A structural break is a discrete change in one or more parameters or relationships in the process represented by a statistical model. The change may affect a mean, trend, regression coefficient, volatility parameter or another modeled relationship, so that one fixed set of parameters no longer describes the entire sample adequately. The defining feature is therefore not simply a large movement in the observed data, but evidence that the relationship generating or governing those observations has changed.
A break changes the relationship, not merely the observation
An isolated shock can create an extreme observation without producing a structural break if the previous statistical relationship subsequently resumes. A structural break instead implies parameter instability across different parts of the sample. It may affect all model parameters or only some of them, and a dataset can contain one break or several. This also distinguishes a structural break from a data revision: a revision changes previously recorded or published observations because the underlying data or methodology has been updated, whereas a structural break concerns a change in the behavior of the process or relationship being modeled. Historical observations can remain exactly as recorded while the model linking them becomes unstable.
How the possible break date is determined matters for inference. If a candidate date is specified independently of the sample being tested, statistical inference can condition on that proposed point. If the break date is unknown and researchers search across many candidate dates, however, they cannot generally select the most dramatic split and then treat it as though it had been specified in advance. Structural-change procedures designed for unknown break points account for that search, while multiple-break methods address situations in which parameters change more than once. An estimated break date should also be understood as a statistical estimate, not necessarily the exact economic moment at which a new regime began.
Why ignoring a break can change the conclusion
Structural breaks matter because many statistical methods rely, explicitly or implicitly, on parameter stability over the estimation window. Pooling observations from different regimes can therefore produce coefficients, correlations, forecasts or long-run relationships that describe neither regime particularly well. One important time-series consequence is that an unmodeled shift in level or trend can make an otherwise stationary process appear unusually persistent or even consistent with a unit root. Structural change can consequently alter unit-root and cointegration inference rather than merely create an obvious discontinuity in a chart.
Detecting a structural break does not by itself identify its economic cause or prove that the latest regime will continue. Policy changes, market structure, liquidity conditions, technology or major shocks may coincide with parameter instability, but establishing causation requires separate evidence. For market research, the practical question is therefore not simply whether the latest observation looks unusual, but whether the historical relationship itself remains stable enough to justify estimating and interpreting it as one process.