Risk assessment

Risk Assessment (Bridgewater style)

Decision Memory Protocol (load first, log after)

Before forming any view, load prior decisions so verdicts stay consistent across sessions:

• mcp__aifolimizer__get_cross_ticker_lessons (max_lessons=3) - portfolio-level win/loss patterns
• For any name you issue a per-ticker BUY/SELL/TRIM/HOLD/ADD on, also load mcp__aifolimizer__get_ticker_decision_history (ticker=…, max_decisions=5) and mcp__aifolimizer__get_ticker_reflection (symbol=…, n=3). If a prior decision exists and this run flips it, state explicitly WHY (new data / catalyst / price); never silently contradict a logged decision.

After output, log every actionable verdict: for each BUY/SELL/TRIM/ADD/HOLD issued, call mcp__aifolimizer__log_recommendation (skill="risk-assessment", ticker, action, conviction, rationale, target_pct, stop_pct). Skipping breaks the cross-session feedback loop and causes drift.

How to run

1. Call mcp__aifolimizer__get_profile - account context
2. Call mcp__aifolimizer__get_personal_context - province, marginal tax rate, risk tolerance, time horizon, account waterfall. Use the live risk tolerance/horizon instead of the generic buckets below, and ground §11 tax-aware rebalancing. If present=false, note the risk framing is generic and suggest profile-setup
3. Call mcp__aifolimizer__get_portfolio - current holdings
4. Call mcp__aifolimizer__get_risk_metrics - vol, Sharpe, Sortino, VaR, expected shortfall
5. Call mcp__aifolimizer__get_correlation_matrix - which positions move together
6. Call mcp__aifolimizer__get_concentration_warnings - over-allocation flags
7. Call mcp__aifolimizer__get_factor_exposure for the top 3-5 holdings by weight - multi-factor betas (market/size/value/profitability/investment/momentum) + annualized alpha. Surfaces hidden FACTOR concentration that name/sector diversification hides (e.g. 4 different names all loaded on momentum = one factor bet)
8. Call mcp__aifolimizer__get_factor_snapshot - current factor regime. A book heavy on a factor that is rolling over is a live risk even if the names look uncorrelated
9. Use Ray Dalio's all-weather/radical-transparency framework

Investor profile

• Canadian retail investor
• Multi-risk-profile investor (conservative, moderate, aggressive buckets)
• Exposure goals: growth stocks, index ETFs, dividends, crypto
• Account types and capital: always read from get_profile - never hardcode

Output structure

1. Correlation analysis - holdings moving together (cluster risk)
2. Sector concentration - % breakdown, max recommended vs actual
3. Geographic + currency risk - CAD vs USD vs international exposure
4. Interest rate sensitivity - most rate-sensitive positions
5. Recession stress test - estimated drawdown in -30% equity bear market
6. Liquidity risk rating per holding (high/medium/low)
7. Single-asset concentration warnings - any position >10% of portfolio
8. Factor concentration - shared factor loadings across holdings (from get_factor_exposure); flag if 3+ names share a dominant factor beta
9. Top 3 tail-risk scenarios with probability estimates
10. Hedging strategies to reduce top 2 risks
11. Rebalancing suggestions with specific target allocations. Before recommending any ADD / increase, call mcp__aifolimizer__get_positioning_signals on those names and gate on crowding: crowding_score >= 70 = consensus-crowded → defer the add; <= 30 = contrarian, favor when fundamentals support

Rules

• Format as risk management report with heat-map table at top
• Under 600 words
• Use actual ticker data from MCP - never invent positions

Gotchas

• get_risk_metrics cached 1h - state as-of timestamp; never claim "real-time" risk.
• get_correlation_matrix covers top-N holdings only; small positions excluded - do NOT claim full-portfolio correlation.
• VaR/ES are historical-distribution estimates - fail in regime changes. State this when stress-testing.
• Sharpe/Sortino use yfinance daily closes - illiquid tickers (e.g. some .TO microcaps) produce unreliable annualized vol. Flag tickers with <60 days of returns as low-confidence.
• Crypto positions NOT in get_risk_metrics - pull separately via get_crypto_data and discuss qualitatively.
• get_factor_exposure uses US Fama-French factors - loadings for .TO/non-US names are directional only; low R² (<0.2) means the factor model doesn't explain that name, so don't over-read its betas.