Forecasting age-specific mortality rates in the Philippines: A maximum entropy approach using statistical moments

• Main Authors: Dayag, John Carlos V., Lugtu, Rani Faith S., Santos, Rafael Lawrence L.
• Format: text
• Language: English
• Published: Animo Repository 2024
• Subjects: Mortality > Philippines > Statistics; Forecasting; Life expectancy > Philippines; Maximum entropy method
• Online Access: https://animorepository.dlsu.edu.ph/etdb_math/40; https://animorepository.dlsu.edu.ph/context/etdb_math/article/1041/viewcontent/2024_Dayag_EtAl_Forecasting_Age_Specific_Mortality_Rates_in_the_Philippines__A_Ma.pdf
• Institution: De La Salle University

Mortality is a crucial factor for different sectors of a nation, namely social security, economic prospects, and the insurance industries. In the Philippines, mortality has improved. However, the constant changes in mortality impose problems in having accurate mortality projections. Inaccurate mortality projections could cause underestimation or overestimation of health care costs, pension schemes, social security policies, and other economic plans. In addition, several mortality models have been created to forecast mortality; however, the data needed are usually unavailable for developing countries such as the Philippines. Moreover, the Philippines lacks recent life table data, which could also result in inaccurate mortality projections. Thus, there is a need to fit more accurate mortality models for the Philippines. This study aims to fit the Maximum Entropy Mortality Model (MEM) using the Philippine data and forecast age-specific mortality rates using the said model. Death counts from the Philippine life tables found in the Developing Countries Mortality Database were used. Since the only life tables available were from 1970 to 2010, the data were split into train and test using the 80/20 method. MEM was fitted to the death counts from the years 1970 to 2002, which resulted in having models up to MEM-6 for males and up to MEM-4 for females. The fitted data were then used to forecast the mortality rates from 2003 to 2010. The accuracy of the models were also compared to benchmark models, Lee-Carter and Hyndman-Ullah, using age-specific life expectancy. MEM models performed better than the said benchmark models. MEM-6 was the best model for males with having the lowest values for MAE and MSE. On the other hand, MEM-4 was the best model for females with the lowest MAE, MSE, and MAPE values. Furthermore, the study also found that higher ages tend to have closer predicted mortality rates to the observed than the lower ages. This is due to MEM failing to capture death counts at age 0.