Agency MBS: Thinking caps in floaters

Research evidence excerpt

Agency MBS: Thinking caps in floaters

risk associated

with floaters created in a given month. As such, we can break margin out into option cost

components such as curve slope and rate volatility (rate level is already captured by the

short-term variable). To capture residual dependence on current coupon levels, we can add

in an option-adjusted spread term as well; this term may also capture the prepay risk embed-

ded in a floater’s cap. If our functional form for describing weighted average floater cap

levels holds true, we can also say that these four variables—short-term rates, curve slope,

rate volatility, and spread—describe cap levels as well.

To translate this into a regression, we begin by including 1 mo. t-bill yield and CC TOAS

variables. Then, to choose the best curve slope and rate volatility components, we perform

a grid search to select at most one curve slope instrument and at most one swaption instru-

ment, picking the combination that maximizes the average 36-month rolling r-squared for

the observation periods ending Jan. 2022 through June 2026. We choose from 2s5s, 2s10s,

2s30s, 5s10s, 5s30s, and 10s30s for curve slope; and 1m5y, 1y5y, 3y5y, 1m10y, 1y10y, and

3y10y for swaptions.

The strength of this multivariate regression has been reasonably strong from 2020 onwards,

though not consistently so before this (Figure 20R-squaredbetwencap-levelandmultivariateregresionandcap-levelandindividualfactors). We show the multivariate regression’s

factor betas in Figure 21Betasforcap-levelmultivariateregresion.

Figure 20: R-squared between cap-level and multivariate regression and cap-level and individual factors

36-month rolling r-squared between floater caps for that issue month (excluding ReRemics, re-securitizations, inverses,

complexes, and IRCs, %) vs.