Bayesian Structural Break Detection for the Identification of Effective Climate Policies

In this study, we introduce a Bayesian approach to detect structural breaks in panel data in the context of step-shift indicator saturated models. We focus on identifying the magnitude of the impact of climate policies with uncertain timing. Traditional methods for climate policy evaluation often rely on precise knowledge of when interventions occur, comparing outcomes between treated and non-treated groups. However, the exact timing of policy implementation is often ambiguous, posing challenges to these conventional approaches.

We propose an estimation method for the parameters of indicator saturated model such as those proposed by Pretis and Schwarz (2022) (see also Hendry et al., 2020; Yao and Zhao, 2022) employing Bayesian methods. Our methodological framework offers a coherent probabilistic framework for the detection of structural breaks with unknown timing in panel data. The setup naturally facilitates the quantification of uncertainty around the estimated break dates without additional computational costs. Our approach applies a spike-and-slab prior (using a Dirac spike and a non-local-piMOM slab component) which allows for model consistency with a linearly growing number of parameters, while allowing for Cauchy tails to avoid bias.

In simulation studies, we show that our approach is competitive with existing frequentist approaches when it comes to identifying breaks in time series, and outperforms them in terms of false positive rates. We apply our method to assess the effectiveness of climate policies in the European transport sector (Koch et al., 2022). Our findings largely confirm previous results, but also uncover additional policy effects that merit further investigation.