Position Sizing

Two people can make the same price forecast and have very different outcomes because one takes ten times the exposure. Position sizing is the decision about how much of your account should depend on an idea.

The size depends on how much the market can move, what would make the trade wrong, and how much confidence the evidence deserves. The maximum leverage available on the ticket tells you none of these things.

Volatility in dollars

Volatility describes the size of price fluctuations. A common measure is the standard deviation of returns: how widely percentage changes are spread around their average over a chosen interval. Historical volatility is calculated from past prices. It describes what happened, giving you a starting point for thinking about movement.

Suppose one market's recent daily volatility is 1% and another's is 4%. On $2,000 of exposure, those figures correspond to daily P&L standard deviations of about $20 and $80. Equal dollars invested therefore need not mean equal price risk. Halving the position halves this measure of its dollar fluctuation.

You can inspect daily percentage changes on a chart or calculate them from closing prices: today's close divided by yesterday's, minus one. Compare ordinary days with the largest moves and with event days. If using a volatility indicator, check its lookback and whether the number is daily or annualized. CME's volatility introduction explains the role of the time interval.

A one-standard-deviation move is not a worst case, and historical calm does not put a ceiling on tomorrow. An earnings release, policy decision, or loss of liquidity can produce a much larger jump. Volatility measures movement; risk also includes the possibility of being unable to exit, losing collateral value, or having several positions fail together.

From a planned loss to a position size

Give the exit a reason tied to the idea. A trade expecting stronger demand after an announcement may lose its case if the evidence shows weaker demand. A price-based setup may need room for normal fluctuations. Moving a stop closer solely to permit a larger position does not improve the idea.

Consider a hypothetical $5,000 account assigning $25 to the planned loss on a trade. These are teaching inputs. Entry is $100, the intended exit is $98, and estimated round-trip costs and execution uncertainty add $0.20 per unit.

Units = loss budget / (loss per unit at the planned exit + estimated costs).

$25 / ($2 + $0.20) = approximately 11.36 units. If the permitted increment is 0.1, rounding down gives 11.3 units, $1,130 of initial exposure, and a planned loss of $24.86.

If the idea instead needs an exit at $96, the same budget supports $25 / $4.20, rounded down to 5.9 units. A wider price allowance is paired with a smaller position. Neither budget guarantees the eventual loss: a gap to $97 in the first example loses $33.90 on price alone.

What Kelly adds to the decision

The Kelly criterion asks how much to risk when repeating a favorable opportunity. Its objective is long-run compounded growth. It recognizes that betting too little wastes an advantage, while betting too much makes losses so damaging that even a favorable game can shrink wealth.

Consider an idealized game with a known 60% chance of winning. A win earns one dollar for every dollar staked; a loss forfeits the stake. The opportunities repeat independently with the same odds and no costs. For this binary game:

Kelly fraction = p − (1 − p) / b.

Here, p is the win probability and b is profit per dollar staked on a win. With p = 0.60 and b = 1, the fraction is 0.20, or 20% of current wealth at risk in this idealized game. Edward Thorp's discussion develops the growth criterion.

The compounding effect explains why a positive edge does not justify unlimited size. Stake 40% and one win followed by one loss multiplies wealth by 1.4 × 0.6 = 0.84. You lose 16% over that pair. At a 10% stake, the same pair leaves 1.1 × 0.9 = 0.99. This pair illustrates compounding, rather than the average outcome of the 60%-win game. More exposure amplifies the damage of reversals as well as the value of favorable outcomes.

In a real trade, the probability is estimated, losses are not necessarily capped at a fixed stake, and positions can be correlated. If the true win probability in the example were 52%, the binary Kelly fraction would be 4%, far below the estimate based on 60%. This sensitivity is a reason to study fractional Kelly—using a portion of a calculated allocation—and the consequences of estimation error. Research on Kelly's limitations also examines drawdowns and approximation errors.

The fraction in the toy game is a fraction of wealth exposed to its full-loss outcome. It is not a percentage to enter in a leverage box or a rule to post 20% as margin. Use our Kelly interactive to see how the result changes with the assumed odds before trying to map any model to a real payoff.

Collateral and the rest of the portfolio

Once the exposure is chosen, decide how much capital supports it. The $1,130 example starts at 5× with $226 of margin or 2× with $565. Both hold 11.3 units. The extra collateral changes the liquidation buffer, while the position's dollar sensitivity stays the same.

Then add up related exposures. Three trades each with a $25 planned loss can all rely on one AI spending announcement. Their combined planned loss is $75 if they all reach their exits, with a shared gap potentially making it larger. Separate tickers do not establish independent opportunities of the kind assumed in the simple Kelly game.

Record the thesis, expected movement, intended exit, quantity, collateral, and cost estimate. After closing, compare those assumptions with the outcome. A trading journal gives you evidence about your own sizing and execution decisions, including the ideas that looked stronger in advance than they turned out to be.