הערה מקדימה: בדף זה תמצאו הסבר על חברות ביטוח. מהי מהותה הכלכלית של חברת ביטוח, כיצד היא מרוויחה כסף ומאילו מקורות, כיצד יש לנתח חברות ביטוח, מהם היחסים הפיננסיים החשובים, מהן הבעיות האינהרנטיות ועוד. כל ההסברים המופיעים כאן נלקחו ממכתביו השנתיים של וורן באפט לבעלי המניות של חברת ההשקעות שלו - ברקשייר האת'ווי. במכתבים אלו, פורס באפט את משנתו ומסביר, מדי שנה לאורך כ-40 השנים האחרונות, אספקט מסויים על חברות ביטוח. מה שניסיתי לעשות בדף זה הוא לאגד הסברים אלו, שנפרסים כאמור על-פני עשרות מכתבים שנתיים, לדף אחד. ניסיתי לארגן את הדברים בסדר הגיוני מסויים שיוביל את הקורא להבנת הנושא. גם אם משפט מסויים או מושג מסויים לא יובנו בתחילה, סביר להניח שניתן יהיה להבינם לאחר קריאה מלאה של דף זה. הערות והארות בנוגע לאגד זה, יתקבלו בברכה.

הערת הבהרה: ברקשייר האת'ווי, זרוע האחזקות של וורן באפט שהחלה את דרכה כחברת טקסטיל, מחזיקה בשלוש חברות ביטוח מרכזיות: (1) חברת National Indemnity שנרכשה בשנת 1967, מנוהלת כיום בידי Ajit Jain ומתמחה בפוליסות ביטוח חריגות, (2) חברת GEICO שאת מניותיה רכש באפט לראשונה בשנת 1951 (ובשנת 1995 הצליח להשיג שליטה מלאה בחברה), מנוהלת כיום בידי Tony Nicely ומתמחה בפוליסות ביטוח רכב, (3) חברת ביטוח המשנה General Re שנרכשה בשנת 1998, מנוהלת כיום בידי Joe Brandon ו-Tad Montross והיוותה בתחילת דרכה אחזקה בעייתית במקצת.

Float and Cost of Float

What counts in insurance, our core business, is the amount of "Float" and its cost over time.

Float is the total of loss reserves, loss adjustment expense reserves and unearned premium reserves minus agents balances, prepaid acquisition costs and deferred charges applicable to assumed reinsurance. And the cost of float is measured by our underwriting loss.

"Float" is money that doesn't belong to us but that we temporarily hold. Float arises because premiums are received before losses are paid, an interval that sometimes extends over many years. loss events that occur today do not always result in our immediately paying claims, because it sometimes takes many years for losses to be reported (asbestos liability losses would be an example - the problems they signify lie dormant for decades, before virulently manifesting themselves.), negotiated and settled. During that time, the insurer invests the money. Insurance has provided a fountain of funds with which we've acquired the securities and businesses that now give us an ever-widening variety of earnings streams. This pleasant activity typically carries with it a downside: The premiums that an insurer takes in usually do not cover the losses and expenses it eventually must pay. That leaves it running an "underwriting loss", which is the cost of float.

An insurance business has value if its cost of float over time is less than the cost the company would otherwise incur to obtain funds. But the business is a lemon if its cost of float is higher than market rates for money. If this cost (including the tax penalty) is higher than that applying to alternative sources of funds, the value is negative. If the cost is lower, the value is positive - and if the cost is significantly lower, the insurance business qualifies as a very valuable asset.

We hold an exceptional amount of float compared to premium volume. Float is wonderful - if it doesn't come at a high price. Its cost is determined by underwriting results, meaning how the expenses and losses we will ultimately pay compare with the premiums we have received. When an insurer earns an underwriting profit, float is better than free. In such years, we are actually paid for holding other people's money. For most insurers, however, life has been far more difficult: In aggregate, the property-casualty industry almost invariably operates at an underwriting loss. When that loss is large, float becomes expensive, sometimes devastatingly so.

Overall, our results have been good. True, we've had five terrible years in which float cost us more than 10%. But in 18 of the 37 years Berkshire has been in the insurance business, we have operated at an underwriting profit, meaning we were actually paid for holding money. And the quantity of this cheap money has grown far beyond what I dreamed it could when we entered the business in 1967 (Berkshire purchased National Indemnity ("NICO") in 1967).

e.g. During 1990 we held about $1.6 billion of float slated eventually to find its way into the hands of others. The underwriting loss we sustained during the year was $27 million and thus our insurance operation produced funds for us at a cost of about 1.6%. There are important qualifications to this calculation - the fat lady has yet to gargle, let alone sing, and we won't know our true 1967 - 1990 cost of funds until all losses from this period have been settled many decades from now.

Since our float has cost us virtually nothing over the years, it has in effect served as equity. Of course, it differs from true equity in that it doesn't belong to us. Nevertheless, let's assume that instead of our having $3.4 billion of float at the end of 1994, we had replaced it with $3.4 billion of equity. Under this scenario, we would have owned no more assets than we did during 1995. We would, however, have had somewhat lower earnings because the cost of float was negative last year. That is, our float threw off profits. And, of course, to obtain the replacement equity, we would have needed to sell many new shares of Berkshire. The net result - more shares, equal assets and lower earnings - would have materially reduced the value of our stock. So you can understand why float wonderfully benefits a business - if it is obtained at a low cost.

