Target-Constrained Pooling for Trajectory Extrapolation
Target-Constrained Pooling (TCP) improves trajectory extrapolation by separating a forecast into an observed boundary response and an unknown continuation shape. It estimates the finite-horizon endpoint using two complementary source signals—uniform pooling for levels and matched-increment pooling for change. TCP then projects these estimates onto a target-consistent region to correct the free forecast, significantly reducing prediction error.