Decoding the Black Box: Using Advanced Data Analytics to Evaluate Non-Performing Consumer Debt


The use of advanced data analytics has not yet filtered through to the market in non-performing consumer debt (NPLs), but its advent spells profound change. In NPL markets, simplistic portfolio-level evaluation techniques remain the norm, with NPL portfolios valued as virtual “black boxes”. This leads to chronic mispricing of NPLs, and the ability to distinguish between more and less valuable accounts remains challenging. 

Heka Global is leading the way in the development of evaluation models that value NPL portfolios on a line-by-line basis. These models have already dramatically improved Heka’s pricing strategies, enabling Heka to discover different repayment patterns in seemingly identical portfolios. Broadly applied, this methodology will have major ramifications for the NPL investments landscape.

The Transformative Potential of Advanced Data Analytics

Since 2010, tech-driven lenders have leveraged Big Data to create new statistical evaluation models. These models operate by making tens of millions of observations about consumers, across a wide range of parameters, both traditional (e.g. credit history) and non-traditional (e.g. employment history). The resulting edge is the ability to distinguish between seemingly similar credit applicants who actually have different repayment profiles. 

In contrast to the origination market, the NPL industry has been slow to adopt these advances. In many cases, non-performing paper is still bundled into portfolios that are priced at a purchase price multiple that accounts for a few portfolio-wide metrics, such as type of product and charge-off date.

The shortfalls of this simplistic evaluation approach are clear, with roughly 50% of NPL transactions being mispriced. And, while it is a long-standing truism that 20% of accounts will return 80% of the portfolio’s value, portfolio-level evaluation is incapable of identifying that elusive, valuable segment. Granular evaluation promises to decode black-boxed portfolios by delivering a refined understanding of accounts and, in turn, pricing accuracy.

Heka’s Use of Advanced Data Analytics

Heka is at the forefront of advanced statistical evaluation of NPL portfolios. Its models assess each and every account through AI models that feed on macro and micro, encompassing loan details, historical context, consumer attributes, local legal settings, market conditions, and broader economic variables.

Figure 1 presents a pinpointed example of the value of such modeling. Heka’s model was applied to evaluate two portfolios with virtually identical high-level features and the same market price. What’s more, all of the underlying NPLs were originated by the same originator and in the same state. Nonetheless, the model discovered that Portfolio 1’s liquidation rate is actually 1.5x that of Portfolio 2.

While the model accounts for multiple feature-interactions, its output can also be substantiated by common sense. The key difference between Portfolios 1 and 2 seems to be the underlying consumers’ proximities to efficient courts. Portfolio 1’s underlying consumers are on average only 2.17 miles from an efficient court, versus 6.40 miles for Portfolio 2


Characteristic (avg.) Portfolio 1 Portfolio 2
Conventional evaluation Observations
Years since charge-off 1.46 1.24
Loan balance $2,640 $2,070
Consumer age 43.9 41.5
Year Placed 2022 2022
Statistically-Driven evaluation Observations
Consumer distance from ‘efficient’ court (miles) 2.17 6.40
Liquidation Rate 10.86% 7.09%

Figure 1. The major characteristics and liquidation rates of portfolios 1 and 2.

Further examination provides additional assurance that the model can be effectively applied in a forward-looking manner. Legal regulation and venue rules dictate cases are assigned to nearby adjacent courts. As illustrated by figure 2, there is a near perfect relationship between the distance of a consumer from a court and the assignment of a case to that court. Evidently, consumers’ proximity to efficient courts can in and of itself serve as a predictor of liquidation.

Figure 2. Underlying portfolio distribution of distance from designated court.

Nonetheless, each portfolio has its own performance drivers. Although the distance from an efficient court was identified as a valuable predictor in this case, other portfolios may have different driving forces. It is certainly improbable that an investment strategy based only on consumer proximity to efficient courts will deliver notable excess returns. Instead, Heka’s model usefulness stems from the fact that it is alive to a whole host of such parameters and their aggregate impact on performance.

Aligned with this understanding, Heka’s supervised learning model is constantly improving. Every time it confronts a new portfolio, it widens its range of observations and refines its predictive capabilities. To date, the model has been trained on millions of NPLs accumulated over three decades under different economic environments.

The Ramifications for Consumer Credit Investments

The ability to meticulously evaluate NPL portfolios on a granular basis promises significant commercial benefits. In transactional processes, more accurate pricing technologies empower buyers and sellers to avoid unintentional concessions and to ensure value for money. Line-by-line evaluation also makes possible more sophisticated differentiated pricing. Higher-value line items can be priced competitively, while lower-value line items – especially those which may require enhanced work-out processes – are priced appropriately.

Beyond the immediate transactional benefits, account-level evaluation helps to upgrade operational efforts. By identifying the higher-value and lower-value line items at the outset, it improves the allocation of collection resources, speeds up portfolio resolution – thereby significantly impacting IRR – and can shape the logic for follow-on sale processes.

These are only the most foreseeable potential benefits. Increased transparency and market participant intelligence inevitably spells wide-ranging improvements in financial markets of all sorts, unsettling old practices and offering new means of realizing value.


About the Authors: 

Polina Dovman

VP Data Sciences, Heka Global

Ph.D. Finance and Economics, Columbia University

MA Financial Economics, The Hebrew University of Jerusalem


Max Lack

Business Development Manager, Heka Global

MA Law, University of Oxford

MA History University of Oxford

Related Posts