Accounting irony: Though our float is shown on our balance sheet as a liability, it has had a value to Berkshire greater than an equal amount of net worth would have had.

The downward trend of interest rates in recent years has transformed underwriting losses that formerly were tolerable into burdens that move insurance businesses deeply into the lemon category. Some years back, float costing, say, 4% was tolerable because government bonds yielded twice as much, and stocks prospectively offered still loftier returns. Today, fat returns are nowhere to be found (at least we can't find them) and short-term funds earn less than 2%. Under these conditions, each of our insurance operations, save one, must deliver an underwriting profit if it is to be judged a good business.

Combined Ratio and Long Tails

The combined ratio represents total insurance costs (losses incurred plus expenses) compared to revenue from premiums.

הערת הבהרה: היחס המשולב הוא מדד רווחיות מקובל במגזר הביטוח. היחס מייצג את האחוז מכל דולר של פרמייה שמוצא על תביעות והוצאות. היחס המשולב מהווה סיכום של יחס ההפסדים ויחס ההוצאות. יחס ההפסדים משקף באחוזים את היחס בין ההפסדים לבין הפרמיות. יחס ההוצאות משקף באחוזים את היחס בין ההוצאות לבין הפרמיות. למעשה, היחס המשולב משקף את היחס בין ההפסדים וההוצאות לבין הפרמיות שנגבו. כאשר היחס המשולב הוא מעל 100, העסק הביטוחי נמצא בהפסד חיתומי, הווי אומר שתשלומי התגמולים וההוצאות שהוצאו לצורך יישוב התביעות (לדוגמא, הוצאות בית-משפט), היו גבוהים מהפרמיות שנגבו. במרבית השנים, הכנסות מהשקעות על ה-float מקזזות את ההפסדים החיתומיים ובלבד שהללו בגדר הסביר (לא חורגים הרבה מעל 100).

Because the funds are available to be invested, the typical property-casualty insurer can absorb losses and expenses that exceed premiums by 7% to 11% and still be able to break even on its business. Again, this calculation excludes the earnings the insurer realizes on net worth - that is, on the funds provided by shareholders. i.e. when the investment income that an insurer earns from holding on to policyholders' funds ("the float") is taken into account, a combined ratio in the 107-111 range typically produces an overall break-even result, exclusive of earnings on the funds provided by shareholders.

However, many exceptions to this 7% to 11% range exist. For example, insurance covering losses to crops from hail damage produces virtually no float at all. Premiums on this kind of business are paid to the insurer just prior to the time hailstorms are a threat, and if a farmer sustains a loss he will be paid almost immediately. Thus, a combined ratio of 100 for crop hail insurance produces no profit for the insurer.

At the other extreme, malpractice insurance covering the potential liabilities of doctors, lawyers and accountants produces a very high amount of float compared to annual premium volume. The float materializes because claims are often brought long after the alleged wrongdoing takes place and because their payment may be still further delayed by lengthy litigation. The industry calls malpractice and certain other kinds of liability insurance "long-tail" business, in recognition of the extended period during which insurers get to hold large sums that in the end will go to claimants and their lawyers (and to the insurer's lawyers as well).

In long-tail situations a combined ratio of 115 (or even more) can prove profitable, since earnings produced by the float will exceed the 15% by which claims and expenses overrun premiums. The catch, though, is that "long-tail" means exactly that: Liability business written in a given year and presumed at first to have produced a combined ratio of 115 may eventually smack the insurer with 200, 300 or worse when the years have rolled by and all claims have finally been settled.

The pitfalls of this business mandate an operating principle that too often is ignored: Though certain long-tail lines may prove profitable at combined ratios of 110 or 115, insurers will invariably find it unprofitable to price using those ratios as targets. Instead, prices must provide a healthy margin of safety against the societal trends that are forever springing expensive surprises on the insurance industry. Setting a target of 100 can itself result in heavy losses; aiming for 110 - 115 is business suicide.

What's the preferable measure? Combined Ratio or Cost of Float?

What should the measure of an insurer's profitability be? Analysts and managers customarily look to the combined ratio - and it's true that this yardstick usually is a good indicator of where a company ranks in profitability. We believe a better measure, however, to be a comparison of underwriting loss to float developed.

This loss/float ratio, like any statistic used in evaluating insurance results, is meaningless over short time periods: Quarterly underwriting figures and even annual ones are too heavily based on estimates to be much good. But when the ratio takes in a period of years, it gives a rough indication of the cost of funds generated by insurance operations. A low cost of funds signifies a good business; a high cost translates into a poor business.

Insurance is a Commodity-like product

Insurers have generally earned poor returns for a simple reason: they sell a commodity-like product - a non-proprietary piece of paper containing a non-proprietary promise. Anyone can copy anyone else's product. No installed base, key patents, critical real estate or natural resource position protects an insurer's competitive position. Unlike the situation prevailing in many other industries, neither size nor brand name determines an insurer's profitability.

