How to Size Positions in Systematic Trading

Position sizing is the part of trading most people underestimate. You can have a model that correctly identifies 65% of profitable trades, and still blow up your account if you size positions wrong. In systematic trading, how much you put into each trade is not a gut call — it is a rule, applied consistently to every single position.

This guide covers the core sizing approaches, how machine learning models change the equation, and how Quant-Builder.ai handles sizing in practice.

Here's a quick look at the platform (31 seconds):

Quant-Builder.ai — Simplifying Quant Trading: Try a Free Demo at quant-builder.ai/learn

Why Position Sizing Matters More Than You Think

Two traders with identical trading signals can produce wildly different results based on sizing alone. A 2% loss on a 1% position is trivial. A 2% loss on a 20% position is account-threatening. The edge you get from a good model can be completely erased by inconsistent sizing.

Systematic sizing solves this by removing emotion from the allocation decision. You define the rule once and apply it the same way every time — whether you're feeling confident or nervous, whether the market is calm or volatile.

Core Position Sizing Methods

Equal-weight (fixed percent per position)

The simplest approach: every position gets the same percentage of portfolio capital. If you take 20 positions with 1% each, you deploy 20% of your portfolio. If you take 50 positions at 1% each, you deploy 50%.

This is the baseline used at Quant-Builder.ai: 1% of portfolio per pick. It is transparent, easy to calculate, and scales naturally — more picks mean more deployment, fewer picks mean you're naturally holding more cash.

Fixed fractional sizing

Fixed fractional sizing ties position size to your stop loss distance. If your stop loss is 5% below entry and you want to risk 1% of portfolio on the trade, you size the position at 20% of the stop distance — meaning you own enough shares that a 5% drop costs exactly 1% of your total capital.

This approach is theoretically sound because it normalizes risk across positions with different volatility. A tight stop means a larger position; a wide stop means a smaller one. The risk per trade stays constant.

Kelly criterion

The Kelly criterion calculates the optimal bet size given your win rate and average win-to-loss ratio. The formula: Kelly % = win rate − (loss rate / (average win / average loss)).

In practice, most systematic traders use "fractional Kelly" — 25–50% of the Kelly-optimal size — because full Kelly is too aggressive for noisy stock markets. Kelly sizing grows positions as your model's historical performance improves, which aligns capital allocation with edge.

Confidence-based scaling

When your models output a confidence score for each pick, you can size positions proportionally to confidence. A 70%-confidence pick gets twice the allocation of a 45%-confidence pick. This approach rewards the model's highest-conviction signals with more capital.

This only works when the confidence scores are well-calibrated — meaning a 70% confidence pick actually wins 70% of the time historically. Quant-Builder models train on historical win rates, which means confidence scores correspond to real predicted probabilities.

How Systematic Sizing Interacts with ML Models

Machine learning models add a dimension that fixed-rule systems don't have: variable conviction. A rule-based system fires when RSI crosses 30 — every time with the same signal strength. An ML model outputs a probability estimate for each stock, every day. Some picks score 75%. Others score 45%. Sizing by confidence is only possible when your signals have this property.

The pick count as a deployment signal

On Quant-Builder.ai, the number of picks your model generates each day is itself a signal. When the model fires 40 picks in a day, it has found many qualifying setups. When it fires 3, conditions are not right and you deploy very little capital.

With equal-weight 1% sizing, 40 picks = 40% deployed, 3 picks = 3% deployed. The model's conviction about the current environment is automatically reflected in your allocation level — without any separate macro call from you.

Stop Losses and Position Sizing Together

Position sizing and stop losses are two halves of the same risk control framework. Sizing determines how much capital is at risk per position. Stop losses determine how much of that capital you actually lose if the trade goes against you.

A practical systematic framework:

• Enter every position at a fixed percent of portfolio (e.g., 1%)
• Set a stop loss at a fixed distance below entry (e.g., -5%)
• Maximum loss per position = 1% × 5% = 0.05% of portfolio
• Even with 50 simultaneous positions, your max loss if all stops hit = 2.5% of portfolio

This math makes it easy to understand your worst-case exposure before you place a single trade.

The Most Common Sizing Mistakes

• Sizing up on "high conviction" trades by feel. If your rule says 1%, bet 1% — not 5% because "this one is obvious." The edge is in the rule, not in the exceptions.
• Not accounting for correlation. Thirty Consumer Cyclical picks that all move together act like one large correlated position. Your effective single-position risk is higher than the sizing rule implies. This is why sector diversification matters.
• Changing sizing during drawdowns. Reducing size when you're down and increasing when you're up is the opposite of systematic behavior. Set your size rule before you start and stick to it.
• Sizing too large to hold through normal volatility. A 10% position in a volatile stock will feel terrible on a bad day. Size small enough that you can sit through normal volatility without second-guessing every hour.

How Quant-Builder.ai Handles Sizing

The default framework at Quant-Builder is 1% per pick with a -5% stop loss. This is conservative by design — it keeps max position loss at 0.05% of portfolio, allows you to hold 50–100 positions simultaneously without dangerous concentration, and makes the math simple to verify every morning.

You can apply confidence-based scaling if your model's confidence scores are well-calibrated. The platform shows you historical win rates by confidence bin so you can verify before you size up high-confidence picks.

See the platform in action (51 seconds):

Quant-Builder.ai — Simplifying Quant Trading: Try a Free Demo at quant-builder.ai/learn

Getting Started

Quant-Builder.ai starts at $25/month. Build your first model, review daily picks with confidence scores, and apply consistent sizing rules to every trade — all without writing a single line of code. Try the free demo at quant-builder.ai/learn.