context/themes.md and screened against your portfolio holdings. The system checks which of your positions fall under each theme and flags concentration risk when a single theme exceeds 50% of portfolio value.run_research.sh). No weekend refresh — markets are closed. The "staleness" indicator shows how recently the thesis was reviewed. Themes older than 30 days should be re-evaluated — markets move and narratives evolve.Approach: Expert-judgment scenario analysis (not VaR, not Monte Carlo).
Each scenario defines explicit percentage shocks per ticker based on historical analogues. For example, the Recession scenario uses 2008 GFC and 2020 COVID as reference points — NVDA gets -40% (semiconductor cyclicality), VOO gets -30% (broad market), SGOV gets +2% (flight to safety).
How it works: For each position, the current market value is multiplied by (1 + shock). The net portfolio impact is the sum of all position-level impacts plus any 401(k) impact modeled separately. Tickers without an explicit shock use the scenario's default (typically -30% for downside scenarios).
8 scenarios spanning: recession, oil spike, AI bust, Fed cuts, geopolitical de-escalation, liquidity crisis, sector rotation, and correlation spike.
Limitations: Shocks are static estimates, not probability-weighted distributions. No correlation modeling between positions — each ticker's shock is independent. No time-horizon component (these are instantaneous shocks). The approach trades statistical rigor for transparency and auditability — every number is traceable to a specific historical analogue.
Why not VaR? VaR requires long price history and assumes returns follow a known distribution. With a concentrated portfolio of 10-15 positions, parametric VaR gives false precision. Scenario analysis is more honest about what we don't know and more useful for a long-horizon investor who cares about "what if 2008 happens again" more than "what's my 95th percentile daily loss."
Sorted by biggest overspend. Click a bar to see transactions.