← All lessons

Fractional Kelly

Bet a fraction of Kelly, because your edge is only an estimate. · 7 min

Decide first

You sized at full Kelly, 10%, for an edge you measured as 40% wins at 2R. The true win rate is 34%. What is your real Kelly fraction?

Kelly is extremely sensitive to the edge. For 2R trades, every percentage point of win rate moves the Kelly fraction by 1.5 points. A backtest that overstates the win rate by six points, which is easy (module C3), puts a "full Kelly" trader at ten times the real Kelly fraction.

A fraction of Kelly is insurance against that error. Half Kelly gives up about a quarter of the growth and removes most of the pain; quarter Kelly gives up about half the growth and survives a large overestimate. Many traders end up near 0.5% to 2% per trade, well below any Kelly they could compute.

10,000 simulated traders

Simulated data · not real prices

Every trader takes the same 200 trades of a 40% win rate, 2R edge (expectancy +0.2R, Kelly 10.0%). The wins and losses are identical across the four groups; only the size differs. Switch on the error to see what happens when the true win rate is 34%, not the 40% the sizes were chosen for.

Sizing (true edge 40% at 2R)RiskShare of Kelly growth10 losses in a row cost
Full Kelly10%100%65%
Half Kelly5%76%40%
Quarter Kelly2.5%45%22%
1% fixed1%19%10%

Worked example: choosing a fraction

  1. Measured edge: 45% wins at 2R. Kelly = 0.45 − 0.55 ÷ 2 = 17.5%.
  2. If the true win rate could be as low as 38%, true Kelly is 0.38 − 0.31 = 7%.
  3. Quarter of the measured Kelly: 4.4%, still below the pessimistic Kelly, so growth stays positive in both cases.
  4. Then check the drawdown: ten losses at 4.4% cost 36%. If that is too much to live with, size lower still.

Common mistake: trusting the backtest win rate

A win rate from 50 or 100 trades has a wide uncertainty band (module C1). Sizing to the point estimate treats a guess as a fact. Size to the low end of what the edge could be.

Check yourself

  1. Ten losses in a row happen. How much of the account is gone at full Kelly (10% risk) and at quarter Kelly (2.5%)?
  2. A backtest says 45% wins at 2R, so you size at Kelly, 17.5%. Live, the win rate turns out to be 40%. What is the true Kelly fraction?
  3. Edge: 40% wins at 2R. What is half Kelly?

Closing round

Pick a risk per trade you could hold through ten losses in a row, then play a mystery market at that size.

Play the round in the simulator →

The simulator's markets are generated, not real prices; the round practises the arithmetic, it does not predict anything.

Further reading: Poundstone, Fortune's Formula. The lesson teaches the idea in its own words and exercises.