deals and orders, and the journal and agent logs. You can export the report as HTML from the context menu on the results tab. Exporting is worth doing, because it freezes the exact settings used and lets you compare runs side by side later.

Step 1: Audit the test setup before the results

The top of the report lists the conditions of the test. This block decides whether everything below it can be trusted. Spend your first two minutes here, and refuse to read further if any item fails.

Next, look at the data block. MT5 reports the amount of history used: the bars and ticks processed, and a history quality percentage. Treat these as sanity checks. A very low tick count for a multi-year period suggests gaps. A quality percentage that reads high does not prove the data resembles your broker's live prices, which is a point we cover in the tick data guide.

If you are reading a seller's published report rather than your own, you often cannot see all of this. That in itself is information. A seller who hides the date range, the spread assumption or the modelling mode is asking you to take the result on trust. Ask for the full report, or run the test yourself.

Step 2: Read the headline numbers, in the right order

Now you may look at results. Resist the urge to start with total net profit. Start with how much the strategy risked to earn it. The table below lists the main figures in the profit and risk block, what each one measures, and how a sceptic reads it.

Notice the pattern. Balance and equity versions of drawdown appear side by side for a reason. Balance drawdown only records closed losses. A grid or averaging system can sit on an enormous floating loss that never shows in the balance until it is closed, so the equity figure is the honest one.

Step 3: Understand drawdown properly

Drawdown is the distance between the equity peak and the lowest point afterwards. Depth is only half of the story. The other half is duration: how long the account stayed below its previous high. A 15% drawdown that heals in three weeks is a different experience from a 15% drawdown that lasts fourteen months, and the second is the one that makes people switch off a sound system at the bottom.

To read the underwater view from a real report, export the deals to a spreadsheet and compute the running peak of equity. You can also read it from the graph tab: find the longest stretch between two equity highs, and ask yourself whether you could keep your hands off the system for that long without cutting risk or stopping it.

Then apply the arithmetic of recovery. Losses are not symmetrical. After a 50% loss you need a 100% gain to return to the starting level. Use the calculator below to see the gain required for any drawdown, and how many consecutive losing trades it would take to reach it at your risk per trade.

Next, compare the backtest drawdown with a realistic worst case. The maximal drawdown in a backtest is the worst thing that happened in one specific history. The future can contain something worse, and most professionals assume the live maximum will exceed the tested maximum, sometimes by half again or more. If the backtested drawdown already hurts you on paper, the live one will hurt more.

• Depth: is the maximal equity drawdown within what you can stomach, doubled?
• Duration: what is the longest time between equity highs, and could you wait that long?
• Frequency: how many drawdowns of at least half the maximum occurred? One in five years is different from one every six months.
• Cause: was the drawdown a run of ordinary losses, or one position growing out of control?

Step 4: Judge the quality of the trades

A strategy makes money when the average trade is positive after costs. That average is called expectancy, and it depends on three things: how often you win, how much you win, and how much you lose. Beginners fixate on the first. A 90% win rate sounds wonderful, until you learn that the average loss is twenty times the average win.

Try the extremes in the calculator. Set the win rate to 90, the average win to 1 and the average loss to 10. The system wins nine times out of ten and still loses money on average. Now set the win rate to 35, the average win to 3 and the average loss to 1. The system loses most of its trades and is clearly profitable. Win rate describes how a strategy feels. Expectancy describes whether it works.

The histogram above shows why the shape of the distribution matters. System A has many small losses and a long tail of larger wins. That is typical of trend following and breakout approaches. System B wins nearly every trade and carries a very rare, very large loss. That is the signature of martingale and uncapped grid systems. In a backtest of limited length, System B can look better than System A for years.

To check the shape in your own report, read these fields together:

• Largest profit trade and largest loss trade. If the largest loss is many times the average loss, the distribution has a heavy tail.
• Average profit trade and average loss trade. Their ratio is the payoff ratio. Combine it with the win rate to compute expectancy by hand.
• Maximum consecutive losses (count and amount). Compare the loss streak with your own risk limit. A system that can lose eleven in a row needs position sizing that survives eleven.
• Long and short trade results. A big imbalance suggests the test period favoured one direction, for example a long gold rally. The strategy may not survive a reversal.
• Profit trades as a percentage of total. Use it only to compute expectancy, never as a quality score by itself.

Step 5: Ask whether the sample is large and varied enough

Statistics reward patience. A result built on thirty trades can be reproduced by a coin that is slightly biased, or simply a lucky streak. As the number of trades grows, the average settles and the noise shrinks. There is no universal threshold, but as a working rule, treat fewer than a hundred trades as an anecdote, a few hundred as the minimum for a first opinion, and more than a thousand as a decent sample for a high-frequency strategy.

