An analysis of the credit rating model of firms in Vietnam - NCS. Phạm Quốc Hải

Tên đề tài: An analysis of the credit rating model of firms in Vietnam
Ngành: Quản trị kinh doanh
Mã số ngành: 9340101
Họ và tên NCS: Phạm Quốc Hải
Tập thể hướng dẫn: TS. Hồ Điệp, TS. Sarod Khandaker
Cơ sở đào tạo: Trường Đại học Quốc tế – ĐHQG-HCM
Những nội dung chính của luận án
Theoretical significance
In the literature over time, there are two main ways to increase the accuracy of credit rating model including improving the independent variables and improving the prediction methodology (Altman and Sabato 2007; Psillaki, Tsolas and Margaritis 2010; Wu, Gaunt and Gray 2010; Duan, Sun and Wang 2012; Giordani et al. 2014; Jones, Johnstone and Wilson 2017; Cornée 2019; Muñoz‐Izquierdo et al. 2019; Sigrist and Hirnschall 2019). This thesis aims to build a credit rating model for small, medium and large firms in Vietnam in the period 2008–2018 and investigates the effects of the local accounting system on the credit rating model.
This thesis contributes significantly to the existing literature for several reasons. First, this research is significant because most of the influential studies analyse data before 2016 and mostly in US and European contexts (Altman and Sabato 2007; Psillaki, Tsolas and Margaritis 2010; Wu, Gaunt and Gray 2010; Duan, Sun and Wang 2012; Giordani et al. 2014; Jones, Johnstone and Wilson 2017; Sigrist and Hirnschall 2019; Cornée 2019; Muñoz‐Izquierdo et al. 2019). Since Altman’s Z-score model (1968), the accuracy of credit rating models has needed updating and rebuilding because it varies significantly based on country and period (Altman and Sabato 2007; Tsai et al. 2009; Altman et al. 2017; Jones, Johnstone and Wilson 2017; Pham, Do and Vo 2018). This study builds a credit rating model of Vietnamese small, medium and large firms in the period 2008–2018. The dataset includes 39,162 small, medium and large firms in Vietnam in the period 2008–2018 drawn from the Orbis database. Pham, Do and Vo (2018) built a default prediction model for Vietnamese listed firms in the period 2003–2016 that focused only on default and non-default (two classes for the dependent variable) prediction. This research investigates a credit rating model with ten classes for the dependent variable that is more informative and powerful in predicting the default probability of firms in Vietnam in advance as well as supporting lending decisions. Also, this research uses small, medium and large firms in Vietnam in the period 2008–2018 to develop the model, which better represents a generalisation of firms in Vietnam. As a result, the dataset for this research has significant advantages compared to the study by Pham, Do and Vo (2018) by using (1) a more informative dependent variable with ten categories; (2) a bigger dataset including small, medium and large firms in Vietnam on which to develop and test the model; and (3) a more current period 2008–2018.
Another significant contribution to the literature is the independent variables. In the literature, many possible ratios have been identified as helpful in predicting the credit rating of firms. Independent variables are usually accounting ratios derived from financial statements and include measures of profitability, liquidity and leverage in a credit rating model (Altman and Sabato 2007; Wu, Gaunt and Gray 2010; Giordani et al. 2014; Pham, Do and Vo 2018; Sigrist and Hirnschall 2019; Muñoz‐Izquierdo et al. 2019). Some studies also include market-based variables and macroeconomic indicators such as stock volatility and past excess returns, GDP or annual interest rates (Wu, Gaunt and Gray 2010; Jones, Johnstone and Wilson 2017; Pham, Do and Vo 2018). The recent literature concludes that quantitative variables are not enough to predict business default and qualitative variables can improve the predictive power of a model (Altman and Sabato 2007; Psillaki, Tsolas and Margaritis 2010; Wu, Gaunt and Gray 2010; Duan, Sun and Wang 2012; Giordani et al. 2014; Muñoz‐Izquierdo et al. 2019; Sigrist and Hirnschall 2019; Cornée 2019). In this research, the independent variables include four main categories: financial indicators, market indicators, firm characteristic indicators and macroeconomic indicators. This thesis also proposes three main models from three different methodologies; ANNs, MDA and ordered logistic regression (OLR). Each model brings a different list of significant independent variables to the credit rating model of Vietnamese small, medium and large firms in the period 2008–2018. This thesis study also ranks each independent variable based on their impacts on each model, so that this research can identify which independent variable is the most significant and which is less significant to the credit rating model.
The third significant contribution to the literature is the research methodology. This study builds a credit rating model with ten classes for the dependent variable of firms in Vietnam using three different models; ANN, MDA and OLR, to maximise the predictive power of the models. To illustrate, several researchers in the credit rating field have used two categories of dependent variable, ‘default’ and ‘non-default’ (two classes) (Altman and Sabato 2007; Psillaki, Tsolas and Margaritis 2010; Wu, Gaunt and Gray 2010; Duan, Sun and Wang 2012; Giordani et al. 2014; Muñoz‐Izquierdo et al. 2019; Sigrist and Hirnschall 2019; Cornée 2019). This research classifies credit ratings into ten classes (from AAA to D), which are more informative compared to a two-class dependent variable. However, the challenge in building more than two classes for the dependent variable in a credit rating model is the significantly larger amount of data required to build the model. Machine learning models such as NNs and SVMs are arguably the most robust in term of predictive accuracy, but these models are hard to interpret and apply because they involve a ‘black box’ in the analysis process (Jones, Johnstone and Wilson 2017; Sigrist and Hirnschall 2019). This research opens the ‘black box’ of ANN using the Simulink tool of MATLAB R2019a  to increase the interpretive power of the model. To illustrate, the Simulink tool provides by MATLAB R2019a can visualise the ‘black box’ of ANN model, so that this research can effectively interpret the processing data and model result. This research also uses the most popular models in the literature, which are MDA and OLR, for comparison purposes (Altman et al. 2017; Pham, Do and Vo 2018; Sigrist and Hirnschall 2019). The results of this thesis show that the ANN model is the most powerful in terms of accuracy rate, then the OLR is the second and the MDA is the weakest.