The insurance industry is cursed with a set of dismal economic characteristics that make for a poor long-term outlook: hundreds of competitors, ease of entry, and a product that cannot be differentiated in any meaningful way.

What finally determines levels of long-term profitability in such industries is the ratio of supply-tight to supply-ample years. When shortages exist (when capacity is short), even commodity businesses flourish. In some industries, capacity-tight conditions can last a long time. Sometimes actual growth in demand will outrun forecasted growth for an extended period. In other cases, adding capacity requires very long lead times because complicated manufacturing facilities must be planned and built. But in the insurance business, capacity can be instantly created by capital plus an underwriter's willingness to sign his name. In this industry, unlike most, capacity is an attitudinal concept, not a physical fact. Insurance managers can write whatever amount of business they feel comfortable writing. That's exactly what was going on in 1985. About 15 insurers raised well over $3 billion, piling up capital so that they could write all the business possible at the better prices then available. The capital-raising trend had accelerated dramatically in 1986. (Even capital is less important in a world in which state-sponsored guaranty funds protect many policyholders against insurer insolvency). Insurance, therefore, would seem to be a textbook case of an industry usually faced with the deadly combination of excess capacity and a "commodity" product.

How do Berkshire's insurance operations overcome the dismal economics of the industry and achieve some measure of enduring competitive advantage? We've attacked that problem in several ways. In such a commodity-like business, only a very low-cost operator or someone operating in a protected, and usually small, niche can sustain high profitability levels. The critical variables, therefore, are managerial brains, discipline and integrity.

Profitability and Low Cost is all that counts (rather than underwriting volume (sales))

Property/casualty companies are judged by their cost of float. If our insurance operations are to generate low-cost float over time, they must:

(a) underwrite with unwavering discipline (accept only those risks that you are able to properly evaluate (staying within your circle of competence) and that, after you have evaluated all relevant factors including remote loss scenarios, carry the expectancy of profit. Ignore market-share considerations and be sanguine about losing business to competitors that are offering foolish prices or policy conditions); and
(b) reserve conservatively; and
(c) avoid an aggregation of exposures that would allow a supposedly "impossible" incident to threaten their solvency.

No matter what others may do, we will not knowingly write business at inadequate rates. We unwittingly did this in the early 1970's and, after more than 20 years, regularly receive significant bills stemming from the mistakes of that era. My guess is that we will still be getting surprises from that business 20 years from now. A bad reinsurance contract is like hell: easy to enter and impossible to exit. I actively participated in those early reinsurance decisions, and Berkshire paid a heavy tuition for my education in the business.

We wish to write only properly-priced business, whatever the effect on volume. Size simply doesn't count.

Can you imagine any public company embracing a business model that would lead to the decline in revenue that we experienced from 1986 through 1999? What CEO wants to report to his shareholders that not only did business contract last year but that it will continue to drop? Anyone examining the table can scan the years from 1986 to 1999 quickly. But living day after day with dwindling volume - while competitors are boasting of growth and reaping Wall Street's applause - is an experience few managers can tolerate. That colossal slide, it should be emphasized, did not occur because business was unobtainable. Many billions of premium dollars were readily available to NICO had we only been willing to cut prices. But we instead consistently priced to make a profit.

Another way to prosper in a commodity-type business is to be the low-cost operator. A man named Leo Goodwin had an idea for an efficient auto insurer and, with a skimpy $200,000, started GEICO in 1936. Goodwin's plan was to eliminate the agent entirely and to deal instead directly with the auto owner.

Geico and State Farm

State Farm is one of America's greatest business stories. Studying counter-evidence is a highly useful activity, though not one always greeted with enthusiasm at citadels of learning. State Farm was launched in 1922, by a 45-year-old, semi-retired Illinois farmer, to compete with long-established insurers - haughty institutions in New York, Philadelphia and Hartford - that possessed overwhelming advantages in capital, reputation, and distribution. Because State Farm is a mutual company, its board members and managers could not be owners, and it had no access to capital markets during its years of fast growth. Similarly, the business never had the stock options or lavish salaries that many people think vital if an American enterprise is to attract able managers and thrive.

In the end, however, State Farm eclipsed all its competitors. In fact, by 1999 the company had amassed a tangible net worth exceeding that of all but four American businesses. If you want to read how this happened, get a copy of The Farmer from Merna.

Despite State Farm's strengths, however, GEICO has much the better business model, one that embodies significantly lower operating costs. And, when a company is selling a product with commodity-like economic characteristics, being the low-cost producer is all-important. This enduring competitive advantage of GEICO - one it possessed in 1951 when, as a 20-year-old student, I first became enamored with its stock - is the reason that over time it will inevitably increase its market share significantly while simultaneously achieving excellent profits. Our growth will be slow, however, if State Farm elects to continue bearing the underwriting losses that it is now suffering.

True Profitability in Long Tailed Business

We should point out again that in any given year a company writing long-tail insurance (coverages giving rise to claims that are often settled many years after the loss-causing event takes place) can report almost any earnings that the CEO desires. Too often the industry has reported wildly inaccurate figures by misstating liabilities. Most of the mistakes have been innocent. Sometimes, however, they have been intentional, their object being to fool investors and regulators.

