Bayesian Indicator-Saturated Regression for Climate Policy Evaluation

Abstract

Structural break identification methods are an important tool for evaluating the effectiveness of climate change mitigation policies. In this paper, we introduce a unified probabilistic framework for detecting structural breaks with unknown timing and arbitrary sequence in longitudinal data. The proposed Bayesian setup uses indicator-saturated regression and a spike-and-slab prior with an inverse-moment density as the slab component to ensure model selection consistency. Simulation results show that the method outperforms comparable frequentist approaches, particularly in environments with a high probability of structural breaks. We apply the framework to identify and evaluate the effects of climate policies in the European road transport sector.

Figures (7)

• Figure 1: Probability density function of $iMOM(0,1,1,\tau)$ prior for the cases $P(|\gamma_{i,s}|\leq \sigma|\tau)=0.05$ and $P(|\gamma_{i,s}|\leq \sigma|\tau)=0.01$ corresponding to $\tau=1.92$ (-----) and $\tau=3.32$ (- - -), respectively.
• Figure 2: Performance metrics (BISAM, GETS, ALASSO) for the simulation setting with a sparse break environment, for varying relative break size (measured in standard deviations of the error term).
• Figure 3: Performance metrics (BISAM, GETS, ALASSO) for the simulation setting with a dense break environment, for varying relative break size (measured in standard deviations of the error term).
• Figure 4: Performance metrics (BISAM, GETS, ALASSO) for simulations assuming a relative break size of three standard deviations of the error term ($\sigma^2$) and varying number of breaks.
• Figure 5: Break detection in EU transport emissions. Breaks detected by GETS (black crosses) and BISAM (vertical dashed lines). Green (orange) shaded areas indicate periods in which BISAM detects positive (negative) breaks. Transparency of shading corresponds to (inverse) interval length in which a break has been found for $\Bar{t}_{\{1,2,3\}}$.