Crypto Position Sizing: Kelly Criterion, Fixed Fractional and the Math That Keeps You Solvent

Risk 10% of your account per trade and a routine seven-trade losing streak, something every strategy eventually produces, cuts your capital roughly in half. Most liquidations on crypto perps are not bad directional calls. They are crypto position sizing failures dressed up as bad luck. The math below is not exciting, but it is the difference between a drawdown and an ended account.

Why Position Sizing Beats Entry Signals

Drawdowns are asymmetric. Lose 20% and you need 25% to get back to even. Lose 50% and you need 100%. Lose 80% and you need 400%. No entry signal is good enough to dig out of the bottom of that curve, which is why sizing errors compound in a way entry errors never do.

Losing streaks are also not a tail risk, they are a certainty. A strategy that wins 50% of the time will almost certainly hit a run of seven straight losses somewhere inside 1,000 trades. Sizing has one job: make sure that inevitable streak is an inconvenience instead of an obituary.

Fixed Fractional: The 1-2% Baseline

The standard baseline is fixed fractional risk: never lose more than 1% to 2% of equity on a single trade. The formula is simple. Position size equals account risk divided by stop distance.

Concrete example: $10,000 account, 1% risk, so $100 on the line. You buy BTC at $62,400 with a stop at $61,200, which is 1.9% away. Position size is $100 divided by 0.019, roughly $5,200 of notional. On 5x leverage that requires about $1,040 of margin.

Notice what leverage did in that sentence: nothing. Leverage decides your liquidation distance and capital efficiency, not your risk. The same $5,200 position at 2x or 20x loses the same $100 at the stop. Traders who size by leverage instead of by stop distance are choosing a random risk number and calling it a strategy.

Kelly Criterion: Sizing to Your Actual Edge

The Kelly criterion sizes positions to the edge you can prove. For a strategy with win rate W and average win-to-loss ratio R, the optimal fraction is f = W minus (1 minus W) divided by R.

Example: 45% win rate with 2R winners gives f = 0.45 minus 0.275, so 17.5% of equity per trade. That number is technically growth-optimal and practically suicidal, because Kelly assumes you know W and R precisely and that they never change. Crypto edges decay, regimes flip, and your backtest overstates both inputs.

The professional convention is fractional Kelly, usually a quarter to a half of the full figure, which in this example lands back around 4% to 8%. Two useful properties remain. If your Kelly fraction is negative, you have no edge and the correct size is zero. And if your measured edge shrinks, Kelly forces your size down before your account does it for you.

Volatility-Adjusted Sizing: Let ATR Set the Stop

Fixed percentage stops get wicked out in high volatility and waste room in low volatility. The fix is to define stops as a multiple of ATR, commonly 1.5x to 2x, and let the sizing formula do the rest. When ATR doubles, your stop widens, and the same 1% account risk automatically halves your position size.

This is regime awareness in its simplest form. The same setup in a quiet range and in a post-liquidation volatility spike does not deserve the same size, and ATR-based sizing encodes that without any discretion required.

Sizing With Flow: Cut Before the Market Makes You

Orderflow adds a forward-looking layer. When VPIN, the flow toxicity measure, runs hot, informed traders are active, spreads widen and slippage grows. Elevated toxicity is a rational trigger to trade smaller or stand aside, because your realized stop will be worse than your planned one.

Some quant desks go one step further and size off their own results: when the strategy's equity curve drops below its moving average, size is cut automatically until performance recovers. It is a humility rule, and it works because it assumes the thing most traders never do, that the edge might be gone.

The practical loop: ATR for the stop, fixed fractional or fractional Kelly for the size, VPIN and regime for the override. All three inputs sit on one screen in the Buildix deep view (buildix.trade/pair/BTC), next to the flow that generated them. Entries are opinions. Size is survival.