The Premium

What does a premium buy? Begin with the barest case. A person owns something — a house, a ship, a life's earning power — that some hazard may destroy: a fire that strikes with probability p and, if it strikes, costs L. Facing this peril, the person bears not a known cost but a gamble: most likely nothing, yet with probability p a loss of L. The expected loss — the average across many such years, or many such houses — is the product pL, and this product is the first and most basic price of the risk. It is called the actuarially fair premium: the sum that, collected from each of many who face the same peril, would exactly cover the losses of the unlucky few. An insurer charging precisely pL, with no other costs, would break even in the long run; a wager at those odds is, in the gambler's term, fair. The fair premium is the anchor of all that follows — the floor beneath every real premium, the expected loss made into a price.

But a premium is rarely merely fair, and the reason is that the peril is not merely its average. To carry a one-in-a-hundred chance of losing a fortune is not the same as paying one-hundredth of a fortune for certain, though the two bear the identical expected cost. The gamble carries, besides its average, a spread — the dread possibility of the whole loss — and it is that spread, not the average, that the anxious owner most wishes to be quit of. The premium is therefore the price of two things at once: the expected loss, which any insurer must recover, and the certainty, which the risk-averse will pay extra to obtain. The whole of the next plate is given to that extra — the risk premium proper, the amount above the fair price a person will surrender simply to turn a gamble into a sure thing. For now it is enough to see that a premium buys the removal of a peril, and that the removal is worth, to one who dreads the spread, more than the mere average of what is removed.

Not every uncertain prospect, however, is a peril of this insurable kind, and the distinction was drawn sharply by Frank Knight in 1921. There is risk, where the odds are known or can be estimated — the roll of a fair die, the failure rate of a machine, the mortality of a great population, all governed by probabilities that long experience or sound theory can pin down. And there is uncertainty, where the odds are not known and perhaps cannot be — the prospect of a wholly novel venture, the chance a new product finds a market, the shape of a future that has no precedent to be counted. Insurance lives in the realm of risk: it presupposes a probability p that the insurer can estimate, a peril recurring often enough, across enough cases, for the law of large numbers to bite. True uncertainty, by contrast, is uninsurable — no premium can be set on a hazard whose likelihood no one can name — and it is left, Knight argued, to the entrepreneur, whose profit is the reward for bearing the unmeasurable, and whose return the ninth plate will meet again as the premium markets pay for risk. The premium of this volume is the price of risk, and risk is uncertainty tamed enough to be counted.

With the object so defined, the volume divides as the trade itself divides, into two sides and the price between them. On the side of the one who pays — the demand for insurance — the question is why anyone gives up a sure sum to be spared a merely probable loss, and how much they will give; and the answer, drawn in the next plate, is risk aversion, the curvature of value that makes certainty worth a premium. On the side of the one who is paid — the supply of insurance — the question is how anyone can afford to take on a peril that would ruin the person shedding it; and the answer, drawn in the plate after, is pooling, the law of large numbers that renders a multitude of risks a near-certainty. Between the two lies the premium, which the fourth plate decomposes. The peril, the aversion, the pool, and the price: these four are the foundations, and the eight plates beyond them build on these.

The first plate left a puzzle: why would anyone pay more than the fair premium — more than the expected loss — to insure? At the fair price the trade is, on average, a wash; pay a loaded premium and it is, on average, a loss. A being that cared only for the average of its wealth would never insure at a loaded price, and would be indifferent to insuring at a fair one. Yet people insure eagerly, and pay dearly to do so. The resolution is that people do not care only for the average of their wealth. They care for the satisfaction wealth brings, and that satisfaction does not rise in step with wealth but more slowly — and from that single curvature the whole demand for insurance flows.

The idea is old, and its origin is a paradox. In 1738 Daniel Bernoulli considered a wager — the St. Petersburg game — whose expected monetary value is infinite, yet for which no sane person would pay more than a trifling stake. Why? Because, Bernoulli proposed, what a person values is not money but the use of money, and the use of each added coin is less than the use of the coin before it: a hundred ducats mean far more to a pauper than the hundred-thousand-and-first to a prince. Utility, the satisfaction wealth yields, rises with wealth but at a diminishing rate — the curve of utility against wealth bends, concave, flattening as it climbs. Two centuries later von Neumann and Morgenstern set this intuition on rigorous foundations, showing that a person whose choices among gambles obey a few reasonable axioms acts exactly as if maximizing the expected value not of wealth but of some such utility of wealth. And the concavity of that utility curve is precisely risk aversion. For if utility bends, then the utility of a sure average exceeds the average of the utilities of the gamble's outcomes — the loss of a sum costs more utility than the gain of the same sum yields — so the person prefers the certain average to the uncertain prospect, and will pay to make the swap. The figure shows why.

