As Asia begins to emerge from the financial crisis, there is consider-able talk about freeing banks from the legacy of bad lending but much less about how to prevent the bad lending itself. A bath of bad loans is being emptied, yet the current lending behavior of many banks suggests that not enough is being done to turn off the taps.

Many banks failed in the crisis, and many others are technically bankrupt, largely as a result of poor lending practices—particularly in commercial lending, the dominant element of most loan books. To improve the quality of such loans, banks must have suitable risk assessment systems.

Experience suggests that generic solutions don’t work; any system must therefore be designed for the country where it will operate (see sidebar, "First things first."). To create such a system, it is necessary to answer two key questions: how good is the judgment of the current credit officers, and what is the quality and availability of the information (financial and nonfinancial) that can be used to assess risk quantitatively? The responses largely determine the type of risk assessment system that should be developed.

Banks in emerging markets generally employ credit officers with limited skills and do not have high-quality financial information on customers. The most feasible solution for a majority of Asian banks is thus a system that standardizes decision making about credit risk. Once such a system is in place, a half dozen credit officers rating the same company should all produce the same result.

A system of this sort has two main parts: qualitative and quantitative. The qualitative component is based on the way a bank’s best credit officers, drawing on experience, determine credit risk. By studying those loan officers’ judgments, and the way they were reached, a bank should be able to articulate implicit criteria that other loan officers can follow. The second, quantitative component of the system produces a risk score resulting from an assessment of the customers’ often meager financial data. A statistical examination of old loan applications for those factors predictive of default in a given market is the foundation of any scoring system.

Banks should resort to qualitative, subjective risk assessment only when the incremental improvement is worth the incremental cost

Once the qualitative and quantitative components have been developed, they must be combined into an integrated system that minimizes the cost of assessing a loan application. Start with the relatively cheap (quantitative) scoring system. Resort to qualitative, subjective risk assessment, which requires scarce and expen-sive credit officers, only when the incremental improvement is worth the incremental cost. This economical approach usually produces three possible outcomes. "Green" applications are approved purely on the basis of a good quantitative score; "red" applications are rejected purely on the basis of a poor one; and "yellow" applications are subjected to further (that is, qualitative) assessment. For those Asian banks with relatively poor financial data, the yellow band is usually quite wide; that is, the score alone decides only a small proportion of cases.

Less than perfect is good enough

Banks often labor under the misapprehension that risk assessment systems for emerging markets should be based entirely on judgment because quantitative scoring doesn’t work. Although data are undoubtedly of poorer quality in many emerging markets, it is possible to develop risk assessment systems that perform surprisingly well there; systems designed for banks in three emerging markets—South Korea, China, and the Czech Republic—accurately distinguish good from bad customers (Exhibit 1). The right method of developing such systems is a huge topic, but two points can be made.

First, perfect models are not necessary. The separation of good and bad loans in the exhibit is far from perfect; not all the loans that remain good are in the lowest risk category, and not all those that turn bad are in the highest. Each band, in short, contains a mixture of loans that will go bad and loans that will remain good. But this doesn’t matter. Consider the example of a Chinese bank. Under its old system, all of the loans shown in the exhibit would have been approved. But use of the risk assessment system, whose output the exhibit depicts, would result in the rejection of all applications in the fourth (or highest-risk) class, representing 48 percent of the bad volume, with losses after workout amounting, on average, to about 75 percent of the sums loaned. Although this improvement would come at the cost of losing 16 percent of the good volume (on which Chinese banks typically make about 1 to 2 percent on each loan), the overall impact of the system would be to improve the bank’s lending margins by 100 to 200 basis points (1 to 2 percent).

Second, the system’s developers should be skilled in both statistics and lending, since scoring systems are very easy to get wrong. Banks that lack people with the necessary skills should require members of the lending organization to observe the process of developing the first model so that they can administer future ones.

The journey to better credit risk management can take as long as two years to complete, but even Asian banks with limited credit skills and inadequate data can improve their results considerably in only six months.

Laying the foundation

Although the ability to pinpoint counterparty risk (that is, the probability of default) lies at the core of a good credit assessment system, it isn’t enough to achieve a lasting change in credit performance. Laying the foundation for such a change calls for improvements in five areas: loan pricing, the credit process, the credit organization, incentives, and the process for changing the system over time.1 Without these basic changes, the new system will probably be unusable.

1. Set the right price

Risk-rating systems determine the probability of a client’s default (counterparty risk) but not the transaction risk—that is, the anticipated loss on a loan given to that client. If, for example, a client’s counterparty risk is 2 percent, but the bank expects to recover 20 percent of the loan amount in the event of default, the loan’s transaction risk is 1.6 percent (2 percent x 80 percent, the loss in the event of default).

As banks develop their rating systems, they should therefore simultaneously devise a way to convert a rating into a transaction risk cost. This method should take into account a number of factors, including the duration of the loan and—by far the most important consideration—the collateral offered. Many Asian banks recover less than 25 percent of the collateral value as originally assessed. This suggests that there is a significant opportunity to improve the recovery process. It also indicates the importance of accurately assessing the real value of the collateral when pricing the loan.