For these reasons, the results in this column simply represent our best estimate at the end of 2004 as to how we have done in prior years. Profit margins for the years through 1999 are probably close to correct because these years are "mature", in the sense that they have few claims still outstanding. The more recent the year, the more guesswork is involved. In particular, the results shown for 2003 and 2004 are apt to change significantly.

In most businesses, insolvent companies run out of cash. Insurance is different: you can be broke but flush. Since cash comes in at the inception of an insurance policy and losses are paid much later, insolvent insurers don't run out of cash until long after they have run out of net worth. In fact, these "walking dead" often redouble their efforts to write business, accepting almost any price or risk, simply to keep the cash flowing in.

In insurance, the urge to keep writing business is intensified because the consequences of foolishly-priced policies may not become apparent for some time. It takes a long time to learn the true profitability of any given year. First, many claims are received after the end of the year, and we must estimate how many of these there will be and what they will cost. (In insurance jargon, these claims are termed IBNR - incurred but not reported). Second, claims often take years, or even decades, to settle, which means there can be many surprises along the way. If an insurer is optimistic in its reserving, reported earnings will be overstated, and years may pass before true loss costs are revealed (a form of self-deception that nearly destroyed GEICO in the early 1970).

Geico 1970 failure and Nicely in 1993

Between 1936 and 1975, GEICO grew from a standing start to a 4% market share, becoming the country's fourth largest auto insurer. But after my friend and hero Lorimer Davidson retired as CEO in 1970, his successors soon made a huge mistake by under-reserving for losses. This produced faulty cost information, which in turn produced inadequate pricing. By 1976, GEICO was on the brink of failure. Jack Byrne then joined GEICO as CEO and, almost single-handedly, saved the company by heroic efforts that included major price increases. Though GEICO's survival required these, policyholders fled the company, and by 1980 its market share had fallen to 1.8%. By 1993 its market share had grown only fractionally, to 1.9%. Then Tony Nicely took charge. And what a difference that's made: In 2005 GEICO will probably secure a 6% market share. Better yet, Tony has matched growth with profitability.

1984-5 losses

Catastrophes were not the culprit in this explosion of loss cost. A partial explanation for the surge in the loss figures is all the additions to reserves that the industry made in 1985. As results for the year were reported, the scene resembled a revival meeting: shouting "I've sinned, I've sinned", insurance managers rushed forward to confess they had under reserved in earlier years. Their corrections significantly affected 1985 loss numbers.

The only bright spot in this picture is that virtually all of the underreserving revealed in 1984 occurred in the reinsurance area - and there, in very large part, in a few contracts that were discontinued several years ago. This explanation, however, recalls all too well a story told me many years ago by the then Chairman of General Reinsurance Company. He said that every year his managers told him that "except for the Florida hurricane" or "except for Midwestern tornadoes", they would have had a terrific year. Finally he called the group together and suggested that they form a new operation - the Except-For Insurance Company - in which they would henceforth place all of the business that they later wouldn't want to count.

A more disturbing ingredient in the loss surge is the acceleration in "social" or "judicial" inflation. The insurer's ability to pay has assumed overwhelming importance with juries and judges in the assessment of both liability and damages. More and more, "the deep pocket" is being sought and found, no matter what the policy wording, the facts, or the precedents.

Reserves Estimates

Loss reserves at an insurer are not funds tucked away for a rainy day, but rather a liability account. If properly calculated, the liability states the amount that an insurer will have to pay for all losses (including associated costs) that have occurred prior to the reporting date but have not yet been paid. When calculating the reserve, the insurer will have been notified of many of the losses it is destined to pay, but others will not yet have been reported to it. These losses are called IBNR, for incurred but not reported. Indeed, in some cases (involving, say, product liability or embezzlement) the insured itself will not yet be aware that a loss has occurred. Sometimes the problems they signify lie dormant for decades, as was the case with asbestos liability, before virulently manifesting themselves.

The loss expense charged in a property/casualty company's current income statement represents:
(1) losses that occurred and were paid during the year;
(2) estimates for losses that occurred and were reported to the insurer during the year, but which have yet to be settled;
(3) estimates of ultimate dollar costs for losses that occurred during the year but of which the insurer is unaware (termed "IBNR": incurred but not reported); and
(4) the net effect of revisions this year of similar estimates for (2) and (3) made in past years.

Such revisions may be long delayed, but eventually any estimate of losses that causes the income for year X to be misstated must be corrected, whether it is in year X + 1, or X + 10.

In effect, insurance accounting is a self-graded exam, in that the insurer gives some figures to its auditing firm and generally doesn't get an argument.

The necessarily-extensive use of estimates in assembling the figures that appear in such deceptively precise form in the income statement of property/casualty companies means that some error must seep in, no matter how proper the intentions of management. In an attempt to minimize error, most insurers use various statistical techniques to adjust the thousands of individual loss evaluations (called case reserves) that comprise the raw data for estimation of aggregate liabilities. The extra reserves created by these adjustments are variously labeled "bulk", "development", or "supplemental" reserves. The goal of the adjustments should be a loss-reserve total that has a 50-50 chance of being proved either slightly too high or slightly too low when all losses that occurred prior to the date of the financial statement are ultimately paid.

