What-if

What it shows. Twenty variables: valuation (the sector ETF), deal count, deal value and IPO count in each of the five sectors, as the % change versus a world without the shock. You can also shock four macro inputs (Fed funds, the 10-year yield, the Baa credit spread and the VIX); they drive the sectors but get no feedback. Every number shows a direction and a rough strength, not a forecast of prices or volumes.

How effects travel. Feature links say how one feature moves another (valuations lift IPOs a quarter later, for example), each with a lag of 0, 1 or 2 quarters. Sector links say how much of a move reaches another sector (half of an Energy move reaches Industrials). Together they form the link matrix A = C ⊗ L; macro links say how each macro input hits each feature in each sector. A shock lands in Q1; each link fires after its lag, and from then on valuations and deal counts persist, deal values fade 15% a quarter and IPO counts fade 30% a quarter (the fades that predicted best in the backtest). Effects of effects are included (3 rounds).

Where the numbers come from. Each link is estimated from 67 quarters of history (2010Q1–2026Q3) and pulled toward its hand-set value, harder where the data is noisy. Feature and macro links come from panel regressions of each sector's quarterly changes on the others, with sector fixed effects and lags of 0–2 quarters. Sector links blend BEA input-output tables (how much of one sector's output another sector's demand drives) with how sector ETFs move together beyond the market. Ranges are 90% intervals from 1,000 bootstrap runs. A link counts as estimated when the data outweighs its hand-set value; the rest, and every result leaning mostly on them, keep the "assumption" badge. Units: feature links are % per 1% move, sector links the share of a move that reaches the other sector, macro links % per +1 percentage point (100 bp) (per +10 VIX points for the VIX).

Advisers. Each bank's share of every sector's deal value over the last 3 years (league-table credit: every adviser on a deal counts its full value), times the sector's average quarterly deal value and its predicted % change, added up over 4 quarters. Levels ("≈ +1.1 deals a quarter") use the last four complete quarters of the monthly history.

Engine. v2 · links estimated from 2010Q1–2026Q3 history (2026-10-13), tuned by backtest. Hand-set links and their reasons live in lib/ripple/links.ts, the estimates in lib/ripple/estimated.json (compared link by link in research/link-estimates.md); the link set, fade, range, rounds and horizon in lib/ripple/settings.ts.