Rate transmission

When the federal funds rate moves, how much of that move reaches the 30-year mortgage rate, and how much housing activity has historically moved with it — at what delay?

The textbook chain runs: policy rate → long-term Treasury yields (via expected future short rates and term premia) → mortgage rates (Treasuries plus an MBS spread) → affordability at prevailing prices → sales and starts → construction hiring and big-ticket spending. Two features matter for estimation. First, expectations: mortgage rates price the expected path of policy, so they often move before the policy rate does. Second, delay: households and builders commit to projects months before the activity is recorded, so the real-side response should appear with a lag — which is why the specifications below lag the rate change by six to twelve months.

The effective federal funds rate (monthly), the 30-year fixed mortgage rate (weekly, averaged to monthly before any differencing so all series share one calendar), housing starts, new single-family home sales, construction employment, and durable-goods spending. Estimation window: 1990–2026, on 12-month changes.

Each link is a two-variable OLS on 12-month changes since 1990: the mortgage-rate change on the policy-rate change (contemporaneous), then 12-month percent changes in each activity series on the mortgage-rate change lagged 6 months (starts, sales) or 12 months (construction employment), and durables spending on the policy-rate change lagged 6 months. The charts below show the first and third links; the full coefficient table, including the judgment ranges the simulator displays, is on the simulator page.

Transmission is real but heavily attenuated and slow. Less than half of a policy move shows up in mortgage rates on this simple measure — partly because mortgage rates price expected policy in advance, so the regression splits credit between anticipation and the move itself. One step further down, a 1pp mortgage-rate rise is associated with roughly 6% fewer starts a half-year later, but with enormous scatter. By the time the chain reaches employment and durable spending, the bivariate link is statistically almost invisible — those outcomes answer to many forces of which financing cost is only one.

The dominant issue is endogeneity, not sampling noise. The Federal Reserve raises rates when housing and spending are already strong and cuts when they weaken, so the estimated slopes blend the effect of rates with the conditions that prompted them — most likely biasing the measured sensitivity of activity toward zero. Pass-through itself varies by regime: it was near-mechanical in the 2022 tightening, and nearly absent during the years at the zero lower bound. These are reasons the simulator wraps every coefficient in a wide judgment range (0.5× to 1.5× the estimate) rather than a confidence interval.

• Association, not causation. None of these regressions identify a causal effect of policy; doing so requires isolating unanticipated policy surprises, which is beyond a two-variable OLS.
• Fixed lags are a simplification. True transmission is spread over many months and varies by episode; a single 6- or 12-month lag is a legible approximation, and the results shift with the lag choice.
• One window, averaged regimes. A single slope over 1990–2026 averages across the Volcker aftermath, the Great Moderation, the zero-lower-bound decade, and the 2022 tightening — regimes with visibly different transmission.

Try the estimates interactively — with every provenance tag and range visible — in the Interest-Rate Transmission Simulator.

The calculation lives in apps/web/src/app/research/rate-transmission/page.tsx (estimation in ../../simulator/associations.ts) and uses lib/econ; every number on this page is recomputed from the committed snapshots on each build. Data series and vintages are listed on the Data page.