Finally, for the first time, this research analyses the effect of the new accounting system on credit rating models of Vietnamese small, medium and large firms. In the literature, several influential papers investigate the effects of International Financial Reporting Standards implementation on significant financial ratios and accounting numbers (Bartov, Goldberg and Kim 2005; Daske and Gebhardt 2006; Hope, Jin and Kang 2006; Ding et al. 2007; Barth, Landsman and Lang 2008; Lantto and Sahlström 2009; Morales-Díaz and Zamora-Ramírez 2018). Besides, the results for each sector are significantly different for most of the cases in the dataset (Morales-Díaz and Zamora-Ramírez 2018). Another influential research study in this field is Tran et al. (2019) that investigates the factors affecting International Financial Reporting Standards adoption in listed firms of Vietnam, but the effects of Vietnamese Accounting Law are still questionable. To illustrate, the new accounting system significantly changes the measure of financial numbers in the financial reports of firms, including assets, liabilities, owner’s equity, revenues and expenses, so this also affects the credit rating model.
This study investigates the new accounting system (Circular 200) implemented on 1 January 2015 in Vietnam, which guides both local and foreign enterprises in accounting policies for financial years beginning 1 January 2015 (Phan, Joshi and Tran-Nam 2018). Circular 200 enhances the comparability and transparency of corporate financial statements and brings the Vietnamese accounting system and international accounting system closer. Circular 200 significantly affects the measure of financial ratios (Phan, Joshi and Tran-Nam 2018). In order to investigate the effects of this new local accounting system on the credit rating models, this research separates the dataset into two sub-datasets, comprising dataset 2008–2014 (before the implementation of Circular 200) and dataset 2015–2018 (after the implementation of Circular 200). After analysis of these two sub-datasets, the results of this study show that the new local accounting system significantly affects the credit rating models in several ways: (1) changing the inputs to the models; (2) changing the model performance; and (3) changing the exact values of the weights and biases of the models.
Practical significance
There are different functions of credit ratings in the financial markets; firstly, they are a technology tool to alleviate asymmetric information problems between borrowers and lenders (Kumar and Bhattacharya 2006; Benmelech and Dlugosz 2009). Secondly, a reliable credit rating model can help investors with low-cost and useful information which can minimise the level of investment risk significantly. To illustrate, credit ratings can classify firms into their credit class so that investors can maximise their investment decisions based on the firm credit rating. Credit ratings can also help to build a company’s brand and image significantly. If companies have good credit ratings, they can get significant advantages in raising funds in the financial markets, including more accessible bank loans, lower loan interest rates from creditors and being attractive to investors (Kumar and Bhattacharya 2006; Benmelech and Dlugosz 2009). From a macroeconomic view, credit ratings can play a role as a regulatory tool in the financial markets. As a result, these ratings will motivate and strengthen the economic growth of a country (Kumar and Bhattacharya 2006). Credit ratings also help to minimise asymmetric information problems between companies and investors in capital markets, which can address the problem of collective action between dispersed investors (Kumar and Bhattacharya 2006; Benmelech and Dlugosz 2009).
Currently, the commercial banking system in Vietnam is still struggling in the early stages of researching and developing projects, as well as applying the international Basel II standards (Mui 2015). The majority of commercial banks are implementing the Basel II standards into their internal credit rating model, and that is considered to be very expensive in terms of time, workforce and financial costs. Also, the internal credit rating model of Vietnamese commercial banks mainly focus on classifying firms into default and non-default groups (Pham, Truong and Bui 2018). However, they have been unable to estimate the credit rating for the firms. Therefore, the resulting model from this study has high practical significance and makes essential contributions as a reference model to the risk management activities of Vietnamese commercial banks.
The statistical model constructed in this study can help Vietnamese commercial banks calculate a specific figure as the credit rating of a borrower, so that the bank may offer an appropriate credit policy and interest rate level for each client following the principle of ‘high risk, high return’. This model will provide more detailed credit rating information to Vietnamese commercial banks so that it will help to limit subjective errors by credit officers in evaluating customers. In fact, the internal credit rating of Vietnamese commercial banks depends more on qualitative factors that are rated by credit officers, such as management level of the firm or the relationship between the firm and the commercial banks, than financial factors of the firms (Pham, Truong and Bui 2018). These qualitative evaluations are subjective because they depend on the experience of the credit officers. Besides, this study will also help Vietnamese commercial banks’ risk administrators to verify the degree of influence of each indicator (the independent variables) on the credit rating of the firm (the Dependent Variable). The risk officers can then judge clearly which elements have the most significant impact on the repayment capacity of the borrower and will make more accurate decisions about granting credit. These results can be used as reference information for granting credit decisions.