Applied Stochastic Models in Business and Industry最新文献

Prediction uncertainty quantification has become a key research topic in recent years, with applications in both scientific and business problems. In the insurance industry, assessing the range of possible claim costs for individual drivers improves premium pricing accuracy. It also enables insurers to manage risk more effectively by accounting for uncertainty in accident likelihood and severity. In the presence of covariates, a variety of regression-type models are often used for modeling insurance claims, ranging from relatively simple generalized linear models (GLMs) to regularized GLMs to gradient boosting models (GBMs). Conformal predictive inference has arisen as a popular distribution-free approach for quantifying predictive uncertainty under relatively weak assumptions of exchangeability, and has been well studied under the classic linear regression setting. In this work, we leverage GLMs and GBMs to define meaningful non-conformity measures, which are then used within the conformal prediction framework to provide reliable uncertainty quantification for these types of regression problems. Using regularized Tweedie GLM regression and LightGBM with Tweedie loss, we demonstrate conformal prediction performance with these non-conformity measures in insurance claims data. Our simulation results favor the use of locally weighted Pearson residuals for LightGBM over other methods considered, as the resulting intervals maintained the nominal coverage with the smallest average width.

In health economics, decision-makers rely on models to assess the cost-effectiveness of healthcare interventions and guide resource allocation. Health Technology Assessment (HTA) agencies employ cost-effectiveness models to determine the approval and market access of new therapies within their respective jurisdictions. Health economists use quantitative techniques to synthesize clinical, epidemiological, and economic data to model the costs and effectiveness of a new drug compared to the current standard of care over the lifetime of the patients. These models frequently integrate a wide range of assumptions and data inputs from various sources, which renders them vulnerable to a significant level of uncertainty. Economic models commonly confront multiple forms of uncertainty, such as stochastic uncertainty (first-order), which differs from parameter uncertainty (second-order), as well as the presence of heterogeneity within patient populations. Additionally, structural uncertainty related to the model itself adds another layer of complexity. Uncertainty assessment is essential in model-based health economic evaluations that inform regulatory and reimbursement decisions. Understanding these sources of uncertainty, taking steps to minimize their impact, and analyzing, quantifying, and reporting these inherent uncertainties are crucial for ensuring that health economic models provide robust and reliable insights for effective decision-making. This article examines different types of uncertainty in health economic models and methods to analyze and quantify them, offering practical guidelines with examples from recent literature.

We analyze the role of leverage, lower and upper tail risks, skewness, and kurtosis of real gold returns in forecasting its volatility over the annual data sample from 1258 to 2023. To conduct our forecasting experiment, we first fit Bayesian time-varying parameters quantile regressions to real gold returns, under six alternative prior settings, to obtain the estimates of volatility (as inter-quantile range), lower and upper tail risks, skewness, and kurtosis. Second, we forecast the derived estimates of conditional volatility using the information contained in leverage of gold returns, tail risks, skewness, and kurtosis using recursively estimated linear predictive regressions over the out-of-sample periods. We find strong statistical evidence of the role of the moments-based predictors in forecasting gold returns volatility over the short to medium term, i.e., till 1–5-year ahead, when compared to the autoregressive benchmark. Robustness of our main result is also validated based on a shorter sample involving higher-frequency data. Our results have important implications for investors and policymakers.

In this article, we propose synthetic analytical approximate pricing formulas for basket and basket spread options within a multidimensional correlated skew Brownian motion framework. For basket options, we derive two analytical approximate pricing formulas by utilizing the partial exact approximation, the moment matching method and convex bounds approximation together and achieve accurate and analytical approximations. For basket spread options, we derive a lower bound approximation using the Bjerksund–Stensland-type approach. Numerical examples demonstrate superior performance with consistent robustness and high precision of our formulas, remarkably maintaining excellent performance for high-dimensional options. We also note that these approximate pricing formulas can serve as powerful control variates for the variance reduction of Monte Carlo simulations.

Detecting directional signals in multiple testing is crucial to take targeted and effective measures. In this article, we consider the directional multiple testing under the dependence problem within a three-group model. Given the assumption that the observed data are generated according to an underlying three-state hidden Markov model, we develop oracle and data-driven procedures to maximize the expected number of true discoveries (ETD) while controlling the false discovery rates (FDRs) of both alternative states at their nominal levels. It is shown theoretically that the proposed directional multiple testing procedures are valid and have certain optimality properties for directional FDR-control. An extensive numerical study shows that our procedures are significantly more powerful than their competitors since the former can accommodate the dependence structure among hypotheses. The proposed procedures also exhibit high flexibility by allowing different nominal levels for the two alternative states, which is appealing in cases when the false discoveries of different alternative states are not equally important. As a demonstration, the proposed data-driven procedure is applied to learn the transcriptomic characteristics of bronchoalveolar lavage fluid in COVID-19 patients.

Repairable systems, crucial in reliability studies, are characterized by recurrent failure times modeled as counting processes with intensity functions. This paper explores models for these failure times incorporating imperfect repairs, addressing unobserved heterogeneity via shared frailty models. In this context, our approach involves scenarios with general imperfect repairs, which offer a more realistic perspective compared to the minimal or perfect repair assumptions commonly employed in the reliability literature. We propose hierarchical Bayesian methods to estimate parameters, leveraging the Power-Law Process for initial intensities and gamma distributions for frailty terms. Bayesian methods are highly flexible and can accommodate complex shared frailty models that include random effects and dependencies between units. Applying Bayesian inference with gamma and beta distribution priors, coupled with Monte Carlo simulations, provides a robust methodology for estimating unknown parameters and deriving posterior distributions. This flexibility is crucial for capturing the underlying structure of the data in repairable systems with imperfect repairs. Our hierarchical Bayesian framework accommodates multiple systems, providing insights into failure processes and supporting enhanced maintenance strategies. We demonstrate our approach using a real failure times dataset and evaluate its performance through simulation studies, showcasing its applicability and relevance in practical settings.

In this article analysis of a simple step-stress accelerated life test is considered under progressive type-II censoring. A cumulative exposure model is considered when the latent lifetimes of test units follow the Gompertz distribution with different shape parameters and a common scale parameter. We explore the study by estimating all unknown parameters using classical and Bayesian techniques. The model parameters are estimated using maximum likelihood and Bayesian methods. Subsequently, interval estimates are derived based on the observed Fisher information matrix. Bayesian estimates are obtained using squared error and linear exponential loss functions. Subsequently highest posterior density intervals are also constructed. We examine the efficiency of all estimators through simulation studies. Finally, we provide a real-life example in support of the considered model.