The insurance business conceals an awkward question. Actuarial work is, at its core, pricing the future: how many people will fall ill in thirty years, how many houses will burn, how many ships will sink — compute it all, discount it into today’s premium. But pricing the future requires predicting it, and no one can predict the future. So how does a business model built on the unknowable stay standing for centuries? Read through the industry’s history of failures and the answer comes into focus: the companies that survive never depended on being right.
The Failures Were Never Tail Events
Start with how the actuaries fell. General Re, acquired in 1998, exposed a giant hole in asbestos litigation liabilities: people exposed decades earlier began filing claims decades later, and this ultra-long-tail liability was essentially absent from the previous management’s models. Hurricane Katrina in 2005 broke the entire industry’s models at once — catastrophic flooding and infrastructure destruction exceeded every ceiling the historical data implied. Long-term care insurance died differently: the models had to project healthcare costs and mortality thirty to fifty years ahead, and people lived longer than the models assumed while care costs outran inflation, forcing reserve top-ups again and again.
The systemic cases are AIG and Japan. AIG’s models assumed American house prices could never fall everywhere at once; extrapolating an unprecedented bubble from decades of low-volatility data, the framework collapsed in the subprime crisis and the company became insolvent. Japanese life insurers offered 5–6% guaranteed rates in the 1980s, betting that high growth and high interest rates would persist; after the bubble burst rates went to zero, asset returns could no longer cover policy promises, and century-old institutions like Toho Mutual and Chiyoda Life failed in succession. The pattern is consistent: what kills insurers is not rare events but systemic ones — when risk shifts from isolated cases to a market-wide chain reaction, the law of large numbers fails on the spot.
Berkshire’s Answer: Three Pools of Cash
So how does Buffett’s insurance operation resolve the paradox? Look at how Berkshire’s balance sheet is carved. The enormous cash balance is not one pool but three. The first is float inside the regulatory line: for every unit of risk underwritten, a matching share of highly liquid assets must be held — roughly 60–70% of total cash by estimate. That money is allowed to sit only in short-term Treasuries, for safety rather than return, because if a mega-catastrophe arrives while the market is crashing, being forced to sell stocks at the bottom to pay claims is technical bankruptcy. The second is Buffett’s self-imposed cushion — never less than twenty to thirty billion dollars, above any legal requirement, reserved for end-of-the-world financial crises. Only the third pool is genuinely investable, and when nothing is cheap enough it accumulates into the hundred-billion scale we see today.
This is the direct answer to “nobody can predict the future”: don’t predict; pad. Accept years of trailing the S&P — which is precisely the bill worth naming: the cost of holding cash is the upside genuinely missed — and ensure that no single misjudgment can kill the company. The survival technology of this industry is not a sharper model. It is a thicker cushion, plus the discipline to admit mistakes fast.
Judge the Cushion, Not the Model
Put the failure history and the three pools side by side, and the screening criterion changes. What deserves scrutiny in an insurer is not how elegant its models are but how honest its cushion is — that is what rating agencies are really measuring. Conversely, a company expanding aggressively while offering investment returns far above its peers is almost certainly gambling on the future — trading down its own safety margin for market share, the 1980s Japan script. Buying that high yield means underwriting the day its model fails.
One layer deeper: an insurer’s profit and a policyholder’s return are two faces of the same coin in mathematics. What disciplined underwriting earns is exactly the long-term yield the policyholder gives up. That is why insurance should never sit at the core of an asset allocation — handing money to a model that supposedly cannot be wrong is outsourcing uncontrollable risk. The room to be wrong can only be reserved inside one’s own portfolio. Actuaries cannot predict the future; that fact applies to everyone. The only difference is whether anyone has left a position open for it.
