Methodology — Named Scenario Shelf + Reverse Stress

Transparency is a feature. This page explains what the scenario shelf shows and the honesty rules it follows — in plain terms. The precise conventions live in our internal methodology; the summary here is deliberate, not an omission.

Everything on this panel is a counterfactual of the book you hold now, labeled as such. Nothing here is a forecast, a probability, or anyone's actual history.

Historical replays

Nine named market episodes — from the dot-com bust to SVB — replayed with actual stored total returns and today's weights placed at the episode start and held. Windows are S&P 500 closing peak-to-trough dates (event windows for the short spikes), shown on every row.

Three rules keep the replays honest:

• Coverage floor. Names without price history in an episode are excluded and the rest renormalized. When covered names hold under ~60% of book weight, the replay abstains and says why — a fraction of the book relabeled as the book would be a wrong number.
• Big exclusions are named. Any excluded name holding ≥5% of the book is flagged on the row itself.
• The bias has a direction. Exclusions skew toward younger, typically higher-beta names, and today's book holds only the survivors of these episodes — so replay losses are likely understated. We print that sentence rather than leaving the sign of the error unsaid.

Reverse stress

"When did this book's path FIRST fall 20% from a peak?" — answered with dates and trading day counts from one scanned historical path at today's weights, at round query levels (−10/−20/−30/−40%). A level never reached reads "never breached in the scanned window" — a fact about the window, not a safety claim. No probabilities are attached, ever: one path is one observation. The scan window is disclosed, and shrinks only when going back further would drop covered weight below the floor (the young names responsible are named).

Factor shocks

Simple linear what-ifs: your book's trailing factor betas times a hypothetical factor move ("market −10%"), in factor-return units of the shipped factor proxies only — market, size and value. We do not offer macro shocks like "rates +100bp": no rate factor is fitted, and converting basis points into a return would require a hidden model. Every row shows the fit's R² and a band for the beta's sampling error, labeled "linear model, trailing betas, out-of-sample extrapolation" — the unexplained (idiosyncratic) share of your book's variance is untouched by any factor shock, and the panel says so.