A swing trader risks 2% of a $50,000 account on every trade. The rule has held since the first month of the journal. Then the VIX prints 38 on a Monday open and by Friday the same setup that usually loses 2% has cost 4.2%. Nothing about the strategy changed. Only the air did.
This is where most position sizing methods quietly break, and where the choice between fixed fractional, volatility-based, and Kelly stops being a textbook question. I want to take the three approaches I lean on most and show how they actually behave on a live book, where each one fits, where each one fails, and how I combine a volatility scalar with a hard risk cap to capture the benefits of both without inheriting the worst of either.
What position sizing methods actually decide
Position sizing methods decide the size of a loss before it happens. Entry decides the price. Exit decides the trigger. Sizing decides how much of the account is on the table while those two are resolving. I treat it as the first risk decision in the chain, ahead of the entry signal, because a perfect entry sized at 50% of equity is still a blown account on a 4% gap.
The three methods I will compare answer one question each. Fixed fractional asks “what percentage of equity am I willing to lose?” Volatility-based asks “how wide does the market need me to be?” Kelly asks “what is my edge, and how much should I press it?” None of them is a position size on its own. Each one is an input to a size.
Fixed fractional: risk a constant percentage of equity
Fixed fractional says: pick a risk percentage, apply it to current equity, divide by the stop distance, that is the position. Account at $50,000, risk fixed at 2%, stop $2 below an entry at $40 puts the size at $1,000 of risk divided by $2 per share, or 500 shares. The position is $20,000 of notional. The most I can lose is $1,000.
I default to fixed fractional for a reason. It scales with the account automatically, it ignores correlation noise, and it gives the journal a clean per-trade R number that the strategy can be evaluated against. The ATR-band stop placement I use on swing entries still produces a clean fixed-fractional size as long as the ATR feeds the stop, not the size.
The hidden assumption is that the stop distance already encodes the market’s volatility. When volatility is stable, that holds. When it changes regime overnight, it does not, and the same 2% rule becomes a different bet entirely.
Where fixed fractional breaks down
Fixed fractional does not signal that the strategy is now operating in a higher-risk regime. It signals nothing at all. If a trader sets a $2 stop on an entry at $40 when ATR is $0.80, the stop sits at 2.5 ATR and the math is normal. If the next day ATR rips to $2.20 and the same stop sits at 0.9 ATR, the trade is being stopped on routine intraday noise. The trader takes a string of small losses that the system would not have produced two weeks earlier. Same rule. Different market.
I have watched fixed fractional turn 2R expected drawdowns into 4-5R realised drawdowns during VIX-above-30 weeks for exactly this reason. The percentage stayed constant. The probability that the stop got hit by noise instead of by adverse trend roughly doubled. Fixed fractional alone does not see this.
Volatility-based sizing: scale by ATR
Volatility-based sizing flips the dependency. Set the stop at a multiple of ATR. Size the position so that one ATR equals a fixed dollar amount of risk. The same $50,000 account targeting $1,000 of risk with ATR at $1.50 and a 2-ATR stop sizes 333 shares at the $40 entry, a $13,330 position. When ATR drops to $0.80 the next month, the same $1,000 risk allows 625 shares, a $25,000 position. The size scales inversely with volatility.
This is the method I lean on for trend-following and breakout systems where the holding period is measured in weeks. The book stays roughly constant in dollar-risk terms even as the underlying instruments cycle through their own volatility regimes. The VIX regime sizing I have written about elsewhere is a coarser version of the same principle applied at the portfolio layer.
What volatility sizing does NOT do: it does not protect against gap risk, it does not adjust for correlation across positions, and it does not stop a trader from running a 50% gross book on low-vol days just because the per-trade dollar risk looks reasonable.
Fractional Kelly: size by edge
Kelly is the formal answer to “how much should I bet on a positive-edge proposition.” For a trade with win rate W, reward-to-risk ratio R, and loss probability L (where L = 1 – W), the Kelly fraction is (W * R – L) / R. A strategy with 55% wins at 1.5R per winner gives (0.55 * 1.5 – 0.45) / 1.5, or 25% of equity per trade. That is the size that maximises long-run geometric growth if the inputs are exact.
Few traders run full Kelly. The standard practitioner approach is fractional Kelly: half-Kelly at 12.5%, quarter-Kelly at 6.25%, or smaller. The longer walkthrough of Kelly for swing traders shows why halving the fraction cuts variance roughly 75% while only giving up about 25% of long-run growth. Ralph Vince’s optimal-f work is the formal extension when the bet distribution is not a clean binary.
Where Kelly fails on real strategies
Kelly fails when the edge estimate is wrong, and on most swing strategies the edge estimate is wrong by more than the trader realises. Out of 30 closed trades, a win rate of 55% has a standard error of about 9 percentage points. The true rate could be anywhere from 46% to 64%. At 46%, the Kelly fraction collapses to about 9%. At 64%, it rises to 40%. The same input data justifies position sizes that differ by a factor of four.
I have seen this kill more swing books than any other risk failure: a strategy with 80 closed trades and a 60% win rate that does not realise its true win rate is 53%, sizes by half-Kelly off the 60% estimate, gets a string of normal losses, and discovers the drawdown is twice what the backtest predicted. Kelly is a multiplier on edge, and a multiplier on a noisy edge multiplies the noise.
Combining a volatility scalar with a hard risk cap
The synthesis I run looks like this. Start with a fixed fractional cap, typically 2% per trade. Compute a volatility-adjusted size: target $1,000 of risk at a 2-ATR stop using current ATR. Take the smaller of the two as the final size. In normal volatility, the ATR-based size and the fixed-fractional cap come out close. In a vol spike, the ATR scaler shrinks the position automatically while the 2% cap still acts as the ceiling. In dead-quiet vol, the ATR scaler would happily size up to 50% gross notional, and the 2% rule prevents that.
On top of that I run a portfolio-level hard drawdown stop. If the equity curve drops 8% from peak, all new sizing gets halved until equity recovers to within 4% of the prior peak. The mechanics of a hard drawdown stop are simple to code but psychologically hard to honour. The combined rule, fixed fractional cap plus ATR scalar plus drawdown halver, sees the three things a single method cannot see at once: account size, current volatility, and the trader’s recent track record on this strategy.
How to pick a method for the strategy you actually trade
Mean-reversion strategies with short holding periods do best on fixed fractional. The edges are usually thin, the per-trade dollar risk is what the operator can verify daily, and the holding period is too short for a regime shift to invalidate the stop. Trend-following and breakout systems with weeks-to-months holding periods do best on volatility-based sizing. The whole point of the strategy is to ride trends through their own volatility expansion, and a constant percentage rule will either size you out of the move or size you so far in that one gap closes the year.
Fractional Kelly is appropriate when the edge is verifiable from outside the strategy, not from the strategy’s own back-test. Index-arbitrage spreads, event-driven equity setups with a known catalyst window, and option-selling strategies with a measurable IV-rank edge are the categories I would consider for fractional Kelly. Discretionary chart-pattern trading is not. The edge is too noisy and the sample too small.
The sizing decision is the strategy
A 55%-win-rate strategy sized wrong loses money. A 45%-win-rate strategy sized right with a 2R reward-to-risk makes money. The sizing decision is doing more of the work than the entry signal does on most operators’ books, and it is the part that gets the least screen time. The traders I have learned the most from spent twice as long on the sizing rule as they did on the entry rule, and zero of them ran a single static method across every regime.
Learn the pattern. Ride the trend. Keep the gains.
Educational content only. Not investment advice. Trading involves risk. You are responsible for your decisions.