The figure names the two quantities on which the rest of the volume turns. The expected utility of the gamble, read off the chord, falls below the utility of the expected wealth, read off the curve above it — that gap is the cost of risk, the utility lost merely by facing the spread rather than its average. To express that loss in money rather than utility, drop from the expected-utility level back to the curve: the wealth at that point, the certainty equivalent, is the sure sum that would leave the person exactly as well off as the gamble. It lies below the expected wealth, and the distance between them — the risk premium — is the price, in money, of the risk: the most a person would pay, over and above the expected loss, to be rid of the spread entirely. A risk-averse person's maximum willingness to pay for full insurance is just the expected loss plus this risk premium; the steeper the curvature of utility, the larger the risk premium, the more eagerly and dearly the person insures. How sharply utility bends — the measure of risk aversion itself — Kenneth Arrow and John Pratt made precise, as the rate at which the marginal utility of wealth declines; and that single curvature, varying from person to person and growing as the stakes rise toward the whole of one's wealth, governs the entire demand side of the market for risk.

If the second plate explained the demand, the third must explain the supply, and it faces the mirror puzzle. A peril that would ruin the household shedding it — the fire that takes the house, the illness that takes the earnings — does not vanish when an insurer assumes it; someone must still bear the loss when it falls. How, then, can the insurer afford to carry what the insured could not? The answer is that the insurer does not carry one risk. It carries many, and in the carrying transforms them. The single fire is a catastrophe; ten thousand fires, of which a known fraction will occur, are a budget. The peril that is wild and ruinous to its bearer alone becomes, when pooled with thousands of its independent kin, tame and very nearly certain — and it is this transformation, not any greater appetite for risk, that lets the insurer take on for a small price what was insupportable to hold alone.

The transformation is worked by the law of large numbers, the deepest theorem in the economics of insurance. Suppose each of n households faces the same independent peril — the same probability of the same loss — and an insurer pools them all. The total claims the insurer must pay will vary from year to year, but the average claim per household varies far less, and the more households in the pool, the less it varies: the spread of that average, measured by its standard deviation, shrinks in proportion to the square root of the number pooled, dwindling toward nothing as the pool grows large. With ten households the average loss is still uncertain; with ten million it is almost exactly the expected loss, year upon year. The insurer, holding that average, faces a near-certainty where each household faced a gamble — and so can charge each little more than the expected loss and still, with overwhelming probability, cover its claims. This is the engine. Pooling does not abolish risk; it cancels the independent part of it against itself, the bad years of some offset by the good years of others, until what remains, spread across the multitude, is small. The figure draws the cancellation.

Two readings of the engine deserve separating, for the volume will need both. The first is pooling proper, the law of large numbers just drawn: many independent risks, gathered into one portfolio, shrink in their average to a near-certainty, and this is how the mutual society and the joint-stock insurer alike turn perils into budgets. The second is risk spreading, the transfer of a risk from those who feel it sharply to those who feel it little. Even a risk that cannot be pooled away — a single great risk, or a correlated one — can be made bearable by slicing it among many backers, each taking a small share of a loss that would crush any one of them, as the syndicates of Lloyd's have underwritten the unpredictable for three centuries, and as reinsurers and capital markets stand behind the insurers themselves. Pooling reduces the total risk; spreading merely divides it; and the great institutions of insurance combine the two, pooling what is independent and spreading what is not. But the whole apparatus rests on a single load-bearing word in the law of large numbers — independent — and when the risks are not independent, when they fall together, the engine seizes. That failure is grave enough to claim a plate of its own, the tenth; here it is enough to mark that the magic of the mutual is the magic of independence, and no broader.

The second plate fixed the most a buyer would pay — the expected loss plus his risk premium — and the third fixed the least a seller could accept — the expected loss plus the cost of bearing and administering the risk. The fourth plate sets the price between them, and in doing so decomposes the premium into its parts. At the bottom is the actuarially fair premium, the expected loss, which the insurer must on average pay out and so must on average collect: this is not a charge for service but a pass-through of the peril itself, the money flowing from the lucky many to the unlucky few. Above it sits the loading, everything the insurer adds beyond the expected loss — the cost of running the business, the cost of the capital it must hold against bad years, the cost of the asymmetries the later plates will name, and its margin of profit. The premium a buyer actually pays is the sum of the two: fair premium plus loading, the peril plus the price of relief from it.

Why does the buyer consent to pay more than the expected loss — to pay the loading at all? Because, as the second plate showed, the expected loss is not what the risk is worth to him. The risk-averse buyer would give the expected loss plus his risk premium to be rid of the peril; the insurer will take it on for the expected loss plus its costs; and provided the buyer's risk premium exceeds the insurer's loading — provided, that is, the buyer dreads the risk by more than it costs the insurer to carry it — there opens between them a zone of agreement, a range of premiums at which both come out ahead. The buyer, paying a loaded premium, is still better off than bearing the spread; the insurer, collecting it, is paid more than its cost. This is the crux, and it dissolves a common suspicion. The loading is not a fleecing, not value taken from the buyer; it is the price of a genuine service — the pooling and the bearing of risk — and within the zone of agreement the trade is a true exchange, leaving both parties better off than before. The figure lays the bargain bare.

