Members Academic Journals

Access The Geneva Papers

Machine learning in long-term mortality forecasting

Abstract

We propose a new machine learning-based framework for long-term mortality forecasting. Based on ideas of neighboring prediction, model ensembling, and tree boosting, this framework can significantly improve the prediction accuracy of long-term mortality. In addition, the proposed framework addresses the challenge of a shrinking pattern in long-term forecasting with information from neighboring ages and cohorts. An extensive empirical analysis is conducted using various countries and regions in the Human Mortality Database. Results show that this framework reduces the mean absolute percentage error (MAPE) of the 20-year forecasting by almost 50% compared to classic stochastic mortality models, and it also outperforms deep learning-based benchmarks. Moreover, including mortality data from multiple populations can further enhance the long-term prediction performance of this framework.

Using the Taiwan National Health Insurance Database to explore the need for long-term care

Abstract

Several factors contribute to the lack of long-term care (LTC) insurance in Taiwan, insufficient data and an absence of unified definitions of LTC are two of them. In this study, we use LTC-related catastrophic illness (CI) as the assessment criteria to investigate the demand for LTC insurance. We selected 13 categories of CI and explored the spatial–temporal properties of LTC incidence rates and mortality rates from the National Health Insurance Research Database. The study shows that the incidence rates did not change much, while mortality rates decreased significantly. Taiwan’s LTC population, which was 0.29 million in 2013, is accordingly expected to triple before 2040 based on the proposed cohort change ratio approach. Currently, Taiwan’s government has planned to fund LTC insurance via a pay-as-you-go system. Furthermore, the increasing LTC population indicates that commercial insurance can play a vital role as a supplement to social LTC insurance.

A spatial analysis of the health and longevity of Taiwanese people

Abstract

Initiated in 1995, Taiwan’s National Health Insurance (NHI) programme now covers over 99.6% of its residents, ensuring widespread medical access. Despite this, regional disparities in medical resource allocation persist. This study investigates the potential urban–rural divide in life expectancy and healthcare utilisation. Drawing data from the Ministry of the Interior (population registration records), NHI Research Database (medical utilisation) and Ministry of Health and Welfare (leading causes of death), we employ spatial analysis, visualisation tools and the standardised mortality ratio for assessing regional disparities. Our findings reveal distinct regional mortality differences in Taiwan, with lower rates in northern counties and higher ones in mountainous regions. However, healthcare utilisation shows no significant regional variations. Notably, patterns of overall mortality rates and primary death causes demonstrate spatial clustering.

Access The Geneva Review

Risk preferences and risk perceptions in insurance experiments: some methodological challenges

Abstract

The ability to run experiments, or to see natural data as a quasi-experiment, does not free one from the need for theory when evaluating insurance behavior. Theory can be used to motivate the experimental design, evaluate latent effects from the experiment, or test hypotheses about latent effects or about observable effects that could be confounded by latent effects. The risk, evident in the broader behavioral literature in general, is the attention given to “behavioral story-telling” in lieu of rigorous scholarship. Such story-telling certainly has a role in fueling speculation about possible casual forces at work generating the data we see, but should not be mistaken for the final word. There is also a severe cost in terms of the heroic assumptions needed for identification. Again, such identifying assumptions can have a valuable role, but many general claims rely critically on those assumptions. Controlled laboratory experiments and Bayesian econometric methods should play a complementary role to field experiments and quasi-experiments. One clear lesson from the evaluation of methodological challenges is to use theory more, to explore the ability of “standard economics” to explain behavior. The time has long passed where straw men theories are set up to fail when confronted with behavior. Just as we want to consider flexible parametric functional forms when appropriate, we should be open to conventional economics applied more flexibly.

Evaluating sustainability actions under uncertainty: the role of improbable extreme scenarios

Abstract

An optimality condition for sustainability actions under discounted expected utility is that, ex post, we should almost surely regret having adjusted them too much for risk. In other words, ex post, one would almost surely feel regret for "excess" precautionary saving, excess insurance and hedging coverage, or for excess risk-bearing. Moreover, for marginal investments whose impacts materialize in t years, t tending to infinity, their state-contingent present value tends to zero almost surely, in spite of the fact that their expected value is one. The value of sustainable actions is thus mostly derived from very improbable extreme scenarios.

Welfare analysis in insurance markets

Abstract

“Efficiency” in economics can be employed in two distinct ways: as a statement about the class of policies that all policy advisors would agree to, regardless of their views about distributional preferences; or as something valuable that policy advisors potentially need to trade-off against equity goals. This distinction can be safely put to the side in some settings, for instance when information is symmetric, individuals have linear-in-consumption preferences, and the planner can implement person-specific taxes and transfers. In asymmetric information settings like insurance markets, it cannot. The efficiency notion employed by Einav and Finkelstein (J Econ Perspect 25(1):115–138, 2011) for studying competitive insurance markets (EF-efficiency) can therefore only be understood in the second way, and policy recommendations based on EF-efficiency alone thus amount to a tacit expression of indifference to distributional concerns.