NON-LINEAR MODEL OF AGGREGATE CREDIT RISK FOR BANKING SECTOR OF PAKISTAN: A Threshold Vector Autoregressive Approach

Abstract

The Global Financial Crisis highlighted the failure on part of risk managers and regulators to adequately determine and account for the buildup of risks in the financial systems. It also highlighted the inherent inadequacy of conventional models to capture and predict the tail risks in the financial sector and provide reliable forecasts under stressed scenarios. Much recently, focus has shifted towards building innovative models to account for these risks. This study develops a Threshold Vector Autoregressive model for the banking sector of Pakistan and compares its accuracy to conventional linear counterpart in terms of forecasting Gross Non Performing Loans ratio, a key financial stress indicator. The results suggest the presence of a significant threshold in the data generating process and the estimated threshold model as faring better at predicting the Gross Non-Performing Loans ratio with much lower forecasting errors for up to four period ahead forecasts, particularly at longer horizons