Skip to content

Shadow rate model

How a latent interest rate and a floor shape the yield curve, with a two-scenario example separating expected short rates from priced forwards.

A shadow rate model is a short-rate model in which a latent rate can keep moving below a floor while the model’s nominal short rate stays at that floor. When the overnight market barely moves, changes elsewhere on the yield curve can still help estimate the latent rate and the distribution of future rates.

What the floor hides

Write sts_t for the shadow rate, rtr_t for the model short rate and ℓ\ell for the floor. The model applies the floor at every date.

rt=max⁡(st,ℓ)=ℓ+(st−ℓ)+.r_t=\max(s_t,\ell)=\ell+(s_t-\ell)^+.
Above the floor, the shadow rate and the model short rate coincide. At the floor, every shadow value at or below ℓ\ell gives the same short rate. A single overnight reading near the floor therefore carries little information about how far below it the shadow rate lies. The positive-part term in the formula has the shape of a call option payoff. Black motivated the zero-floor case with the option to hold currency at zero nominal interest. Black, Interest Rates as Options
The floor is an assumption about the particular rate and market being modelled. An effective lower bound can be below zero because storing and handling large amounts of cash costs money. For example, the ECB introduced a negative deposit facility rate of −0.10% in June 2014. BIS analysis of the lower bound · ECB decision
Fictional zero-floor model. A shadow rate of minus 0.75% gives a model short rate of 0%, and a shadow rate of plus 1.00% gives a model short rate of 1.00%.
Fictional zero-floor model. A shadow rate of minus 0.75% gives a model short rate of 0%, and a shadow rate of plus 1.00% gives a model short rate of 1.00%.
The horizontal segment shows why different shadow rates can produce the same model short rate. Above the kink, the two rates move one for one.

Apply the floor before averaging

In the fictional zero-floor model above, suppose next year’s shadow rate has two equally likely outcomes under the risk-neutral measure. The −0.75% shadow rate maps to 0%, and the +1.00% shadow rate maps to 1.00%. With annual rates throughout, the average after flooring is 0.50%. The average shadow rate is 0.125%, which is already positive, so flooring that average leaves it at 0.125%.

This order matters because the maximum is a convex function. For an integrable future shadow rate sTs_T, Jensen’s inequality gives
Et[max⁡(sT,ℓ)]≥max⁡(Et[sT],ℓ).\mathbb{E}_t[\max(s_T,\ell)]\geq \max(\mathbb{E}_t[s_T],\ell).

The inequality is strict when there is positive probability on both sides of the floor. The downside scenarios stop at the floor while the upside scenarios retain their full effect.

Price the rate path before quoting a forward

The expected floored rate averages possible rates at one date. A zero-coupon bond instead prices the whole short-rate path. In continuous time, the price of one unit paid at TT is the risk-neutral average of the discount factors formed from those paths.
P(t,T)=EtQ ⁣[exp⁡ ⁣(−∫tTmax⁡(su,ℓ) du)].P(t,T)=\mathbb{E}^{\mathbb{Q}}_t\!\left[ \exp\!\left(-\int_t^T \max(s_u,\ell)\,du\right) \right].
Continue the two-scenario example. The first year’s short rate is known to be 0%, and at the start of the second year the model selects a short rate of 0% or 1.00%, each with risk-neutral probability one half, and holds it for that year. The one-year bond price is 1. The two-year bond price is the average of 11 and e−0.01e^{-0.01}. The continuously compounded one-year forward rate for the second year is
f(0;1,2)=log⁡ ⁣(P(0,1)P(0,2))=−log⁡ ⁣(1+e−0.012)≈0.49875%.f(0;1,2)=\log\!\left(\frac{P(0,1)}{P(0,2)}\right) = -\log\!\left(\frac{1+e^{-0.01}}{2}\right) \approx 0.49875\%.
The difference from 0.50% comes from averaging the discount factors before taking the logarithm. Wu and Xia use an analytical forward-rate approximation that also accounts for the variability of the cumulative rate path. Wu and Xia, Appendix A
The SelfMadeQuant lesson on the fundamental pricing equation derives path discounting and works through a two-state bond price. The preview is public and the full lesson is for members.

Reading the Wu–Xia series

Wu and Xia estimate the shadow rate from one-month Treasury forward rates at several horizons, using a model with three latent factors and a 0.25% floor in their benchmark specification. A decline in their estimate while the short rate is pinned can summarise easing reflected in the curve, including changes around asset-purchase announcements or forward guidance. The estimated level depends on the floor and the fitted dynamics. It can also reflect moves in the term premium. Wu and Xia, §§2 and 5–6
As of 8 November 2026, the Atlanta Fed still listed its Wu–Xia Shadow Federal Funds Rate updates as suspended. Its 8 April 2022 notice says updates are suspended while the federal funds target range remains above 0–0.25%. The Federal Open Market Committee’s 28 October 2026 statement maintained the range at 3.75–4.00%. Atlanta Fed series notice · October 2026 FOMC statement

Sources

Last updated
  • short rates
  • yield curve
  • monetary policy
  • fixed income
← Back to the glossaryBack to top ↑