Sizing from backtest drawdown
Lesson 27 · about 10 min
The backtest gives you a drawdown in R. Your tolerance gives you a drawdown in percent of account. The ratio between them is the risk per trade you can afford. This is one of the most useful things a backtest produces, and it is more reliable than the return figure, because drawdowns in R depend on the sequence of wins and losses, which is roughly what Monte Carlo reshuffling estimates, rather than on the exact size of the edge.
The calculation
risk per trade (% of account) = maximum tolerable drawdown (%) ÷ expected worst drawdown (R)
Two inputs. The first is personal: the largest peak-to-trough loss you can sit through without abandoning the system, which for most people is between 10% and 25% and is almost always lower than they think. The second comes from Module 5: not the backtest's observed max drawdown, but the 95th percentile of the Monte Carlo reshuffle, and then some margin on top, because live will be worse than reshuffled history.
| Input | Value |
|---|---|
| Observed backtest max drawdown | 9.8R |
| Monte Carlo 95th percentile | 18.2R |
| Safety multiplier for live degradation | 1.25 |
| Planning drawdown | 18.2 × 1.25 ≈ 23R |
| Tolerable account drawdown | 15% |
| Risk per trade | 15% ÷ 23 = 0.65% |
At 0.65% risk per trade, a 23R drawdown costs about 15% of the account, and the median reshuffled drawdown of 11R costs about 7%. Both are survivable. Had you sized from the observed 9.8R instead, you would have chosen 1.5% per trade, and the same 23R drawdown would have cost about a third of the account.
A sizing table
The same calculation across tolerances and planning drawdowns:
| Planning DD (R) \ Tolerable DD | 10% | 15% | 20% | 25% |
|---|---|---|---|---|
| 10R | 1.0% | 1.5% | 2.0% | 2.5% |
| 15R | 0.67% | 1.0% | 1.33% | 1.67% |
| 20R | 0.5% | 0.75% | 1.0% | 1.25% |
| 25R | 0.4% | 0.6% | 0.8% | 1.0% |
| 30R | 0.33% | 0.5% | 0.67% | 0.83% |
Most robust retail systems, once the Monte Carlo and margin are applied, land at planning drawdowns of 15R to 30R, which puts risk per trade in the 0.5% to 1% range for ordinary tolerances. This is why "risk 1% per trade" is such a common rule: it is roughly what falls out of the arithmetic for a typical system and a typical stomach.
Compounding versus fixed
Fixed-fractional sizing (a percent of current equity) shrinks positions in drawdowns and grows them in run-ups. It makes the percent drawdown smaller than the R drawdown times the risk percent would suggest (because size falls as you lose), and it slows recovery for the same reason. For planning purposes the simple multiplication is close enough and errs on the conservative side.
Key idea: Risk per trade = tolerable drawdown ÷ planning drawdown in R, where the planning drawdown is the Monte Carlo 95th percentile with a margin, never the backtest's observed figure. This one line converts the backtest into the only sizing decision you need.
Starting smaller
For the first live period, halve the computed figure. Three reasons:
- Incubation slippage and cost estimates are still being measured, and they only move in one direction.
- The first drawdown will arrive before you have any live profit to absorb it, so it comes straight out of starting capital.
- You do not yet know how you behave when this specific system loses; that is worth learning at half size.
Scale up to the computed figure once you have 50 or more live trades with slippage in line with the plan and no rule breaches. The Risk Management course covers scaling rules in detail.
Multiple systems
If you trade several systems, their drawdowns do not simply add, because they are unlikely to happen at once, but they are also not independent, because bad markets are bad for many things. A conservative planning assumption is that the two largest systems' drawdowns can coincide. Size each so that the sum of the two worst planning drawdowns stays within your tolerance.
| System | Planning DD (R) | Risk per trade | Account DD if it happens |
|---|---|---|---|
| A: trend, daily | 24R | 0.5% | 12% |
| B: mean reversion, intraday | 16R | 0.5% | 8% |
| Both at once | 20% |
If 20% is above tolerance, both come down.
Try it: Take your strategy's Monte Carlo 95th percentile drawdown, multiply by 1.25, and divide your tolerable account drawdown by it. Compare the answer with the risk per trade you were planning to use. If you were planning more, the backtest has just told you something the equity curve did not.
Recap
- Risk per trade = tolerable account drawdown ÷ planning drawdown in R.
- Planning drawdown is the Monte Carlo 95th percentile times a margin, not the observed backtest figure.
- Typical results land at 0.5% to 1% per trade, which is where the common 1% rule comes from.
- Start at half the computed size and scale up after 50 live trades with costs in line.
- With several systems, assume the two worst drawdowns can coincide and size to that.
See it drawn
Original diagrams for the ideas on this page. Illustrative, not real market data.