Position Sizing: The Most Underrated Part of Quantitative Trading

An Experiment

Same strategy. 72% win rate. Average win 1.5%, average loss 2%.

Run 100 trades with different position sizing methods:

Starting capital $500. Same 100 trades. Results differ by 14x.

Why Such a Big Difference?

Fixed Amount ($50/trade): Regardless of account size, bet $50 each time. When the account grows, the bet proportion becomes too small — compounding is wasted. When the account shrinks, the proportion becomes too large — losses accelerate.

Fixed Percentage (10%): Better than fixed amount. Bet more when the account is bigger, less when it’s smaller. But where does 10% come from? It’s an arbitrary number.

Risk 2%: Maximum loss per trade never exceeds 2% of account balance. Note: this is not 2% position size, it’s 2% maximum loss. If your stop loss is 3% away, position size = 2% / 3% ≈ 67% of balance.

Kelly Criterion: The mathematically optimal betting fraction. Formula: f* = (bp - q) / b, where p=win rate, q=loss rate, b=win/loss ratio. Full Kelly maximizes growth but also maximizes volatility. Half Kelly is the practical compromise.

Risk-Based Calculation

Our system uses Risk 2% as the baseline. The math is straightforward:

1

Position = (Balance × Risk%) / |Entry Price - Stop Loss|

Suppose you have $1,000, going long BTC @ $90,000, stop at $88,200 (2% away):

1
2
3
4

Position = ($1,000 × 0.02) / ($90,000 - $88,200)
         = $20 / $1,800
         = 0.0111 BTC
         ≈ $1,000 notional

Maximum loss = 0.0111 × $1,800 = $20 (2% of account).

If the stop is tighter (1%):

1

Position = $20 / $900 = 0.0222 BTC ≈ $2,000

This is the core of Risk-Based sizing: tighter stops = larger positions; wider stops = smaller positions. The risk per trade stays consistent.

Dynamic Adjustments

Fixed 2% is already good, but we layer two more adjustments:

1. Losing Streak Protection

Consecutive losses trigger automatic risk reduction:

Psychology research shows that decision quality degrades after consecutive losses. Rules-based risk reduction protects you from tilt.

2. Session Adjustment

We analyzed 400+ trades by time of day:

Asia morning has the highest win rate — good liquidity and clean trends. US late session performs worst — high volatility but unclear direction.

Final risk = Base 2% × Losing streak factor × Session factor.

After 3 consecutive losses, during US late session:

1

Actual risk = 2% × 50% × 80% = 0.8%

The system protects you automatically.

Kelly Criterion

The Kelly formula gives the mathematically optimal bet size:

1

f* = p - q/b

• p = win rate
• q = 1 - p
• b = win/loss ratio (average win / average loss)

Our Pipeline strategy: p=0.72, b=0.75 (small wins, larger losses)

1

f* = 0.72 - 0.28/0.75 = 0.72 - 0.37 = 0.35

Full Kelly suggests 35% per trade. But Full Kelly’s volatility is extreme (P95 drawdown 55%). In practice, Half Kelly (17.5%) is more stable.

Half Kelly is a compromise: you give up half the expected return in exchange for significantly lower drawdown risk. In real trading, psychological stability matters more than maximizing returns.

Our Choice

We chose Risk 2% + losing streak protection + session adjustment over Kelly:

1. Risk 2% is more robust: Kelly requires precise win rate and payoff ratio estimates. If these are wrong, it’s catastrophic. Risk 2% doesn’t depend on these estimates.

2. More controllable risk: Risk 2% caps single-trade loss at 2% of account. Kelly during a losing streak can lose 10-20%.

3. Better psychology: Losing 2% on a trade won’t keep you up at night. Losing 17.5% might.

4. More flexible layering: Losing streak and session adjustments stack naturally on Risk 2%. Layering them on Kelly is mathematically complex.

Takeaway

Position sizing isn’t a supporting character in trading — it’s the key factor that determines whether you survive.

A 60% win rate strategy + good position sizing > an 80% win rate strategy + bad position sizing.

If you can only improve one part of your trading system, don’t tweak indicator parameters. Get position sizing right first.