Why does position sizing matter more than the strategy?
A profitable edge and a bad sizing rule still ends in zero. That is the uncomfortable part. Two bots can trade the exact same signals; the one that risks 20% per trade dies in a streak that the one risking 1% barely notices.
The math is not symmetric. A 50% drawdown needs a 100% gain to recover. A 20% loss only needs 25% back. Every extra percent you risk per trade widens your worst-case drawdown geometrically, not linearly. Sizing is the single lever that turns a decent strategy into a survivable one, which is why it belongs in the same conversation as how deep a drawdown is too deep.
Bots make this sharper because they trade often and without hesitation. A human might skip a marginal setup after three losses. A bot takes all of them at the same size unless you tell it not to.
How does fixed-fractional sizing work?
Fixed-fractional means you risk the same percentage of current equity on every trade. Not the same dollar amount, the same fraction. As the account grows, position sizes grow with it; as it shrinks, they shrink automatically. That built-in de-risking is why it is the most common professional default.
The formula is simple:
- Risk per trade ($) = Equity × risk fraction
- Position notional = Risk per trade ($) ÷ stop distance (%)
- Units = Position notional ÷ entry price
The stop distance is doing the heavy lifting. A tight stop lets you hold a bigger position for the same dollar risk; a wide stop forces a smaller one. This is the mechanism that keeps risk constant even when volatility changes your stop placement.
Typical fractions for automated systems run 0.5% to 2% per trade. At 1%, you can lose 20 trades in a row and still have roughly 82% of your capital. At 5%, that same streak leaves you around 36%. Small numbers, very different outcomes.
A worked risk-per-trade example
Say the account is $10,000 and you risk 1% per trade, so $100 is on the line. The bot wants to long BTC at $60,000 with a stop at $58,800 (a 2% move against you).
- Risk per trade: $10,000 × 0.01 = $100
- Stop distance: ($60,000 − $58,800) ÷ $60,000 = 2%
- Position notional: $100 ÷ 0.02 = $5,000
- Units: $5,000 ÷ $60,000 = 0.0833 BTC
So a $10,000 account takes a $5,000 position, which is 0.5x notional leverage. If the stop hits, you lose $100 before fees. Notice that if the stop were only 1% away, the same $100 of risk would justify a $10,000 position (1x). The stop, not your conviction, sets the size.
Do not forget costs. On major perp venues, taker fees run roughly 2-5 basis points per side, and funding is charged in periodic intervals (often hourly or every eight hours). A high-frequency bot paying 4 bps a side round-trips at 8 bps; if your average edge per trade is 15 bps, more than half of it is gone before you count funding. A position sizing calculator that ignores fees will flatter every backtest. This is also where backtest overfitting quietly hides.
What is the Kelly criterion and should a bot use it?
The Kelly criterion answers a precise question: what fraction of bankroll maximizes long-run compounded growth, given a known edge? For a simple win/loss bet:
f* = W − (1 − W) ÷ R
where W is win probability and R is the payoff ratio (average win ÷ average loss). If your bot wins 55% of trades with equal-sized wins and losses (R = 1), then f* = 0.55 − 0.45 = 0.10, meaning full Kelly says risk 10% of the account per trade.
Ten percent per trade is punishing. Full Kelly produces the highest growth rate in theory but also stomach-churning drawdowns; a full-Kelly system routinely spends time down 40-50%. Worse, the formula assumes you know W and R exactly. In live markets you are estimating them from noisy history, and overestimating your edge by even a little pushes you past the growth-optimal point into pure risk.
That is why most quant-minded bot traders use fractional Kelly — half-Kelly or quarter-Kelly. Half-Kelly keeps about three-quarters of the growth rate while roughly halving the volatility. It is a much more honest bet on your own numbers, which are always fuzzier than a backtest suggests.
How does volatility-based sizing differ?
Volatility-based sizing sets position size from how much the asset is actually moving, not from a fixed stop. The idea: risk a constant amount of expected dollar movement, so a calm week and a violent week put the same real risk on the book.
The common approaches:
- ATR sizing: Position units = (equity × risk fraction) ÷ (ATR × multiplier). When Average True Range widens, the position shrinks automatically.
- Volatility targeting: weight = target volatility ÷ realized volatility. If you target 15% annualized volatility and BTC is realizing 60%, you hold a 0.25 weight. When realized vol falls to 30%, the weight rises to 0.5.
Vol targeting is what keeps a portfolio's risk roughly steady through regime changes, and it pairs naturally with momentum bots, which suffer badly when volatility spikes against a crowded trend. The cost is turnover: as vol moves, you rebalance, and every rebalance pays fees.
Fixed-fractional vs Kelly vs volatility: which to pick?
They are not mutually exclusive. Many production bots use fixed-fractional risk per trade, cap total exposure with a Kelly-derived ceiling, and scale the raw position by volatility. But if you are choosing a primary method, the trade-offs look like this:
| Method | Best for | Main risk | Typical setting |
|---|---|---|---|
| Fixed-fractional | Most bots, clear stops | Ignores changing volatility | 0.5-2% risk per trade |
| Kelly (fractional) | Known, stable edge | Overestimating your edge | Half or quarter Kelly |
| Volatility-based | Trend/momentum, portfolios | Turnover and fee drag | 10-20% annualized target |
For someone starting out, fixed-fractional at 1% is the honest default. It is simple to code, hard to blow up, and forces you to define a stop before every trade. Add volatility scaling once you can measure realized vol reliably, and treat Kelly as a ceiling you never fully use.
One rule that survives all three methods: cap total portfolio risk, not just per-trade risk. Ten uncorrelated 1% bets is very different from ten correlated ones, and crypto correlations snap toward 1 in a selloff. If you run multiple bots or copy several traders, size the whole book, which is the core of copy trading portfolio allocation.
Where does leverage fit in?
Leverage does not change your risk per trade if your stop distance stays the same; it changes how close liquidation sits to your entry. In the worked example above, a $5,000 position on $10,000 is 0.5x, nowhere near liquidation. Push that to 10x notional and a routine 2% wick can force liquidation before your stop even fills, converting a planned $100 loss into a total loss on the position. Position sizing and leverage are two dials on the same machine, and getting the interaction wrong is covered in how much leverage is too much.
Frequently asked questions
How much should a bot risk per trade?
A common default is 0.5% to 2% of account equity per trade, with 1% being a sensible starting point for most automated systems. At 1%, a run of twenty consecutive losses still leaves roughly 82% of your capital, giving the strategy room to recover. Lower the fraction if your win rate or edge is uncertain.
Is the Kelly criterion good for trading bots?
Full Kelly is mathematically growth-optimal but far too aggressive in practice, producing 40-50% drawdowns and assuming you know your edge exactly. Most bot traders use half-Kelly or quarter-Kelly, which keeps most of the growth while roughly halving volatility. Treat Kelly as an upper bound on size, not a target.
What is a position sizing calculator?
It is a simple tool that turns your account size, risk percentage, entry, and stop price into a position size. The core math is: risk dollars equal equity times risk fraction, and position notional equals risk dollars divided by stop distance. Any version that ignores fees and funding will overstate how much you can safely trade.
Does position sizing change with volatility?
It should. Volatility-based sizing shrinks positions when the market moves violently and grows them when it calms, keeping real dollar risk steady. You can do this directly with ATR or volatility targeting, or indirectly with fixed-fractional sizing by placing wider stops in volatile conditions, which naturally produces smaller positions.