The Implication Of Fair Value Accounting For General Insurance Companies
By P. K. Clark, P. H. Hinton, E. J. Nicholson, L. Storey, G. G. Wells, and M. G. White
British Actuarial Journal, 9, V, 1007-1059 (2003)
This paper was originally presented by a Working Party (WP) of the General Insurance Research Organisation (GIRO) in October 2002. It has since been updated through February 2003 and presented to the Institute of Actuaries (IA) on March 24, 2003. Besides being interesting and informative, it offers the reader a veritable festival of acronyms. It is not, however, a quantitative study like the recent work commissioned by CAS, which look at the effect of fair value prescriptions on realistic reserve numbers.
The fair value accounting initiative now being spearheaded by the International Accounting Standards Board (IASB) is aimed at ensuring, to the extent possible, that the assets and liabilities appearing on an enterprise balance sheet carry values that can be realized if assets are sold or liabilities ceded in the open market. In the case of assets this hinges on current market prices for these assets. In the case of liabilities, it hinges on the price of guarantees. This is an important distinction, and ignoring it leads one straight into a quagmire. We shall see that IASB is quite ready to ignore it by contriving a false equivalence, a confusion which underlies the bumpy path that fair value has so far taken.
The authors point out that IASB has tabled its attempt to publish a comprehensive standard for financial instruments. IAS 39, recently released with a January 1, 2005 effective date, prescribes that the most important financial instruments will be recognized at fair value but amortized at a fixed rate. This is a rather small change to current practice. This setback exposes insurance contracts to heightened attention from IASB in hopes that they will present a softer target.
The paper is devoted to exploring the implications of the Draft Statement of Principles (DSOP), first exposed for discussion in November 2001. Though considerable time has passed since this article was current, we need not be concerned since events in this arena unfold with great deliberation. The most significant intervening event is the IASB's issuance, on March 31, 2004 with a January 1, 2005 effective date, of International Financial Reporting Standard No. 4 (IFRS 4), which contains no material surprises. Among the issues raised in the DSOP are:
- Reconfiguration of accounts. As a consequence of the fair value valuation paradigm, reporting will be on the basis of closed cohorts of contracts. This has the effect of eliminating unearned premium reserves and deferred acquisition costs. Actuaries have always known that this is a superior approach for tracking underwriting performance. Our current reliance on accident year accounting is a compromise based partly on the need to break out cancellation reserves and the desire to have cohorts that close in one year rather than two.
- A strong preference for stochastic reserving methods over deterministic methods.
- And, closely tied to 2. above, a strong insistence that loss reserves be reported with Market Value Margins (MVMs), a concept very close to the actuarial one of risk load. IASB has decreed that the MVM reflect both diversifiable and systematic risk. All in all, both the GIRO WP and their IA discussants regard these as very muddy waters indeed and make their perplexity very clear.
- Entity specific value. This, as the GIRO WP remark, is a somewhat vague and elusive concept. When direct market information is not available for valuation purposes, fair value dictates using information for a typical market player. Entity-specific allows use of the company's own information. For general insurance (P&C), this boils down to some technical points in the evaluation of MVMs.
The WP discusses these issues at some length and in detail, producing a fairly comprehensive catalog of problems and concerns. The overall impression is that insurance fair value is no closer to prime time than financial institutions fair value. The GIRO WP is to be commended for a thorough and thoughtful job. I note, however, that a number of dots remain unconnected and some avenues remain unexplored, being arguably outside the WP's scope. Being relatively untrammeled by scope definitions, I will devote my remaining space to filling those voids.
Fair Value Accounting
It helps to be clear that fair value is a very large paradigm shift for the accounting community. Accountants are expected to transform almost overnight. From people who record and summarize transactions according to a set of rules, accepting whatever valuations come out the tailpipe, they are to become people who do valuations - like actuaries or financial economists. Naturally, they want their new world to look as much like the old world as possible. One thing they would very much like to survive is the traditional accounting for debt. In fact they want all other forms of valuation to work the same way.
Herein lies a rub, for traditional debt accounting prescribes recording the proceeds of the obligation, net of credit penalties, as the cash flow in the transaction. This makes sense at the transaction end because it avoids the inconvenience of recording a gain or loss as part of the debt transaction. When, however, it emerges from the tailpipe as a liability valuation, it has some strange properties. Each such value turns out to be discounted for the borrower's own credit risk, as if the contractual obligation were somehow premised on the borrower's ability to pay. This naturally sticks in the craw of any actuary who has spent his/her entire career doing liability valuation. However, many accountants have no trouble with it at all. Even worse, some financial economists have found an argument to support this position, imputing the value of the insolvency put (the owner's option to put the company to the creditors in case of default) as an asset of the corporation. Alas, some financial economists have become so preoccupied with equity valuation that they forget there is a corporate interest distinct from the ownership interest. It is the corporate interest, of course, that must be reported in public financials.
