How Big Should Your Next Trade Be?

The risk playbook gave the formula that sizes a single trade: position = money at risk ÷ stop distance. That answers one trade. It quietly leaves open the harder, slower question: what number belongs in the "money at risk" slot, and how should that number change as your account grows, shrinks, and moves between a sleepy major and a perp that swings 12% before lunch? That question has four established answers, called sizing models, and they differ less in their good weeks than in how they behave when things go wrong. Consistency beats model choice, but you should still choose on purpose.

Fixed size, the honest simpleton

The oldest model: trade the same size every time, a fixed dollar risk or a fixed position, regardless of account balance. Its virtue is statistical honesty, since every trade is directly comparable and your results are pure signal about your edge. Its flaw is that it ignores the account entirely. On the way up there is no compounding; a doubled account still risks the old fifty dollars. On the way down it is quietly lethal, because a fixed dollar risk becomes a growing percentage of a shrinking account, exactly the wrong direction. The honest use case is evaluation: testing a new setup, a new market, or yourself, where clean statistics matter more than growth. As a permanent policy it is a slow leak with good bookkeeping.

Fixed fractional, the default for a reason

Risk a constant percentage of current equity, the 1% from the playbook. This is the model most professionals mean when they say risk management, and its power is that it breathes with the account in both directions. After winners, the same 1% is more dollars, so gains compound without any decision. After losers, the same 1% is fewer dollars, so the model automatically trades smaller exactly when your edge is most in doubt. That built-in flinch is why ten straight 1% losses cost 9.6% rather than 10, and why the model mathematically cannot ride to zero. Its weakness is at the small end: 1% of an $800 account is eight dollars of risk, which fees and slippage eat alive. Below a certain account size, fixed fractional keeps you safe and keeps you tiny, which is the honest price of safety. If this post ends here for you, use fixed fractional at 1% and you are ahead of most of the market.

Fixed ratio, the small account's gamble

Ryan Jones's fixed ratio model targets exactly that smallness problem: instead of tying size to equity, it ties size increases to accumulated profit, stepping up a unit every time profits grow by a chosen delta. The result is aggressive early compounding, since the first won trades escalate size quickly, which is why the model appeals to small accounts trying to become medium ones. The bill arrives in two ways. Early drawdowns hit a small base hard, because the step-down lags the step-up. And after a good run you find yourself trading a size that your recent profits purchased but your skill has not yet confirmed, at which point one normal losing streak returns several steps of progress. It is a legitimate model for a small account run with open eyes and a hard drawdown line, and a common way for a lucky month to erase itself.

Volatility sizing, the crypto-native model

The fourth model changes the other input. Keep the risk slot constant, but derive the stop distance from the market's current temperament, measured by ATR, the average true range. The trade's invalidation still comes from structure, the level or the swing the levels series teaches, but the buffer beyond it scales with ATR, and the position size follows from the formula as usual. The effect is that one trade always buys one unit of market heat: the same $100 risk automatically holds a large position in a calm BTC and a small one in a meme perp doing 12% days, without you deciding anything per trade. All the familiar advice about "sizing down in volatile conditions" is this model done with arithmetic instead of vibes. It matters double in crypto, where the gap in daily range between the calmest and wildest pairs on the same screen is not 20% but severalfold. The number itself is one click away, since the ATR panel ships with the chart indicators on any Osiris chart.

Kelly, the ceiling that is not a setting

Sooner or later someone discovers the Kelly criterion, the formula that computes the growth-optimal risk fraction from your win rate and payoff ratio, and concludes that mathematics endorses risking 15% per trade. Two problems. Kelly assumes you know your edge precisely, and you do not; your win rate is an estimate from a sample, drifting with regimes. And full Kelly tolerates drawdowns beyond 50% as a matter of course, which no human trades through with their judgment intact. Practitioners who use it at all use a quarter of it. Kelly's real job is as a ceiling: estimate it pessimistically, and if your current risk percentage exceeds even that, you are not aggressive, you are miscalibrated. It is a boundary marker, not a dial to set.

The ten-loss test

Whatever model you pick, it must pass one exam before touching money: ten consecutive losses, which any real strategy will eventually deliver, per the streak math in the playbook. Under fixed fractional, the arithmetic is fixed: at 0.5% risk, ten straight losses cost 4.9% of the account. At 1%, they cost 9.6%. At 2%, 18.3%. At 3%, 26.3%, and a quarter of the account is where judgment starts making the losses bigger. The old rule of thumb survives translation to any model: if ten losses in a row put you down more than about 20-25%, the sizing is too hot, whatever the backtest says about how unlikely the streak is. The streak is not unlikely. It is scheduled, and the model's whole job is to make its arrival boring.

The two commandments, model-independent

First: the stop is never adjusted to afford a size. The invalidation comes from the chart, the size comes from the formula, in that order, every time; a trader who widens a stop to justify a bigger position has replaced both inputs with appetite. Second: models get changed consciously, on a calendar, with a written reason, never in the middle of a losing week. Switching to fixed ratio because fixed fractional "recovers too slowly" after a drawdown is not a strategy decision, it is tilt wearing a spreadsheet. The journal is where a model earns or loses its job, over months, in R, the way the playbook measures everything.

Matching model to trader is the short version of this whole post. Testing something new: fixed dollar, for the clean statistics. Running a real account: fixed fractional at 1%, the boring default that survives everything. Small account, high conviction, hard rules: fixed ratio, with the drawdown line written down before the first trade. Scalping across a screen of pairs with wildly different personalities: volatility-weighted stops on top of fixed fractional, so every trade costs one unit of heat. And over all of them, the ten-loss test as the entrance exam. The formula sizes the trade. The model sizes the career.