Quantity is only part of it. Variety matters as much. Five hundred trades taken during one gold bull run still describe one regime. Ask whether the sample spans rising, falling and sideways markets, quiet and volatile years, and different parts of the trading day.

A monthly heat map is the quickest way to check this. Export the deals, group profit by month, and colour the cells. You are looking for three things. First, how many months are positive, and by how much. Second, whether the total depends on one or two outstanding months, which would mean you are betting on a rare event recurring. Third, whether there are periods, such as a particular quarter or volatility regime, where the strategy consistently loses. Those are the conditions in which you should expect to suffer, and perhaps switch the system off.

Step 6: Open the deals list and look for hidden risk

The summary can hide how profits were made, but the deals list cannot. Export it and scan it with the specific goal of finding the behaviours that blow up accounts. This is the step that separates a careful reader from a casual one, and it takes about ten minutes.

1. Lot size progression. Sort by time and look at the volume column. Lot sizes that rise after losses (1.0, 2.0, 4.0 and so on) mean martingale or recovery logic.
2. Number of simultaneous positions. If many positions in the same direction stay open together, it is a grid or averaging system. Check the maximum number of open positions and whether the EA caps it.
3. Holding time of losers versus winners. Winners closed in minutes while losers are held for days is a sign of a system that refuses to take losses until forced.
4. Presence of stop losses. Look at whether every deal closes at a stop, a target, or an exit signal. Positions closed by stop-out or at huge losses are the warning.
5. Trading around rollover and news. Check whether results around the daily rollover or big releases look different. Wide spreads there can turn a profitable strategy into a losing one.
6. Time-of-day and weekday concentration. If almost all profit comes from one session or one weekday, the edge is specific, and you should check it against your broker's server time.

Do not skip the biggest losses. Sort the deals by profit and read the ten worst. Ask what the market was doing, how long the position was open, how large it was relative to the others, and whether it could have been worse had the test period contained slightly different data. A system whose worst trade is less than three times its average loss is behaving very differently from one whose worst trade equals months of profit.

Step 7: Test whether the result survives change

A real edge is robust. It survives a small change in parameters, a small change in costs, a different data feed and a different period. A fitted one does not. The final stage of research is to deliberately disturb the result and see whether it holds.

Parameter sensitivity: look for plateaus, not peaks

If you have access to the optimisation results, plot the profit against two parameters. A robust setting sits in the middle of a broad region where neighbouring values also work. A fitted setting sits on a lone spike surrounded by losses.

Cost and delay stress

Run the same test three times: the base case, then with spreads about 30% higher, then with the delay doubled. A robust system degrades gracefully. A fragile one flips from profit to loss, which tells you the edge was smaller than the friction all along. Scalpers and breakout systems are the most sensitive, and their reports deserve the harshest stress.

Different data, different period, different symbol

Repeat the run on a second data source, such as your broker's real ticks against an imported Dukascopy-based custom symbol. Expect small differences in trade count and profit. Large differences mean the result depends on the quirks of one feed. Then test a related symbol if the logic is generic, or a period the optimiser did not see. Positive results across independent tests are much stronger evidence than one spectacular run.

Walk-forward and Monte Carlo thinking

Walk-forward analysis optimises on one window, tests on the next unseen window, then slides forward and repeats. If out-of-sample windows keep profiting, the edge is more likely to be real. A simple Monte Carlo check shuffles the order of trades or randomly skips a fraction of them, and shows how deep and how long the drawdown could be with the same average results. MT5 does not provide this natively, but exporting the trade list to a spreadsheet or a short script makes it easy.

A worked example: two reports with the same net profit

Numbers make the method concrete. Below are the headline figures from two hypothetical reports. Both come from a five-year test on a 10,000 deposit, and both show a net profit of about 6,000. A casual reader would call them equal. A careful one would not.

Start with expectancy. Report X wins 37% of the time. Its expectancy is 0.37 multiplied by 93 minus 0.63 multiplied by 40, which is about 9.2 per trade. Report Y wins 91% of the time, so its expectancy is 0.91 multiplied by 15 minus 0.09 multiplied by 92, which is about 5.4 per trade. Y's profit factor is higher and its win rate is dazzling, but its average loss is six times its average win.

Now read the risk. Report Y's balance drawdown is only 6%, yet its equity drawdown is 41%. The gap says the system carries large floating losses, and the maximum of fourteen open positions says why: it averages into losing trades. Its largest single loss of 2,400 is over a third of its entire net profit. One more bad sequence, or a slightly different data feed, could remove years of gains in a week.

Report X looks ordinary by comparison. Its win rate is low, its curve will be bumpier, and its drawdown is moderate on both measures. It never has more than one position open, and its worst loss is small relative to its profit. If the remaining checks pass, Report X is the stronger candidate, even though Report Y has the prettier headline and the higher profit factor. This is exactly the case where reading the whole report changes the decision.

