Capital Manager Proportional Limits

Capital Manager Proportional Limits

Experiment Overview

Context

The capital manager was initialized with min_cash_buffer=1000.0 - a fixed dollar amount that:

1. Blocked 20% of a $5,000 account - Leaving only $4,000 for trading
2. Blocked 10% of a $10,000 account - Still significant
3. Blocked only 1% of a $100,000 account - Negligible

The 30% safety buffer (percentage-based) already provides proportional protection. The extra $1,000 was redundant and harmful for smaller accounts.

Verified Workflow

Problem Pattern

# WRONG: Hard-coded dollar amount
capital_mgr = CapitalManager(
    safety_buffer_pct=0.30,
    trading_allocation_pct=0.70,
    max_position_pct=0.20,
    min_cash_buffer=1000.0  # <-- Blocks 20% of $5k account!
)

Solution: Use Percentage-Based Limits

# CORRECT: Let percentage-based safety buffer handle it
capital_mgr = CapitalManager(
    safety_buffer_pct=0.30,       # 30% safety buffer (proportional)
    trading_allocation_pct=0.70,  # 70% available for trading
    max_position_pct=0.20,        # 20% max per position
    # min_cash_buffer defaults to 0.0 - safety buffer is proportional
)

Capital Manager Design

The CapitalManager class is correctly designed with percentage-based defaults:

class CapitalManager:
    def __init__(
        self,
        safety_buffer_pct: float = 0.30,      # 30% reserved
        trading_allocation_pct: float = 0.70,  # 70% for trading
        max_position_pct: float = 0.20,        # 20% max per position
        min_cash_buffer: float = 0.0,          # Additional cash buffer (default: none)
        min_trade_value: float = 1.0           # Supports fractional shares
    ):

The min_cash_buffer parameter exists for edge cases but should rarely be used since the percentage-based safety buffer already scales with account size.

Impact Analysis

Note: The 30% safety buffer still applies to all accounts, so actual trading capital is 70% of these values.

Failed Attempts (Critical)

Key Insights

Percentage-Based vs Dollar-Based

Other Hard-Coded Values to Watch

These were identified during investigation (not fixed yet):

# In live_trader.py - consider making configurable:
confidence = 0.6       # Default fallback confidence
rl_confidence < 0.4    # Minimum RL confidence threshold
confidence < 0.70      # Extreme volatility threshold
base_breakeven = 0.02  # 2% breakeven stop
base_trailing = 0.05   # 5% trailing stop

# In harmonized.py:
min_cash_buffer = 50.0  # Should also be removed

The 30% Safety Buffer

The existing safety buffer provides strong protection:

# How it works:
conservative_value = account_value - unrealized_profits + unrealized_losses
safety_buffer = conservative_value * 0.30  # 30% reserved
available_capital = conservative_value * 0.70  # 70% for trading

# For a $10,000 account:
# - Safety buffer: $3,000 (never touched)
# - Trading capital: $7,000
# - Max per position: $1,400 (20% of $7,000)

Files Modified

scripts/​live_trader.py:
  - Line 1751-1756: Removed min_cash_buffer=1000.0
  - Added comment explaining proportional protection

References

• alpaca_trading/​risk/​capital_manager.py: Lines 74-100 (class design)
• scripts/​live_trader.py: Lines 1750-1757 (initialization)
• CLAUDE.md: Risk Controls section