The Kelly Criterion gives you the mathematically optimal fraction of your account to risk on each trade. Most explanations stop at the formula. This guide goes further: it shows you how to compute your personal Kelly fraction from your own journal data, why the raw output is almost always too aggressive to use directly, and what practical limits apply in live trading.
This guide is for intermediate traders who already track their trades and want a rigorous, data-driven approach to position sizing.
Step 1: Understand the Kelly Formula
Developed by John Kelly Jr. at Bell Labs in 1956, the formula is:
f* = (b × p − q) / b
Where:
- p = win rate (fraction of trades that are winners)
- q = 1 − p (loss rate)
- b = average win divided by average loss (reward-to-risk ratio)
- f* = the fraction of your account to risk per trade
Walk through a concrete example. A trader has a 55% win rate and an average winner of $312 against an average loser of $208, giving b = 312 / 208 = 1.50.
f* = (1.5 × 0.55 − 0.45) / 1.5
f* = (0.825 − 0.45) / 1.5
f* = 0.375 / 1.5
f* = 0.25
Full Kelly says to risk 25% of the account per trade. On a $25,000 account that is $6,250 of risk per trade. Four consecutive losers cuts the account to roughly $12,700 — a 49% drawdown. This is not a malfunction of the formula; it is a direct consequence of log-utility optimization. Full Kelly results in a 50% drawdown with approximately 1-in-3 probability over a sufficiently long trading horizon.
Step 2: Extract Your Stats from Journal Data
The Kelly formula is only as accurate as the statistics you feed into it. Pull your last 50-100 closed trades from your journal. Fewer than 50 trades leaves too much estimation error: a ±5 percentage-point uncertainty in win rate is common at small sample sizes, and as Step 3 shows, that level of error changes the resulting position size by 3x or more.
From your trade history, record:
| Metric | How to Calculate |
|---|---|
| Total trades (N) | Count all closed positions |
| Winners | Count trades with P&L above $0 |
| Win rate (p) | Winners / N |
| Average winner | Sum of winning P&L / count of winners |
| Average loser | Sum of losing P&L / count of losers (use absolute value) |
| b | Average winner / average loser |
For the profit factor relationship: profit factor equals b × p / q. A profit factor above 1.0 guarantees a positive Kelly fraction.
Step 3: Calculate Your Kelly Fraction
Using the SPY example: 80 closed trades imported into JournalPlus show 44 winners (55% win rate), average winner +$312, average loser −$208.
b = 312 / 208 = 1.50
p = 0.55, q = 0.45
f* = (1.5 × 0.55 − 0.45) / 1.5 = 0.25 (25%)
Now test the formula’s sensitivity. Drop the win rate by just 5 percentage points — from 55% to 50%:
f* = (1.5 × 0.50 − 0.50) / 1.5 = 0.25 / 1.5 ≈ 0.083 (8.3%)
A 5-point drop in estimated win rate reduces the Kelly fraction from 25% to 8% — a 3x swing in position size. This is why accurate trade logging is not just good practice; it directly controls how much capital you deploy.
Negative Kelly is a stop signal. If the formula returns a negative number, the strategy has negative expectancy. For a trader with a 45% win rate and a 1.0 reward-to-risk ratio: f* = (1.0 × 0.45 − 0.55) / 1.0 = −0.10. That is not a sizing problem — the strategy should not be traded at all.
Marginal profitability check: A trader with a 45% win rate and a 1.33R average gets f* ≈ 3.6%. Quarter Kelly is then 0.9% — nearly identical to the standard 1% rule. This validates traditional risk guidelines as a reasonable default for strategies with thin edges.
Step 4: Apply Fractional Kelly in Practice
Full Kelly is theoretically optimal for a gambler with infinite time and no utility concern for drawdowns. Traders have neither. The practical standard is 25-50% Kelly (fractional Kelly).
Why fractional Kelly works mathematically: Half Kelly retains approximately 75% of the maximum compound growth rate of full Kelly. This is a derived mathematical property, not a rule of thumb. The compound growth loss from halving your fraction is small; the variance reduction is large.
Applied to the SPY example on a $25,000 account:
| Kelly Fraction | % of Account | Dollar Risk Per Trade |
|---|---|---|
| Full Kelly | 25% | $6,250 |
| Half Kelly | 12.5% | $3,125 |
| Quarter Kelly | 6.25% | $1,562 |
The trader buys SPY at $580 with a stop at $574 — a $6/share risk. Quarter Kelly ($1,562 risk) allows 1,562 / 6 = 260 shares, a notional position of roughly $150,760, requiring margin. In a cash account capped at $25,000, the maximum is about 43 shares with a $258 actual risk — approximately 1% of account. Account-size constraints naturally enforce sensible limits when the Kelly output exceeds available capital.
