How to Evaluate the Performance of Algorithmic Trading Strategies
To evaluate an algorithmic trading strategy, look past total profit to how it was earned: the maximum drawdown (worst fall from a peak), risk-adjusted return (Sharpe or Calmar ratio), profit factor and expectancy per trade, and whether the win rate fits the size of wins and losses. Then check the result holds up across periods and after costs.
Why total profit is the wrong place to start
Imagine two strategies that each made ₹1,00,000 in a year on ₹5,00,000 of capital. Strategy A never fell more than ₹20,000 from a peak. Strategy B fell ₹1,50,000 at one point before recovering. They have the same return, but B nearly lost a third of the account on the way, and most people would have switched it off before it recovered.
The key metrics, at a glance
| Metric | What it measures | Formula | What to look for |
|---|---|---|---|
| Maximum drawdown | Worst fall from a peak to a low | (Peak − Trough) ÷ Peak | Small enough that you'd keep trading through it |
| Sharpe ratio | Return per unit of volatility | (Return − risk-free rate) ÷ standard deviation of returns | Higher is better; compare like with like |
| Sortino ratio | Like Sharpe, but only counts downside volatility | (Return − risk-free rate) ÷ downside deviation | Higher is better |
| Calmar ratio | Annual return per unit of drawdown | Annual return ÷ maximum drawdown | Higher is better |
| Win rate | Share of winning trades (or days) | Wins ÷ total trades | Only meaningful next to win/loss size |
| Payoff ratio | Size of wins compared with losses | Average win ÷ average loss | Read together with win rate |
| Profit factor | Gross profit per rupee of gross loss | Gross profit ÷ gross loss | Above 1 to make money; more is better |
| Expectancy | Average result per trade | Win rate × average win − loss rate × average loss | Positive after costs |
A worked example (hypothetical trades)
Here are ten trades from an imaginary strategy:
| Trade | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 |
|---|---|---|---|---|---|---|---|---|---|---|
| P&L (₹) | +3,000 | −1,000 | −1,000 | +2,500 | −1,200 | −800 | +4,000 | −1,000 | +1,500 | −1,000 |
| Running total (₹) | 3,000 | 2,000 | 1,000 | 3,500 | 2,300 | 1,500 | 5,500 | 4,500 | 6,000 | 5,000 |
- 40% win rate (4 wins of 10)
- 2.75 payoff ratio (₹2,750 average win ÷ ₹1,000 average loss)
- 1.83 profit factor (₹11,000 gross profit ÷ ₹6,000 gross loss)
- ₹500 expectancy per trade
Expectancy checks out two ways: 0.4 × ₹2,750 − 0.6 × ₹1,000 = ₹500, and ₹5,000 total ÷ 10 trades = ₹500. The maximum drawdown is ₹2,000: the running total fell from ₹3,000 to ₹1,000, and later from ₹3,500 to ₹1,500.
Six losses out of ten feels bad while you live through it. The numbers say the strategy is sound, because each win is nearly three times each loss. That is why you read win rate and payoff together.
Win rate: the most misread number
A high win rate looks reassuring, but what matters is whether the wins pay for the losses. The break-even win rate depends only on the payoff ratio:
Break-even win rate = 1 ÷ (1 + payoff ratio), before costs
Maximum drawdown: the number that decides if you'll stick with it
Maximum drawdown is the largest fall from a high point of your equity to a later low, before a new high. Look at two things alongside it:
- Its size in rupees at the size you'll trade. A 15% drawdown on ₹10,00,000 is ₹1,50,000. Picture that loss on your screen.
- How long it lasted. A drawdown that took eight months to recover from is harder to sit through than a deeper one that recovered in three weeks.
Plan for worse than the backtest shows. History contains only the drawdowns that happened, not the ones that could. Our post on reducing drawdown covers how to control it.
Risk-adjusted return: Sharpe, Sortino and Calmar
These ratios answer: how much return did I get for each unit of risk?
- Sharpe ratio divides excess return by the volatility of returns. Volatility counts both up and down swings. Read the Sharpe ratio explained for how it is calculated and its limits.
- Sortino ratio only counts downside volatility, so a strategy with big up days isn't penalised for them.
- Calmar ratio divides annual return by maximum drawdown. It's the most intuitive for traders: a strategy returning 18% a year with a 12% maximum drawdown has a Calmar of 1.5 (hypothetical figures).
Consistency and robustness checks
A strategy can score well on every metric above because of one lucky period. Before trusting it, check:
- Month-by-month P&L. Is profit spread across many months, or did two months make it all?
- Different market conditions. How did it do in a sharp fall, a sideways stretch and a strong trend?
- Number of trades. Thirty trades is a small sample; a few hundred tells you much more.
- Nearby settings. If a 25% stop works but 20% and 30% don't, the edge is fragile.
- Results after costs. Apply brokerage and slippage. High-frequency strategies are the most sensitive.
- Backtest vs live. Once it runs live, compare live results with the backtest for the same period. A growing gap means something in the test was unrealistic. See why backtests and live trading differ.
Which metrics should you monitor once a strategy is live?
| Check | How often | Warning sign |
|---|---|---|
| Current drawdown vs backtest maximum | Weekly | Deeper than anything in testing |
| Win rate and payoff ratio | Monthly | Both drifting down together |
| Slippage per trade | Monthly | Fills consistently worse than expected |
| Live vs backtest P&L for the same period | Monthly | Gap widening |
| Trade count | Weekly | Far more or fewer trades than usual (often a logic or data issue) |
When a strategy that used to work stops working, read strategy fatigue before you change it.
Frequently asked questions
What are the most important metrics for evaluating a trading strategy?
Maximum drawdown, a risk-adjusted return measure such as the Sharpe or Calmar ratio, profit factor, and expectancy per trade after costs. Win rate is useful only together with the average win and average loss.
What is a good Sharpe ratio for an algo strategy?
Higher is better, and a ratio above 1 is commonly treated as decent for a strategy after costs. But Sharpe ratios depend on how they are calculated and on the period tested, so compare them within the same report or tool, and be sceptical of very high values from short backtests.
What is a good profit factor?
Above 1 means gross profits exceeded gross losses. Many traders look for comfortably above 1 after costs, because a profit factor close to 1 can turn negative with a little extra slippage.
Is a high win rate good?
Not by itself. What matters is win rate together with the payoff ratio. A 90% win rate loses money if the rare losses are much bigger than the frequent wins; a 35% win rate makes money if the wins are more than about twice the losses.
How do I calculate maximum drawdown?
Track your running equity. At each point, measure the fall from the highest value so far. The largest such fall is the maximum drawdown: (peak − trough) ÷ peak, as a percentage.
How many trades do I need to evaluate a strategy?
As many as you can get across different market conditions. A few dozen trades can be luck; a few hundred over several years gives a much more reliable picture.