The following table shows the results from insurance underwriting as we have reported them to you, and also gives you calculations a year later on an "if-we-knew-then-what-we think-we-know-now" basis. I say "what we think we know now" because the adjusted figures still include a great many estimates for losses that occurred in the earlier years. However, many claims from the earlier years have been settled so that our one-year-later estimate contains less guess work than our earlier estimate:

	Underwriting Results as Reported to You	Corrected Figures as Reported After One Year's Experience
1980	$ 6,738,000	$ 14,887,000
1981	1,478,000	(1,118,000)
1982	(21,462,000)	(25,066,000)
1983	(33,192,000)	(50,974,000)
1984	(45,413,000)	?

It is important that you understand the magnitude of the errors that have been involved in our reserving. You can thus see for yourselves just how imprecise the process is, and also judge whether we may have some systemic bias that should make you wary of our current and future figures. As you can see from reviewing the table, my errors in reporting to you have been substantial and recently have always presented a better underwriting picture than was truly the case.

You should be very suspicious of any earnings figures reported by insurers (including our own, as we have unfortunately proved to you in the past). The record of the last decade shows that a great many of our best-known insurers have reported earnings to shareholders that later proved to be wildly erroneous. In most cases, these errors were totally innocent: The unpredictability of our legal system makes it impossible for even the most conscientious insurer to come close to judging the eventual cost of long-tail claims. We want to emphasize that we are not faulting auditors for their inability to accurately assess loss reserves (and therefore earnings). We fault them only for failing to publicly acknowledge that they can't do this job.

Because our business is weighted toward casualty and reinsurance lines, we have more problems in estimating loss costs than companies that specialize in property insurance. (When a building that you have insured burns down, you get a much faster fix on your costs than you do when an employer you have insured finds out that one of his retirees has contracted a disease attributable to work he did decades earlier).

The loss-reserving errors of other property/casualty companies are of more than academic interest to Berkshire. Not only does Berkshire suffer from sell-at-any-price competition by the "walking dead", but we also suffer when their insolvency is finally acknowledged. Through various state guarantee funds that levy assessments, Berkshire ends up paying a portion of the insolvent insurers' asset deficiencies, swollen as they usually are by the delayed detection that results from wrong reporting. There is even some potential for cascading trouble. The insolvency of a few large insurers and the assessments by state guarantee funds that would follow could imperil weak-but-previously-solvent insurers. Such dangers can be mitigated if state regulators become better at prompt identification and termination of insolvent insurers, but progress on that front has been slow.

Discounting Reserves? No!

Because of this one-sided experience, it is folly to suggest, as some are doing, that all property/casualty insurance reserves be discounted, an approach reflecting the fact that they will be paid in the future and that therefore their present value is less than the stated liability for them. Discounting might be acceptable if reserves could be precisely established. They can't, however, because a myriad of forces - judicial broadening of policy language and medical inflation, to name just two chronic problems - are constantly working to make reserves inadequate.

Reasonably Knowing Insurance Costs is Vital (Under-Reserving)
("Loss Development") ("Reserve Strengthening")

It's clearly difficult for an insurer to put a figure on the ultimate cost of all such reported and unreported events. But the ability to do so with reasonable accuracy is vital. Otherwise the insurer's managers won't know what its actual loss costs are and how these compare to the premiums being charged. GEICO got into huge trouble in the early 1970s because for several years it severely underreserved, and therefore believed its product (insurance protection) was costing considerably less than was truly the case. Consequently, the company sailed blissfully along, underpricing its product and selling more and more policies at ever-larger losses.

I can promise you that our top priority going forward is to avoid inadequate reserving. We are making every effort to get our reserving right. But I can't guarantee success. The natural tendency of most casualty-insurance managers is to underreserve, and they must have a particular mindset - which, it may surprise you, has nothing to do with actuarial expertise - if they are to overcome this devastating bias. Additionally, a reinsurer faces far more difficulties in reserving properly than does a primary insurer. Nevertheless, at Berkshire, we have generally been successful in our reserving, and we are determined to be at General Re as well.

Any insurer that has no idea what its costs are is heading for big trouble. Not knowing your costs will cause problems in any business. In long-tail reinsurance, where years of unawareness will promote and prolong severe underpricing, ignorance of true costs is dynamite.

When it becomes evident that reserves at past reporting dates understated the liability that truly existed at the time, companies speak of "loss development." In the year discovered, these shortfalls penalize reported earnings because the "catch-up" costs from prior years must be added to current-year costs when results are calculated.

This is what happened at General Re in 2001: a staggering $800 million of loss costs that actually occurred in earlier years, but that were not then recorded, were belatedly recognized last year and charged against current earnings. At yearend 2001, General Re attempted to reserve adequately for all losses that had occurred prior to that date and were not yet paid - but we failed badly. Therefore the company's 2002 underwriting results were penalized by an additional $1.31 billion that we recorded to correct the estimation mistakes of earlier years. The mistake was an honest one, I can assure you of that. Nevertheless, for several years, this underreserving caused us to believe that our costs were much lower than they truly were, an error that contributed to woefully inadequate pricing. Additionally, the overstated profit figures led us to pay substantial incentive compensation that we should not have and to incur income taxes far earlier than was necessary.

