Risk Management for AI Portfolios

AI stocks are among the most volatile in the equity market. A concentrated position in the wrong layer during a supply chain shock or valuation reset can erase months of gains in days. Effective risk management is not optional for AI-focused portfolios — it is the difference between surviving drawdowns and being forced out of positions at the worst time.

Drawdown Regimes

Early Signal uses a four-state drawdown regime system that dynamically adjusts gross portfolio exposure as losses accumulate:

• Normal (drawdown < 5%) — full 100% exposure
• Caution (drawdown ≥ 5%) — exposure reduced to 80%
• Defensive (drawdown ≥ 10%) — exposure reduced to 60%
• Crisis (drawdown ≥ 20%) — exposure reduced to 30%

The system is deliberately asymmetric: entering a worse regime happens immediately when the threshold is breached, but returning to a better regime requires five consecutive days below the stricter threshold. This hysteresis prevents whipsaw behavior during volatile recovery periods. Full details are on the methodology page.

Sleeve-Based Allocation

Rather than putting all capital into a single strategy, the engine splits the portfolio into three sleeves with distinct risk profiles:

• Core (60%) — high-conviction, quarterly-driven positions in up to 15 names. These are the structural bets with minimum signal thresholds of 0.50.
• Satellite (30%) — medium-conviction, weekly-signal positions in up to 25 names. This sleeve captures medium-term rotational opportunities.
• Tactical (10%) — daily-signal short-term trades in up to 20 names. Small position sizes limit the impact of any single short-term call being wrong.

This structure means that even if the tactical sleeve suffers a complete loss, the portfolio impact is capped at 10%. The core sleeve, backed by the strongest multi-timeframe signals, carries the bulk of the allocation.

Concentration Limits

No single position can exceed 8% of portfolio value (10% for mega-caps). Per-layer caps are set at 1.5 times each layer's typical annualized volatility. These hard limits prevent the optimization from loading up on correlated bets in a single part of the supply chain.

Stress Testing

The engine runs predefined stress scenarios that model specific macro and geopolitical shocks: interest rate spikes compressing growth multiples, Taiwan foundry disruptions cascading through the chip supply chain, hyperscaler CapEx cuts reducing infrastructure demand, and broad valuation compression in high-multiple AI names. Each scenario applies layer-level shocks and rolls up estimated P&L across the current portfolio. Explore the live results on the stress test dashboard.

Monitoring Your Risk

Free accounts see stress scenario summaries. Operator accounts get detailed contributor breakdowns, real-time risk alerts, and digest notifications when regime changes occur.