2010issue C0967-69
Treat risk of ruin, drawdown limits, and Kelly sizing as consistent pre-trade filters
Risk of ruin, a drawdown limit, and the Kelly criterion only work as filters when volatility, stop distance, and exposure stay consistent before entry. An inconsistent standard-deviation input, a reward-to-risk comparison that barely moves ruin, and size inferred after a loss can break that bound.
- A standard-deviation formula that approaches the win size at a 100% win rate can make the risk-of-ruin percentages that depend on it unreliable.
- A worked case with a 65% win rate moved only from 8.6% to 8.4% when reward-to-risk was cut from 3 to 1, which contradicts the requirement that ruin probability fall as reward-to-risk rises.
- Illustrated peak drawdowns of 60%, 74%, and 82% sat beside ruin-style readings of 16.2%, 27.6%, and 19.4%, above a commonly cited 10% ruin bound and a 30% drawdown pain threshold.
- Position size has to be set from known entry and stop-loss prices before the trade is placed; later lot cuts and leftover profit treated as house money do not repair the filter.
Why these filters have to match the stop
Risk of ruin, a drawdown limit, and the Kelly criterion are used here as filters. The shared inputs are account equity, volatility, stop distance, and exposure. The shared job is to keep a loss or exposure decision bounded before a trade is placed and while the position is open.
TradersWeek editorial view: those filters are not a later commentary on an illustrated equity curve. They only hold if the volatility input, the stop distance, and the exposure stay consistent with one another before the trade is placed.
An inconsistent volatility input
A standard-deviation formula used to feed a risk-of-ruin calculation can be internally inconsistent. If the win rate is expanded to 1.0, that formula approaches the win size instead of collapsing to zero.
Because that standard-deviation input is then used to compute risk of ruin, errors in the volatility formula call the resulting ruin percentages into question. An alternative quadratic volatility formula goes to zero at both a 0% and a 100% win rate and produces different standard-deviation results than the original expression.
Ruin percentages that barely move with reward-to-risk
A worked comparison with a 50% loss bet, reward-to-risk of 3, and a 65% win rate produced a risk-of-ruin-style percentage of 8.6% versus a Balsara-style result of 1.6%.
Setting reward-to-risk to 1.0 in that same comparison produced 8.4%, which contradicts the requirement that ruin probability should fall as reward-to-risk rises for a fixed win rate.
Drawdown and ruin can disagree
Three illustrated sequences were calculated to have peak equity drawdowns of 60%, 74%, and 82%, with associated ruin-style percentages of 16.2%, 27.6%, and 19.4%.
A commonly cited pain threshold for maximum drawdown is 30%, and a commonly cited upper bound for reasonable risk of ruin is 10%. TradersWeek editorial view: the largest peak drawdown is not the sequence with the largest ruin-style percentage, so the two filters are not interchangeable.
Kelly-style size still starts at the stop
High illustrated returns can still coincide with high risk when exposure is run near a Kelly-style optimum under very high leverage.
Position size must be set from known entry and stop-loss prices before the trade is placed. Treating later lot reductions as if the program already knew a large loss was coming is not a valid sizing method.
Calling leftover profit house money does not make a large concurrent drawdown inconsequential for risk control.
All readings on this track · 9 readings
- 1982Three gates for a futures book: equity risk, expected value, and shrinking pyramids
- 1995A Kelly-style leverage grid and reshuffled paths
- 2004Bound the loss before leverage changes size
- 2010Treat risk of ruin, drawdown limits, and Kelly sizing as consistent pre-trade filters
- 2010Fixed-fractional forex position sizing
- 2013Kelly fraction versus risk of ruin
- 2016Expected value versus leverage, drawdown, and Kelly sizing
- 2017Fixed-fraction sizing versus a theoretical pattern edge
- 2018Evaluating double-bottom breakouts as a testable system