Structural break tests (Chow + supF + CUSUM)

What is Structural break tests (Chow + supF + CUSUM)?

A regression's coefficients are often assumed constant across the sample — structural-break tests check that. The Chow test splits at a pre-specified date and F-tests whether the two sub-sample fits differ. When the break date is unknown, the supF (Quandt-Andrews) statistic takes the largest Chow F over all candidate break points, with a corrected null distribution.

The OLS-CUSUM test looks at the cumulative sum of recursive residuals: systematic drift away from zero signals parameter instability without committing to a single break date.

When should I use Structural break tests (Chow + supF + CUSUM)?

• Testing whether a regression relationship is stable over time (regime change, policy shift).
• Locating an unknown break with the supF statistic.

What data does it need?

Dependent variable + numeric regressor(s), time-ordered + a candidate break fraction for the Chow test.

What does it report?

Chow F at the chosen break, supF (unknown break), and OLS-CUSUM statistics, each with a p-value and a significance badge.

What does it assume?

• Observations are in time order.
• Standard OLS assumptions within each regime; the tests lose power with very small sub-samples.

How do I interpret the result?

A small p rejects coefficient stability — the relationship changes across the sample. supF also implies where (the argmax break point).

See also

• Linear regression
• Stationarity tests (ADF + KPSS)
• Vector autoregression (VAR) + Granger causality

References

• Chow (1960). Tests of equality between sets of coefficients in two linear regressions. Econometrica 28.
• Andrews (1993). Tests for parameter instability and structural change with unknown change point. Econometrica 61.
• Zeileis et al. (2002). Testing for structural change. JSS 7.