"Loss development" suggests to investors that some natural, uncontrollable event has occurred in the current year, and "reserve strengthening" implies that adequate amounts have been further buttressed. The truth, however, is that management made an error in estimation that in turn produced an error in the earnings previously reported. The losses didn't "develop" - they were there all along. What developed was management's understanding of the losses (or, in the instances of chicanery, management's willingness to finally fess up). A more forthright label for the phenomenon at issue would be "loss costs we failed to recognize when they occurred" (or maybe just "oops"). Underreserving, it should be noted, is a common - and serious - problem throughout the property/casualty insurance industry.

Reserves Shenanigans in Mergers and Restructuring

In the acquisition arena, restructuring has been raised to an art form: Managements now frequently use mergers to dishonestly rearrange the value of assets and liabilities in ways that will allow them to both smooth and swell future earnings. Indeed, at deal time, major auditing firms sometimes point out the possibilities for a little accounting magic (or for a lot). Getting this push from the pulpit, first-class people will frequently stoop to third-class tactics. CEOs understandably do not find it easy to reject auditor-blessed strategies that lead to increased future "earnings".

An example from the property-casualty insurance industry will illuminate the possibilities. When a p-c company is acquired, the buyer sometimes simultaneously increases its loss reserves, often substantially. This boost may merely reflect the previous inadequacy of reserves - though it is uncanny how often an actuarial "revelation" of this kind coincides with the inking of a deal. In any case, the move sets up the possibility of "earnings" flowing into income at some later date, as reserves are released.

A preliminary tally by R. G. Associates, of Baltimore, of special charges taken or announced during 1998 - that is, charges for restructuring, in-process R&D, merger-related items, and write-downs - identified no less than 1,369 of these, totaling $72.1 billion. That is a staggering amount as evidenced by this bit of perspective: The 1997 earnings of the 500 companies in Fortune's famous list totaled $324 billion.

Berkshire has kept entirely clear of these practices: If we are to disappoint you, we would rather it be with our earnings than with our accounting. In all of our acquisitions, we have left the loss reserve figures exactly as we found them. After all, we have consistently joined with insurance managers knowledgeable about their business and honest in their financial reporting. When deals occur in which liabilities are increased immediately and substantially, simple logic says that at least one of those virtues must have been lacking - or, alternatively, that the acquirer is laying the groundwork for future infusions of "earnings".

Loss growth Vs. Premium growth

We expect the industry's incurred losses to grow by about 10% annually, even in years when general inflation runs considerably lower. If premium growth meanwhile materially lags that 10% rate, underwriting losses will mount, though the industry's tendency to underreserve when business turns bad may obscure their size for a time.

---

Reinsurance: Strength of Promise to Pay Means a Lot

Insurance sold to other insurers who wish to lay off part of the risks they have assumed - should not be a commodity product. At bottom, any insurance policy is simply a promise, and as everyone knows, promises vary enormously in their quality. When insurers purchase reinsurance, they buy only a promise - one whose validity may not be tested for decades - and there are no promises in the reinsurance world equaling those offered by Gen Re and National Indemnity.

Reinsurance is characterized by extreme ease of entry, large premium payments in advance, and much-delayed loss reports and loss payments. Initially, the morning mail brings lots of cash and few claims. This state of affairs can produce a blissful, almost euphoric, feeling akin to that experienced by an innocent upon receipt of his first credit card.

The solvency risk in primary policies, pales in comparison to that lurking in reinsurance policies. When a reinsurer goes broke, staggering losses almost always strike the primary companies it has dealt with. This risk is far from minor: GEICO has suffered tens of millions in losses from its careless selection of reinsurers in the early 1980s. Here's one footnote to GEICO's 2002 earnings that underscores the need for insurers to do business with only the strongest of reinsurers. In 1981-1983, the managers then running GEICO decided to try their hand at writing commercial umbrella and product liability insurance. The risks seemed modest: the company took in only $3,051,000 from this line and used almost all of it - $2,979,000 - to buy reinsurance in order to limit its losses. GEICO was left with a paltry $72,000 as compensation for the minor portion of the risk that it retained. But this small bite of the apple was more than enough to make the experience memorable. GEICO's losses from this venture now total a breathtaking $94.1 million or about 130,000% of the net premium it received. Of the total loss, uncollectable receivables from deadbeat reinsurers account for no less than $90.3 million (including $19 million charged in 2002). So much for "cheap" reinsurance. "Cheap" reinsurance is thus a fool's bargain: When an insurer lays out money today in exchange for a reinsurer's promise to pay a decade or two later, it's dangerous - and possibly life-threatening - for the insurer to deal with any but the strongest reinsurer around.

Some reinsurers, particularly those who, in turn, are accustomed to laying off much of their business on a second layer of reinsurers known as retrocessionaires - are in a weakened condition and would have difficulty surviving a second mega-cat. When a daisy chain of retrocessionaires exists, a single weak link can pose trouble for all. In assessing the soundness of their reinsurance protection, insurers must therefore apply a stress test to all participants in the chain, and must contemplate a catastrophe loss occurring during a very unfavorable economic environment. After all, you only find out who is swimming naked when the tide goes out. At Berkshire, we retain our risks and depend on no one. And whatever the world's problems, our checks will clear.