The central ambiguity here is in the timing of interest payments. In the traditional approach, the credit penalty is expressed as a higher interest rate and is amortized over the term. However, the credit penalty is obviously an immediate expense deducted from the proceeds at inception. If it is treated as prepaid interest, and the liability amortized at the default-free rate, the resulting liability valuation will, apart from frictional charges, satisfy a guarantor if one is desired, always supposing that the debt instrument is free of uncertainty in timing and amount.
IASB's stubborn refusal to reexamine this controversial subject is arguably the chief obstacle to adoption of fair value proposals in the industries that actually value their liabilities analytically. Discounting for own credit risk, the authors note, was not part of the current DSOP at the time of writing, nor was it included in the CAS fair value studies; but the issue is not dead, only dormant. In the meantime, P&C insurance companies will report liabilities properly - at full, default-free value - while competing in the equity markets with industries that rely on debt financing and report their liabilities discounted for own credit risk. This inequity makes it clear that that the issue must eventually be resolved. After all, accounting trumps reality.
Market Value Margins
In the more general case, e.g. casualty loss reserves, the liability is uncertain in timing and/or amount; and the guarantor, real or putative, is harder to satisfy. Suppose the guarantor is assembling a diversified portfolio of guarantees. This portfolio will not be very large when market risk begins to dominate. In fact the market risks facing guarantors are typically enormous. The pricing will work like the Capital Asset Pricing Model (CAPM) only in reverse. I don't have a reference for the theorem, nor any assurance that one exists; but one can expect it to work like the derivation in Bob Butsic's 1988 Michelbacher Prize paper. One can calculate some MVMs from stochastic reserving models to test the supposition that market risk dominates. I strongly suspect that the guarantor's market risk is what deserves attention. Question: Can one justify a large MVM in property policy reserves, perhaps based analytically on catastrophe exposure, to be taken down as policies expire?
MVMs, as prescribed in the DSOP, are widely seen as a stumbling block, both by the GIRO WP and the authors of the CAS studies. There is little doubt that the concept will have to be reexamined.
The DSOP's expression of favor for stochastic loss modeling very likely arises from the perceived need to calculate MVMs based on reserve volatility. We have argued above that the dominant factor in MVMs, properly understood, is the systematic risk faced by the real or putative guarantor. This risk has more to do with the systematic response of casualty reserves to the state of the insurance and financial markets than with volatility. If volatility matters, it is guarantor's that matters and not the insurer's. What we mean is that the insurer's volatility is relevant to assessing the insurer's capital adequacy, not to valuing its liabilities. The very ability to measure capital reliably depends on having public valuation standards that are free of such idiosyncrasies as own volatility (and own credit standing). That said, stochastic modeling of insurance losses is still a worthwhile undertaking. I maintain, however, that it is still in the exploratory stage. Too much remains yet to be gotten right.
One obstacle to success is the measure problem. The first (Some would say zeroth.) rule of stochastic modeling is to choose the right measure. Nevertheless, it is not hard to detect a cavalier, even dismissive, attitude toward this issue among some practitioners. Cohort loss development is manifestly a jump process with finite, episodic increments. Frequency grades to zero at long durations, but average loss size does not. A lognormal diffusion process, with infinitesimal increments accruing continuously, does not answer the case. It will handle situations where claims are small and numerous, but it will fail and embarrass the user where he/she needs it most: in the development tail. Furthermore, use of the wrong measure will compromise any probability statements about the distribution of reserves.
Modeling a jump process adds to the parametric burden, putting a heavy strain on the meager information surviving in a triangle summary. Finding a suitable error structure for lines where a claim may pay out over several valuation periods is also a daunting challenge. Most daunting is the systematic variability arising from the claims administration process itself.
In the present state of development, results from stochastic models are certainly fit to be shown alongside more conventional results. However, in my opinion, there is no such method extant that is ready to be canonized as a gold standard for public reporting.
The working party is to be commended for a thorough and insightful presentation of the issues. The IA discussion was also timely and illuminating. Fair value implementation has not been, and will not be, a smooth process. In my own opinion, many of the issues that must be resolved are general accounting issues and involve insurance only incidentally. Actuaries should be wary of calls to surrender their intuition in the name of accounting uniformity. Actuarial intuition in liability valuation issues has had longer to develop and is arguably more reliable than that of accountants, for whom valuation is a new thing.
Philip E. Heckman September 2, 2004