Change point estimation in panel data with time‐varying individual effects

Abstract

This paper proposes a method for estimating multiple change points in panel data models with unobserved individual effects via ordinary least-squares (OLS). Typically, in this setting, the OLS slope estimators are inconsistent due to the unobserved individual effects bias. As a consequence, existing methods remove the individual effects before change point estimation through data transformations such as first-differencing. We prove that under reasonable assumptions, the unobserved individual effects bias has no impact on the consistent estimation of change points. Our simulations show that since our method does not remove any variation in the dataset before change point estimation, it performs better in small samples compared to first-differencing methods. We focus on short panels because they are commonly used in practice, and allow for the unobserved individual effects to vary over time. Our method is illustrated via two applications: the environmental Kuznets curve and the U.S. house price expectations after the financial crisis. * We are grateful to

Development of Retail Banking Services in the Context of Digital Transformation

“…The pooled OLS model does not assume individual or time-specific effects, meaning that the coefficients are the same for all individuals and time periods. This can be called the assumption of homogeneity of coefficients across individuals and time periods [33].…”

Structural Breaks in Interactive Effects Panels and the Stock Market Reaction to COVID-19

“…It is, however, still the same breaking constant-only model that is being considered, and many models of interest involve more general regressors. Antoch et al (2019), Baltagi et al (2016), Boldea et al (2020), Hidalgo and Schafgans (2017), and Li et al (2016) do for a general linear panel data regression model what Horváth and Hušková (2012), Kim (2014), and Westerlund (2019) do for the constant-only model. In particular, while Antoch et al (2019), and Hidalgo and Schafgans (2017) propose tests for the presence of a structural break, Baltagi et al (2016), Boldea et al (2020), and Li et al (2016) take the existence of a break as given and focus instead on the breakpoint estimation problem.…”

“…Antoch et al (2019), Baltagi et al (2016), Boldea et al (2020), Hidalgo and Schafgans (2017), and Li et al (2016) do for a general linear panel data regression model what Horváth and Hušková (2012), Kim (2014), and Westerlund (2019) do for the constant-only model. In particular, while Antoch et al (2019), and Hidalgo and Schafgans (2017) propose tests for the presence of a structural break, Baltagi et al (2016), Boldea et al (2020), and Li et al (2016) take the existence of a break as given and focus instead on the breakpoint estimation problem. But while highly complementary in terms of the methods they propose, the assumptions employed are materially different.…”

“…The same critique applies to the paper of Li et al (2016), which uses the principal components method instead of CCE to estimate factors. Boldea et al (2020) also do not consider the asymptotic distribution of their estimated breakpoint, although in their paper T is fixed. Their approach is similar to the one of Antoch et al (2019) in the sense that the estimation is carried out while ignoring the factors.…”

“…This simplicity does, however, come at a cost in terms additional restrictive assumptions. Boldea et al (2020) do not go as far as Antoch et al (2019) and require negligible loadings, but they do assume that the omitted variables bias caused by the factors is time-invariant, up to the breakpoint, which limits the type of factors that can be permitted. Motivated by the above discussion, the present paper develops tools that enable researchers to both test for the presence of a common structural break, and to infer the breakpoint of an existing break.…”