Non-Performing Loans in South Africa: a Scoping paper for Future Model Development

Abstract:

Elevated non-performing loans (NPLs) are a recurrent characteristic of banking crises, with an important two-way connection between credit conditions and NPLs. Modelling such credit risk indicators could be highly relevant for informing monetary and macroprudential policy in SA, by strengthening model linkages between the financial sector and real economy. This paper surveys international literature on identifying, measuring and modelling NPLs. SA is not immune from the inconsistency of concepts across countries and jurisdictions, and within countries between different institutions and across time. Our clarifying typology details the evolution of NPL concepts in SA, heavily affected by regulatory definitional changes. We propose how pre- and post-2008 data on three different NPL concepts might be joined to permit an analysis of data from 2001 on reasonably consistent definitions (the volatile period, 2001-2007, can help draw robust insights). The SARB publishes a time series for the related credit risk indicator: ‘credit impairments’, a loan loss provision, not an NPL concept. We develop a new empirical model for the ratio of ‘credit impairments’ to gross loans and advances from 2001. Major drivers are the ratios of mortgage debt and house prices to income, credit conditions in the previous three years measured by credit spreads, and the GDP growth rate in two prior years. We allow for regulatory data breaks in 2008 and 2018. For potential NPL models, we expect similar key drivers with different relative weights in the long-run solution, and different short-run dynamics. We further propose panel studies of NPLs for individual banks and loan classes, even if only post-2008. Before institutional memory is lost, we urge a concerted effort from the Prudential Authority and Financial Stability Department to publish time series data for at least two of the NPL measures for 2001-2007, and from 2008, with clear documentation, to benefit in-house and more general modelling efforts.