ADVANCED: COURSE 2 | LESSON 3
Metrics that matter: reviewing yourself like a fund reviews a PM
Learning objectives
Compute and interpret the core performance statistics: expectancy (in R), profit factor, maximum drawdown, and the R-multiple distribution
Judge when a sample is large enough for a metric to mean anything, and when a change in the numbers is signal rather than noise
Run a monthly/quarterly self-review structured the way a fund reviews a portfolio manager: numbers first, narrative second
The numbers a risk desk would pull on you
When a fund reviews a portfolio manager, the conversation starts from a standard panel of statistics, not from war stories. This lesson gives you that panel for your own trading. Every metric below is computed from a journal of trades logged in R-multiples — profit or loss divided by the amount risked at entry — which normalises across position sizes and account growth. A $50 win risking $100 and a $500 win risking $1,000 are both +0.5R; without this normalisation, your statistics mostly measure your position sizing history, not your skill.
Expectancy — the average R per trade:
Example: 40% win rate, average win +2.2R, average loss −0.9R → (0.40 × 2.2) − (0.60 × 0.9) = 0.88 − 0.54 = +0.34R per trade. Multiply by trade frequency for expected R per month. Expectancy is the number: everything else qualifies it.
Profit factor — gross profits ÷ gross losses. A PF of 1.0 is breakeven; below ~1.15 is within rounding error of no edge after costs; sustained PFs above 2 on large samples are rare and worth suspicion (usually a short sample or a strategy that hasn't met its bad regime yet).
Maximum drawdown (MDD) — the largest peak-to-trough decline in equity, in % and in R. Pair it with drawdown duration (time from peak to recovery), which traders consistently underestimate psychologically. Recall the asymmetry from P2.3: a 20% drawdown needs +25% to recover; 50% needs +100%.
Win rate and payoff ratio together, never alone — a 70% win rate with a 0.4 payoff ratio loses money (0.70 × 0.4 − 0.30 × 1.0 = −0.02R); a 35% win rate with 2.5 payoff makes it comfortably (+0.225R).
The R-multiple distribution: your trading fingerprint
Averages hide what histograms reveal. Plot every trade's R-multiple as a histogram and read it like a diagnostician:
- A clipped right tail (nothing beyond +1R despite planned 2R+ targets) is the disposition effect from A2.1 — you are cutting winners.
- A fat left tail (losses at −2R, −3R when every plan said −1R) means stops are being widened, skipped, or slipping badly at news (A3.2). One −4R outlier erases twelve +0.33R days.
- A spike at exactly −1R is what discipline looks like. It should be the tallest bar on the loss side.
- Clusters at tiny ±0.1R suggest overtrading and impatient scratching.
The histogram is also where strategy and behaviour separate cleanly: compare the distribution of your A/B-grade trades against your C/D-grade trades (A2.2). Most traders discover their entire negative tail lives in the C/D pile — which is genuinely good news, because discipline is fixable in a way that a strategy without edge is not.
Sample size: when do numbers mean anything?
The uncomfortable statistics, stated plainly. A win rate estimated from 25 trades has a 95% confidence interval of roughly ±20 percentage points; from 100 trades, about ±10; you need ~400 trades to pin it within ±5. Expectancy is noisier still, because it depends on the tails of the R distribution, which are exactly where samples are thinnest.
Practical rules of thumb: treat fewer than 30 trades as anecdote; 30–100 as a rough sketch — enough to spot catastrophic problems (PF below 0.8, fat left tail), not enough to certify an edge; 100+ per strategy before you let the numbers make big decisions (sizing up, going live from demo, retiring a system). And when comparing two periods ("my expectancy fell from +0.3R to +0.1R this quarter"), remember both figures carry error bars that likely overlap — persistent structural changes (the left tail growing, average loss creeping past −1R) are more trustworthy than wiggles in the headline number. A2.4 applies this directly to the "is my edge dead or is this variance?" question.
The review meeting: run it like a fund
Once a month (statistics review quarterly if you trade fewer than ~30 times a month), hold a formal review — same template, written record, calendar-protected. Agenda:
- The panel first. Trades, win rate, payoff ratio, expectancy, PF, current and max DD, R-histogram — overall, then split by strategy, by A/B vs C/D grade, by session, by instrument. Numbers before narrative, so the story must fit the data rather than the reverse.
- Distribution inspection. What changed shape since last review? New left-tail entries get individually named and explained — every trade beyond −1.5R gets one line: what happened, which rule would have prevented it, does that rule exist?
- Process metrics (from A2.2): grade distribution, plan-adherence rate, circuit-breaker triggers. A falling adherence rate with flat P&L predicts future losses — it is your earliest warning light.
- Behavioural cross-tabs. Expectancy after a losing day vs after a winning day; first-hour trades vs rest of session; trades taken during scheduled news. These cuts locate when you are a worse trader, which is more actionable than knowing that you occasionally are.
- Decisions and experiments. Every review ends with at most three written changes ("stop trading the first 15 minutes after CPI", "reduce size 25% until adherence is back above 90%") — each framed as an experiment with a metric and a review date. Changing five things at once makes the next review uninterpretable.
The tone to import from the institutional world is clinical, not judgmental. A risk desk doesn't shame a PM for a drawdown inside mandate; it acts fast when limits are breached or when behaviour changes. Extend yourself the same deal — and hold yourself to the same enforcement.
Key takeaways
Log everything in R-multiples; expectancy = (WR × avg win) − (LR × avg loss) is the headline number, qualified by profit factor, max drawdown and drawdown duration
The R-histogram is diagnostic: clipped right tail = cut winners; fat left tail = broken stop discipline; a tall bar at −1R is what discipline looks like
Fewer than 30 trades is anecdote; ~100+ per strategy before numbers justify big decisions; a 25-trade win rate carries a ±20-point confidence interval
Split every statistic by process grade (A/B vs C/D) — most traders' entire negative tail lives in their rule-breaking trades
Review monthly on a fixed template: panel → distribution → process metrics → behavioural cross-tabs → at most three written experimental changes