The underwriting of loan risk accounts for 40% to 50% of non-interest expense at commercial banks. This includes the cost of maintaining appropriate reserves against expected losses and paying FDIC premiums.

Adding the incremental or variable impact on interest expense (including cost of capital) brings the total cost of intermediating credit risk to 300 basis points for many institutions.

Despite the benefits that might accrue from an improved understanding of the causes, timing, and magnitude of credit losses, the banking industry has been reluctant to move away from its rules of thumb. The rules of thumb have, for the most part, been developed over time by bankers and their regulators.

Judgments by lending officers, documented according to the industry conventions, are typically evaluated by credit committees. Regulators then audit banks against a subset of the conventions that has been codified into regulations.

For example, the Comptroller of the Currency's much talked-about "black box," used in examinations of New England banks, was an application of the conventional wisdom that if the discounted-cash-flow value of a property is estimated to be lower than its carrying value, a loss should be recorded, especially if the borrower is in a weakened condition.

More Tools Available

More recently, securitization has spawned a variety of techniques that permit comparison of borrowers - frequently small." or privately held companies - with public companies, whose default experience is known.

Given the wealth of readily available credit data, more reliance should be placed on statistics, the workhorse of applied social sciences, to explain and forecast credit losses.

Statistics has rarely been used in banking, despite the relatively low cost.

With the recent exception of certain investment traders, banks have not had personnel with training in quantitative data analysis. The most important barrier may simply be senior management's unwarranted fear that quantitative projections are intended to replace rather than supplement expert judgment.

Data Gaps Not a Deterrent

In using the statistical approach, one must start with the conviction that not having ideal data should not be a deterrent to analyzing the adequate data available. Thus, the data analysis plan is based on the information that is readily accessible.

Historical information on losses, delinquencies, accrual status, and foreclosed real estate by product is usually available on a monthly or quarterly basis. Historical information on individual loans can sometimes be obtained from loan operations, but that depends on how long computer backups are maintained.

For all banks, there exists a wealth of public data on outstandings by product, delinquencies and losses. At any time, an enormous amount of data exists on current loans both in the operating and subsidiary systems as well as in the loan files. This data base is rarely fully analyzed, because it is disorderly.

Once assembled, this wealth of credit data can be used in a number of statistical analyses that can be rapidly executed.

Filtering Out the Noise

Multiple studies can compensate for the weakness of a particular analysis. Conclusions from multiple studies are more likely to bracket the future losses in question because the sources of error or bias for each study will vary. Rather than focus on the results of a particular analysis, the intention is to triangulate on estimated losses.

This argument for multiple studies can be made more concretely.

Predicting losses from an analysis of a single institution can result in distortion. If loan losses have anything to do with regional economics, systematic bias, or random events such as fraud, then single-institution studies are risky.

Analyses that look at data from many banks overcome this difficulty. However, they introduce another. The data available for such analyses are highly aggregated and omit many of the variables helpful in explaining and predicting losses. For example, detailed analyses of the loan files of a bank's worst losses can be rich with insights about causes of losses.

Another alternative - analyzing losses across time - may provide the best basis on which to estimate the timing and magnitude of future losses. One can come closer to predicting the future by combining estimates from a variety of studies and weighting each by an estimate of its accuracy than by relying on the estimates of a single "best" study.

While the statistical analyses can be complex, they are low in cost compared to the consequences of not doing them.

Betting the Bank

Fundamentally, banks get in trouble when their reserves do not cover the embedded loss of their current portfolio and current earnings are insufficient to cover the annual losses.

While the root cause may be an unwillingness to correctly price credit risk in an effort to build market share, the consequence is a massive gamble. Unrealistic hopes are in no one's long-term interest. As the industry begins to build a more actuarial perspective on risk, the following observations are important to keep in mind:

* Losses, even in a credit crisis, are relatively rare. Special methods of analysis are required for dealing with a rare phenomenon.

* Assumptions about the types and stages of loan decay need to be tested before being used. It makes no sense to predict losses from delinquencies, changes in credit ratings, or product type until there is evidence that they matter.

* Losses do not decline in a straight line. Furthermore, events at previous periods influence losses at subsequent periods. For this reason, the statistical technique of linear regression does not suffice.

Time-series models using time lags best capture what every bank management understands about the timing of losses. Such loan amortization analyses are necessary for building predictive models.

* A key to being able to combine results of different studies is to have error estimates for each.

The principals of my firm and six other investors own a small Maryland thrift, Reisterstown Federal Savings Bank. The institution's experience illustrates the value of statistical analysis of projected losses.

Reasonable Goals Set

Reisterstown Federal's management regards 20% return on equity (on a 10% capital ratio) and 2% return on assets as reasonable performance goals.

The managers are highly focused on residential construction loans to small builders in a five-county area. They originated 1,550 loans totaling $230 million in the most recent fiscal year.

On the basis of its loan concentration, Reisterstown Federal would appear to have a very high risk profile. However, when acquired in January 1990, the thrift had only one loss to its record - its only out-of-state participation - and almost no reserve for future losses.

The quality of Reisterstown Federal's underwriting and credit skills have been borne out as management managed down its nonperforming assets from a high of 4.8% to 1.14% of total loans plus real estate owned, placing them in the 79th percentile of commercial banks and thrifts.

Net chargeoffs including writedowns and gains on sale were nonexistent until 1990, when they were $107,000. Net chargeoffs were $406,000 in 1991 and $1.65 million to date for fiscal 1992. Reserves will stand at $2.5 million at fiscal yearend, amounting to 119% coverage of nonperforming loans and foreclosed real estate.

Two questions linger: What losses remain embedded in the portfolio? What should reserves and consequently provisions be, given that management can identify no loans in the current portfolio likely to result in a chargeoff?

Three statistical analyses assisted in determining embedded losses.

The first, a time-series analysis of levels of outstandings and delinquencies, resulted in an embedded loss estimate of $4.5 million.

The second analysis, which tracked all individual loans over 14 months, produced a loss estimate of $2.6 million.

The third, an institutional analysis comparing Reisterstown to other institutions, resulted in a loss estimate of $2.1 million.

Weighting each estimate by statistical measures of their reliability, an estimate of embedded losses of $3.5 million over the remaining life of the portfolio was agreed upon.

By repeating the analyses each quarter, management can review three simple charts summarizing results, and determine whether to increase or decrease the provision.

Few other institutions with a similar risk profile would have fared as well as Reistertown Federal. While the results of its analyses were highly contrarian, they enabled the bank to build appropriate reserves in advance of the estimated future losses.

Reistertown's earnings were simply reduced over the last few years from the highs of the 1980s, which produced returns on equity in excess of 20%. Other institutions that did not reserve sufficiently in earlier periods will not be able to catch up.

Ms. Abt is managing partner of the Kellett Group, a Boston-based investment company.

Subscribe Now

Access to authoritative analysis and perspective and our data-driven report series.

14-Day Free Trial

No credit card required. Complete access to articles, breaking news and industry data.