What fills the loading, and how large it grows, is itself a study. Part is the plain cost of administration — the agents, the actuaries, the offices, the adjusters who settle claims. Part is the cost of capital: because pooling tames the spread but does not abolish it, the insurer must hold a buffer of capital against the bad years, and that capital demands a return, which the loading must supply — the very risk premium of the ninth plate, paid now by the policyholder to the insurer's shareholders for bearing the residual, undiversifiable risk. Part, in markets afflicted by them, is the cost of the asymmetries: the screening against adverse selection, the monitoring against moral hazard, the claims that fraud and carelessness inflate, all of which the honest must help fund. And part is profit. Competition presses the loading down toward the true cost of the service, and in a well-functioning market the ratio of loading to fair premium — what the trade calls the expense ratio — is the measure of the industry's efficiency and the intensity of its rivalry. But the deepest lesson of the plate is the one the figure shows: that because the buyer is risk-averse, insurance creates value even when its price exceeds the expected loss, so that the gains from the trade in risk are real and not a mere redistribution — the world is wealthier in welfare, if not in money, for every premium struck within the zone.

The first four plates settled why insurance exists and what it costs; the fifth asks how much of it to buy. The simplest theory seems to counsel buying all of it: if shedding risk is good, and the risk-averse buyer gains by insuring, why not insure completely, against every peril, to the last dollar of possible loss? Yet no one does, and no insurer expects them to. Real policies are riddled with partial coverage — the deductible, the sum the insured bears himself before the policy pays; the coinsurance, the fraction of each loss he continues to carry; the policy limit, the ceiling above which he is again on his own. The household that insures its house against fire still pays the first thousand dollars of any claim; the patient still pays a share of the bill. Partial coverage is not a defect of these markets but their normal condition, and the plate must explain why the rational buyer, who gains from insurance, nonetheless declines to buy all of it.

The decisive result is Jan Mossin's, of 1968, and it turns on the loading. Were insurance sold at the actuarially fair premium — the expected loss and nothing more — a risk-averse person would insure fully, buying complete coverage and converting his gamble into perfect certainty, for at a fair price every dollar of coverage is a free reduction of risk and he wants all of it. But insurance is not sold at a fair price; it carries a loading, and each dollar of coverage therefore costs slightly more than the expected loss it covers. Now the buyer faces a trade-off at the margin: more coverage means less risk but a larger loading paid, and the rational stopping point is short of full. The reason is exact and instructive. The last increments of coverage insure against the small, likely losses — the part of the loss distribution near zero — and against small losses a person is very nearly risk-neutral, his utility curve there almost straight, so the risk premium he would pay to shed them is tiny, smaller than the loading they cost. He therefore declines to insure the small losses and bears them himself, insuring only the large ones whose spread he truly dreads. A loaded premium buys partial coverage; only a fair one buys full; and the deductible is the mark of the loading made visible.

The forms of partial coverage are not arbitrary, and the most common is, under a natural condition, the most efficient. Kenneth Arrow proved that if the loading is a fixed proportion of the expected loss, the policy a risk-averse buyer most prefers is precisely a deductible: full coverage of every loss above a threshold, no coverage below it. The logic is the second plate's, sharpened — concentrate the coverage where the dread is greatest, on the large losses, and bear the small ones yourself, because the small ones cost more in loading than they are worth in relief. The deductible has, besides, three further virtues the later plates will prize: it spares the insurer the cost of processing a swarm of tiny claims, so shrinking the loading itself; it leaves the insured exposed to the first slice of every loss, and so preserves his incentive to take care, blunting the moral hazard of the seventh plate; and a menu of deductibles, offered at different prices, can sort the high-risk from the low, easing the adverse selection of the sixth. Coinsurance — sharing each loss in fixed proportion — serves the same incentive ends by a different geometry. The optimal coverage, in short, is rarely all and rarely none; it is the interior point where the relief from the last dollar of coverage just balances its loading, and the shape of the policy that achieves it is the deductible.

The first five plates assumed the insurer knew the risk it priced. The sixth abandons that assumption, and with it the volume crosses into the territory of the twelfth volume, on asymmetry, where one party to a trade knows what another does not. In insurance the asymmetry is acute and characteristic: the person seeking coverage commonly knows his own risk — his health, his habits, the soundness of his house, the care he takes — far better than the insurer who must price it. And this private knowledge bends the demand in a fatal direction. Those who know themselves most at risk are the most eager to insure, and will pay the most; those who know themselves safe are the most inclined to go without. The buyers an insurer attracts are therefore not a fair sample of the population but its worst slice, weighted toward the very risks the insurer would least like to hold. The selection of who insures is adverse — adverse, that is, to the insurer — and it is the master malady of these markets.

