Insurance is, in large part, an exercise in making decisions about an uncertain future.
A life insurance policy may remain in force for decades. An annuity can create obligations extending well into a policyholder’s retirement. Before those obligations unfold, insurers must estimate mortality, policyholder behavior, investment returns and future cash flows, then determine how much capital and liquidity mayb be needed to support the risks they assume.
These calculations are familiar territory for actuaries. What has changed is the amount of information available and the speed at which insurers are expected to make sense of it.
Large reinsurance transactions make the challenge especially visible. When substantial insurance liabilities change hands, the analysis depends on assumptions about events that have not yet happened. Models can estimate those events, but they cannot make the uncertainty disappear.
Yujia “Emily” Jiang encountered that challenge while working on life and annuity reinsurance transactions. Her role included developing actuarial models that projected liability cash flows and testing how those projections changed when assumptions involving mortality, lapse rates, investment yields and foreign exchange moved.
Multiple stress scenarios were used to examine the transactions from different angles. A change in lapse behavior, for example, could alter when cash leaves or remains within a portfolio. Changes in mortality could affect the timing and amount of future benefits. Investment yields introduced another variable into the relationship between the assets supporting a transaction and the liabilities being assumed.
The resulting analysis fed into pricing and valuation work, as well as discussions around assumptions and potential capital impacts. The objective was less about finding a single correct forecast than understanding how far outcomes could move when the underlying assumptions changed.
But a transaction does not stop generating risk once its initial models have been completed.
After a reinsurance transaction closes, the underlying policies begin producing a stream of new information. Policyholders keep policies, surrender them, pay premiums and submit claims. Investment conditions change. The question then becomes whether the assumptions embedded in the original analysis continue to resemble what is happening in the portfolio.
This is where Jiang’s work moved from projection to observation.
She developed an automated Actual-to-Expected monitoring process using Python and Excel for policy and transaction data received from ceding companies. The process compared actual movements in the portfolio with expected experience and was used to track changes in liabilities and identify areas where emerging experience deviates from modeled assumptions. Automation significantly reduced manual work and the time required for the monitoring process.
The distinction between forecasting and monitoring may sound narrow, but it points to a recurring problem in insurance.
An assumption does not necessarily become obviously wrong overnight. Differences may accumulate gradually. Policyholder behavior can shift as interest rates change. Mortality experience can move away from historical expectations. Investment performance can diverge from the conditions assumed when a transaction was evaluated.
For a large portfolio, persistent differences can eventually affect valuations, earnings, reserves and capital assessment.
Actual-to-Expected analysis provides one way to identify and investigate that divergence. Rather than asking only what a model predicts, it asks how the prediction compares with the experience that subsequently arrives.
There is a common thread running through these projects. The first attempts to describe future liabilities, and the second checks those expectations against actual policy experience.
That same sequence becomes relevant at the balance-sheet level.
Jiang now works in quantitative risk analysis, contributing to enterprise risk and capital assessments involving reinsurance, asset-liability management and liquidity. She also monitors reinsurance portfolios, examining investment yields and policy economics in the context of capital adequacy and earnings stability.
Liability modeling and risk managemnent both require an understanding of how insurance obligations may change. Both depend on assumptions about policyholders and financial markets. And both eventually lead to questions about whether an insurer has sufficient resources available when its obligations come due.
The difference is one of perspective. Transaction modeling looks closely at the economics and risks of a particular block of business. Enterprise risk analysis steps further back and considers how liabilities, investments, liquidity and capital interact across an institution.
That movement from transaction-level modeling toward balance-sheet analysis reflects a broader develoopment in insurance risk management. Actuarial projections, portfolio experience, reinsurance analysis and capital assessment are increasingly connected because information generated in one area can change decisions made in another.
The consequences ultimately extend beyond an insurer’s quarterly results. Life insurance and annuity companies make long-term commitments to households that use these products for retirement income, protection against mortality risk and financial planning. Those commitments require insurers to maintain reserves and capital through economic conditions that may look very different from those prevailing when a policy was originally issued.
Across the United States, life insurers and annuity providers manage trillions of dollars in assets while making long-term commitments to policyholders. No single model, analyst or monitoring process can determine whether those commitements will be met. Collectively, however, the actuarial models, experience studies and capital assessments used across the industry form part of the less visible infrastructure supporting them.
As the volume of available information grows, that infrastructure is becoming increasingly data-intensive. Automation makes it possible to examine larger volume of policy and transaction data more frequently, but processing more information does not eliminate uncertainty. In some respects, it creates another challenge: determining which changes are meaningful enough to warrant attention.
Jiang’s experience across reinsurance modeling, portfolio monitoring and risk analysis offers one view of how that challenge is being approached. A projection establishes an expectation. Incoming data tests it. Risk analysis considers what the difference may mean for the balance sheet.
The process is iterative rather than predictive in an absolute sense. No model can know precisely how policyholders or markets will behave years from now. What insurers can do is to improve how they compare expectations with emerging experience and shorten the time between a meaningful change and their understanding of it.
For an industry built around promises extending decades into the future, that may be a more practical measure of progress.