How do I test for a structural break using the Chow test?

Question

AcadiFi · Accepted Answer

The Chow test evaluates whether a regression's coefficients are stable across two known subsamples. Rejecting the null means the structural relationship changed, and pooling the data into one regression would be misleading. Procedure: run the regression separately on each subsample, obtaining SSE_1 and SSE_2. Run the regression on the combined sample, obtaining SSE_P. The test statistic is F = [(SSE_P - SSE_1 - SSE_2) / k] / [(SSE_1 + SSE_2) / (n_1 + n_2 - 2k)], where k is the number of parameters. Under the null of stability, F follows an F distribution with k and n_1+n_2-2k degrees of freedom. Imagine Valentina Okonjo modeling Redhollow Retail monthly returns with three regressors across 96 months. She splits at March 2020 (n_1=30, n_2=66). SSE_P = 0.42, SSE_1 = 0.11, SSE_2 = 0.22, k=4. F = [(0.42 - 0.33)/4] / [0.33/88] = 0.0225/0.00375 = 6.0. The critical F(4, 88) at 5% is about 2.48, so she rejects stability. ```mermaid flowchart LR A[Full
sample] --> B[Split at
break date] B --> C[SSE
pre-break] B --> D[SSE
post-break] A --> E[SSE
pooled] C --> F[Chow F
statistic] D --> F E --> F ``` Limitations: you must specify the break date a priori. If the date is unknown, use the Quandt-Andrews or Bai-Perron tests that search over possible break points. The test also assumes constant variance across subsamples; violating this requires adjustment.

How do I test for a structural break using the Chow test?

Master Level II with our CFA Course

Related Questions

Practice Questions