Varun Nakra’s “Downturn LGD model” was designed to predict potential losses on residential mortgage portfolios during economic crises. Now widely adopted by major Australian banks, it has become a benchmark in financial risk management. We spoke with Nakra about his model’s applications in banking and real estate.
With a career spanning multiple geographies, including the U.S., Singapore, and Australia, Varun Nakra is a leading expert in developing machine learning and statistical models for credit risk management.
His research has been instrumental in maintaining global banking system stability, with a focus on myriad other applications of machine learning and AI – detecting anomalies in cyber defense systems, AI applications for urban planning, and more.
Nakra’s most notable achievement is his “Downturn LGD model”, which predicts mortgage credit losses.
Development and technical challenges
The main challenge in developing the “Downturn LGD model” was the lack of downturn data.
Before the COVID-19 pandemic, Varun worked in Australia, where real estate had been robust since the market crash of the 1990s. Without sufficient historical data on downturn events, Varun’s team faced a significant technical challenge.
“We had to develop a model that could still be accurate and reliable without having direct data from past downturns to rely on,” Nakra said.
Reputation and regulatory scrutiny
With property prices in Sydney and Melbourne rising rapidly, regulators identified a systemic risk in the mortgage market.
This required a predictive model to ensure that banks could hold sufficient capital and self-insure against potential losses.
The approval process was rigorous, with Nakra’s team receiving and responding to over 100 questions from regulators across two rounds of evaluation.
“This extensive scrutiny was necessary to prove the model’s robustness and accuracy,” Nakra explained.
Despite the challenges, Nakra’s model was approved by regulators, providing a precise calculation of the capital required to be held in reserve.
Approval and impact
Currently, one of the major Australian banks uses Nakra’s model for regulatory capital planning. It is also being used by the regulators to ensure that capital requirements for housing lending are sufficient to withstand losses through the economic cycle.
Before the model’s implementation, banks were required to hold a capital reserve using a fixed LGD floor of 20%. With Nakra’s model, this requirement was reduced to around 15% on average, resulting in a 5% average capital saving for the bank, which could then be utilized for other purposes.
For Nakra, the approval was a milestone. The model not only earned him outstanding performance ratings and an internal promotion, but also garnered appreciation from the regulators.
Continuous adaptation and future challenges
The Downturn LGD model’s success showed the need for continuous innovation in the banking industry. Nakra highlighted the importance of updating models to account for new risks, such as post-pandemic pressure from high inflation and interest rates.
“In a higher interest rate environment, payments swell when loans mature, posing significant risks if a bank has a large proportion of such loans,” Nakra explained. “Models must be redesigned and recalibrated to address these evolving risks, incorporating new stress-testing scenarios and quantitative analyses.”
Commercial real estate risks
According to Nakra, one area of particular concern for banks is commercial real estate, especially in major hubs like New York and Los Angeles.
The pandemic led to a surge in vacant office spaces, boosting the risk of defaults on office building loans. Regulators have been keen to monitor this risk, to prevent it from becoming systemic.
“There are still many empty buildings in places like Manhattan,” Nakra pointed out. “With significant vacancies, there’s a heightened risk of delinquencies as property owners struggle to meet their financial obligations.”
As the U.S. banking industry faces emerging risks, Nakra is researching new models for sustainable finance, incorporating environmental, social and governance risks.
“At the moment, there’s no standard methodology or simple guidelines on developing these models,” he said. “But that’s where the future lies.”
Nakra believes that integrating AI will be essential and inevitable for the development and implementation of these innovative models.
Featured image credit: Freepik
