TL;DR. A strategy's quality lives in a small set of numbers, not the dozens shown in the analytics panel. CAGR, max drawdown, Sharpe, profit factor, expectancy, and the rolling-window version of each — read in that order, with the right context — give you 90% of the truth about a strategy. The rest is decoration.
What you'll be able to do
- Read any backtest result and know which numbers to trust, which to verify, and which to ignore.
- Compute expectancy from scratch and use it to compare strategies that have very different win rates.
- Recognize when a metric is misleading (small sample, fat-tail luck, regime concentration) and what to look at instead.
Why this matters
The single most expensive mistake in evaluating a strategy is anchoring on the wrong metric. A trader who picks the strategy with the highest Sharpe ratio without checking the drawdown might commit capital to a 0.9-Sharpe strategy with a 60% max drawdown. A trader who picks the highest win rate might pick a strategy that wins 90% of the time and 1R per win, then loses 10R when it loses.
Knowing what each number measures — and what it doesn't — is the bedrock of every evaluation that follows.
The metrics that matter (in order)
CAGR — Compound Annual Growth Rate
The annualized return of the strategy over the test period. "How fast does this strategy grow money?"
A CAGR of 15% means the strategy grew the account at an effective rate of 15% per year, compounded. It's the most quoted number and the most over-emphasized — useful only when paired with the drawdown that produced it.
Maximum drawdown (MDD)
The largest peak-to-trough drop in equity, expressed as a percentage of the peak. "What's the worst the account went underwater?"
This is the number that determines whether you can actually trade the strategy. A 20% CAGR with a 15% max DD is tradable; a 20% CAGR with a 50% max DD requires nerves that 95% of people don't have, and the strategy will probably get killed by the trader long before it kills the account.
Always read CAGR and max DD together. Their ratio — the Calmar ratio — is one of the best one-glance quality metrics.
Sharpe ratio
Average return divided by standard deviation of returns, annualized. "How much return per unit of total volatility?"
Useful as a comparison metric between strategies on similar assets. A Sharpe above 1.0 is decent for a retail strategy; above 2.0 is rare; above 3.0 is suspicious (usually indicates overfitting or short sample).
Caveat 1: Sharpe penalizes upside volatility the same as downside. A strategy with big winners gets penalized for those wins, which is silly.
Caveat 2: Sharpe assumes returns are normally distributed. Trading returns aren't. Strategies with rare-but-huge losses (selling naked options, betting against tail events) can show fantastic Sharpe ratios right up until the moment they blow up.
Sortino ratio
Same as Sharpe, but only penalizes downside volatility. "How much return per unit of downside risk?"
For most trading strategies, Sortino is the more honest measure. If Sortino is much higher than Sharpe, the strategy has fat upside tails (good). If Sortino and Sharpe are similar, the volatility is roughly symmetric.
Profit factor
Gross profit ÷ gross loss. "For every dollar I lose, how many do I win?"
A profit factor of 1.0 is break-even. Below 1.0 the strategy loses money. Above 1.5 is the rough floor for "real edge after costs." Above 2.5 in a backtest on retail-accessible markets is usually a flag for overfitting or look-ahead bias.
Expectancy
The average dollar P&L per trade, including costs. "What do I expect to make on the next trade?"
Expectancy = (Win Rate × Avg Win) − (Loss Rate × Avg Loss)
Example. A strategy wins 45% of the time, averaging $250 per winner. It loses 55% of the time, averaging $180 per loser.
Expectancy = (0.45 × $250) − (0.55 × $180) = $112.50 − $99 = $13.50 per trade
Multiply by trades-per-year for an annualized P&L estimate. If expectancy is negative after costs, the strategy is mathematically guaranteed to lose money over a long sample — full stop, no exceptions.
Expectancy is the single most important per-trade metric because it combines win rate, win size, and loss size into one number. Two strategies with wildly different win rates can have identical expectancy and behave identically over a long horizon. Anchor on expectancy, not on win rate.
Win rate
Percentage of trades that were profitable. "How often does this strategy win?"
The most over-emphasized number in retail trading and the least diagnostic of strategy quality. A 30%-win-rate strategy with 3R wins and 1R losses dominates a 70%-win-rate strategy with 0.5R wins and 1R losses, but only the second one feels good to trade.
