How did one company use Big Data to reduce payment defaults and improve revenue recognition’
Credit risk management is a top priority for many financial companies, for both ROI and regulatory reasons. This typically includes managing reputational risks, which entails the real-time monitoring of various transactions across multiple channels, identifying suspicious activity, and compiling a ?blacklist? of doubtful entities.
However, it?s incredibly important to avoid placing customers on a blacklist erroneously. Not only does this take away business from the company, it can be extremely frustrating for the customer and damaging to the customer-company relationship.
It?s also vital to prevent customer attrition. Balancing risk with potential revenue requires an approach that can be personalized for each customer. It requires Big Data. This becomes clear when we consider how credit risks are assessed.
How Lending Institutions Calculate Risk
Nearly every large company, regardless of industry, is vulnerable to credit risk. The need to protect themselves from undue risk impacts company operations and overall revenue. Therefore, both companies and lending organizations have an interest in computing expected losses from nonpayment. Here are the current factors in this computation:
- Exposure at Default (EAD) is the total exposure to credit risk. In lending, this is the amount owed at the time of default.
- Loss Given Default (LGD) is the fractional loss due to default.
- Probability of Default (PD) is the likelihood that the loan (or transaction amount, etc.) will not be repaid. This number is based on the known credit history of the borrower and the nature of the borrower?s investments.
- Recovery Rate (RR) is defined as 1 LGD. This applies to financial institutions that can sell assets seized from defaulting customers as a way to offset some of their loss.
The formula for calculating expected loss is PD x LGD x EAD.
In the past, various methods were used to determine a PD, but recent advancements in data analytics and especially machine learning have made LGD as important a factor as PD. Today, analytics can be used to correctly predict LGD and to calculate PD values.
Let?s see how one company used Big Data to successfully manage credit risks.
Managing Credit Risks with Big Data
Our client, a very large manufacturer of consumer essentials, wanted to calculate an expected loss due to default value for all its vendors. They have a global presence, but we chose to focus on the American and Canadian markets because they are the most stable.
As the client is not a banking institution, their recovery rate was zero; they could not sell defaulting vendors? assets to cover any losses. For all vendors, the LGD was fixed at ?1?. Accounts were determined to be in default when no payments were received for 30 days.
PD is a binary classification? it separates elements into two groups. There are many ways to approach such a problem; we chose to use logistic regression to start. Then we built a detailed model to analyze 36 months? worth of transactional data. The goal was to identify which American and Canadian vendors were most likely to default on their payments to our client and to calculate the firm’s expected loss due to nonpayment.
And that?s exactly what the model did. Once riskier accounts were identified, the client could take steps to reduce payment defaults and improve their overall revenue recognition.
Even though the client wasn’t a financial company, this experience still holds a valuable lesson for those in the banking and finance sector: It’s time to change how we calculate risks. Instead of relying on cumbersome old ways, it’s time to embrace the newer, more agile, more efficient options. In a world where decision quality must go up while operating costs must come down, an investment in data analytics technology is a necessity.
