<![CDATA[Indonesia’s geographical location along the Pacific Ring of Fire makes it one of the most seismically active countries in the world, with earthquakes causing recurrent and significant economic losses. To address the need for more accurate and regionally sensitive insurance pricing, this study develops a Bayesian spatial quantile regression model that estimates the 90th percentile of earthquake-induced economic losses. Unlike conventional models that focus on mean losses, this approach captures the upper tail of the loss distribution, which is essential for designing risk financing instruments that can withstand catastrophic events. The model incorporates two main predictors: earthquake magnitude (on the Richter scale) and a provincial risk exposure index constructed from population and GDP per capita. Spatial effects are modelled using a Gaussian kernel with multiple bandwidths. Based on Leave-One-Out Cross-Validation, a bandwidth of 500 kilometers yields the best model performance, effectively capturing regional dependence in earthquake loss data. Historical data from 1930 to 2024 are used to estimate parameters via Markov Chain Monte Carlo sampling with the No-U-Turn Sampler. Results indicate that both earthquake magnitude and socioeconomic exposure are significant drivers of high-end losses. For instance, the model estimates that West Sumatra and Yogyakarta could experience annual benefit payouts exceeding USD 300,000 in high-severity scenarios. Earthquake insurance premiums are then derived using the expected payout values and a 10% premium loading factor. Premium estimates range from USD 0 to over USD 50,000 across provinces, with 20 out of 34 provinces requiring positive premiums. This study contributes a novel modelling framework that integrates quantile regression, spatial weighting, and exposure-based risk assessment. The results provide a data-driven basis for setting premiums and allocating disaster risk financing more equitably across regions. Limitations include reliance on proxy variables for exposure and the exclusion of building-level vulnerability data, which may affect precision in highly localized assessments.]]>
Copyrights © 2025
