Lompat ke konten Lompat ke sidebar Lompat ke footer



One of the main tasks of commercial banks is to offer loans, and their main source of risk is credit risk, that is, the uncertainty associated with borrowers‟ repayment of these loans. A non-performing loan (NPL) may be defined as a loan that has been unpaid for ninety days or more. 

For the purpose of this study, we analyse the non-performing loan ratios of the commercial banking sector calculated by dividing gross classified debt by total loans. 

The commercial banking sector of Barbados consists of six commercial banks which are currently all foreign owned, and presently the aggregate NPL ratio is approximately 3.138%. The magnitude of non-performing loans is a key element in the initiation and progression of financial and banking crises. 

Ahmad (2002), in analyzing the Malaysian financial system, reported a significant relationship between credit risk and financial crises and concluded that credit risk had already started to build up before the onset of the 1997 Asian financial crisis, and became more serious as NPLs increased. Li (2003) and Fofack (2005) also found this relationship to be significant. 

Further, the current global financial crisis, which began in the United States, is attributed to the August 2007 collapse of the sub-prime mortgage market. In fact, there is evidence that the level of NPLs in the US started to increase substantially in early 2006 in all sectors. 

NPLs are therefore a measure of the stability of the banking system, and thereby the financial stability of a country. Given the above discussion, it is not difficult to see why the ability to forecast non-performing loans is important. 

Generally, previous empirical studies have modeled NPLs through the use of various multivariate analyses. For example, Chase et al. (2005) used OLS to forecast non-performing loans using the treasury bill rate, the consumer price index, real gross domestic product (GDP) and a lagged dependent variable. 

This study contributes to the existing literature by modeling the NPL ratio of the commercial banking sector in Barbados, not only on an aggregate level but also on an individual bank level. 

This research paper therefore attempts to use a multivariate model to forecast non-performing loans using quarterly bank specific data, as well as macroeconomic factors The structure of the paper is as follows: section 2 provides an overview of non-performing loans in Barbados; section 3 provides a review of existing literature; section 4 then presents the model estimates and results; section 5 offers a discussion of results and concludes with a summary of the findings, including limitations and policy implications.

Overview of Non-Performing Loans in Barbados

This section reviews the evolution of NPLs in the banking system of Barbados. As a precursor to the discussion, it should be noted that the Barbadian financial sector is well developed and encompasses a wide range of financial institutions. 

There are currently six commercial banks, 13 non-bank financial institutions, 34 credit unions, 11 life insurance and 16 general insurance companies2 . At end-2008, assets of commercial banks accounted for 142% of GDP and about 80% of the assets of all deposit-taking institutions. 

In addition, commercial banks accounted for 82% of all deposits and around 74% of loans and advances. Our study utilises quarterly data spanning the period 1996 to 2008. Prior to1995 there was no standard treatment or interpretation of nonperforming loans. 

Information was received on past-due loans that did not include all the features of what is now termed as classified debt. Each bank employed its own rating system, and some still retain their own internal classification system which runs parallel to that instituted by the Central Bank of Barbados. 

The Asset Classification and Provisioning guidelines, which are based on the Basle Committee‟s Core Principles, were written into law in 1996. Over time there has been general adherence to these guidelines and standardisation has been largely achieved. 

Therefore, figures on classified debt are available on a quarterly basis from 1996. However, since complete adherence to the new provisioning guidelines was not immediately achieved, figures may have been misrepresented in the earlier stages. 

In fact, during this period, it was not unusual for examiners to adjust the level of classified debt reported by banks on conclusion of an on-site examination. However, these adjustments were usually minor.

Literature Review

Despite the importance of the examination and monitoring of nonperforming loans, forecasting these ratios has only received moderate attention in the literature. There is a general consensus that the level of NPLs experienced by banks is determined by internal and/or external factors. 

For instance, Keeton and Morris (1987) pointed out that local economic conditions and the poor performance of certain industries explain the variation in loan losses. However, commercial banks with greater risk appetite and that are more willing to make loans with a higher probability of default, tend to record higher losses. 

Sinkey and Greenwalt (1991) also shared this general view, and posited that NPLs reflect realized credit risk for banks arising either from external factors such as depressed economic conditions, or internal factors such as poor lending decisions or both. 

