Tóm tắt Luận án A study on accounting - Based approach and market-based approach in corporate defaults in Vietnam

recast model by Altman (1983), prediction model by Deakin (1972), logit and probit analysis models including Ohlson's bankruptcy prediction model (1980), forecast model by Zmijewski (1984), Fulmer's model (1984), forecast model by Zavgren (1985).... It can be seen that foreign researchers who studied default models via accounting-based approach primarily focused on solving two objectives: (1) Evaluate conditions at which the model works effectively. (2) Concentrate on explanatory variables that can be used to design the suitable model. 1.1.2. Accounting-based approach in Vietnam: 6 In Vietnam, many studies on accounting-based approach were carried out. They are Dinh The Hien (2008); Lam Minh Chanh (2007); Nguyen Trong Hoa (2009); Hoang Tung (2011), Nguyen Thanh Cuong, Pham The Anh (2010); Khong Thanh Hoa (National Economics University, 2008); Dao Thi Thanh Binh (Hanoi University, 2011); Nguyen Phuc Canh and Vu Xuan Hung (2014); Tram Thi Xuan Huong et al (2015); Pham Thi Tuong Van (2016); Nguyen Duy Collect (2016); Tristan Nguyen and Trung Dung Doan (2018)... Although many studies aimed at improving default prediction by using Z-score model, these research with a concentration on accounting-based approach have these following strengths and shortcomings: ❖ Strengths + Data was easily collected from financial statements and public information. + Data was widely accepted and used in reality. ❖ Shortcomings + Data in financial statements are data from the past which does not possess prediction features for the future. + Data can be forged and overstated which leads to results being inaccurate. + Different methods of accounting can produce different outcomes. 1.2. Corporate defaults in market-based approach 1.2.1. Market-based approach in the world: An overview of research using market-based approach via traditional methods and option pricing shows that there are many studies following this orientation such as option pricing – Moody’s – KMV (1993), research by Bharath (2004); Chan et al. (Chan, Faff, and Koffman, 2008; Gharghori, Chan and Faff, 2007); Lu (2008)... 1.2.2. Some research using KMW-Merton model in Vietnam Some notable studies are Lam Chi Dung and Phan Dinh Anh (2009), Le Long Hau (2010), Nguyen Thi Canh and Pham Chi Khoa (2014)... 1.3. Research comparing accounting-based approach and market-based approach Some studies focusing on comparing accounting-based approach and market- based approach in corporate default forecast are Beaver et al. (2005); Agarwal and Taffler (2006), Liu et.al (2010); Illegeist et al. (2004); Mensah (1984) and Hillegeist et al (2004); Saunders and Allen (2002)... . Research on corporate default forecast whether using which approach all had strengths and weaknesses and required to be modified to suitably apply with each context. In Vietnam, there has not been a study that is able to clarify the efficiency of 7 corporate default prediction through using two above approaches, therefore to provide conclusion on the efficiency of forecast model usage as well as recommendations for application. Through the overview of above studies, the author discovered several shortcomings as follows: Firstly, corporate default forecast models are studied and applied widely in countries with developed market economy while this is not a similar case in Vietnam. A few research on this subject were conducted but still suffer various limitations. Secondly, there is a lack of legal documents, suitable forecast models and methods to identify corporate defaults in the context of applying default prediction. Also, there have not been standard methods as well as regulations for the identification of default point for Vietnamese enterprises in order to use them as the basis for practical application. Thirdly, studies on corporate default prediction in Vietnam only researched the application of only one forecast model and not multi-approaches or models to compare and select the most appropriate method. Taken all these above limitations into consideration, the author decided to conduct “A study on accounting-based approach and market-based approach in corporate defaults in Vietnam” with the aim of providing solutions to them. Additionaly, the thesis can be considered to be significant theoretically and practically, especially in the context of Vietnam’s regulations related to corporate default forecast being modified to be complied with international standards. CHAPTER 2: THEORETICAL FRAMEWORK ON CORPORATE DEFAULT FORECAST 2.1. Corporate defaults 2.1.1. Definition of corporate defaults and bankruptcy The thesis clarifies the definition of default risk, default and bankruptcy. Law on Bankruptcy promulgated on June 19, 2014 stated that “Bankruptcy means a status where an enterprise or cooperative becomes insolvent and is subject to a people's court's decision declaring bankruptcy”. It also clarified that insolvent enterprise or co-operative means an enterprise or cooperative failing to perform an obligation to repay a debt within three (3) months from the maturity date. The research stated that corporate default is a company’s failure to meet the legal obligations (or conditions) of a loan, tax and employee payments Theoretically, it is possible to use one or mix with other criteria to identify default which can possibly lead to bankruptcy. According to Law on Bankruptcy 2014, bankruptcy means a status 8 where an enterprise or cooperative becomes insolvent and is subject to a people's court's decision declaring bankruptcy. 