Technology Category
Reasoning under Uncertainty: Relational Probabilistic ModelsMachine Learning: Calibration & Uncertainty QuantificationPlanning, Routing, and Scheduling: Model-Based Reasoning
Dynamic Count Models with Flexible Innovation Processes for Irregular Maritime Migration
🤖 AI Summary
This study addresses the modeling challenges of heteroskedastic, zero-inflated, and nonstationary count time series arising from maritime migration flows in the Mediterranean (2015–2025) and the English Channel (2018–2025). We propose a Bayesian dynamic count model featuring a latent log-intensity random walk, stochastic volatility, and heavy-tailed innovations, explicitly distinguishing structural zeros from sampling zeros. Inference is performed via MCMC, and predictive performance is evaluated using proper scoring rules—specifically, the logarithmic score and quantile coverage. Our key contribution is the first integration of stochastic volatility into a dynamic count framework, which substantially improves calibration of extreme quantile forecasts (e.g., 99th percentile). Empirical results demonstrate that the model effectively captures pronounced stochastic volatility in migration data, delivering robust probabilistic support for cross-border migration risk early warning and policy decision-making. The framework further holds promise for extension to other zero-inflated, nonstationary count applications, such as infectious disease surveillance.
Technology Category
Application Category
📝 Abstract
Motivated by the dynamics of weekly sea border crossings in the Mediterranean (2015-2025) and the English Channel (2018-2025), we develop a Bayesian dynamic framework for modeling potentially heteroskedastic count time series. Building on theoretical considerations and empirical stylized facts, our approach specifies a latent log-intensity that follows a random walk driven by either heavy-tailed or stochastic volatility innovations, incorporating an explicit mechanism to separate structural from sampling zeros. Posterior inference is carried out via a straightforward Markov chain Monte Carlo algorithm. We compare alternative innovation specifications through a comprehensive out-of-sample density forecasting exercise, evaluating each model using log predictive scores and empirical coverage up to the 99th percentile of the predictive distribution. The results of two case studies reveal strong evidence for stochastic volatility in sea migration innovations, with stochastic volatility models producing particularly well-calibrated forecasts even at extreme quantiles. The model can be used to develop risk indicators and has direct policy implications for improving governance and preparedness for sea migration surges. The presented methodology readily extends to other zero-inflated non-stationary count time series applications, including epidemiological surveillance and public safety incident monitoring.
Problem
Research questions and friction points this paper is trying to address.
Modeling irregular maritime migration count time series
Separating structural zeros from sampling zeros
Evaluating forecasting performance for extreme quantiles
Innovation
Methods, ideas, or system contributions that make the work stand out.
Bayesian dynamic framework for count time series
Heavy-tailed or stochastic volatility innovations
Markov chain Monte Carlo for posterior inference
🔎 Similar Papers
No similar papers found.
