Multi-Strategy Manager

There's no fixed maximum, but most retail traders find 3-6 strategies optimal. Beyond that, complexity increases faster than diversification benefit. Each strategy requires monitoring, understanding, and occasional intervention. Start with 3-4 and add more only when you've mastered current ones. Institutional managers may handle 10-20+ strategies but with dedicated teams and sophisticated systems.

Yes, you can add new strategies at any time. The recommended approach is: 1) Add in 'Incubation' status with small allocation (5-10%), 2) Monitor for 1-2 months alongside existing strategies, 3) Check for unexpected correlations with existing strategies, 4) Gradually increase allocation if performance is satisfactory. The Multi-Strategy Manager is designed to accommodate dynamic addition and removal of strategies.

When capital decreases, your actual allocation amounts shrink proportionally. A S$500,000 portfolio with 30% trend-following means S$150,000 allocated. If the portfolio drops to S$400,000, trend-following now has S$120,000. Percentage allocations remain the same, but absolute dollar amounts decrease. This is actually a built-in risk management feature - you trade smaller after losses. You may need to rebalance if allocations drift significantly.

Running strategies on different instruments is a good way to minimize conflicts. For example, trend-following on STI, mean-reversion on specific stocks, and options selling on China A50 rarely conflict. However, some overlap is fine - the conflict resolution mechanisms handle it. The key is having clear resolution rules so conflicts are handled consistently. Don't artificially force separation if it compromises strategy effectiveness.

Check if strategies tend to win and lose together. In AlgoKing, the correlation matrix shows pairwise correlation. Correlation above 0.7 means strategies move very similarly - you're not getting full diversification. Correlation around 0-0.3 is ideal. Negative correlation (-0.2 to -0.5) is great for diversification. If correlation is high, consider: 1) Replacing one strategy, 2) Running on different instruments, 3) Different timeframes.

Several approaches: 1) Rule-based reduction: Automatically reduce allocation when drawdown exceeds threshold (e.g., halve allocation at 50% of limit), 2) Wait and observe: Drawdowns are normal, don't knee-jerk react to temporary underperformance, 3) Evaluate cause: Is it market conditions (will recover) or strategy failure (needs intervention)? The Multi-Strategy Manager can automate progressive responses. Avoid completely removing a strategy during temporary drawdown - you may remove it just before recovery.

Capital-intensive strategies can coexist with others using careful planning: 1) Allocate based on margin requirements, not notional exposure. Options selling might use S$500,000 margin but have S$2,000,000 notional - allocate based on margin. 2) Use cross-margining benefits - hedged positions require less margin. 3) Maintain buffer - don't allocate 100% of margin capacity. 4) Stagger expiries - not all positions expiring simultaneously. 5) Monitor margin utilization, not just capital allocation.

Consider dynamic allocation when: 1) You have 6+ months of multi-strategy live data to inform dynamics, 2) You understand why strategies perform differently in different conditions, 3) You can define clear, testable rules for allocation changes, 4) You have systems to implement and monitor dynamic changes. Start with regime-based allocation (simple rules for bull/bear/range) before complex performance-based dynamics. Poorly designed dynamic allocation can hurt more than help.

Per-strategy limits should: 1) Reflect the strategy's nature - volatile strategies need looser limits to function, 2) Allow for normal strategy behavior - don't set so tight that normal drawdowns trigger limits, 3) Prevent catastrophic impact on portfolio - no single strategy should be able to lose more than portfolio can tolerate. A common approach: per-strategy max drawdown at 1.5-2x expected typical drawdown, and no more than 1/3 of portfolio max drawdown. Example: If portfolio limit is 15%, no single strategy should be able to lose more than 5% of total portfolio.

Monthly review is standard, but with attention to regime changes. Check: 1) Rolling 30-day correlation vs 90-day - is correlation changing? 2) Correlation during recent stress periods - did it spike? 3) Compare to historical - is current correlation unusual? Correlation shifts often precede or accompany market regime changes. If you notice correlation rising, it may indicate both strategies are responding to same market factor - consider if diversification is weakening.

Implementation steps: 1) Calculate equilibrium returns: Use current allocation weights as 'market' weights, derive implied returns using risk aversion parameter and covariance matrix. 2) Define views: Express your beliefs about relative or absolute strategy performance (e.g., 'Trend will outperform mean-reversion by 5% with 70% confidence'). 3) Build view matrices: P (view portfolio), Q (view returns), Ω (view uncertainty). 4) Calculate posterior: Blend equilibrium with views using Black-Litterman formula. 5) Optimize: Use posterior expected returns in mean-variance optimization. Libraries like PyPortfolioOpt in Python have Black-Litterman implementations. The key skill is formulating meaningful views and appropriate confidence levels.

This is an opportunity for internal netting: 1) Identify matching quantity (e.g., Strategy A wants +1000 STI, B wants -800 → net order is +200), 2) Execute only the net order to market, 3) Allocate fills internally: A gets +1000 virtual (800 from internal match + 200 from market), B gets -800 virtual (all from internal match), 4) Benefit: Lower transaction costs, zero spread on matched portion. For timing, aggregate orders in a short window (500ms-2s). If strategies have different price sensitivities, more sophisticated matching considers limit prices. Track virtual positions separately from actual net position for accurate strategy attribution.

Multi-layered approach: 1) Strategic allocation to long volatility: Maintain small (5-10%) allocation to strategies that profit from tail events (long puts, long VIX calls). Accept this as 'insurance cost' with negative expected return. 2) Dynamic hedging: Increase hedge allocation when risk indicators elevate (VIX term structure inverted, high correlation). 3) Scenario-based sizing: Stress test to find tail exposure, size hedges to offset specific loss level. 4) Cross-strategy natural hedges: Design strategy mix so some strategies benefit from crisis (e.g., long gamma options). 5) Operational hedges: Maintain cash buffer and margin headroom for crisis flexibility. The goal is controlled tail loss (e.g., max -25% even in 2008-style crash) at acceptable cost.

Key framework elements: 1) Governance: Define decision authority levels (what requires approval, what's discretionary), review cadence (daily P&L check, weekly strategy review, monthly rebalance, quarterly deep-dive), escalation procedures for problems. 2) Processes: Document all recurring activities - strategy onboarding, rebalancing, risk response, incident handling. Use checklists to ensure consistency. 3) Systems: Ensure technology supports framework - dashboards, alerts, automated reporting. Integrate systems to avoid manual reconciliation. 4) Monitoring: Real-time for risk, daily for performance, weekly for strategy health. Define what to monitor and thresholds for action. 5) Reporting: Standard reports for each cadence with defined distribution. 6) Continuous improvement: Regular retrospectives on what worked, what didn't, and how to improve. Start simple and enhance based on actual needs rather than theoretical completeness.

Correlation stress testing approach: 1) Historical correlation stress: Use correlation matrices from past crisis periods (2008, 2020) instead of calm-period correlations. Calculate portfolio VaR with crisis correlations. 2) Parameterized stress: Assume correlations increase by a fixed amount (e.g., +0.3) across all pairs, calculate impact. 3) Copula-based stress: Use non-normal dependence structures (t-copula with low degrees of freedom) that capture tail dependence. 4) Scenario-specific correlation: For each scenario (market crash, volatility spike), define appropriate correlation assumptions based on historical behavior. 5) Correlation transition: Model how portfolio behaves as correlations shift from normal to stressed over time. Key insight: Size positions assuming crisis correlations, but track returns assuming normal correlations. If actual correlations spike, portfolio should still survive.