Corporate lending in India awaits its Moneyball moment

Big-ticket corporate lending blowups are not limited to state-run banks. They are regular occurrences in private banks, mutual funds and private-equity sponsored non-bank finance firms. One wonders if there are chronic systemic shortcomings in the risk processes of corporate lending. Only a fraction of such defaults may be attributed to borrower fraud and corrupt bankers. However, undue generalization of this narrative may have prevented an introspection of existing risk practices. Incremental changes have taken place, such as higher data usage, but only a few top banks have fundamentally re-imagined the lending process. For a lot of banks, the corporate lending process remains highly subjective, at times driven by unverified rules-of-thumb and selective readings of data. Even the bespoke ‘expert models’ that are sometimes used have limited risk prediction ability.

Moneyball in corporate lending: The Billy Beanes of corporate lending need to come forward. Billy was the legendary manager of the Oakland Athletics baseball team. His exploits of using analytics to select players for his team and its subsequent success is the subject of the book, Moneyball: The Art of Winning an Unfair Game, and a movie by same name. Baseball teams have a budget to buy players for the season, on the lines of our Indian Premier League. Before Billy introduced analytics, the selection and prices of players were determined by baseball experts with decades of talent-spotting experience.

While choosing players, apart from basic performance statistics and their immediate success, the players’ personality traits as perceived by these experts impacted their selection. Parallels may be drawn between how underwriters look at financial statements and perceived management quality to select clients for credit. In the movie, a talent spotter who dislikes the analytical approach says, “Baseball isn’t about numbers, it’s not science,” and, “There are intangibles that only baseball people understand.” If one replaces baseball with corporate lending, it may reflect the thoughts of a hypothetical underwriter with doubts about analytic intervention in big-ticket lending. The movie did dramatize certain facts. The data analyst had an economics degree but limited prior knowledge of baseball. In reality, the analyst Paul DePodesta was himself a baseball scout. Ultimately, sports analytics improved the decision-making skills of talent spotters like Billy Beane, but did not replace them.

The human element versus expertise: It may be worthwhile to note that predictive models by themselves are unlikely to beat the best underwriter working at peak attention span, with no pressure of meeting quarterly credit targets. But consistency issues may crop up. Research suggests that in marginal cases, with no black-and-white answer, the mood of the underwriter matters. On high-sentiment days, more loans are approved and higher defaults are observed for those loans. (In The Mood For a Loan: The Causal Effect of Sentiment on Credit Origination, Agarwal et al, 2012)

The challenges of ‘expert models’: Certain lenders are aware of the decision inconsistency problem, but continue to depend heavily on qualitative aspects. At times, this leads to the use of quasi-quantitative assessment techniques, such as expert judgement models. Such models use hard information (financial statements) and soft information (quality of management). The soft aspects, being qualitative, are hard to verify and their interpretation is individual- dependent. Research suggests they may lead to worse lending decisions, particularly if the person collecting it is under time pressure. One study discovered behavioural nuggets: such as, when loan officers have earlier sales experience, the soft information collected tends to be interpreted with a more optimistic bias. (Making Sense of Soft Information: Interpretation Bias and Loan Quality, Campbell et al, 2018)

As such, the hybrid models that use qualitative and quantitative variables may not always be statistically robust. The qualitative inputs often do not stand statistical scrutiny of their predictive power. If one adds inconsistently collected soft information, it is not surprising that it has less predictive power than models using hard information.

Start with measuring decision quality: One may first like to identify and measure the shortcomings of the traditional underwriting approach and ascertain what is working and what is not. The final underwriting decision of ‘go’ versus ‘no go’ is preceded by several components of prediction and judgement. Some of these components are the industry outlook, financial statement projections, future debt requirements and the quality of management. These components’ values keep evolving as the credit call goes from the initial underwriter recommendation to a credit committee and then higher. While most lenders are able to track the final decision quality by the loan delinquency rate, some are unable to assess the quality of it at each stage. For instance, if the financial projections are way off from actual results, even if the account is non-delinquent, it suggests a weakness in the process.

A high delinquency rate can be a result of bad underwriting or a bad economy. Likewise, a low delinquency rate may not always mean high-quality underwriting decisions. Knowing the shortcomings will help design a precise analytical intervention. The goal, as in Moneyball, is not to replace the underwriter, but take better decisions.

Deep Mukherjee is visiting faculty of finance at IIM Calcutta and risk management consultant