Reading results honestly
Backtests are the most misused tool in systematic trading. A Grade A result on Stretus is meaningful. And it is not a promise. This page is about the difference.
What the backtest is
Your strategy's exact entry and exit rules, run over real historical OHLCV data for the tested period, with every trade logged at the actual candle prices when the signal fired and with costs charged on both legs.
An earlier version of this page described the result as "an accurate simulation ... with perfect order execution and no slippage."
That undersold the platform in a way that could make you misjudge your own results. The engine models slippage, commission and applicable statutory charges. A user who mentally discounted the result for costs was double-counting.
What remains genuinely un-modelled is execution mechanics (candle-close fills, no partial fills, no latency, no queue position) and those are the right things to be sceptical about. See Backtest limitations.
What it is not
It is not future performance. Markets change regime. A momentum strategy that worked through a trend behaves differently in a range, and nothing in the strategy knows the regime changed. A grade says the rules performed a certain way on past data. It says nothing about persistence.
It is not fill-accurate. Fills come at candle boundaries. Live, you get real latency, real queue position, partial fills and a spread that widens exactly when you want out. On a 1-minute strategy, being filled 0.2% worse than the simulation across hundreds of trades changes the outcome materially.
It is not free of survivorship bias. The instrument universe is the universe that exists now. Equities that were delisted and tokens that went to zero during the window are absent from it.
It is not one era-independent test. Every backtest here draws on data from 1 January 2024 onward. That window contains particular conditions, and a strategy validated only on it has not been tested against anything outside it.
It is not a portfolio result, if you tested multiple symbols. Each symbol ran independently with the full capital. Two 20% returns are not a 20% portfolio return.
The four questions to ask before deploying
1. Are there enough trades to mean anything?
Under 20 and the rest of the page is noise. Sample size is one of five grading components. But it carries only five points, so a small-sample A is possible. Read the trade count yourself.
Ask also: is the trade count consistent with the strategy's intent? An "intraday" strategy with 11 trades over two years is not doing what you designed.
2. Is the drawdown tolerable in practice: in money?
A -22% max drawdown looks manageable as a percentage. On a ₹10 lakh allocation it is ₹2.2 lakh.
Ask the question in its honest form: if my capital were down ₹2.2 lakh right now and the strategy were still holding, would I let it run? Most traders say yes in theory and no in practice.
Then read the two companions:
- Recovery time. A 10% drawdown recovered in three weeks and one still unrecovered at the end of the window are not the same experience.
- Longest losing streak. Six consecutive losses feels different from a gradual -12%, at the same P&L.
If the drawdown is uncomfortable, reduce the allocation. That is a real fix, and the allocation is the only control that scales your loss directly.
3. Is the test period representative?
A 30-day backtest on a volatile crypto in a trending market is not a meaningful test.
| Timeframe | A reasonable minimum |
|---|---|
| 1m - 5m | 90 days |
| 15m - 1h | 6 months |
| 4h - 1d | The full available window |
And check the composition, not just the length: did the window contain both trend and consolidation? If the strategy only traded in a directional run, you do not know how it behaves in a range. The market alignment label in the assessment is one input; looking at the index over the window is another.
4. Were the parameters chosen before you saw the data?
If you ran twenty variations and kept the best Sharpe, you have overfitted. The A grade reflects the best configuration of the past.
The remedy is procedural. Run the first version, accept the result, and iterate against the reason for weakness ("drawdown too high, so tighten the stop") rather than optimising the metric directly. Then re-read the result knowing you have used up some of your evidence.
Choosing the period that produces the best number is the same error, and it is harder to notice because it feels like diligence. Pick the window for a stated reason and accept what it gives you.
The gross-versus-net check
The one diagnostic on the page that is not subject to any of the above.
Compare gross return to net return. The gap is what execution cost you, and it is arithmetic on the trades that actually happened.
| Gross return, then net | Verdict |
|---|---|
| 12% becomes 10.5% | Costs are a normal drag |
| 12% becomes 4% | The strategy is trading too much. Reduce frequency or widen targets |
| 12% becomes -1% | The edge is entirely consumed by execution. This is a clear answer |
A strategy failing this test does not need more analysis. It needs to trade less.
Using the grade
| Grade | A reasonable response |
|---|---|
| A | Paper it. Two to four weeks in current conditions is the cheapest information you will buy |
| B+ / B | Identify the weakest component, not the weakest metric, and make one change. Re-test |
| C+ / C | Read which components failed. If it is drawdown, the idea may be fine and the risk controls wrong. If it is return, the idea is probably weak |
| D | If zero trades, the condition is unreachable, check that first. Otherwise, a new idea usually beats patching this one |
Two B grades can be entirely different problems: a strong return with an ugly drawdown, or a modest return with excellent risk control. The letter compresses that away. The breakdown is on the Overview tab.
Never skip paper
Deploying a Grade A backtest straight to live is the single most common expensive mistake on any systematic platform. A paper run costs you time and nothing else, and it tests the two things a backtest cannot: current market conditions, and whether your risk limits actually work.
Both paper and live run the nine risk gates identically. A paper run that trips your daily-loss cap has told you something true that no backtest could. See How risk is enforced
Always paper first, even for a week. Always.
Next
- Backtest limitations, the complete list
- Paper trading
- How risk is enforced, before any live capital
- Risk disclosure