Trace what the asymmetry does to the price. An insurer that cannot tell the high-risk from the low must charge a single premium, and the natural one is the average cost of all who might buy. But at that average premium the low-risk are overcharged — they are paying for the high-risk pooled with them — and being overcharged, some of them decline and drop out. Their departure leaves the pool riskier than before, so the average cost rises, so the premium must rise; and the higher premium drives out the next-safest tier, who were worth covering at the old price but not the new, which raises the average again. Round and round the spiral turns, each increase in price worsening the pool that sets the price, in the insurance form of the unraveling that Akerlof found in the market for used cars — the good risks driven out by the bad until, in the extreme, only the worst remain and the market for the rest has collapsed. The figure draws the equilibrium this produces. Its novelty is the cost curve: because the eager buyers are the high-risk, the average cost of those who buy falls as a lower price coaxes in the safer, so the average-cost curve slopes downward, and the marginal buyer — always safer than the average already in — lies below it. Competition drives the premium to the average cost of the pool; but efficiency would set it at the cost of the marginal buyer, lower and further out, and the gap between them is insurance that ought to be written and is not.

The maladies of asymmetry breed their own medicines, and adverse selection has three. The first is screening, the insurer's attempt to learn what the buyer knows — the medical examination, the questionnaire, the inspection — and, where direct knowledge fails, the cleverer device of the self-revealing menu. Michael Rothschild and Joseph Stiglitz showed, in 1976, that an insurer facing two hidden types can offer two policies designed so that each type chooses the one meant for it: full coverage at a high price, which the high-risk take, and partial coverage at a low price, which the low-risk prefer because their lesser fear makes the cheaper, thinner policy worth more to them than to their riskier neighbors. The menu sorts the pool by making the buyers sort themselves — a separating equilibrium, in which the low-risk buy too little coverage, a real cost, but the pool no longer unravels. The second medicine is experience rating: pricing each by his record, the claims he has made, the accidents he has had, so that the hidden risk reveals itself over time and the premium comes to track it. And the third is the most radical — compulsion. If exit is what drives the spiral, forbid exit: mandate that all must buy, the safe pooled with the unsafe at a common rate, and the death spiral cannot begin because the low-risk cannot leave. The individual mandate of a national health scheme is, in the language of this plate, an engineered defeat of adverse selection, and it carries the volume to the doorstep of the fairness contest the eleventh plate will join — for to forbid the low-risk to leave is to make them subsidize the high, which is solidarity to its friends and injustice to its critics.

The sixth plate's asymmetry was of knowledge — the insured knew his risk and the insurer did not. The seventh plate's is of action: the insured, once covered, does things the insurer cannot see and would not wish. For insurance changes behavior. A man who has insured his warehouse against fire is a little less diligent about the wiring; a shipper whose cargo is covered watches the weather a little less anxiously; a patient whose every visit is paid for sees the doctor a little more freely than one who pays from his own purse. The coverage that was bought against a loss comes, perversely, to raise the loss it was bought against — and because the insurer cannot observe the slackened care, it can neither price nor forbid it. This is moral hazard, a phrase inherited from the early fire underwriters, who used it for the temptation a policy created in its holder, and given its modern economic form by Kenneth Arrow and Mark Pauly: the hidden action of an insured who no longer bears the full consequences of his own conduct.

The mechanism is a matter of incentive, not of wickedness. Care is costly — the lock, the inspection, the caution, the forgone pleasure — and a person takes care up to the point where its marginal cost equals the marginal benefit it brings him. Uninsured, that benefit is the whole reduction in his expected loss; insured, it is only the small part of the loss he still bears, the rest having passed to the insurer. So the insured, weighing his private cost against his now-diminished private benefit, takes less care than he would unprotected, and less than is efficient, for the care he forgoes would have spared a loss that someone — the pool — must still pay. Carried to its limit the logic is stark: a person fully insured, who bears none of the loss, has no private reason to take any care at all, and full coverage would invite the very ruin it indemnifies. The same logic has an after-the-loss form — the patient, fully covered, who consumes treatment up to the point of no further benefit because it costs him nothing, inflating the claims of all — and a systemic form the later volumes will press, in the bank that gambles the more boldly for knowing it will be rescued. In every case the insurance, by severing conduct from consequence, breeds the loss.

The cure is the same instrument the fifth plate already met, now pressed into a second service. If full coverage destroys the incentive to take care, then leave the insured bearing a slice of every loss — a deductible, a coinsurance, a co-payment — and his incentive partly revives, for he again has a private stake in the outcome. But the slice that restores his care is risk he must carry, and risk-bearing was the very thing insurance was meant to relieve; so the contract is caught in a genuine tension, between spreading risk and preserving incentive, and can serve neither fully without failing the other. The optimal policy is therefore, once again, partial — but now for a reason quite distinct from the fifth plate's loading. There, partial coverage was the buyer's free choice, declining to insure the small losses he scarcely feared; here, it is the insurer's imposition, forcing the insured to keep a stake so that he will keep up his guard. The deductible serves both masters at once, which is why it is ubiquitous; experience rating, which prices next year by this year's claims, restores the incentive across time; and direct monitoring, where it is feasible, lets the insurer reward the care it can see. None abolishes the tension. When action is hidden, there is no contract that both fully insures and fully preserves the incentive to take care — and that impossibility, not mere cost, is why the world's insurance is everywhere partial.