The study found a significantly positive relationship between the level of loan defaults and high interest rates, excessive lending and volatile funds. The existing literature however suggests a variety of determinants and approaches to be used in the forecasting of non-performing loans. 

Graham and Humphrey (1978) presented one of the early attempts at predicting non-performing loans. The authors suggested that, in general, banks with larger amounts of classified loans (loans with more than normal risk) will experience greater amounts of future losses, and hence classified loan data should be included as an indicator of these loan losses. 

The authors therefore evaluated whether taking classified loan data into account improves forecasts of future net loan losses. Subsequent models are of a more complex nature and include a greater selection of variables for the forecasting of non-performing loans. 

For example, Barr et al. (1994) argued that bank failure prediction studies have continually concluded that the level of efficiency of a bank‟s management is the leading cause of failure, yet few researchers have attempted to quantify management quality or incorporate it into predictive models. 

Seballos and Thomson (1990) and Hsing et al. (1991) also supported the view that a key determinant is management‟s ability to operate efficiently and manage risks. 

Barr et al. (1994) therefore attempted to incorporate management quality as an explanatory variable through the use of a data-envelopment analysis (DEA), which combines multiple inputs and outputs to compute a scalar measure of efficiency. 

In addition, the authors included variables representing Capital Adequacy, Asset Quality, Earnings Ability and Liquidity Position, to complete the CAMEL rating, as well as a proxy for local economic conditions. 

The performance of the DEA management variable is assessed using a Probit regression model to develop one- and two-year ahead forecasts. Their results supported the claim that management‟s efficiency is indeed important in forecasting bank failure. 

More recently, Chase et al. (2005) modelled non-performing loans using the Treasury bill rate, the consumer price index, real GDP and a lagged dependent variable. The authors use a similar technique to Graham and Humphrey (1978), where Ordinary Least Squares (OLS) is employed to forecast the NPL to total loans ratio for the banking system in Barbados. 

All of the explanatory variables were found to be significant. Subsequent research conducted in the Caribbean includes that of Khemraj and Pasha (2009), who examined the determinants of non-performing loans in Guyana. 

Using a panel dataset and a fixed effect model, the authors regressed the NPL ratio on the GDP growth rate, inflation rate, real effective exchange rate, and the bank specific variables, loans to total assets ratio, size, real interest rate and annual growth in loans. 

The empirical results revealed that with the exception of the inflation rate and bank size, all other factors have a significant relationship with the NPL ratio. 

However, note should be made of an earlier argument by Smith and Lawrence (1995) that macroeconomic variables have limited predictive powers in explaining loan defaults, and that explicitly including them in the forecasting model is unlikely to improve its effectiveness for forecasting purposes. 

They specified a mortgage-loan-default forecasting model based on a Markovian structure, as an extension of the work of Lawrence et al. (1992), who examined the determinants of default risk for mobile home loans. 

Smith and Lawrence‟s findings suggested that payment history, the geographical area in which the home is located, and the number of months expired and remaining in the loan‟s term, are the main contributions to loan default. 

The authors also noted that several papers have concentrated on the identification of factors that help in the prediction of default, but neglect issues in the development of long-term forecasts of losses on loan portfolios. 

Nonetheless, Betancourt (1999) remarked that although the Markov Chain technique is a reasonable approach for estimating loan losses, a common problem with these models is the requirement of very strong assumptions regarding stationarity and homogeneity, which are not usually satisfied. 

The author estimated loan losses from a portfolio of mortgages, where in any month, a mortgage could be classified into one of the following categories: 
  • Active, 
  • Thirty days delinquent, 
  • Sixty days delinquent, 
  • Ninety plus days delinquent, 
  • foreclosure, 
  • Real estate owned (REO) and 
  • Paid off. 
If B0 represents a start vector of mortgages at time 0, then multiplying the vector B0 times the transition matrix P yields a forecast B1 of how the mortgages in the start vector will be distributed at time 1. 

A forecast of loan losses (REO acquisitions) at time t can be generated by simply observing the number of loans expected to transition to REO at time t. The authors concluded that when using the most recent information on transition probabilities, the Markov Chain approach could provide a more accurate forecast of loan losses than a random walk model.
Bona Pasogit
Bona Pasogit Content Creator, Video Creator and Writer