A Model-Based Approach to Determine the Number of Scenarios and Scenario Probabilities for Loan Loss Provision Calculations Under the Accounting Standards of IFRS 9 and US-GAAP CECL

Abstract

The accounting standards of the International Financial Reporting Standards (IFRS) and the United States Generally Accepted Accounting Principles (US-GAAP) require from financial institutions to consider multiple macroeconomic scenarios when calculating loan loss provisions. At present, however, it is unclear how to determine the number of scenarios and scenario probabilities without resorting to - often subjective - expert judgement. The paper discusses a model-based approach and proposes to use hidden Markov models to determine the number of relevant scenarios and scenario probabilities. The tool of the hidden Markov model allows to use established model selection criteria, such as the Akaike information criterion, to decide on the number of scenarios. Hidden Markov models also provide estimates of the transition matrix of the hidden states, which constitute the required conditional scenario probabilities. The tool of the hidden Markov model is discussed by using a time series of defaults from Standard & Poor's.