Skip to content
Time-Series
EconometricsIan Helfrich PhD ↗

Lecture 13 summary · April 16, 2027

When the model or the measurement changes

Complete chapter · Laboratory ch13 · Practice · All summaries

The economic question

A model has a period of usefulness. Shorter windows can adapt to change while sacrificing precision; neither window policy wins under every mechanism.

Central relation

d‾=T−1∑t(ℓM,t−ℓB,t),\overline d=T^{-1}\sum_t(\ell_{M,t}-\ell_{B,t}),
SE^(d‾)=Ω^/T\widehat{\mathrm{SE}}(\overline d)=\sqrt{\widehat\Omega/T}

Individual problem

A rolling method wins after a simulated break. What general conclusion follows? What is uncertain about a small loss advantage on twelve later targets?

Worked reasoning

The advantage is conditional on that mechanism and sample; shorter windows can lose precision in stable systems. Twelve dependent losses leave substantial uncertainty about persistent improvement.

A consequential failure

Continuation outcomes select the fitting window, or changing coverage is interpreted as changed economic propagation.

Repair

Choose on an earlier segment, freeze the later rule, inspect measurement provenance, and report paired dependent-loss uncertainty.