portfolio-risk-management

name	portfolio-risk-management
description	Guidelines for portfolio-level risk management in the Argus backtest system. Covers circuit breakers, correlation handling, position sizing, and market regime detection. Use when modifying engine.py, risk_manager.py, or config.yaml.

Portfolio Risk Management

Current State (as of May 2026)

What's Implemented

• Risk Manager (backtest/risk_manager.py):

  • Daily/Weekly/Monthly drawdown limits (3%/8%/15%)
  • Position halt when limits exceeded
  • Per-symbol exposure limit (max 2 positions per symbol)
  • Correlation check (threshold 0.8) via correlation_matrix

• Backtest Engine (backtest/engine.py):

  • Bankruptcy detection (<5% initial balance = bankrupt, stops all trading)
  • Max drawdown detection (>50% from peak balance = bankrupt)
  • Breakeven logic at 1R profit
  • Partial close at 1.5R (50% of position)
  • 3x leverage cap per trade
  • Commission, slippage, spread modeling

What's Missing (Key Findings)

1. No dynamic circuit breaker: Once DD limits are hit, trading halts permanently until next day. There's no gradual position scaling based on current DD level.

2. No volatility-based position sizing: Lot size is based on fixed 1% risk_percent_per_trade, regardless of market volatility regime.

3. All strategies lose simultaneously: In Q1-Q2 2022, ALL strategies lost money, indicating a systemic market regime problem that per-strategy tuning cannot fix.

4. Bankruptcy ends everything: Once bankrupt, no more trades execute. The system doesn't reduce positions gradually or switch to defensive mode.

5. No market regime override: The regime_mask exists per-strategy but there's no global "high-volatility regime" mode that reduces position sizes across all strategies.

Proposed Improvements (Not Yet Implemented)

Circuit Breaker Enhancements

- Graduated DD response:
  - DD > 10%: reduce position size to 75%
  - DD > 20%: reduce position size to 50%
  - DD > 30%: reduce position size to 25%
  - DD > 40%: close all positions, stop trading, enter recovery mode
- Recovery mode: trade only the 2 best-performing strategies at 0.5x size until DD < 15%

Dynamic Position Sizing

- Base risk = 1.0% per trade
- Volatility multiplier = max(0.5, 14_ATR / 20_ATR)  # shrank when vol spikes
- Correlation penalty: if 3+ correlated strategies enter same direction same day, reduce size by 20% each
- Drawdown multiplier: linear reduction as DD increases (1.0 at 0% DD, 0.2 at 40% DD)

Market Regime Detection

- Trend Strength Index (ADX) across portfolio:
  - ADX > 30: Trending regime → favor trend-following strategies
  - ADX 20-30: Normal → all strategies
  - ADX < 20: Ranging regime → favor mean-reversion strategies
- Global volatility regime:
  - Average ATR(14) across all symbols vs 50-period average
  - High vol: reduce all position sizes by 50%

Lessons Learned

1. Fix the portfolio first, then individual strategies. If all strategies lose money in the same period, no amount of per-strategy tuning will help.
2. Correlation neglect is dangerous. When correlated strategies all enter similar positions, a small adverse move hits the portfolio disproportionately hard.
3. The best defense is not trading. A circuit breaker that halts trading during high DD is worth more than any single strategy improvement.
4. Partial closes help but aren't enough. The 1.5R partial close protects profits on winning trades but doesn't help during sustained losing streaks.