The seventh plate completes the volume's account of insurance proper; the eighth begins its account of the wider thing of which insurance is one instance — the pricing of risk wherever it is borne. The bridge is the third plate's own engine, pooling, seen now from the side of the one who holds the risks rather than the one who sheds them. An investor who places his wealth in a single venture bears that venture's whole risk; an investor who spreads it across many ventures is, like the insurer pooling many policies, holding a portfolio, and the same law of large numbers that tamed the insurer's book now works upon the investor's. The question of the plate is exactly which risks the spreading tames and which it cannot — for the answer divides all risk into two kinds, and that division is the deepest idea in the pricing of risk.

Harry Markowitz, in 1952, gave the division its modern form. Consider the risk of a portfolio — the spread of its return — as more and more assets are added to it. Part of each asset's risk is peculiar to itself: the firm whose factory burns, the venture whose product fails, the mine that floods, hazards independent from one holding to the next. This idiosyncratic risk behaves exactly like the insurer's independent policies — pooled across many assets, the good and bad surprises cancel, and the portfolio's exposure to any one of them shrinks toward nothing. But part of each asset's risk is not peculiar to itself; it is shared with the whole market — the recession that depresses every business, the panic that drains every price, the rise in rates that burdens every borrower — and this systematic risk does not cancel, because it does not differ from one holding to the next: when it strikes, it strikes all at once and all together. So as the portfolio grows, its risk falls, but not to zero; it falls to a floor, the systematic risk that no diversification can remove. The figure draws the descent. The shrinking gap above the floor is the idiosyncratic risk, erased for free by spreading; the floor itself is the systematic risk, which spreading leaves untouched. Total risk is the sum of the two, and only one of them can be diversified away.

From this division the central principle of the pricing of risk follows, and it is a principle of startling reach. A holder who is well diversified bears almost none of any asset's idiosyncratic risk — he has cancelled it against the rest of his portfolio — and so he will neither pay to be rid of it nor demand a reward for holding it: diversifiable risk, to a diversified holder, is free, and free risk earns no premium. What he cannot escape, and what he must therefore be paid to bear, is the systematic risk, the part that moves with the whole and survives all spreading. Hence the rule that governs the price of risk in every market: only undiversifiable, systematic risk is compensated; idiosyncratic risk, poolable to nothing, commands no return. This single principle reaches back to unify the volume. The insurer's pooling and the investor's diversification are one act under two names, each cancelling independent risk against itself; the perils that insurance handles cheaply — the scattered fires, the individual deaths, the separate accidents — are cheap precisely because they are idiosyncratic and pool away, while the perils that defeat insurance, the correlated catastrophes of the tenth plate, are dear or uninsurable precisely because they are systematic and do not. And it reaches forward to the ninth plate, which must now ask the question the principle poses: if only systematic risk is priced, what exactly is its price?

The eighth plate ended with a question: if only systematic risk is paid for, what is its price? The answer is the Capital Asset Pricing Model, built in the 1960s by William Sharpe and others upon Markowitz's foundation, and it is at once a theory of how risk is priced and a measure of how much of it an asset carries. The measure is beta. An asset's beta is the degree to which its return moves with the return of the market as a whole — how violently it rises in the booms and falls in the busts that no one can diversify away. An asset with a beta of one moves, on average, in lockstep with the market; one with a beta of two amplifies the market's swings; one with a beta of zero is uncorrelated with the market, its risk wholly idiosyncratic and so, by the eighth plate's principle, unpriced. Beta, in short, isolates exactly the systematic part of an asset's risk — the only part a diversified holder bears — and discards the idiosyncratic part that diversification has made free. It is the quantity of priced risk in an asset.

The price of that risk is drawn by the security market line. The model holds that the expected return on any asset is the risk-free rate — the return on a holding with no systematic risk, the third volume's rate — plus a reward proportional to the asset's beta: the more systematic risk it carries, the more return the market demands to hold it. Plot expected return against beta and the relation is a rising straight line, anchored at the risk-free rate where beta is zero and climbing with a slope equal to the market risk premium, the extra expected return the market grants for bearing one full market's worth of undiversifiable risk. The figure draws the line. That slope is the price of risk itself, the reward per unit of systematic exposure; expressed against the spread rather than the beta, it is the famous ratio of William Sharpe, the reward-to-risk of the best attainable portfolio, and it is the same quantity the eighth plate's frontier would yield, tangent to the line drawn from the risk-free asset. The equity premium — the margin by which the return on the broad market of stocks exceeds the risk-free rate — is this price made concrete: the going rate, in the market for risk, for holding the one risk that cannot be shed.

