Analytics is critical to the insurance business—it allows us to interpret the data that drives the industry.
What is insurance analytics? For some, analytics means advanced data-mining tools, like neural networks, classification trees, and nonparametric regression. Others think about the traditional triad of statistical inference: hypothesis testing, parameter estimation, and prediction. Medical statisticians test the efficacy of a new drug, and econometricians estimate the parameters of an economic relationship. In insurance, the emphasis is on the prediction of unforeseen events, like an automobile accident or a damage to a home.
I recently wrote an article for the Annual Review of Financial Economics 2015 in which I explained how the insurance industry uses analytics in decision-making. The process of gathering data, comprehending models of uncertainty, making inferences, and communicating results helps insurance companies use data to make decisions.
Since its inception, the insurance industry has worked to limit risk. Insurance and related risk-sharing agreements have existed for over 2,000 years. While early insurance contracts charged premiums that intuitively varied by the risk insured, the use of data for insurance pricing began in the 1600s with research conducted on births and deaths over a 60-year period in London and outlying areas; and, with Edmund Halley’s Breslau life tables, the first systematic way of computing annuity values was published, a cornerstone of the modern life insurance industry. Other pioneers of probability theory and statisticians used marine insurance and life tables to motivate their advances.
Statistician W. Edwards Deming pioneered the use of quality improvement techniques for business processes in the 1980s. But while the use of experimental design techniques for production workers’ small data sets drove the development of business statistics, today analytics is geared toward massive data sets that are external to the analyst and even the company.
When an insurance contract is initiated, the premium income is specified in the contract, but the contract payout, or benefit, is an unknown random variable that must be predicted. When the company is valued, the insurer’s claim obligations aren’t completely known and must be predicted. This prediction is critical for insurance analysts, and proper use of stochastic models is the key to improving its accuracy.
Here’s how stochastic modeling works, in terms of basic insurance processes and aggregations of these processes into operational units in which an analyst may participate:
- Initial underwriting and ratemaking: At this stage, the company makes a decision as to whether to take on a risk (the underwriting stage) and assign an appropriate premium (or rate). Insurance analytics has its actuarial roots in ratemaking, in which analysts seek to determine the right price for the right risk and avoid adverse selection.
- Renewal underwriting and ratemaking: Many contracts, particularly in general (property and casualty) insurance, have relatively short durations, such as 6 months or 1 year. Although there is an implicit expectation that such contracts will be renewed, the insurer has the opportunity to decline coverage and to adjust the premium. Analytics is also used at this policy renewal stage, with the goal to retain profitable customers.
- Claims management: Analytics has long been used to detect and prevent claims fraud; manage claim costs, including identifying the appropriate support for claims handling expenses; and understand excess layers for reinsurance and retention.
- Reserving: Analytic tools are used to provide management with an appropriate estimate of future obligations and to quantify the uncertainty of the estimates.
- Solvency and capital allocation: Deciding on the requisite amount of capital and ways of allocating capital to alternative investment activities is another important analytics activity.
Companies in the financial services sector are linked to data and models of uncertainty in the insurance market, due to the randomness of insurance contract payouts even when controlling for the time value of money. Plus, the aggregation and spreading of risk, known as “risk pooling,” is one of the central tenets of insurance. To truly understand the benefits of diversification and risk pooling, it’s important to use stochastic modeling to examine correlations, associations, and other ways of measuring dependencies among random outcomes.
To learn more about the history of analytics and insurance, as well as to hear more about stochastic modeling, see the article “Analytics of Insurance Markets” in the Annual Review of Financial Economics 2015.
Author’s note: There is a tremendous overlap between insurance analytics and the discipline of actuarial science. Some might view the two fields as synonymous with each other, while others might view the processes of collecting data (e.g., through databases and data warehouses) as squarely part of the analytics purview but outside the traditional domain of actuarial science. In the same vein, actuarial science has historically embraced retirement systems and health care as part of its domain, areas that are outside the insurance analytics I reviewed for my article.
Edward Frees is John and Anne Oros Distinguished Chair for Inspired Learning in Business Fund; Hickman-Larson Professor of Actuarial Science; and professor of risk and insurance at the Wisconsin School of Business.