What Is at Stake If an Insurance Company's Models Aren't Particularly Good at Predicting Risk?

The Short Answer

If an insurance company’s risk models are inaccurate, two things can go wrong — in opposite directions. The company may underprice risk, accepting policies where expected claims will exceed premiums collected, leading to losses and potential insolvency. Or it may overprice risk, setting premiums too high and losing customers to competitors with better models. In either case, the company’s viability is threatened. Beyond the individual firm, poor risk models contribute to adverse selection — where the company disproportionately attracts high-risk customers — which can spiral into a market failure. At the systemic level, flawed risk models across an industry have contributed to financial crises: the 2008 financial crisis is the most prominent recent example of what happens when institutional risk models fail at scale.

Underpricing Risk: When the Model Is Too Optimistic

Underpricing is the more immediately dangerous failure mode. If an insurance company’s model predicts that a certain category of policyholders will generate $500 in average annual claims, and the company charges premiums based on that prediction, but actual claims average $800 — the company is losing money on every policy in that category.

For a few policies, this is manageable. For thousands or tens of thousands of policies, sustained underpricing creates a structural deficit that accumulates over time. Premium income does not cover claims and operating costs. Reserves — the funds insurers are required to maintain to pay future claims — are depleted. If the gap is large enough, the insurer cannot meet its obligations and faces insolvency.

Insurance insolvency is not just a corporate failure; it is a harm to policyholders who purchased protection and may now find themselves uncompensated for legitimate claims. State insurance regulators maintain guaranty funds to provide partial coverage when insurers fail, but these do not fully replicate the protection that policyholders thought they had purchased.

The 2008 financial crisis offers a large-scale version of this dynamic. Companies like AIG had written enormous volumes of credit default swaps — instruments that functioned like insurance on mortgage-backed securities — based on models that dramatically underestimated the risk of correlated mortgage defaults. When housing prices fell nationwide and mortgage defaults spiked simultaneously, the models’ assumptions collapsed and AIG’s losses were so large that the firm required a federal bailout to avoid failure.

Overpricing Risk: When the Model Is Too Pessimistic

The opposite error — overpricing risk — is less catastrophic in the short term but damaging in a competitive market. If an insurance company charges premiums that are too high relative to actual expected claims, it remains profitable on the policies it sells, but it will lose customers to competitors whose models are more accurate and whose premiums are consequently lower.

In a well-functioning competitive insurance market, more accurate risk models confer a competitive advantage: they let a company price more precisely, attract low-risk customers with competitive premiums, and avoid the losses that come with underpricing high-risk ones. A company with consistently overpriced premiums will either lose market share to better-modeled competitors or find itself with a shrinking, self-selecting pool of customers who have fewer alternatives — typically high-risk customers who cannot get coverage elsewhere.

Over time, persistent overpricing either forces model improvement, market exit, or a narrowing of the customer base toward the segment that cannot shop effectively for competing offers.

Adverse Selection and the Market Death Spiral

One of the most significant risks from poor risk models is adverse selection — a market failure described in George Akerlof’s classic “market for lemons” analysis and central to insurance economics.

Adverse selection occurs when buyers have better information about their own risk than the insurer does. If an insurer cannot accurately distinguish high-risk from low-risk customers, it must charge a single average premium. Low-risk customers — who know their risk is below average — find this premium overpriced relative to their expected claims and drop coverage or decline to purchase. High-risk customers — who know their risk is above average — find the same premium attractive and purchase coverage enthusiastically.

The result is a pool of insured customers that is systematically riskier than the general population. Claims rise, the insurer raises premiums to compensate, which drives out more low-risk customers, which raises the average risk of those remaining — and so the spiral continues.

If adverse selection goes uncorrected, it can unravel an insurance market entirely: the equilibrium becomes one where only the highest-risk individuals hold insurance, at extremely high prices that accurately reflect only the worst-case risk pool. This is not a hypothetical — it was a significant feature of pre-ACA individual health insurance markets in the US, where people with pre-existing conditions were priced out of coverage or denied it entirely, creating the exact adverse selection dynamic theory predicts.

Better risk models that can distinguish risk categories — and price each appropriately — reduce adverse selection by offering prices that low-risk customers find fair.

Systemic Risk: When Bad Models Cause Industry-Wide Failures

When multiple insurers use similarly flawed models, the risk becomes systemic rather than isolated to individual firms. The 2008 financial crisis demonstrated this at a large scale, but the principle applies to property and casualty insurance as well.

If the insurance industry as a whole underprices catastrophe risk — flood, wildfire, hurricane — because historical models do not account for changing climate conditions, multiple insurers simultaneously face losses larger than their models predicted when a major catastrophe occurs. Reserves that looked adequate based on historical loss distributions prove insufficient when actual losses exceed historical precedents.

This dynamic is currently visible in the US property insurance market, particularly in California and Florida, where multiple major insurers have withdrawn from or dramatically curtailed coverage. Their existing models — calibrated on historical loss data — did not adequately price the risk of increasingly frequent and severe wildfire and hurricane events. Losses have exceeded premiums, and the response has been exit or extreme premium increases, leaving some markets functionally uninsurable at prices homeowners can pay.

How Insurers Try to Improve Their Models

The core challenge is that risk models are built on historical data, but the future may not resemble the past — particularly under conditions of structural change (new technologies, climate change, demographic shifts, pandemic risk).

Insurers invest heavily in actuarial science, data collection, and increasingly in machine learning approaches that can identify risk factors and correlations invisible to traditional actuarial methods. Better data granularity — increasingly available through telematics in auto insurance, smart home devices, electronic health records in life and health insurance — allows more precise risk segmentation.

Reinsurance also serves as a check on individual model risk: primary insurers cede a portion of their risk to reinsurers, who maintain their own independent risk assessments. When a primary insurer’s model is wrong in one direction, the reinsurance relationship provides a form of error correction and capital buffer. Regulatory capital requirements and mandatory reserve levels serve a similar function — requiring insurers to hold more capital than their own models might suggest is necessary, as a margin for model error.

The deeper lesson from insurance economics is that accurate risk prediction is not just commercially valuable — it is what makes insurance markets function. An insurance market where risk is systematically mispriced is one where coverage either disappears, becomes inaccessible to those who most need it, or contributes to broader financial instability when the mispricing is eventually corrected by reality.