Banks face challenges in making ECL-based provisioning work

Loss given default estimates the extent of the loss the bank will incur if a default takes place, subject to the quality of the security collateral, efficiency of the recovery process and recovery institutions and the strength of the legal frameworks.

Reliable Measure

With these two estimates, the bank should be able to compute the present value of expected losses over the life of a loan. ECL, a more reliable and forward-looking measure of potential loan loss, then becomes the basis of provisioning and reporting for the bank.

What is the information required to assess the probability of a default and the loss if a default takes place?

For probability of default, it is the past performance of the borrower. This could be in the form of the bank’s historical data on loan performance and defaults. Or it could be data on borrower ratings for bonds and loans, which may range from AAA for the most creditworthy to D for those already in default. Or it could be data from credit bureaus on loan repayments and defaults.

For loss given default, it is data on recoveries made through mechanisms such as liquidation of security collateral or enforcement through the courts.

Four common features of good information for ECL estimation are: granularity, frequency, width, and standardisation. Information that is at the level of loan records and borrower characteristics and can be aggregated to different cuts of the lending portfolio is better.

Similarly, a higher periodicity of information refresh is desirable. Pooling of information from across lenders is richer than a bank’s own historical information. And a certain degree of standardisation is key to information aggregation.

Given this, the information requirements for ECL models possess some characteristics akin to a "public good." For instance, their procurement and aggregation may entail substantial private costs, yet the communal benefits derived from their dissemination are substantial.

Rating companies publicly disclose default probabilities for the bonds they rate, based on historical data. However, these companies do not extend the same transparency when rating bank loans.

While bond default probability data for loans may serve as a reasonable approximation, it will not be entirely accurate. In the absence of publicly available data from the rating companies on default probabilities for bank loans, banks are left with no alternative but to rely on their own customer data to compute these probabilities.

Inadequate Datasets

Further, information aggregation may not take place organically. Lenders that otherwise compete for business may not be able to come together to pool information without a regulatory nudge.

The largest Indian bank holds about a quarter of the market share in bank loans, and most of the approximately 150 scheduled commercial banks possess single-digit market shares. Consequently, the individual datasets of banks are woefully inadequate to generate reliable estimates of default probabilities.

This issue is especially acute for smaller banks. Incorrect estimations of default probabilities can lead banks to miscalculate their capital requirements. Moreover, aggregated data on loan ratings and associated default probabilities are exclusively accessible to rating companies, which refrain from public disclosure.

An additional issue pertains to the lack of standardised ratings. Loans rated the same by different companies can exhibit significantly different default probabilities. This disparity becomes more pronounced as one descends the rating scale, particularly for BBB and BB-rated bonds.

Lack of standardisation of the rating scale, combined with the limitation of data, would lead banks to miscalculate default probabilities, ultimately resulting in inaccurate capital levels.

Hence, it stands to reason that regulatory agencies and their affiliated institutions have a role to play in enabling the collection, standardisation and distribution of such information. Currently, there is no regulatory mandate to publicly disclose the rating mix of a bank's loan book.

Due to the lack of standardisation, it is impossible to compare the actual quality of the loan portfolio even for banks that voluntarily disclose the rating mix of their loan books. The regulator must take a multi-pronged approach to fix this, which includes standardising rating categories by defining threshold default probabilities, and compelling rating companies to publicly disclose loan rating default probability data. Banks should be obliged to disclose information about the rating companies they utilise, including details such as rating transitions.

There could also be a model where the regulator itself aggregates loan rating data across all agencies and computes default probabilities, thereby establishing a more robust basis for estimation. However, this is subject to issues of regulatory capacity and capture.

Limited Utility

The other facet of ECL-based provisioning, loan given default, is also fraught with estimation challenges for banks. Much like the default probabilities, individual banks have limited data sets for estimation.

The enactment of the Insolvency and Bankruptcy Code (IBC) in 2016 ushered in a fundamental shift in loan recovery processes, rendering pre-2017 data less useful. The Insolvency and Bankruptcy Board of India provides "overall" recovery rates, but these are of limited utility for ECL-based provisioning, which necessitates recovery rates across various categories of loan security (including collateral).

Further, there are some non-IBC recovery processes that are still used by banks, such as through the invocation of the Securitisation and Reconstruction of Financial Assets and Enforcement of Securities Interest Act, 2002, (SARFAESI Act) or the use of debt recovery tribunals, for which there are only high-level aggregates available on recovery rates.

In this context, collaboration between the regulator, the ministry of finance and the IBBI could become vital. They can work together on building a reporting framework at the level of categories of debt and security collateral, which is then implemented across the tribunals, SARFAESI actions and the IBC.

In the absence of estimation based on aggregated, sector-wide data, individual banks are likely to struggle in accurately estimating loss given default.

The adoption of ECL-based provisioning could usher in a transformative era for Indian banking, with the potential to rival the impact of the entry of private banks in the 1990s and the enactment of the IBC. It has the power to instil stability within the Indian banking sector, averting the tumultuous cycles of non-performing assets and capital fluctuations witnessed in recent decades.

However, the effectiveness of its implementation hinges on overcoming informational constraints. The information and data required for ECL implementation possess characteristics of a public good, emphasising the crucial role of the regulator in collecting and disseminating this vital information. Without this support, this monumental reform may fall short of achieving its full potential.

Harsh Vardhan is a management consultant and researcher based in Mumbai and Anjali Sharma, is research director at TrustBridge. The authors thank Prateek Mankad of World of Finance for valuable comments. Views are personal, and do not represent the stand of the publication.