Unlike most reinsurers, we retain virtually all of the risks we assume. Therefore, our ability to pay is not dependent on the ability or willingness of others to reimburse us. This independent financial strength could be enormously important when the industry experiences the mega-catastrophe it surely will.

Berkshire is sought out for many kinds of insurance, both super-cat and large single-risk, because: (1) our financial strength is unmatched, and insureds know we can and will pay our losses under the most adverse of circumstances; (2) we can supply a quote faster than anyone in the business; and (3) we will issue policies with limits larger than anyone else is prepared to write.

Our willingness to put such a huge sum on the line for a loss that could occur tomorrow sets us apart from any reinsurer in the world. There are, of course, companies that sometimes write $250 million or even far more of catastrophe coverage. But they do so only when they can, in turn, reinsure a large percentage of the business with other companies. When they can't "lay off" in size, they disappear from the market. Berkshire's policy, conversely, is to retain the business we write rather than lay it off. When rates carry an expectation of profit, we want to assume as much risk as is prudent. And in our case, that's a lot.

We will accept more reinsurance risk for our own account than any other company because of two factors: (1) by the standards of regulatory accounting, we have a net worth in our insurance companies of about $6 billion - the second highest amount in the United States; and (2) we simply don't care what earnings we report quarterly, or even annually, just as long as the decisions leading to those earnings (or losses) were reached intelligently.

Reinsurance and Mega-Cats (Super-Cats)

"CAT covers" are reinsurance contracts that primary insurance companies (and also reinsurers themselves) buy to protect themselves against a single catastrophe, such as a tornado or hurricane, that produces losses from a large number of policies. Say, terrorists detonated several large scale nuclear bombs in an attack on the U.S. A disaster of that scope was highly improbable, of course, but it is up to insurers to limit their risks in a manner that leaves their finances rock-solid if the "impossible" happens. In these contracts, the primary insurer might retain the loss from a single event up to a maximum of, say, $10 million, buying various layers of reinsurance above that level. When losses exceed the retained amount, the reinsurer typically pays 95% of the excess up to its contractual limit, with the primary insurer paying the remainder. (By requiring the primary insurer to keep 5% of each layer, the reinsurer leaves him with a financial stake in each loss settlement and guards against his throwing away the reinsurer's money.)

CAT covers are usually one-year policies that also provide for one automatic reinstatement, which requires a primary insurer whose coverage has been exhausted by a catastrophe to buy a second cover for the balance of the year in question by paying another premium. This provision protects the primary company from being "bare" for even a brief period after a first catastrophic event. The duration of "an event" is usually limited by contract to any span of 72 hours designated by the primary company. Under this definition, a wide-spread storm, causing damage for three days, will be classified as a single event if it arises from a single climatic cause. If the storm lasts four days, however, the primary company will file a claim carving out the 72 consecutive hours during which it suffered the greatest damage. Losses that occurred outside that period will be treated as arising from a separate event.

Depending upon many variables, a CAT premium might generally have run 3% to 15% of the amount of protection purchased. For some years, we've thought premiums of that kind inadequate and have stayed away from the business.

But because the 1989 disasters left many insurers either actually or possibly bare, and also left most CAT writers licking their wounds, there was an immediate shortage after the earthquake of much-needed catastrophe coverage. Prices instantly became attractive, particularly for the reinsurance that CAT writers themselves buy. Just as instantly, Berkshire Hathaway offered to write up to $250 million of catastrophe coverage, advertising that proposition in trade publications. Though we did not write all the business we sought, we did in a busy ten days book a substantial amount.

Reinsurance is a business that is certain to deliver blows from time to time. Were a true mega-catastrophe to occur in the next decade or two - and that's a real possibility - some reinsurers would not survive. The largest insured loss to date is the World Trade Center disaster, which cost the insurance industry an estimated $35 billion. Hurricane Andrew cost insurers about $15.5 billion in 1992 (though that loss would be far higher in today's dollars). Both events rocked the insurance and reinsurance world. But a $100 billion event, or even a larger catastrophe, remains a possibility if either a particularly severe earthquake or hurricane hits just the wrong place. Four significant hurricanes struck Florida during 2004, causing an aggregate of $25 billion or so in insured losses. Two of these - Charley and Ivan - could have done at least three times the damage they did had they entered the U.S. not far from their actual landing points. Many insurers regard a $100 billion industry loss as "unthinkable" and won't even plan for it. But at Berkshire, we are fully prepared. Our share of the loss would probably be 3% to 5%, and earnings from our investments and other businesses would comfortably exceed that cost. When "the day after" arrives, Berkshire's checks will clear.

Super-Cats Accounting

When viewing our quarterly figures, you should understand that our accounting for super-cat premiums differs from our accounting for other insurance premiums. Rather than recording our super-cat premiums on a pro-rata basis over the life of a given policy, we defer recognition of revenue until a loss occurs or until the policy expires. We take this conservative approach because the likelihood of super-cats causing us losses is particularly great toward the end of the year. The bottom-line effect of our accounting procedure for super-cats is this: Large losses may be reported in any quarter of the year, but significant profits will only be reported in the fourth quarter.

