Chow Test for Structural Change: Concept, Steps and a Real-World Example

Structural change is one of the most important issues that an econometrician must check for before relying on the results of a regression model. Economic relationships are rarely fixed forever; wars, financial crises, policy reforms, and other major shocks can alter the way variables relate to one another over time.

When such a shift occurs but is ignored, a single regression estimated over the whole period can give misleading coefficients and unreliable forecasts. This article explains one of the most widely used tools for detecting such a shift: the Chow test.

Table of Contents

What Is Structural Change?

By structural change, we mean that the values of the parameters of a regression model do not remain the same throughout the entire sample period. In other words, the intercept, the slope coefficients, or both may change at some point in time.

Structural change is usually associated with major external events such as oil price shocks, wars, changes in government policy, or financial crises, all of which can alter the underlying relationship between the dependent and independent variables.

If such a change is not accounted for, fitting one regression line to the whole sample forces a single set of parameters onto what are, in reality, two (or more) different relationships. This can distort the estimated coefficients, inflate the residual sum of squares, and lead to incorrect conclusions.

The Chow Test

The Chow test, developed by Gregory Chow, is an F-test used to determine whether a structural change has occurred in a time series regression at a known point in time. It tests whether the intercept, the slope, or both coefficients differ before and after a specific break point.

Setting Up the Test

Suppose we have data on a dependent variable $Y_t$ and an independent variable $X_t$ over a certain time period, and we suspect that a structural break occurred at some point during that period. We divide the full sample into two sub-periods — before and after the suspected break — and estimate three separate regressions.

$Y_t = \lambda_1 + \lambda_2 X_t + u_{1t}$ , for the pre-break period, with $n_1$ observations (eq. 1)

$Y_t = \gamma_1 + \gamma_2 X_t + u_{2t}$ , for the post-break period, with $n_2$ observations (eq. 2)

$Y_t = \alpha_1 + \alpha_2 X_t + u_t$ , for the entire period, with $n = n_1 + n_2$ observations (eq. 3)

Equation 3 assumes that $\alpha_1 = \lambda_1 = \gamma_1$ and $\alpha_2 = \lambda_2 = \gamma_2$ , that is, it assumes there is no structural change between the two sub-periods. The Chow test essentially checks whether this assumption is statistically valid.

Hypotheses of the Chow Test

H0: $\lambda_1 = \gamma_1$ and $\lambda_2 = \gamma_2$ (no structural change)

Ha: $\lambda_1 \neq \gamma_1$ and/or $\lambda_2 \neq \gamma_2$ (there is structural change)

Steps of the Chow Test

State the null and alternative hypotheses, as shown above.
Estimate the regression for the entire sample period (eq. 3), which imposes the restriction that the parameters are the same across both sub-periods. Obtain its residual sum of squares, called the Restricted RSS, denoted $RSS_R$ .
Split the sample into the two sub-periods before and after the suspected break and estimate eq. 1 and eq. 2 separately. Obtain their residual sums of squares, $RSS_1$ and $RSS_2$ .
Add $RSS_1$ and $RSS_2$ together to obtain the Unrestricted RSS: $RSS_{UR} = RSS_1 + RSS_2$ .
Compute the F-statistic using the following formula:

$F = \dfrac{(RSS_R - RSS_{UR})/k}{RSS_{UR}/(n_1 + n_2 - 2k)}$

where $k$ is the number of parameters estimated in each sub-period regression (here, $k = 2$ : the intercept and the slope).

Find the critical F-value from the F-distribution table using numerator degrees of freedom $df = k$ and denominator degrees of freedom $df = n_1 + n_2 - 2k$ .
If the calculated F-statistic exceeds the critical F-value, we reject H0 of no structural change and conclude that a structural change has occurred in the model.

Assumptions of the Chow Test

The error variances in both sub-period regressions must be equal, that is, $\sigma_1^2 = \sigma_2^2$ .
The Chow test does not identify the source of the structural change — it does not tell us whether the difference between the two regressions is due to the intercept, the slope, or both.
The test assumes that the point (or points) of structural break is already known in advance.

A Textbook Illustration

Applied econometrics texts commonly illustrate the Chow test using US personal savings and disposable income data from 1970 to 1995, with the 1982 recession treated as the structural break. The sample is split into 1970–1981 ( $n_1 = 12$ ) and 1982–1995 ( $n_2 = 14$ ), and three regressions are estimated:

$\widehat{Y}_t = 1.0161 + 0.0803\,X_t$ , $RSS_1 = 1785.032$ , $n_1 = 12$

$\widehat{Y}_t = 153.4947 + 0.0148\,X_t$ , $RSS_2 = 10{,}005.22$ , $n_2 = 14$

$\widehat{Y}_t = \alpha_1 + \alpha_2 X_t$ , $RSS_R = 23{,}248.30$ , $n = 26$

Applying the Chow test formula gives:

$F = \dfrac{(23{,}248.30 - 11{,}790.252)/2}{11{,}790.252/22} = 10.69$

For 2 and 22 degrees of freedom, the 1 percent critical F-value is 5.72, and the associated p-value is about 0.00057. Since the calculated F exceeds the critical value, the null hypothesis of no structural change is rejected: the savings–income relationship for the two periods is significantly different.