Two threads run from the line into the rest of the table. The first is a puzzle that has resisted the discipline for forty years. The equity premium that history records — the margin by which stocks have outrun safe bonds over the long run — is very large, far larger than the modest curvature of any plausible utility curve can justify; to rationalize it within the second plate's framework would require a risk aversion so extreme as to be absurd in every other setting. Rajnish Mehra and Edward Prescott named this the equity premium puzzle in 1985, and its many proposed solutions — rare disasters, habits, a distaste for ambiguity, the very behavioral distortions of the eleventh plate — have each illuminated a corner without closing the case, so that the price of market risk remains, embarrassingly, higher than the theory of risk aversion can explain. The second thread ties the plate to the third volume, on the rate: the risk-free rate is the floor from which the security market line rises, the equity premium the height it climbs, and the return on any risky asset the sum of the two — the pure price of time and the price of risk, added together, which is why this volume and the volume on the rate are read as a pair. And beneath both threads lies the unification the plate completes: the insurer's loading of the fourth plate and the market's risk premium of the ninth are, at bottom, the same payment under two names — the reward demanded by whoever agrees to hold the undiversifiable risk that someone else is anxious to shed. Insurance and finance are one subject, and its name is the price of risk.

The third plate rested the whole edifice of insurance on a single word — independent — and warned that when it failed, the engine would seize. The tenth plate is that failure. Some perils do not strike at random, one here and one there, the misfortunes of the few covered by the premiums of the many. They strike all at once and all together: the earthquake that levels a city, the flood that drowns a valley, the hurricane that flattens a coast, the pandemic that sickens a continent, the financial panic that ruins every holder in the same season. Against such correlated perils the law of large numbers offers no protection, for there is no average to be tamed — the losses do not cancel but compound, ten thousand claims arriving in a single dreadful month rather than scattered across the years. Correlated risk is the systematic risk of the eighth plate wearing its most violent face; it is the risk that diversification cannot diversify and pooling cannot pool, and it is the natural limit of the insurable.

From this the conditions of insurability follow, the checklist by which underwriters have always sorted the coverable peril from the uncoverable. A risk is readily insurable when its probability can be estimated, when the losses are independent across the insured so that pooling tames them, when no single event is so large as to swallow the insurer's capital, when the loss is definite and verifiable, and when adverse selection and moral hazard are kept within bounds. The catastrophic perils fail the conditions that matter most. Their losses are correlated, so pooling fails; and any one event is enormous, so it threatens the insurer with ruin rather than a bad year. To insure such a peril at all, the insurer must hold vast capital against the rare calamity, and the cost of that idle capital — the ninth plate's price of systematic risk, demanded now by those who fund the insurer — drives the premium far above the expected loss, often so far that no buyer will pay it. Where the premium that would make the insurer solvent exceeds the premium any buyer will bear, the zone of agreement of the fourth plate closes, and the peril passes out of the realm of private insurance altogether. It is not that catastrophes are uninsurable in principle; it is that their correlation makes their honest price unbearable.

What lies beyond the boundary is not nothing, but three further institutions, each a way of bearing what the simple pool cannot. The first is to spread the correlated risk wider than any one insurer — reinsurance, by which insurers insure one another across the whole globe, so that a catastrophe local to one is diluted among many; and the catastrophe bond, by which the risk is sliced and sold to the capital markets, whose vast pool of investors can each take a sliver of a peril that would break any insurer, and who hold it precisely because, to a globally diversified portfolio, even an earthquake may be idiosyncratic. The second is to restore independence by mixing perils and places — a book of risks drawn from many regions and many hazards, uncorrelated with one another, recovering across the portfolio the cancellation that any single peril denies. And the third, where private spreading still falls short, is the state as insurer of last resort. For the government commands what no insurer does: it can spread a loss across the entire population and, by borrowing, across the generations; it can compel participation and so defeat selection; and it can stand behind the floods, the crops, the bank deposits, the acts of terror, and the pandemics that the private market, left alone, would leave bare. The supreme instance is the gravest correlated risk of all — the changing climate, whose losses fall together across whole regions and whole decades, the eleventh volume's externality of externalities here wearing its insurer's aspect — and which is, for that reason, straining the boundary of the insurable as nothing before it. Beyond independence lies the reinsurer, the capital market, and the state; and beyond even these lies the uninsurable, the risk that must simply be borne.

The volume has, until now, granted its central character one steady virtue: the buyer of the second plate weighed his perils coolly, dreading each in exact proportion to its probability and its size, and paid the premium his honest utility named. The eleventh plate withdraws that grant, and in doing so turns the volume from the mechanics of the premium to the two great doubts that hang over it — whether the price is true, and whether it is fair. The first doubt is the buyer's. People do not weigh probabilities as the theory assumes. They overweight the small ones and underweight the large, attending to the vivid possibility rather than its true frequency, as Daniel Kahneman and Amos Tversky showed in the work that founds the twenty-fourth volume, on bias. And this misweighing falls with peculiar force on the demand for insurance, whose whole business is the weighing of chance.