Different ways of negotiating prices can have a big effect on lending profits. Consider two different methods. Under one, the loan officer calculates the loan’s breakeven price, or "manufacturing" cost,2 and then negotiates the price with the customer. A second method involves determining the clients’ expectations at the start: before calculating the manufacturing cost, the loan officer asks customers what price they would be willing to pay. The officer then calculates the manufacturing cost and negotiates the real price.

Experience shows that the second method produces a higher average price and thus much higher profits. The first method forces loan officers to negotiate from a position of weakness because they don’t know how much customers are prepared to pay. Inexperienced salespeople in a weak negotiating position are therefore tempted to price at or close to the manufacturing cost, since it is easier to do so and also yields a price that is more appealing to the customer and thus more likely to result in a deal.

2. Get the process right

Even the best risk systems will not be used correctly unless the frontline people understand them. All too often, the credit process manuals that attempt to explain such systems are thick binders of dense text that few can comprehend. To make the process easy to learn, use, and maintain, the manual ought to be modular—that is, all stages of the credit process should have separate sections, each with standard subsections covering, for example, responsibilities, tools, and forms.

3. Align the organization with the new system

In many banks, the same person sells loans, assesses risk, and attempts to recover the collateral when loans go bad. This creates obvious dangers, including fraud. In one notable case in Eastern Europe, a lending officer gave 100 loans to a single property developer, all in a few months. Each individual loan was small and within the officer’s authority, but the total exposure—and the subsequent loss when none of the loans was repaid—turned out to be very large indeed.

So however good the risk assessment system may be, and however clear the manual, organizational changes will be necessary to minimize bad lending. The most basic of them is splitting up tasks so that different people are responsible for selling, assessing risk, and recovery (workout). Specialization promotes greater focus, greater accountability, and better monitoring of performance.

4. Design the right incentives

New credit systems invariably tighten up lending criteria and therefore cut the proportion of accepted cases. For banks that have measured their lending performance largely by volume, the prospect of an immediate 15 to 20 percent decline in volume in the hope of a future decline in default rates isn’t appealing. We recommend two ways of encouraging management and staff to implement the new system.

First, reward salespeople on the basis of expected profit. Frontline sales employees are much less likely to oppose a drop in volume if their pay reflects lending profit rather than the loan’s total amount. For compensation purposes, the profit must be what is expected (that is, the price charged less the manufacturing cost) rather than the actual profit achieved one or two years later (when the loan has gone bad or remained good). If a loan has been priced badly and appears to be a money loser, a bank using an expected-profitability system can immediately challenge the salesperson about the terms extended.

Second, reward management on the basis of actual profitability. Some countries limit the amount of loans that can be written off annually in profit-and-loss accounts; China, for instance, restricts reported loan losses to 1 percent of the portfolio, even though actual losses typically exceed 6 percent. Under such a system, management has little incentive to improve the default rate of its portfolio by making fewer but better loans, since short-term income is bound to decline, and regulations forbid reporting the improvement until the number of defaults falls below 1 percent. So long as reported loan losses are held down in this way, banks should ensure that their internal performance reports, at least, are based on actual, not reported, profitability.

5. Maintain the system

Finally, since a risk-rating system must be revised every two or three years, banks should create a process for maintaining and updating it by making people responsible for specific aspects of the system and having them report to a central credit unit.

Building basic systems that can turn around the performance of banks in emerging markets usually takes six months. One large emerging-market bank with which we worked reduced its loan loss rates from 4.7 percent in 1997 to 1.7 percent in 1999, a period when its country’s overall default rates increased.

Besides reducing the bad-loan ratio, a risk-rating system can be a powerful way of attracting new customers in many Asian markets. Because many Asian banks assess risk inaccurately, they overestimate the transaction risks of particular clients and end up overpricing their loans. The implications should be clear: with a better understanding of the actual risk profile of each prospective client, banks can take away business from rival institutions that force customers to pay considerably more than they should.

First things first

A tempting way for an Asian bank to overcome its credit problems is to import generic solutions used in more developed markets, such as basing lending decisions primarily on cash flow projections or scoring systems that were developed for them. We would not recommend either of these approaches.

Many banks in developed countries base lending decisions on whether a prospective borrower’s projected cash flow clearly exceeds the planned debt repayments. Although this approach is fine in theory (and in practice in developed markets), companies in emerging markets often manipulate their profit figures to minimize tax payments, thus rendering the starting point of cash flow projections meaningless.

Even without such manipulation, it is very unlikely that a system developed for one country will work in another. To understand why, you have to be familiar with how scoring systems work. Typically they consist of an equation containing five to ten financial factors, each with an associated coefficient. Multiplying the value of the factor by the coefficient and combining the results produces the score, which is a direct indicator of the probability of default. A scoring system for Germany, say, would work in an Asian bank only if the same factors predicted default in both markets and the relative importance of those factors was the same in both. Unfortunately, neither assumption is true, since Germany and the countries of Asia have different accounting standards and the relative importance of the factors changes along with the economic situation, even within the same country.

The exhibit illustrates the impact of transplanting a scoring system developed for one market into another. In this case, 64 percent of the loans that remained good had the same rating as 80 percent of the loans that subsequently went bad. Not a very useful system, it would appear.

Return to reference