Five common misreadings

1. Treating profit factor as a quality score. It rises when wins are small and frequent, which is exactly how tail-risk strategies look. Use it only alongside the largest loss and the equity drawdown.
2. Reading the Sharpe ratio as a grade. The figure depends on how returns are sampled and on the period. Compare it between similar systems, and do not rely on it as a pass mark.
3. Trusting a smooth balance curve. Balance records closed trades only. Check the equity line and the underwater view.
4. Confusing history quality with realism. A high percentage means bars and ticks were available. It does not mean spreads, fills and slippage match your broker.
5. Ignoring the time dimension. A report that covers a short window of sixty trades cannot show whether the system survives a regime change, however good the numbers look.

Check the numbers yourself in a spreadsheet

You do not need specialist software to verify a report. Export the deals list to a CSV file from the tester or from the HTML report, open it in a spreadsheet, and build a small sheet with four calculations. The effort is about an hour the first time and minutes afterwards.

• Cumulative profit and running peak. Add a column of cumulative profit, then a second column with the maximum so far. The difference between them is the drawdown at every trade, and the longest run of non-zero values is the time under water.
• Profit by month. Use a pivot table on the close time. This gives you the monthly heat map, the number of positive months and the share of profit from the best months.
• Expectancy and payoff ratio. Average the winning trades and the losing trades separately, then combine them with the win rate. Compare the result with your estimated spread and commission per trade.
• Position sizing check. Plot volume against time. A staircase pattern after losses is martingale. A cluster of rising volumes inside a single day is grid averaging.

Once you have these four views, you no longer depend on the summary the report chose to show. The same sheet can be reused for every EA you test, and it makes comparisons fair, because every candidate is measured the same way.

Red flags that should end the research early

You will not need every step for every report. Some signs are serious enough to stop right away. Each of the following has an innocent explanation occasionally, but in combination they are a strong reason to walk away or to ask hard questions.

Reading a seller's published report

Sellers, including us, publish reports and screenshots because buyers want to see something. Treat any seller's figures as a starting point for questions rather than as evidence. A good report answers most of the following without being asked, and a good seller answers the rest when you do ask.

• Which broker or data source, which modelling mode, which spread and delay, and which date range?
• Is the equity drawdown shown, and is the largest losing trade stated?
• Is the report from the exact version of the EA and the exact setfile you will receive?
• Is there a longer or different-period test, an out-of-sample window, or a verified live record?
• Does the manual state the strategy family and the conditions under which it is expected to struggle?

Then do the one thing no seller can do for you: run the setfile on your own broker's data over your own date range, and compare it with the published report. Differences are information. Honest sellers welcome the comparison, and you can use our installation and risk guide to set up the test environment properly.

From report to live: the forward check

A backtest tells you what the rules would have done. The only way to learn what they do now is to watch them. Put the validated EA on a demo account for several weeks and log the same statistics you read in the report: trades per week, average trade, win rate, payoff ratio, floating loss and drawdown. Compare them against the backtest.

Expect some difference. Fills, spreads and swaps will not match the tester exactly. What you are checking is whether the live numbers sit inside a reasonable range around the backtest. A useful habit is to write the range down before you start, for instance trades per week between a lower and upper limit, average spread paid, and the maximum floating loss you will tolerate. If the live result drifts well outside that range, stop and investigate before adding real money or size.

After demo comes a small live account, then slow scaling. Once running, keep reviewing weekly. A strategy that has been profitable for two years can stop working because the market regime changed. Pre-defined stop criteria, such as a live drawdown beyond the worst tested drawdown, save you from deciding under stress.

A repeatable research workflow

1. 1

   Step 1: Freeze the setup

   Record the EA version, setfile, symbol, period, modelling mode, spread, delay and data source. Export the HTML report.

2. 2

   Step 2: Audit the setup block

   Reject unrealistic settings. Rerun with real ticks, realistic costs and a delay if they are missing.

3. 3

   Step 3: Read risk before reward

   Check equity drawdown, its duration and margin level before you look at net profit.

4. 4

   Step 4: Compute expectancy by hand

   Combine win rate, average win and average loss. Compare the expected payoff with all-in costs.

5. 5

   Step 5: Check the spread of profits

   Build a monthly table and a trade distribution. Note whether a few events carry the result.

6. 6

   Step 6: Open the deals list

   Look for lot growth, stacked positions, long-held losers and rollover or news clusters.

7. 7

   Step 7: Stress and cross-check

   Add costs and delay, change parameters slightly, switch data source and test an unseen period.

8. 8

   Step 8: Forward test and set stop rules

   Run on demo with written expectations. Start live small, with a pre-agreed drawdown limit.