The figure draws the distortion, and its consequences run exactly counter to good sense. Because small probabilities loom larger than they are, people over-insure against the vivid minor risk — buying the extended warranty on the appliance, the flight insurance at the gate, the low deductible the fifth plate showed to be a poor bargain — paying heavy loadings to be rid of risks they could easily bear, simply because the small chance of loss looms large in the mind. And because large probabilities are underweighted, and remote catastrophes discounted toward zero until they feel impossible rather than merely unlikely, people under-insure against the rare ruinous peril — leaving the home unguarded against the flood, the earthquake, the once-in-a-century storm — precisely the correlated catastrophes of the tenth plate, and precisely the risks the rational dread of the second plate would have them cover first. The buyer over-insures the trivial and under-insures the grave: the demand for insurance is bent, predictably, by the misweighing of chance, and the bend is the opposite of what cold reason would counsel.

If the first doubt is whether the buyer reckons his risk truly, the second is deeper, and it concerns the insurer's reckoning rather than the buyer's. For the premium is not merely a price; it is a measurement, and what it measures is the person. To set a fair premium, the insurer must estimate each buyer's risk, and to estimate it must classify — sorting the applicant by age and sex and health, by the place he lives and the work he does and the record he has run up — and charging each according to the class he falls in. The premium, accurately set, is a verdict on who you are. And against the justice of that verdict stands a rival idea of fairness, old as the mutual societies themselves, which holds that some risks ought not to be priced to the person at all. Here the volume meets its hardest contest, and it is a contest economics can frame but cannot decide.

Set the two ideas of fairness side by side. The first is actuarial fairness: that each should pay a premium equal to his own true expected loss, no more and no less, so that no one subsidizes another and the price tells each the honest cost of his own risk. By this standard, to charge the same premium to a high-risk and a low-risk person is itself unjust — a tax on the careful to relieve the reckless, a hidden transfer the sixth plate's adverse selection would in any case unravel. The second is solidarity: that some misfortunes are not the person's doing or desert — the genes he was born with, the illness he did not choose, the neighborhood he could not leave — and that to price these to the individual is to punish him for his luck, abandoning the unfortunate to premiums they cannot pay precisely when they need cover most. By this standard, the just premium is the common one, the fortunate pooled with the unfortunate, the risk borne by the community rather than charged to its victims. The two are irreconcilable, and the choice between them is not a matter of fact but of value.

The contest is not abstract; it is fought, statute by statute, over what the insurer may be permitted to know and to price. May a health insurer charge the sick more than the well, or must it offer community rating, one price to all, as solidarity demands and adverse selection — checked only by a mandate — allows? May a life insurer demand a genetic test, pricing the buyer on a fate written before his birth, or is some knowledge too fatalistic to be priced? May premiums turn on sex, on race, on the postal code that stands proxy for race — the redlining by which whole neighborhoods were once denied — or are these classifications, however actuarially sound, forbidden as unjust? Each answer draws the line between the priceable and the unpriceable in a different place, and each is a verdict not on the buyer's risk but on the polity's values: how much it will let the cold accuracy of the premium sort its members, and how much it will insist that some risks be shared because they are no one's fault. The premium, here, is revealed as the most intimate of prices — a number presuming to say what a life is likely to cost — and the question of what it may justly contain is among the sharpest at the meeting of the market and the moral.

The volume set out to explain a price stranger than the wage or the rent or the rate — the price of risk itself, the premium a person pays to be rid of a peril he cannot carry. Eleven plates have traced it, and the account may now be gathered. The premium is the price of transferring risk: what the risk-averse pay to turn an uncertain loss into a certain cost, and what the risk-bearing are paid to pool and to carry what others dread. It rests on two foundations, one on each side of the trade — the concavity of utility, which makes certainty worth a premium to those who fear the spread, and the law of large numbers, which makes a multitude of independent risks a near-certainty to those who gather them. From these the price takes its shape: a fair premium, the bare expected loss, and a loading laid upon it for costs, for capital, and for the residual risk the pool cannot shed. And beneath the mechanism lies the volume's deepest finding — that the premium is paid, in the end, only for the risk that cannot be diversified away. Idiosyncratic risk pools to nothing and earns nothing; systematic risk, which moves with the whole and survives all spreading, is the only risk the market troubles to price.

That single principle is the thread on which the whole table of this volume hangs, for it joins what looked like separate subjects into one. The insurer pooling its policies and the investor diversifying his portfolio perform the identical act, each cancelling independent risk against itself; the loading the policyholder pays his insurer and the equity premium the shareholder earns from the market are the identical payment, each the reward for bearing the undiversifiable remainder. Insurance and finance, the mutual society and the stock exchange, are revealed as two dialects of a single language, and the word that language is built to name is the price of risk. To know what a premium is, the volume has argued, is to know this much: that it is a verdict on a hidden contest — between the one who would shed a risk and the one who would bear it, between the risk that can be pooled and the risk that must be priced, and between the two ideas of fairness that quarrel over whether a risk should be priced to its bearer at all.

