How to read a backtest, and why not to take its word
A backtest answers one question: what would have happened had these settings run in the past. It does not answer “what will happen”.
What a backtest does
The app takes closed candles for the pair and runs your grid over them. A decision is made on one candle and filled on the next, with fees and slippage. That is a deliberately pessimistic estimate: in life the order might have filled a little sooner and a little better.
What the figures mean
- Net profit and return — the result for the period, fees already taken out.
- Max drawdown — the deepest fall in the account’s value from a peak. The figure that answers “could I sit through this”.
- Closed trades — how many times the bot bought and sold. The fewer there are, the less everything else weighs.
- Win rate — for a grid it is nearly always about 100%: the bot does not sell at a loss, so there simply are no losing closed trades. A loss, if there is one, sits in the unsold coins.
- Profit factor and Sharpe — profit against losses, and against how much the result swings. When there is nothing to compute them from, the app shows a dash, not a handsome number.
- Fees vs gross profit — the share of the result the exchange took. The higher it is, the less sense the settings make.
- Time in market — the share of the period the bot held coins.
When the figures cannot be trusted
First, few trades. With fewer than thirty closed trades the app says outright that there are too few: over a handful of trades, win rate, profit factor and Sharpe describe that sample, not the strategy.
Second, the period. A backtest over a month of rising prices says all is well. The same settings over a falling month show money tied up and a drawdown. Run them over different periods and timeframes.
Third, fitting. Turn the parameters long enough for the history to look good and you get settings that fit the past perfectly. They say nothing about the future.
Using a backtest well
- Compare settings with each other on the same data rather than trusting an absolute figure.
- Look at the drawdown before the profit.
- After the backtest, run the same settings on paper: a backtest sees candles, paper mode sees the live market.
The point
Past results, simulated or real, say nothing about future ones. A backtest helps throw out plainly bad settings. It cannot find “profitable” ones.
Trading cryptocurrency carries a high risk and may result in the partial or total loss of funds. Nothing on this site is investment advice. Nexora is software, not a broker or an asset manager.