Improving the prediction of the demand for life insurance using artificial neural network estimation procedures
MetadataShow full item record
This study focused on two points: estimation and prediction of the demand for life insurance. Previous studies used diverse perspectives, like actuarial and life-span perspective, in order to understand the demand for life insurance. These approaches have shown inconsistent findings. The inconsistent results were caused by two factors: (a) research models did not always match the theories and (b) an indifference among researchers towards interdependent correlations among influential variables. In order to perform a better estimation and a prediction of the demand for life insurance, this study explored and employed three theories: ecological systemic theory, a transformative consumer research framework, and dynamic nonlinear science framework. A new framework emerged, which was called a dynamic nonlinear systemic framework. A statistical model used for the new framework is artificial neural network (ANN). In the procedure of analysis, data came from the National Longitudinal Survey of Youth 1979. The study split the data into two sets: a training and a testing set. In addition, the study employed clustering analysis to create a sub-sampling of all respondents since the assumption of the study was that consumers do not have the same reason to own life insurance. By executing the clustering analysis, the study yielded three significant groups: (a) low wealth single females, (b) average consumers, and (c) wealthy people with large families. Utilizing the three sub-samples with the two datasets (i.e., training and testing), the study estimated and predicted the demand for life insurance with two different estimations. The first one was a linear estimation using multinomial logistic modeling. The second one was based on nonlinear estimation using ANN modeling techniques. The results were noteworthy. Each cluster had its own influential variables, as well as common influential variables across the three clusters. In addition, prediction using ANN showed a considerably smaller Root Mean of Square Error (RMSE). This means that prediction using ANN was more efficient than linear estimation. As a result, ANN was found to provide better performance when predicting the probability of dropping and purchasing life insurance. Based on the findings, four implications and limitation are discussed.