Reinsurance: how to price Super-Cats (Experience And Exposure)

Experience, of course, is a highly useful starting point in underwriting most coverages. At certain times, however, using experience as a guide to pricing is not only useless, but actually dangerous. Late in a bull market, for example, large losses from directors and officers liability insurance ("D&O") are likely to be relatively rare. When stocks are rising, there are a scarcity of targets to sue, and both questionable accounting and management chicanery often go undetected. At that juncture, experience on high-limit D&O may look great. But that's just when exposure is likely to be exploding, by way of ridiculous public offerings, earnings manipulation, chain-letter-like stock promotions and a potpourri of other unsavory activities. When stocks fall, these sins surface. Consequently, the correct rate for D&O "excess" (meaning the insurer or reinsurer will pay losses above a high threshold) might well, if based on exposure, be five or more times the premium dictated by experience.

Since September 11th, Ajit has been particularly busy. Among the policies we have written and retained entirely for our own account are (1) $578 million of property coverage for a South American refinery once a loss there exceeds $1 billion; (2) $1 billion of non-cancelable third-party liability coverage for losses arising from acts of terrorism at several large international airlines; (3) £500 million of property coverage on a large North Sea oil platform, covering losses from terrorism and sabotage, above £600 million that the insured retained or reinsured elsewhere; and (4) significant coverage on the Sears Tower, including losses caused by terrorism, above a $500 million threshold. We have written many other jumbo risks as well, such as protection for the World Cup Soccer Tournament and the 2002 Winter Olympics. In all cases, however, we have attempted to avoid writing groups of policies from which losses might seriously aggregate. We will not, for example, write coverages on a large number of office and apartment towers in a single metropolis without excluding losses from both a nuclear explosion and the fires that would follow it.

Megacat: Volatility in profits is welcomed (compared with 'smooth' annual profits)

Most of the demand for reinsurance comes from primary insurers who want to escape the wide swings in earnings that result from large and unusual losses. In effect, a reinsurer gets paid for absorbing the volatility that the client insurer wants to shed.

I have known the details of almost every policy that Ajit has written since he came with us in 1986, and never on even a single occasion have I seen him break any of our three underwriting rules. His extraordinary discipline, of course, does not eliminate losses; it does, however, prevent foolish losses. And that's the key: Just as is the case in investing, insurers produce outstanding long-term results primarily by avoiding dumb decisions, rather than by making brilliant ones. We want to emphasize, however, that we assume risks in Ajit's operation that are huge - far larger than those retained by any other insurer in the world. Therefore, a single event could cause a major swing in Ajit's results in any given quarter or year. Reinsurance is a highly volatile business.

That bothers us not at all: As long as we are paid appropriately, we love taking on short-term volatility that others wish to shed. At Berkshire, we would rather earn a lumpy 15% over time than a smooth 12%. We are perfectly willing to lose $2 billion to $2½ billion in a single event (as we did on September 11th) if we have been paid properly for assuming the risk that caused the loss.

When a claims manager walks into the CEO's office and says "Guess what just happened," his boss, if a veteran, does not expect to hear it's good news. Surprises in the insurance world have been far from symmetrical in their effect on earnings.

---

"Retroactive" Insurance - Accountancy

In this line of business, we assume from another insurer the obligation to pay up to a specified amount for losses they have already incurred - often for events that took place decades earlier - but that are yet to be paid - (for example, because a worker hurt in 1980 will receive monthly payments for life). In these arrangements, an insurer pays us a large upfront premium, but one that is less than the losses we expect to pay. We willingly accept this differential because a) our payments are capped, and b) we get to use the money until loss payments are actually made, with these often stretching out over a decade or more. About 80% of the $6.6 billion in asbestos and environmental loss reserves that we carry arises from capped contracts, whose costs consequently can't skyrocket.

When we write a retroactive policy, we immediately record both the premium and a reserve for the expected losses. The difference between the two is entered as an asset entitled "deferred charges - reinsurance assumed". This is no small item: at yearend, for all retroactive policies, it was $3.4 billion. We then amortize this asset downward by charges to income over the expected life of each policy. These charges - $440 million in 2002, including charges at Gen Re - create an underwriting loss, but one that is intentional and desirable. By their nature, these losses will continue for many years, often stretching into decades. As an offset, though, we have the use of float - lots of it.

Clearly, float carrying an annual cost of this kind is not as desirable as float we generate from policies that are expected to produce an underwriting profit (of which we have plenty). Nevertheless, this retroactive insurance should be decent business for us.

Finally, the competitive picture changed in at least one important respect: State Farm - by far the largest personal auto insurer, with about 19% of the market - has been very slow to raise prices. Its costs, however, are clearly increasing right along with those of the rest of the industry. Consequently, State Farm had an underwriting loss last year from auto insurance (including rebates to policyholders) of 18% of premiums, compared to 4% at GEICO. Our loss produced a float cost for us of 6.1%, an unsatisfactory result. (Indeed, at GEICO we expect float, over time, to be free.) But we estimate that State Farm's float cost in 2000 was about 23%. The willingness of the largest player in the industry to tolerate such a cost makes the economics difficult for other participants.