Big data and LASSO improve health insurance risk prediction
KeAi Communications Co., Ltd.
image:
Variable-group selection results indicate which categories of information are most informative for predicting the study’s health-risk proxies.
view moreCredit: Shaoran Li, et al
Insurers must price and underwrite policies with incomplete information, while applicants often know more about their own health risks. This information gap can contribute to adverse selection and inefficient pricing. A new study published in Risk Sciences investigates whether alternative data sources (“big data”) and modern predictor-selection methods can improve health insurance risk assessment — which data sources are most worth collecting.
The researchers, from Peking University and University of International Business and Economics in China, analyzed proprietary critical illness insurance application and claim information from Chinese insurance company InsurTech. In addition to standard policy and demographic variables, the dataset includes applicant-authorized smartphone-related “label” information, such as device signals, location- and app-related indicators, and credit-inquiry related signals, as well as public medical-claim records from hospitals.
“To capture health risk, we used outcomes tied to critical illness claims as well as information derived from individuals' prior public medical-claim history,” explains lead author Ruo Jia. “We found that adding big data and applying LASSO-style methods improves out-of-sample prediction compared with models relying only on traditional underwriting information.”
Notably, big data obtained from smartphone use offer extra-predictive power in addition to past medical histories.
“Because collecting and processing underwriting data can be expensive, we also applied Adaptive Group LASSO to identify which categories of variables are most useful,” says Jia. “We determined that the most fruitful data collection sources for health insurance underwriting are personal digital devices, recent travel experience, and insureds' credit records.”
The authors emphasize that the analysis is predictive rather than causal: “we do not offer causal interpretations.” They also discuss limitations related to the study's coverage and context.
###
Contact the author:
Shaoran Li (corresponding author)
School of Economics, Peking University, China
lishaoran@pku.edu.cn
The publisher KeAi was established by Elsevier and China Science Publishing & Media Ltd to unfold quality research globally. In 2013, our focus shifted to open access publishing. We now proudly publish more than 200 world-class, open access, English language journals, spanning all scientific disciplines. Many of these are titles we publish in partnership with prestigious societies and academic institutions, such as the National Natural Science Foundation of China (NSFC).
Journal
Risk Sciences
Method of Research
Data/statistical analysis
Subject of Research
Not applicable
Article Title
Data-enriched prediction of insurance risk
No comments:
Post a Comment