Application of Cluster Analysis in Building Retirement Wealth Adequacy Profile: A Case Study in Malaysia

Abstract

This study aims to examine retirement wealth adequacy based on the underlying structure of households’ attributes from Household Expenses and Income Survey 2014 (HEIS14) data. In specific, the two-step cluster analysis is carried out to build the profiles that will be used to compute retirement wealth adequacy. In general, the data that contains both numerical and categorical variables do not suit well for pure hierarchical and non-hierarchical cluster analysis. To counter this issue, the twostep cluster analysis employs log-likelihood function to measure the distance between two clusters, where the numerical variables are assumed to follow a normal distribution, meanwhile categorical variables are assumed to follow the multinomial distribution. We have used Schwarz’s Bayesian Criterion (BIC) and Akaike’s Information Criterion (AIC) to determine the number of clusters to be formed. In addition, we employed a Silhouette measure of cohesion and separation to examine the overall goodness-of-fit. The analysis successfully generated a solution of three clusters. The first cluster represents households with the majority of SPM holders, receiving an average gross monthly salary of RM 5058.73 and working in the private sector. The second cluster represents households with the majority of diploma holder/professional certificate, receiving an average gross monthly salary of RM 10154.06 and most of them (92%) working in the private sector. The third cluster represents households who work in the public sector with an average gross monthly salary of RM 7185.74. From the microsimulation process, the results show that 51% of households in Cluster 1 have adequate retirement wealth, while only 13% of households in Cluster 3 have adequate retirement wealth.