A Shifting Endpoint (SE) affine term structure model significantly outperforms standard
Fixed Endpoint (FE) models in forecasting the Brazilian yield curve. By anchoring longrun
expectations to "Focus" survey data, the SE model disentangles cyclical shocks from
structural trend shifts. For forecast horizons of 12 to 24 months, the SE model produces
consistently lower root mean squared errors and statistically superior predictive accuracy
compared to random walk and FE benchmarks. Furthermore, the model resolves the
anomaly of negative term-premia generated by stationary term structure models during
the 2021 monetary tightening, showing these resulted from small sample bias. Correcting
for this bias yields stable, countercyclical risk premiums and accurately identifies the
"moving target" of long-run equilibrium rates in a volatile emerging market like Brazil.