The reason win rate gets emphasis it doesn't deserve: humans hate losing. A strategy that loses 7 times in a row before winning is psychologically punishing, even if its expectancy is excellent. This is real and worth tracking, but as a psychology metric, not as a quality metric.
Drawdown duration
How long the strategy spent in drawdown — measured as the longest stretch between two consecutive equity highs. "How long would I have been underwater?"
A 15% drawdown that lasts 3 weeks is a different psychological experience from a 15% drawdown that takes 10 months to recover. Long drawdowns are when traders abandon strategies that would have eventually worked.
Healthy strategies put in frequent new equity highs (every 1–3 months for active strategies). A strategy that has gone 8+ months without a new high in backtest will probably do the same live, and you should be honest with yourself about whether you'd hold it that long.
Ulcer Index
A composite metric that weights both the depth and duration of drawdowns. Lower is better.
Mostly useful as a tiebreaker between strategies with similar headline metrics. A lower Ulcer Index means the equity curve was more "comfortable" to ride.
A worked example
Three strategies, all tested on MES 5m from 2018-01-01 to 2024-06-30, all with similar CAGR. Which is the best?
Three strategies, same CAGR, three different qualities
What this shows. If you ranked these by CAGR they'd tie. By Sharpe, by Calmar, and by drawdown duration, they'd rank in opposite orders. Picking the right metric to anchor on is part of the analysis, not separate from it. Most readers would prefer Strategy B — the one with the longest losing streak and the lowest win rate — once they understand the full picture.
Common misunderstandings
- "Higher win rate is better." No. Higher expectancy is better. Win rate is a component, not a conclusion.
- "A strategy with 800 trades is statistically valid." Sample size matters, but so does what you're measuring. 800 trades with average profit of 0.5R sample the average decently. 800 trades where most of the profit came from 4 outliers don't sample anything reliably — you need a way bigger sample to call those outliers representative.
- "The summary metrics tell me everything I need to know." They tell you the cover of the book. The analytics view and especially the rolling diagnostics and worst-trade list are where the real story lives.
What to ignore (mostly)
- R-squared of the equity curve. Reads as "how straight is the line." High values usually mean the strategy concentrates profit in one regime; lower values are sometimes healthier than they look. Don't optimize for this.
- Total profit in dollars. Same strategy on a $10k account vs $100k account produces 10× the dollar profit. Percentage metrics are normalized; dollar metrics aren't.
- "Recovery factor." Net profit ÷ max DD. A useful-sounding but redundant metric — Calmar tells you essentially the same thing, with the additional benefit of being annualized.
Sample-size reality check
Most retail strategies are evaluated on tiny samples and the metrics overstate confidence.
A rough rule: 30+ trades per degree of freedom. If your strategy has 4 parameters being tuned, you want at least 120 trades before trusting the metrics. Lower than that and the numbers are largely noise; higher than that and they start to mean something.
For strategies with very few trades per year (position-trading or macro strategies), this means you need many years of data — sometimes a decade or more — before any individual backtest's metrics can be trusted in isolation.
Pro pattern: the 5-metric scan
Before deeper analysis, scan five numbers in order: (1) Calmar ratio above 1.0? (2) Sortino above 1.2? (3) Profit factor above 1.4? (4) Worst-month return survivable? (5) Expectancy positive after realistic costs?
If yes to all five, the strategy is worth deeper analysis. If no to any one, fix it or set it aside. This scan takes 30 seconds and routes you away from strategies that aren't going to work, regardless of how good the CAGR looks.
How this fits the workflow
Every metric in this page is computed by the backtest engine and displayed in the analytics view. The same metrics are computed for the portfolio in the Portfolio Builder and re-evaluated continuously during forward testing. The Level 2 metrics & KPIs lesson covers the same ground at curriculum depth, with more worked examples.
The metrics that matter most — Calmar, Sortino, profit factor, expectancy, drawdown duration — are also the ones that take 30 seconds to scan. Most evaluation mistakes happen when traders anchor on CAGR or win rate instead. Build the habit of starting with the five-metric scan, and you'll filter out 80% of bad strategies before spending real time on them.