The achievement the volume records is real and deep, and it deserves to be named before its limits are. To have seen, with Bernoulli, that what people value is not wealth but its diminishing use, and so to have explained at a stroke why a sure thing is worth more than its average and why anyone insures; to have proved, with the law of large numbers, that risk crushing to one is tame to many, and so to have founded the whole apparatus of mutual security on which the modern world rests; to have shown, with Markowitz and Sharpe, that only undiversifiable risk is priced, and so to have unified the insurance of a cottage with the valuation of a corporation — these are among the genuine triumphs of economic thought, and they are not merely elegant but useful, underwriting the pensions, the mortgages, the ventures, and the protections without which a precarious life would be more precarious still. The economics of risk has made it possible for ordinary people to face an uncertain future without being destroyed by it, and that is no small thing.

Yet the volume has been equally a chronicle of where the apparatus strains and breaks. The two asymmetries — the hidden knowledge that drives the good risks from the pool, the hidden action that breeds the very loss insured against — gnaw at every real market in risk, and the deductibles and screens and mandates raised against them are imperfect and costly cures. Correlation, the failure of the one assumption on which pooling wholly depends, turns the insurable into the catastrophic, and leaves the gravest perils of the age — the flood, the earthquake, the pandemic, the warming climate — at or beyond the boundary of what private insurance can bear. The human mind misweighs the very probabilities the theory rests upon, over-insuring the trivial and leaving the grave uncovered. The price of market risk remains larger than any plausible reckoning of fear can explain. And the contest over fairness — whether the premium may justly sort persons by the luck of their birth and circumstance, or whether some risks must be shared because they are no one's fault — is a question economics can sharpen but cannot settle, for it turns not on what is efficient but on what is just. The premium is a powerful instrument and an honest one; it is not, and the volume has not pretended it to be, an instrument without limits or a price without a conscience.

With this volume the group of the prices nears its completion. The first four of The Prices are now drawn: the wage, the price of labor; the rent, the price of land and of all that is scarce; the rate, the price of time and of waiting; and now the premium, the price of risk and of the certainty that shedding it buys. Four of the five foundational prices stand, each a verdict on a different hidden contest, and each, on inspection, entangled with the others — for the rate is the floor from which the premium's market reward rises, the rent shares the premium's logic of scarcity, and the wage itself carries a premium for the risks a trade exposes its workers to. One price of the group remains unwritten, and it is the natural sequel to this one.

That price is the spread, the subject of the fifth volume and the last of The Prices — the margin the intermediary charges to stand between buyer and seller, to make a market by holding what neither yet wants. The spread is the premium's near kin, for what the market-maker is paid for is precisely risk: the risk of holding inventory that may move against him, and the risk of trading against someone who knows more, the very asymmetry this volume met in its sixth plate. To read The Premium and then The Spread is to complete the price of bearing risk, in the pool and in the market alike. Other debts this volume has incurred will be paid elsewhere in the table. To the third volume, on the rate, it is bound as a pair, the pure price of time and the price of risk together composing every return. To the twenty-fourth, on bias, it owes the misweighing of chance that bends its eleventh plate. To the twelfth, on asymmetry, it owes the two maladies that haunt its markets; to the twenty-third, on choice, the expected utility on which its second plate stands; and to the ninth, on the stock of capital, the diversification its eighth plate borrows. The premium does not stand alone; it is one price among the family of prices, and it has been read here as the price the world sets on the bearing of what cannot be foreseen.

What the premium buys, in the end, is assurance — the conversion of a dreaded possibility into a borne and bounded cost, the transformation of a ruin that might come into a payment that certainly does. It is the price of facing the future without being paralyzed by it: of building the ship knowing it may sink, of planting the crop knowing the hail may fall, of living the single irreplaceable life knowing its hazards, and doing these things anyway because the worst of the risk has been laid off onto a pool, a market, a state, or a contract that will make the loss whole. That is a profound service, and the premium is its honest price. But the assurance is never complete and never wholly fair. Some risks no pool can tame and no premium can honestly price; some are misjudged by the very minds that bear them; and the question of who shall pay for whose misfortune — of whether the price of a risk should fall on its bearer or be shared among all — is a question the premium poses but cannot answer, and that each society must answer for itself.

The premium is the price of risk transferred and the price of certainty obtained — resting on aversion and on pooling, paid only for what cannot be diversified away, and shadowed throughout by the contest between the one who sheds a risk and the one who bears it, between what can be pooled and what must be priced, and between two irreconcilable ideas of what it means for such a price to be fair. The argument remains open.