How to read a backtest: profit, drawdown and trades
Read profit factor, losses, sample size and costs together. A numerical example shows why win rate alone is not enough.
Try this example →Positive net profit starts a review; it does not finish one. Two strategies can end with the same capital after taking very different risks. Read a report in this order: context, trades, losses and sensitivity to execution assumptions.
Start with what was tested
Record the symbol, data source, interval and dates. Check starting capital, position sizing and configured costs. Synthetic candles are useful for learning; historical market data describes a simulation of that particular period.
In the Deepwick demo, open Strategy Tester. The curve and trade table belong to the script that actually ran. After editing it, press Run to refresh the result.
Net profit and profit factor answer different questions
Net profit summarizes what the simulation gained or lost after the costs applied by that engine and configuration. Check whether positions remain open and how the report handles them before equating the figure with an account balance.
Profit factor divides total gains by the absolute value of total losses. For an arithmetic example, gains of 1,200 and losses of 800 produce a factor of 1.5. That does not mean a 50% return on capital or a 50% chance of success. Profit factor definition.
Without losing trades, the denominator is zero. An infinite or missing value in a small sample does not establish low risk.
Win rate needs the size of wins and losses
These are hypothetical numbers, not a published backtest:
| Outcome | Count | Average | Total |
|---|---|---|---|
| Winning trades | 6 | 100 | 600 |
| Losing trades | 4 | −200 | −800 |
| Balance before additional costs | 10 | — | −200 |
The win rate is 60%, yet the balance is negative. Read «Win rate», «Avg win / loss» and «Trades» together. None replaces the others.
Drawdown measures a fall from a previous peak
If an equity curve rises from 10,000 to 12,000 and then falls to 9,000, its drop from that peak is 3,000, or 25%. A later recovery does not remove that experience from the path.
Inspect the curve alongside maximum drawdown. Recovery time and the concentration of losses matter too. Historical drawdown is not a guaranteed ceiling for future losses. Platforms may value the curve at different points, so review their definitions before comparing reports.
How many trades support the result?
An exceptional winner can dominate a small sample. Check the trade count, when trades occurred and the contribution of the largest winners. Several closely spaced entries can respond to the same market episode; they need not be independent observations.
There is no universal trade count that makes a strategy reliable. A Sharpe or profit-factor threshold cannot validate it on its own either. Record what the experiment demonstrated and what remains untested.
Run a second experiment that could contradict you
Keep the code fixed and explicitly change one execution assumption, such as configured commission or slippage. Then evaluate a period you did not use to select the strategy. Changing the code to fit that period turns the test into another development session.
In Deepwick, continue with a saved backtest and walk-forward validation. Compare like-for-like configurations: the demo, saved reports and validation can use different cost assumptions.
A report becomes useful when you can explain which rules produced the trades, which costs were simulated and under what conditions you would stop trusting the hypothesis.