Backtesting & Trading Journal: Test Your Forex Strategy

Why Test Before You Trade

A strategy is a hypothesis: “when I see X, doing Y has a positive expected result after costs.” Backtesting checks that hypothesis against history, and a trading journal checks it against your own live execution. Neither can prove a method will work in future, because markets change. What they can do is reject bad ideas cheaply, show you what normal losing streaks look like, and give you a baseline to compare live results against. Trading without either means you are paying the market to run the experiment for you.

Manual Versus Automated Backtesting

Manual backtesting means scrolling a chart back, advancing it candle by candle, and recording every trade your rules would have taken. It is slow, but it builds screen time and it works for discretionary methods that are hard to code, such as price action setups. Its weakness is hindsight: it is very easy to skip the losing trade you “would not have taken”. Hide the future candles and log every valid signal, no exceptions.

Automated backtesting runs coded rules over years of data in seconds. It removes hindsight bias and makes large samples easy, but it introduces its own traps: poor-quality data, unrealistic fills, fixed spreads that ignore news-time widening, and the temptation to optimise endlessly. Whichever route you take, include realistic costs. Spread, commission and swap can turn a thin edge negative, so be sure you understand spreads and pips before trusting any result.

For the step-by-step mechanics of MetaTrader, including modelling modes, spread settings and how to read the report, see how to backtest an EA in the Strategy Tester. On TradingView, the browser-based charting platform, a Pine strategy gets a similar report; try strategies without repainting flags and check the cost settings, because repainting scripts look better in history than live.

Sample Size and Expectancy

Twenty trades tell you almost nothing; luck dominates small samples. A common rule of thumb is at least 100 trades covering different conditions — trending, ranging, quiet and volatile — and more for methods with a low win rate, whose results are lumpier.

The key output is expectancy, the average amount you make or lose per trade:

E = (win% × average win) − (loss% × average loss)

Take a hypothetical test of 120 trades: 48 winners (40%) and 72 losers (60%), with an average win of 180 USD and an average loss of 100 USD. E = (0.40 × 180) − (0.60 × 100) = 72 − 60 = 12 USD per trade. Over 120 trades that is 1,440 USD. In risk units, with 100 USD as 1R, it is 0.12R per trade. Note that the system loses more often than it wins and is still positive.

Now flip it. A method that wins 70% of the time with an average win of 50 USD and an average loss of 150 USD gives E = (0.70 × 50) − (0.30 × 150) = 35 − 45 = −10 USD per trade. A high win rate with poor risk-to-reward loses money. Margins matter too: if the first system’s costs were understated by just 5 USD a trade, its expectancy falls from 12 to 7 USD, a drop of more than 40%.

Overfitting and Curve Fitting

Overfitting means tuning rules until they describe the past perfectly — and the future not at all. With enough parameters you can make any dataset look profitable. Warning signs:

• Oddly specific values, such as an RSI threshold of 37 or a 23-period average, with no logic behind them.
• Results that collapse when a parameter shifts slightly. If a 20-period setting works but 18 and 22 fail, you have found noise.
• Filters added one by one to remove specific historical losers.
• An equity curve that looks too smooth to be real.

The defence is simplicity: few rules, each with a reason you can explain, that hold up across neighbouring parameter values and ideally across several pairs.

Out-Of-Sample and Forward Testing

Split your data before you begin. Develop the rules on one portion — say the first five years — and keep the most recent two years untouched. When development is finished, run the rules once on the unseen portion. If performance holds up reasonably, that is encouraging; if it falls apart, the method was fitted to the first period. Be honest about the word “once”: every time you tweak after peeking, the out-of-sample data quietly becomes in-sample.

A public forward test looks like ForexR Signals: each trade published before its entry, then tracked and kept, losses included. How to read a signal track record explains how to judge one.

Then forward test on a demo account or at minimum size for a few months, or 30 to 50 trades. This is the first time live spreads, slippage and your own hesitation enter the results.

What a Journal Records, and When to Review It

The journal is a forward test that never ends. For each trade record: date, time and session; pair and direction; setup name; entry, stop and target; planned risk-to-reward; position size; result in R and in money; before and after screenshots; whether you followed your rules; and a line on your emotional state, which ties in with trading psychology.

Review on three cadences. Daily, spend five minutes checking rule adherence. Weekly, go through screenshots and tag mistakes. Monthly, calculate expectancy by setup, pair and session, compare it with the backtest, and use the drawdown calculator to see what any losing run has cost and the gain needed to recover. Change one thing at a time.

Past performance, tested or live, does not guarantee future results. Forex and CFDs carry a high risk of loss; only risk money you can afford to lose.