ForgeXBacktesting
What backtesting is, and what it can and cannot tell you
Backtesting means checking a trading idea against history before you risk money on it. Here is what it proves, what it does not, and where it misleads people.
Backtesting means taking a trading idea and checking it against what the market actually did. You go back in time, apply your rules, and write down what would have happened. The point is simple. You want to know whether the idea has ever worked before you find out with real money.
The basic idea
Say you think gold tends to reverse when it touches the previous day high during the London session. That is a belief. Right now it is worth nothing, because you have not checked it.
To backtest it, you go through months of gold charts, find every time price touched the previous day high in London, and record what happened next. After a few hundred of those you have a number instead of a feeling.
That number might tell you the idea works. It might tell you it does not. Both are useful. The one thing you cannot do is keep trading the idea without knowing.
Two ways to do it
Manual backtesting means you look at the chart yourself, bar by bar, and decide each trade. It is slower, but you see the market as it developed and you learn to recognise your setup.
Automated backtesting means you write the rules as code and let software run them over years of data in seconds. It is much faster, but it only works if your rules can be written down exactly. Most discretionary traders cannot do that, because part of their edge is judgement.
What a backtest proves
A backtest tells you how a set of rules would have performed on the data you tested. That is all. It is a measurement of the past, not a forecast.
That is still worth a great deal. It tells you whether the idea is worth pursuing, roughly how often you win, roughly how big your wins and losses are, and how bad the worst stretch was. Most of all it tells you what a normal bad run looks like, so you do not panic the first time you have one.
What it does not prove
It does not prove the idea will work tomorrow. Markets change. A pattern that paid for three years can stop paying and never come back.
It does not account for how you will behave. In a test you take every signal. In real trading you will skip some, size up after wins, and hold losers longer than you said you would.
And it does not account for costs unless you make it. Spread, commission and slippage turn plenty of profitable looking systems into losing ones. A test that ignores them is not a test.
The trap that catches most people
The biggest danger is not that backtesting is useless. It is that a bad backtest looks exactly like a good one.
If you test twenty small tweaks and keep the best, you have not found an edge. You have found the luckiest of twenty results. FTMO make the same point about adding conditions until the numbers improve, warning that it "can lead to curve-fitting—a strategy that works well on historical data but poorly in live trading" [1].
The fix is to test fewer things, on more data, and to keep some data aside that you never look at while you are building. That last part is covered in its own guide.
Sources
- FTMO Academy, "How to Backtest Trading Strategies". https://academy.ftmo.com/lesson/how-to-backtest-trading-strategies/