4 min read.Updated: 14 Jan 2021, 10:47 PM ISTDeep Mukherjee
The Reserve Bank of India’s stress test model compares well with global benchmarks, but it needs to improve its assessment of tail-risk events, which are no longer as rare as they were
Central banks must continually assess the resilience of the banking system in case of a severe crisis. Globally, stress tests are undertaken to address such systemic risks. If done well, this can prepare a system well in advance to handle an economic crisis, by identifying potential capital requirements, so that it doesn’t need to hunt for money in adverse market conditions.
The Reserve Bank of India’s (RBI’s) half-yearly Financial Stability Report has the results of its stress tests on Indian banking. These tests have been helpful to the system and they tick most of the right boxes against global benchmarks. However, a review may be in order to assess whether RBI needs to increase the severity of the stress scenarios and sharpness of its models.
In these tests, the gross non-performing asset (GNPA) and capital to risk-weighted- asset ratio (CRAR) are estimated in the event of specific economic scenarios playing out. RBI considers six macro-economic variables and projects a baseline scenario. It represents the most likely trajectory of these variables for the subsequent 12-18 months. The baseline GNPA and CRAR estimates represent their most likely future levels. However, adverse scenarios are not predictive in nature and have no specific likelihood of occurrence associated with them. They represent different levels of severity for ‘what-if’ analysis to estimate GNPA and CRAR.
In its most recent report, RBI has gone back to using two adverse scenarios, as it did until December 2019. In last July’s report, three adverse scenarios were used. Additionally, since 2017, it has performed reverse stress testing. For this, it back-calculates the level of stress that will cause system-wide CRAR to fall below the 9% threshold.
System-wide GNPA in a baseline scenario is predicted to rise to 13.5% by September this year from 7.5% a year before that. Since regulatory relief may have suppressed bad loans in the post-March 2020 period, the stress-test models in the latest report used data until December 2019. The September 2020 GNPA is a statistical proxy for GNPA if regulatory forbearance was not offered.
The baseline gross domestic product (GDP) growth assumed is 0%. In a severely adverse scenario, GNPA is predicted to reach 14.8%, with GDP growth being -7.6%. Also, CRAR may drop to 14% in a baseline scenario and to 12.5% in a severe one by September from 15.6% the year before. Given the scale of India’s economic shock, these are not bad numbers.
However, analysing the reverse stress test results of previous reports raises some questions. The December 2019 report, for instance, assumed GDP growth of 5.2% (baseline scenario) and 2.9% (stress scenario). The stress scenario represented a standard deviation of 1.25 to 2. It suggested that a 3.5 standard deviation may bring the system-wide CRAR to 9.0%. This would imply GDP growth that is close to 0% or slightly negative. Assuming the estimates were reasonable, it is hard to understand how the latest report, which has GDP growth scenarios of 0% to -7.6%, has forecast the system-wide CRAR to remain well above the 9% threshold.
Similarly, the growth scenario in the July 2020 reverse stress test also estimated a stress level with standard deviation close to 3. Given these past published numbers, the 12.6% CRAR projected in the latest financial stability report calls for further explanation.
Further, analysing past reports suggests that the baseline scenarios have often been in line with the actual GNPA rates in the following 6-15 months. But there were some misses as well. For instance, as per the December 2015 report, in an extremely stressful scenario associated with GDP growth of 3.7%, GNPA was estimated at 6.9% in March 2017. The actual GNPA that month stood at 9.6%, despite GDP growth exceeding 7%. Likewise, actual GNPA (11.6%) in March 2018 was higher than the stress case GNPA (11.2%) projected in the December 2017 report. This was despite economic activity being in line with the baseline scenario.
A spike following an asset quality review may explain the GNPA underestimation. Such variations have continued to show up since 2016. One hopes that the latest report has addressed these underestimation issues.
A rethink of the stress-testing framework is needed. What constitutes a ‘severe but plausible’ stress scenario needs to be introduced. Evolved implementors of stress tests tend to interpret ‘severe and plausible’ as a tail-risk event that is rare and very severe. For instance, in Japan, the stress scenario replicates the worst recession observed. The UK assumes scenarios such as currency depreciation of 20-25%. In contrast, in India, prior to June 2020, sub-2% GDP growth was considered an implausible stress scenario.
If economic crises followed a bell-shaped normal distribution, a 3 standard deviation event would have been considered severe, with close to one-in-a-thousand odds of occurrence. However, economic shocks do not follow well-behaved distributions. Thus, such a risk scenario should not be rare.
Likewise, the relationship between the macro-economic numbers and GNPA cease to have easy-on-the-eye linear relationships during crises. Such assumptions may cause underestimation of losses and capital requirements. Risk frameworks are tested during crises. Being a prudent and capable regulator, RBI should consider upgrading its stress-testing framework.
Deep Mukherjee is a quantitative risk management professional & visiting faculty of risk management at IIM Calcutta.
Subscribe to Mint Newsletters
* Enter a valid email
* Thank you for subscribing to our newsletter.
Never miss a story! Stay connected and informed with Mint.
our App Now!!