Refining longevity risks for China’s ageing population

How modern mathematics helps policymakers and insurers connect the dots of the country’s fragmented demographic data

Featured faculty: Johnny Li, Chan Wai-sum and Zhu Xiaobai
Written by Putro Harnowo

Just like every dark cloud has a silver lining, insurers are proficient at managing risks and turning them into opportunities. As ageing population has become a pressing issue around the globe, the booming demand for insurance and annuities presents considerable potential.

China, one of the fastest-ageing countries, is expected to see its personal insurance market expand from four trillion Chinese yuan (US$553 billion) in 2021 to between 6.6 and 12.6 trillion Chinese yuan by 2035, according to the Boston Consulting Group 2024 report. Its pension insurance market share is expected to soar from one to 20 per cent, and health insurance is projected to grow from 27 to more than 35 per cent.

The Chinese government has also encouraged commercial insurance, basic pension, annuity, and wealth management funds to participate in listed companies’ private placements as strategic investors. Such a move is a part of a broader strategy to attract long-term capital and enhance the stability of the country’s financial markets.

While a sense of optimism seems to fill the air, there lies a fundamental challenge that could defy expectations. “Mortality data in China is collected through diverse sources and methodologies, resulting in inconsistency,” says Johnny Li, Tan Bingzhao Professor of Actuarial Science at the Chinese University of Hong Kong (CUHK) Business School.

Mortality rate is a fundamental component determining population size and age structure, in addition to fertility and migration rates. Compared to other demographic measures that inform broader societal trends, mortality forecasting has more direct financial and actuarial consequences as it allows insurers and pension funds to price life insurance products, annuities, and pensions appropriately and to maintain sufficient reserves.

Accurate mortality projection is clearly the need of the hour, yet only a handful of attempts have been made to build reliable models. Therefore, Professor Li and his colleagues at the School, Emeritus Professor Chan Wai-Sum and Assistant Professor Zhu Xiaobai, as well as Kenneth Zhou at the University of Waterloo and Felix Chan Wai-Hon at the University of Hong Kong, developed A Bayesian approach to developing a stochastic mortality model for China.

“Bayesian is a technical term, a distinguished approach that we can incorporate as much information as possible and overcome data challenges from different sources,” Professor Li says. “Insurance and financial sectors need a model like this, given the data challenges facing them.”

“For actuaries and insurers, the life expectancy projection is not the only important variable, but the uncertainty surrounding the projection is probably even more important. With the Bayesian approach, we can generate measures of uncertainty.”

Professor Johnny Li

Three criteria for a robust mortality model

As explained in the paper, China’s age and gender-specific mortality data before 1981 are absent, while the data from 1981 to 2014 are either partially or completely missing. The data for 1981, 1989, 2000 and 2010 were obtained from nationwide censuses, but the rest were acquired from surveys of a fraction of the national population, making it arduous to forecast demographic trends accurately.

Many other countries may have similar data issues. However, Professor Li notes that the problem in China is unique due to its significant population size. A reliable mortality model is compulsory to drive more innovative pension policies and healthier ageing strategies, crucial for the current demographic shifts.

For a mortality model to be robust, he underscores three essential criteria. The first criterion is to incorporate as much data as possible so that the model can use any available information. The second criterion is appropriate provision of uncertainty, since the model should consider not only the best estimate but also a range of possible deviations or uncertainty around the estimate.

Uncertainty here refers to the degree of variability produced by the model, quantifying how much the projected mortality rates might deviate from expectations due to various sources and limited data. “For actuaries and insurers, the life expectancy projection is not the only important variable, but the uncertainty surrounding the projection is probably even more important,” he adds. “With the Bayesian approach, we can generate measures of uncertainty.”

The third criterion is parsimonious and biologically reasonable. As the model is basically an approximation of reality, parsimonious refers to a realistic model that captures the necessary details but not overly complicated. Biologically reasonable means the mortality model should produce sensible results, for instance, the mortality rates for the elderly should be higher than those of much younger people.

Wider applications in the markets

Since the initial study proposes a mortality model for the nationwide population, Professor Li and his colleagues at CUHK have developed an enhanced version for different parts of China.

In practice, this enhanced model has been utilised by insurance companies to establish continuing care retirement communities (CCRCs) in Shanghai, Beijing, Guangzhou, Xiamen, and more. CCRC is a residential community providing healthcare, security, social activities, meals, housekeeping, and wellness programmes for the elderly. This concept has been around for more than a century in Europe and the US, and has found its way to China in recent years.

“Our model now includes the extension of geographical specificities that can be used to predict the demands of CCRCs in different cities across China and also to estimate how long people will spend their time in those communities,” Professor Li adds.

Professor Li argues that the enhanced mortality model is also very useful for reinsurance purposes. Insurers that have sold a considerable number of annuity and other insurance products normally share their risks by purchasing insurance policies from reinsurers. Reinsurance companies naturally have larger risk pools than primary insurance companies. Professor Li’s model can be used, for instance, by a reinsurer with portfolios in Shanghai and China’s Greater Bay Area, to leverage diversification opportunities.

Professor Li is currently refining his model to incorporate more factors that may drive life expectancy, such as smoking prevalence, lifestyle and living conditions. This work is expected to improve forecast accuracy.