A key lesson that lenders learned during the 1980s is that active and careful portfolio management is crucial to credit policy.

Furthermore, as the course of the economy is always uncertain, diversification of the loan portfolio is necessary to safeguard lending performance.

But assessing the diversification of a loan portfolio presents difficulties.

A key problem is risk covariance -- the tendency of different segments of a portfolio to behave similarly in response to economic events. This can make a seemingly well-diversified portfolio highly vulnerable to a particular event, rendering it poorly diversified in reality.

Reasons for Covariance

The most obvious source of risk covariance relates to regional economic interdependencies.

The power of regional covariance was dramatically illustrated in the Texas debacle of the mid-1980s, when the collapse of the energy sector triggered widespread difficulties in others.

All industries with principally local markets were adversely affected, including retail, services, and real estate. As a consequence, even banks with modest direct exposure to the energy sector were vulnerable to the oil price shock.

Banks with a national lending scope can also fall victim to shared risks if their loan portfolios contain concentrations in industries that are vulnerable to the same economic shocks.

For example, a major West Coast bank's difficulties during the early and mid-1980s largely resulted from industry positions -- in real estate, agriculture, energy and shipping - that were all vulnerable to disinflation.

Although the portfolio was geographically diversified and contained no single overwhelming industry concentration, the bank was devastated.

Hidden Connections

Mindful of such dangers and responding to regulatory pressure, banks are now seeking ways to identify and manage risk covariance. The potential benefits include more-stable earnings and higher stock prices.

Covariant risks are often hidden from casual scrutiny, so the analytical approach chosen must be powerful enough to identify all significant sources of risk.

The framework must produce information that motivated tactical and strategic actions, integrating the process of making transaction decisions with the broader goals of portfolio management.

Furthermore, past relationships will not necessarily hold in the future; therefore, an effective approach must be predictive.

The major methods employed to date to assess the covariant risks embedded in a loan portfolio fall into three classes: historical correlations, economic structure analysis, and simulation-based approaches.

Each of these methods possess virtues as well as drawbacks and can provide valuable insights. But historical correlations and structure analysis are more limited.

Analyzing the Past

The historical tendency of different sectors to behave similarly can be measured with correlation analysis.

For each segment, correlation coefficients are calculated using some historically available proxy for the credit risk. Risk proxies might include such variables as sales, cash flow, charge-off rates, or market-based measures such as equity prices or bond risk premiums.

This approach is fairly straightforward and scores high on ease of implementation. The chief drawbacks include limited reliability and predictive power.

Reliability suffers because of spurious correlations. During the historical period analyzed, sectors may have behaved similarly because of nonrecurring factors or chance.

Also, industries that will be closely tied in the future may not have been in the past. As a consequence, the predictive power of a historical correlation is often low.

Structure Analysis

Better estimates of economic interrelationships can be derived from analysis of economic structure.

An industry's performance is strongly influence by the health of its principal customers and, to a lesser extent, that of its suppliers. Analysis of patterns in customer-supplier relationships can identify sectors that are likely to be in linked in the future.

Although this method is more forward-looking than historical correlations and less susceptible to spurious relationship, careful interpretation is still required. That is because commonality of customers does not guarantee similar performance.

For example, two sectors with virtually identical customer profiles are oil refining and equipment leasing.

Both sell in similar proportions to a range of industries including airlines and construction. Yet their fortunes are not equally affected by the health of their customers.

An ailing airline might defer leasing additional planes but is less likely to forgo fuel purchases.

Industry and regional performance can be examined through simulation under a range of potential economic events, such as a significant shift in exchange rates, a Japanese financial collapse, a recession, or a pronounced regional downturn.

The performance of the individual segments is then correlated across the range of scenarios developed by the set of potential events. In this manner, segments likely to suffer in the same economic environment can be identified.

This approach is by far the most reliable. The resulting relationships are forward-looking, because they embody the full range of potential economic developments and are not limited to events in recent history.

The chief drawback is the intensity of the analysis needed. This method requires an integrated, consistent economic modeling framework capable of translating economic events to industry and regional financial performance and credit risk. Few banks possess the analytical resources to undertake this effort without outside assistance.

Each of these methods can enhance the management of a loan portfolio. However, the limitations of historical correlations and structure analysis undermine their validity.

Mr. Ross is a principal at DRI/McGraw-Hill in Lexington, Mass.

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.