Structural Breaks

A structural break is a point in a time series at which the statistical properties of the data-generating process change abruptly. Unlike a temporary shock, which perturbs a variable around a stable trend, a structural break indicates that the underlying model itself has changed — the parameters, the functional form, or even the set of relevant variables. It is the econometrician's equivalent of a regime shift: a moment when the past ceases to be a reliable guide to the future.

Structural breaks are pervasive in economic data. The Great Moderation of the 1980s-2000s, the 2008 financial crisis, and the post-pandemic inflation surge all represent shifts in the stochastic process of key macroeconomic variables. Before each break, forecasting models performed adequately; after each break, the same models produced systematic errors because they assumed a stability that no longer held. The adaptive expectations framework, with its single fixed adjustment parameter λ, is particularly vulnerable to structural breaks: it learns the old regime too well and adjusts too slowly to the new one.

The standard approach to detecting structural breaks — the Chow test and its sequential variants — requires knowing the break date in advance or testing all possible dates, which inflates the false positive rate. More sophisticated methods use Bayesian model averaging or change point detection algorithms from machine learning, but these require stronger assumptions about the post-break model than are typically available. The fundamental problem is epistemological: detecting that the model has changed requires a model of change, which is itself subject to uncertainty.

The rational expectations framework assumes that agents know the true model and can therefore anticipate structural breaks. This is not merely descriptively false; it is conceptually incoherent. A structural break, by definition, is a change in the model that agents are assumed to know. If they knew the model would change, they would already be using the new model, and there would be no break. The existence of structural breaks is proof that rational expectations cannot be the general theory of expectation formation — only a special case that holds in the rare intervals between breaks.