Data has rarely been so in-demand by banks, especially for building the required credit models to satisfy regulators and internal compliance requirements. But rigorous risk models, as the backbone of an enterprise-wide risk management framework, require lengthy and well-populated sets of data. And it’s a fact of life that, across banks of all sizes – and for many kinds of reasons – data is sometimes incomplete and difficult to compile.
Difficult, yes, but not impossible. Missing data is rarely an insurmountable problem, and there are several ways in which banks can (and must) fill their information gaps to effectively manage risk.
First, where gaps do exist, it’s important to determine why. Data could be missing for random reasons, such as a loan officer forgetting to record it. In this case, its omission will not affect the risk model either way. But there might be a pattern to the gaps. Do they, for example, occur consistently for borrowers under financial stress, or in a particular region or asset size? If so, the missing data may introduce a systematic bias to risk analysis.
When the omissions in certain borrower cases are random, as in the first example, there is the simple opportunity to drop them from the analysis. In taking this “complete case analysis” approach to missing data, modelers must make sure they have enough remaining data to compensate. If the reasons for missing data are, however, systematic, an alternative, bias-busting approach is “available case analysis.” This uses all borrower cases, however incomplete, and skips over unavailable fields within them.
In practice, banks often adopt a combination of these “complete” and “available” approaches to work around incomplete data. They can also use a “substitution” method: either replacing blank data with an average for the sample as a whole, values predicted by another statistical model, or – most sophisticated of all – the merged results of multiple iterations under a range of models. Other options include developing risk models based on formal surveys of internal experts and finding external sources of detailed, representative data, which can prove difficult and expensive.
Whichever approach a bank takes, dealing with missing risk data should ultimately form a part of its larger data management and improvement strategy. This in turn should promote the importance of complete and clean data, offer incentives to data gatherers to improve data quality, and ensure the continual monitoring of data gaps.
Ultimately, such a strategy can help an organization upgrade its risk modeling capabilities and, as a result, make more informed business decisions, backed by rigorous analysis. In other words, by tackling the problem of missing data, banks can also find new ways to improve their operations – and exceed the expectations of management and regulators alike.
Comments