The Operator’s Playbook for Bankroll Management in Algorithmic Trading

1. The Foundation: Dedicated Trading Capital

Isolate capital you can afford to lose entirely into a separate account or wallet. This enables clear performance benchmarking and prevents emotional decisions like dipping into other funds to recover losses.

2. The Core Engine: Fractional Kelly Sizing

The Kelly Criterion calculates the optimal bet size to maximize long-term growth. For trading:

f = (p b - q) / b

Where p is win probability, q is loss probability (1-p), and b is net win/loss ratio.

Example: A strategy with a 55% win rate, 8% average win, and 5% average loss has b = 0.08 / 0.05 = 1.6. Full Kelly gives f = (0.551.6 - 0.45)/1.6 ≈ 26.9%—too aggressive. Using 1/4 Kelly gives a sane 6.7% risk per trade.

This is a starting anchor, not the final rule.

3. The Critical Safety Layers: Exposure & Position Limits

Real-world conditions require hard limits:

• Total Portfolio Exposure: Cap open positions at 20-25% of bankroll to prevent over-leverage.
• Single Position Limit: Restrict any trade to 1-5% of capital to contain damage from failures.
• Stop-Loss as Risk Governor: Size positions so the stop-loss enforces your risk limit. For a 2% risk cap and 8% stop, position size = (Bankroll × 0.02) / 0.08.

4. The Implementation Architecture

Codify these rules into your bot. In Freqtrade:

In config.json, set global limits:

{
  "tradablebalanceratio": 0.25,
  "maxopentrades": 5
}

In your strategy, use customstakeamount to enforce per-trade risk:

def customstakeamount(...):
    sl_perc = abs(self.stoploss)
    maxriskratio = 0.015  # 1.5% risk cap
    bankroll = self.wallets.gettotalstake_amount()
    allowablestake = (bankroll * maxriskratio) / slperc
    return min(allowablestake, maxstake)

Case Study 1: The Cautious AI Developer

A developer building an AI agent on MindStudio for futures had a strategy with 52% win rate and 1.8 profit factor. They implemented:

• 1/3 Kelly sizing
• 20% total exposure limit
• 5% max position size
• Factored in Claude inference costs (~$5/month)

Result: Survived a 7-trade losing streak with only 9.8% drawdown, then capitalized on the next trend.

Case Study 2: The Freqtrade Spot Trader

An operator used customstakeamount to enforce:

• 2% max loss per trade
• Volatility-adjusted stops (6-10% via ATR)

Result: Smoother equity curve and 40%+ Sharpe improvement vs. fixed sizing.

Common Pitfalls

• Overfitting Sizing: Optimizing Kelly inputs on the same data as strategy → overestimation. Use fractional Kelly as a buffer.
• Ignoring Correlation: Ten 2% positions in correlated assets (e.g., BTC pairs) act as one 20% position. Understand underlying correlations.
• Changing Rules mid-Drawdown: Halving size after losses destroys long-term calibration. Stick to the rules.

Myths vs. Facts

What is a good starting bankroll percentage to risk per trade?

0.5% to 2.0% is standard. Start at 0.5-1.0% while gathering live data, then let fractional Kelly adjust it up if your edge is proven.

How do I calculate win rate and profit factor for Kelly?

Run a robust backtest (500+ trades across cycles) in Freqtrade. Use its output for winrate and profit_factor, but apply a skepticism discount.

Should I reset position size after a loss or win?

No. Always use a percentage of current bankroll. Losses reduce size automatically; wins increase it. This is fixed fractional trading.

How do exposure limits work with compounding?

They scale seamlessly. A 20% limit on a $10k bankroll allows $2k exposure; at $15k, it allows $3k—enabling controlled growth.

What’s the biggest beginner mistake?

Using the same size in live trading as in backtests with infinite simulated capital. Backtests lack real drawdown constraints.