Note, however, that this result alone does not tell us whether the difference arises from the intercept, the slope, or both — a limitation we address using the dummy variable approach in a separate article.

A Real-World Example: GDP and CPI for Pakistan

To see the Chow test applied to real data, consider Pakistan’s annual real GDP (in billions of national currency, constant prices) and Consumer Price Index (CPI) from 1980 to 2025, sourced from the IMF World Economic Outlook database.

We treat GDP ( $Y_t$ ) as the dependent variable and CPI ( $X_t$ ) as the independent variable, and we test whether the relationship between them experienced a structural change around the 2008 global financial crisis, a well-documented shock to inflation and output in developing economies.

The sample of 46 annual observations is split into two sub-periods: 1980–2008 ( $n_1 = 29$ ) and 2009–2025 ( $n_2 = 17$ ). Three regressions were estimated as follows.

Full sample (1980–2025):

$\widehat{Y}_t = 10{,}963.22 + 153.68\,X_t$ , $RSS_R = 830{,}423{,}252.18$ , $n = 46$

Pre-crisis period (1980–2008):

$\widehat{Y}_t = 4{,}296.95 + 421.60\,X_t$ , $RSS_1 = 15{,}781{,}288.87$ , $n_1 = 29$

Post-crisis period (2009–2025):

$\widehat{Y}_t = 21{,}867.11 + 81.60\,X_t$ , $RSS_2 = 95{,}758{,}345.42$ , $n_2 = 17$

The unrestricted RSS is:

$RSS_{UR} = RSS_1 + RSS_2 = 15{,}781{,}288.87 + 95{,}758{,}345.42 = 111{,}539{,}634.29$

Applying the Chow F-test formula with $k = 2$ :

$F = \dfrac{(830{,}423{,}252.18 - 111{,}539{,}634.29)/2}{111{,}539{,}634.29/(29+17-4)}$

$= \dfrac{359{,}441{,}808.95}{2{,}655{,}705.58} = 135.35$

For 2 and 42 degrees of freedom, the 1 percent critical F-value is approximately 5.15, and the calculated p-value is effectively 0.0000. Since the calculated F-statistic of 135.35 is far greater than the critical value, we reject the null hypothesis of no structural change.

We conclude that the relationship between GDP and CPI in Pakistan changed significantly around the 2008 global financial crisis. The coefficient on CPI falls sharply, from about 421.60 before 2008 to about 81.60 after 2008, while the intercept rises substantially, from about 4,296.95 to about 21,867.11.

This is broadly consistent with the structural shift in Pakistan’s economy over this period: GDP growth in the later years became far less responsive to changes in the price level than it had been in the earlier decades, likely reflecting cumulative structural, exchange-rate, and policy changes in the economy.

As the Chow test itself does not tell us whether this difference stems from the intercept, the slope, or both, this is exactly the kind of question the dummy variable regression approach is designed to answer.

Limitations of the Chow Test

It requires the researcher to already know, or at least strongly suspect, the exact point at which the structural break occurred.
It assumes equal error variances across the two sub-periods, which may not always hold in practice.
It only tells us whether a structural change has occurred, not where the difference lies — whether in the intercept, the slope, or both.

Conclusion

The Chow test remains one of the simplest and most widely used tools for testing parameter stability in regression models over time. It provides a clear, formal way to check whether a suspected event — a war, a policy reform, or, as in the Pakistan example above, a global financial crisis — has genuinely altered the relationship between economic variables.

Its main drawback is that it cannot pinpoint the exact source of the structural change. In the next article, we address this limitation using the dummy variable regression approach, which not only tests for structural change but also identifies whether it arises from the intercept, the slope, or both.

References

D. Asteriou and S. G. Hall, Applied Econometrics
D. N. Gujarati, Basic Econometrics.

Share this article

About the author

Muhammad Minhaj Akhtar

Muhammad Minhaj Akhtar is a Lecturer in Economics at Government Graduate College Jauharabad, Pakistan. He holds an M.Phil. in Economics from Quaid-i-Azam University, Islamabad, and an MSc in Economics from the University of Sargodha, where he earned a Silver Medal. His academic passion lies in Econometrics, with a strong focus on applying empirical methods to real-world economic issues. Through MinhajMetrixHub, he shares learning resources, research guidance, and practical econometric insights for students and researchers.