Trading basicsUpdated 3 min readBy the FXNOD team
How to evaluate a trading strategy before using it
Evaluate a trading strategy by testing it over at least a hundred trades and measuring its win rate, average win, average loss, expected value per trade, maximum drawdown and longest losing streak. A strategy is worth using only if its expected value is positive after costs and you could sit through its worst drawdown without changing the rules.
First: can it be tested at all?
A strategy you can evaluate has exact rules: when to enter, how much to stake, when to exit. If two people following it would take different trades, it is a style, and a style cannot be measured. Write the rules down until there is no judgment left in them.
The seven measures
| Measure | How to calculate it | What to look for |
|---|---|---|
| Sample size | Number of trades in the test | At least 100; several hundred if wins or losses are rare |
| Win rate | Wins divided by trades | Above the break-even rate for its payout |
| Average win and loss | Total won over wins; total lost over losses | Read together with the win rate |
| Expected value | Win rate x average win minus loss rate x average loss | Positive after costs |
| Profit factor | Total won divided by total lost | Above 1; a figure far above 2 on a small sample is suspicious |
| Maximum drawdown | Largest fall from a peak in the balance to the next low | Small enough that you would keep going |
| Longest losing streak | Count from the trade list | Your stake sizing must survive it, and a longer one |
A worked example
A test of 200 trades at a flat stake of 1 gives 112 wins with an average win of 0.92 and 88 losses of 1.00. Win rate: 56%. Total won: 103.04. Total lost: 88. Net: 15.04. Expected value: about 0.075 per trade. Profit factor: 1.17. Break-even win rate for that payout: 1 divided by 1.92, or 52.1%. The strategy cleared break-even by about four points in this sample, which is encouraging and not yet proof. Another 200 trades are worth more than any amount of further thinking.
Four ways a test lies
- Too few trades. Over 20 trades luck dominates everything.
- Curve fitting. Settings adjusted until the past looks perfect have learned the past, including its accidents.
- A hidden tail. Recovery strategies such as Martingale show a smooth rising line until the streak that ends them. A test that has not met that streak has not tested the strategy.
- One market condition. A result from a single trending week says little about a sideways month.
Questions to ask of someone else's strategy
- How many trades is this result based on?
- What was the largest drawdown, and the largest single loss?
- Does the stake grow after a loss?
- Can I run it myself on a demo account before paying or depositing anything?
Testing it yourself
Forward testing on a demo account is the simplest honest test: live prices, virtual money, fixed settings. In FXNOD you can build the rules in Bot Forge, run them on a Deriv demo account and read the results from the run history. Keep the settings unchanged for the whole test, and decide in advance what result would make you abandon the idea.
Questions people ask
- How many trades do I need to evaluate a strategy?
- A hundred is the minimum to say anything. Several hundred are needed when the strategy wins very often or very rarely, because its result then hangs on a few rare trades.
- What is a good profit factor?
- Above 1 means the strategy made money in the test. Between about 1.2 and 2 over a large sample is a realistic good result. Much higher on a small sample usually means luck or curve fitting.
- What is maximum drawdown?
- The largest fall in the balance from a high point to the low that followed. It is the best single measure of how painful a strategy is to follow.
- Does a strategy that worked in the past keep working?
- There is no guarantee. A test describes the period it covered. Keep measuring a live strategy and stop when its results fall well outside what the test showed.
Practise it with virtual funds
Trade Pilot gives you a live chart and Deriv's contracts on your demo account. There is nothing to install; check the required package in Discover.
Get started