The risk-of-ruin implications are stark: at full Kelly, ruin is possible and drawdowns are severe. At quarter Kelly, the growth rate slows modestly but the probability of catastrophic loss shrinks to near zero for a strategy with genuine edge.
Step 5: Adjust for Correlated Positions
The Kelly formula assumes each trade is an independent bet. That assumption breaks down when positions are correlated — for example, holding long SPY and long QQQ simultaneously. Both positions respond to the same macro moves. Treating them as independent and applying Kelly to each leg separately will over-size the combined exposure by a factor proportional to their correlation.
The correct approach: treat correlated positions as a single exposure unit. If SPY and QQQ have a 0.90 correlation, size the combined position using Kelly applied to the combined risk, then allocate between legs based on your preferred weighting. This is particularly relevant for futures traders running multiple correlated contracts, and for day traders running sector ETF pairs.
Pro Tips
- Run the Kelly calculation separately for each distinct strategy or setup type. A breakout trader running a mean-reversion strategy simultaneously will have different win rates and R-multiples for each — blending them produces a meaningless average Kelly fraction.
- Recalculate every 50-100 new trades. Win rates drift as market regimes change. A Kelly fraction calculated during a trending market may be dangerously inflated during a choppy one. See market regime identification.
- Use Kelly as a ceiling, not a target. Set your working size at quarter Kelly and treat it as the maximum. Scale up only with a larger, more statistically stable sample.
- Ed Thorp — who applied Kelly to blackjack counting (Beat the Dealer, 1962) and later to convertible bond arbitrage — consistently used fractional Kelly in practice, not full Kelly, even with robust statistical data behind his edge.
- If your Kelly fraction exceeds 20%, treat the result with skepticism. Either the sample is too small, the win rate is being over-estimated, or the strategy has unusual characteristics that violate Kelly’s assumptions.
Common Mistakes to Avoid
-
Using fewer than 50 trades. A win rate estimated from 20 trades can be off by 10 percentage points, making the Kelly output meaningless. Build a statistically reliable sample before trusting the result.
-
Applying full Kelly directly. Full Kelly produces severe drawdowns — roughly a 1-in-3 chance of losing half the account. Use 25-50% of the Kelly fraction as your working position size limit.
-
Ignoring correlation between open positions. Running multiple correlated trades and applying Kelly to each leg separately doubles or triples actual risk exposure relative to the formula’s intent. Always apply Kelly to total correlated exposure.
-
Recalculating Kelly after a losing streak and sizing down to near zero. A short losing streak does not invalidate a win rate calculated from 80+ trades. Distinguish between statistical noise and a genuine change in strategy performance before cutting size dramatically.
-
Treating Kelly as a fixed number. Kelly is a function of current performance statistics. As your edge changes — different setups, different market conditions — the fraction changes. Review it on a rolling basis, not once at account inception.
How JournalPlus Helps
JournalPlus calculates win rate, average winner, and average loser automatically from your closed trade history, giving you the exact inputs needed for the Kelly formula without manual spreadsheet work. The analytics dashboard lets you filter by setup tag or instrument, so you can compute a separate Kelly fraction for each strategy type rather than blending everything into a single misleading average. With R-multiple tracking built in, you can verify your reward-to-risk ratio is stable over time — a key assumption that the Kelly calculation depends on. For traders managing multiple accounts or strategies, the multi-account view lets you apply Kelly-based sizing rules consistently across all positions from one place.
People Also Ask
What is the Kelly Criterion formula?
The formula is f* = (b × p − q) / b, where p is your win rate, q is 1 − p, and b is average win divided by average loss. The result is the fraction of your account to risk per trade.
Why should most traders use fractional Kelly instead of full Kelly?
Full Kelly produces roughly a 1-in-3 chance of a 50% drawdown over a long trading horizon. Half Kelly retains approximately 75% of the maximum compound growth rate while dramatically reducing variance — a much more sustainable path.
How many trades do I need to calculate Kelly reliably?
At least 50-100 closed trades. Below 50 trades, a 5-percentage-point error in estimated win rate is common, which can change your Kelly fraction by 3x or more.
What does a negative Kelly fraction mean?
A negative result means the strategy has negative expectancy — it is expected to lose money over time. The correct response is to stop trading it and review the setup criteria, not to adjust position size.
Does Kelly work for correlated positions?
Not directly. The Kelly formula assumes each bet is independent. For correlated positions like long SPY and long QQQ simultaneously, apply Kelly to the total correlated exposure, not to each leg separately.
