Across the span of recent decades, lenders have established impressively predictable and reliable statistical models that forecast consumer default. There are now changes on the horizon that are likely to dramatically improve our ability to understand and predict consumer behavior. Think of our current understanding as a black and white photo. Adding non-traditional data to credit risk models is like upgrading to color, revealing tremendous insight about the consumer’s situation at that moment. Another benefit is equitable to replacing a single photo with a photo album that provides insight over time. Combined, these changes stand to redefine how consumer risk is evaluated and an acceptable margin of error. For institutions lagging behind, it will likely define their future existence.
Even during the recent recession when consumer priorities shifted dramatically, to paying their cards and defaulting on their homes, the rank ordering of consumers with only traditional credit data remained impressively accurate – though the entire curve shifted to the left. Part of the effectiveness of these models is that they typically ask critical, but limited, questions. For example, what is the likelihood of default or bankruptcy in the next 18 months? Many resources have focused on improving the possible results obtained from the current dataset, but the returns rarely exceed a few hundred basis points. While this improvement is valuable for the largest card lenders, it is hardly worth the cost for anyone else. It is these models that I refer to as a black and white photo—amazing, yet still limited.
Over the past few years data has emerged to respond to demands for richer insights—sometimes called alternative or non-traditional data. The key value of this data is that it brings new perspectives about consumers based on information such as state licenses, payday loan usage, cell phone payments, and address change velocity. While still a snapshot in time, the richer data adds color to our image, revealing if a grey coat is actually grey or if it is red, or perhaps chartreuse. As with traditional data, not all alternative data is the same, some is relevant for only certain populations and some proves interesting, but ultimately useless. Only in the laboratory of a skilled statistician can the relevant details be coaxed forth. Then, prime consumers with few traditional accounts can be uncovered, as well as consumers showing early signs of impending default.
Another new, and less understood form of consumer credit data is also emerging—time series data. The greatest failing of a snapshot is that it only reveals a moment in time. Paying off a large balance the day before a credit card is calculated can actually lower your score. Studying a consumer’s behavior over time reveals a much richer perspective. Imagine a researcher moving from a single photo of a historical figure to a photo album that covers a span of years. Insights that were previously inconceivable become mundane. For example, determining if a consumer is a revolver or transactor is nearly impossible (<35% success) with a traditional credit report, but with a time series report accuracy exceeds 75%. Simply matching this month’s payment to last month’s balance reveals a wealth of consumer insight—and that is only the beginning of what will be possible. One bureau has already released this capability. This is a game changing step in delivering relevant, actionable data. And it will only be a matter of time before the other bureaus adapt.
This improved view of consumers will have the most dramatic impact on underbanked consumers who are interested in additional banking relationships and have prime consumer behaviors. We should see a dramatic increase in credit availability for these individuals and businesses. Those who have already proven themselves a bad credit risk may see some gains as they work to demonstrate newfound responsibility. Among prime consumers, will be a clearer image of those who are improving and those who are losing ground in their financial responsibility. Even more interesting will be the ability to understand what consumers are actually looking for in a new financial relationship and making offers that match their desires. Gone will be the generic offers designed to appeal to everyone. Instead, bank products will be more tightly defined based on each consumer’s actual behaviors and needs.
Given the complexity of managing this additional dimension of data, it will likely be some time before it is widely adopted. Statistical researchers will need to develop new attributes and models that enable predictions of previously impossible attributes. Ability to Pay can be revealed by the consumer’s payment history over the past 2 years (monthly paid in full or only minimum payments made), instead of awkwardly asking the consumer for their income. Currently, only two decision engines are even capable of handling time series data like this for realtime decisions. The initial releases will likely be in the form of additional scores that augment those based on traditional data. In essence, a traditional data score, then alternative data to improve rank ordering, then a time series score to provide trending and product interest data. Combined, they will yield a rich and more accurate view of the consumer.
It will take time for these changes to take hold, but make no mistake—the revolution has begun. These new data sets are available and are being adopted by leading lenders. It will likely take 10 years before they are fully integrated into most credit decisions, but their adoption is underway. Those who are early adopters will be able to identify profitable consumers and make profitable offers with an accuracy that will seem almost magical to their competitors. Some will delay implementation, assuming that “good enough” will work for them. And it will be good enough for awhile, until their profitability begins to lag. One day they will realize that they are studying a black and white still photo to understand a consumer’s behavior while their competitors are watching a high definition color movie.
Eric Lindeen is marketing director for Zoot Enterprises Inc., a provider of loan origination, account acquisition and credit risk management solutions for large financial institutions. You can follow him on Twitter @EricLindeen. Visit Zoot’s Credit Strategy Session and Merchant Acquiring Strategy Session on PaymentsJournal.