2.1.2. Characteristics of default risk: imbalanced distribution of interest rate, asymmetric information on default risk, unsystematic risks. Measurement of default risk is conducted via four primary quantities: ❖ Probability of default ❖ Loss given default ❖ Expected loss ❖ Present value of the expected loss 2.1.3. Factors affecting probability of corporate defaults - Business factors - Non-business factors 2.1.4 Corporate default threshold By carrying out research on regulations related to the identification of default threshold in Vietnam and the world, the study base on these following criteria to classify default companies: (1) The company fails to fulfil obligations to a debt within 90 days from the maturity date (2) Business results suffer loss for three consecutive years or cummulative loss exceeds authorized capital stated in the most recent audited financial statement upon the time of evaluation. (3) The company has negative revenue for a year and share capital in that year is lower than authorized capital. (4) Audit organizations refuse to audit or provide opinions for the company’s most recent financial statement. 2.2. Corporate default forecast models 2.2.1. Model using accounting-based approach * Model by Altman (1968) Z = 1,2X1 + 1,4X2 + 3,3X3 + 0,64X4 + 0,999X5 In which: Z = Composite index X1 = Working capital / Total assets X2 = Retained earnings / Total assets X3 = Earnings before interest and taxes (EBIT)/ Total assets X4 = Market value of share capital / Book value of total debts X5 = Revenue / Total assets 9 The research applies classification if Z-score is smaller than 1.81, the company defaults. * Z-score forecast model by Altman (1993): Z= 6,567X1 + 3,26X2 + 6,72X3 + 1,05X4 In which: Z = Composite index X1 = Working capital / Total assets X2 = Retained earnings / Total assets X3 = Earnings after interest and taxes (EAIT)/ Total assets X4 = Book value of equity capital / Book value of total debts The research applies classification if Z-score is smaller than 1.10, the company defaults. 2.2.2. Model using market-based approach • KMV model According to KMV model, default pount is approximately estimated by short- term debts and half of long-term debts DPT = STD + 0.5*LTD In which + DPT: default point + STD: Short-term debts + LTD: Long-term debts In addition, before calculating probability of default, KMV firstly calculates one more quantity called Distance to Default (DD). Distance to Default is the expected difference between the asset value of the firm relative to the default barrier, after correcting and normalizing for the volatility of assets. DD=  Probability of default has a correlation with Distance to Default. If the larger the Distance to Default is, the smaller the probability of default will be and the company is unlikely to default. 10 Figure 2.1. Definition of KMV model Calculation of distance to default includes two steps: 1. Calculation of Absolute Distance to Default- DD’ 2. Calculation of Relative Distance to Default - DD Absolute Distance to Default (DD’) is expressed as a distance between expected assets and Default Point (DPT) or it can be displayed as a sum of initial distance and the growth of that distance within the period T  =     +  − 1 2    Dissimilar to Merton model, in KMV model µA is no longer risk-free rate but expected rate of the return of the firm's asset and DPT is Default Point instead of nominal value K. Expected growth of assets is equal to  −  2   instead of µA. While rate of return is normally distributed, consequently future value of investment (or effective yield of return) is distributed lognormaly. The relationship between two distributions is as follows:  !"  ~ $ % − 2  2  , √( Dividing absolute value DD’ with calibrated (according to T – usually annualized) volatility of assets, we can calculate DD in relative terms as a multiplier of standard deviation. Therefore, the research can identify the probability of default according to this following formula: PD = 1 - N(d2) = N (-d2) In KMV model, probability of default (PD) is replaced with Expected Default Frequency (EDF). 11 Figure 2.5. Graph demonstrating the correlation between EDF and DD • Methods to identify default point when using KMV model When applying KVM model in each market, at each period and for each subject, a different default point will be used. In the research by Anthony Saunders (2006) in Credit Risk Measurement, out of 40.000 surveyed enterprises, 3.400 default companies defaulted with data collected from 1996 to 2001 and the default point of KMV model in 2011 was 20%. Lopez (2002) in The Empirical Relationship between Average Asset Correlation, Firm Probability of Default studied 2000 populations in credit portfolio investments including companies from the US, Japan and Europe with data collected in 2000 to establish Credit Ranking based default value of KMV. The model was also used by many researchers such as Allen M. Featherstone, Michael R. Langemeier, and Kent Haverkamp (2006) in Credit Quality of Kansas Farms to calculate probability of default and respective credit ranking for 51.382 data inputs from 1980 to 2003. The author used Credit ranking by Lopez (2002) with D = 18.25% ro identify whether companies default or not. Chapter 2 provided the corporate default basis and studied the default threshold for Vietnamese companies. The thesis selected Z-score (1968) and Z- Score model (1999) for accounting-based approach and KMV model for market-based approach. These are representative models requiring theoretical and empirical research for Vietnamese enterprises. 12 CHAPTER 3 RESEARCH METHODS 3.1. Research design The application is implemented through these four following steps: Step 1: Financial and market data are collected from secondary sources including financial accounting reports, market data archived at Stoxplus Company, reports of companies listed on Ho Chi Minh Stock Exchange from 2014-2016. Data was collected an inserted into fields to calculate default index: - In terms of finance: Working capital, short-term debts, long-term debts, earnings after taxes, return on assets at each period of time, total assets, retained earnings, earnings before interest and taxes, market value of share capital, book value of total debts, revenue. - In terms of market: prices in each transaction day of the year, transaction volume in each transaction day of the year to calculate the market’s asset capitalization. Step 2: Application of default forecast models via accounting-based and market-based approaches to predict corporate defaults. * Application of accouting-based approach via Z-score model + Applying Z-score model (1968), discriminant function is as follows: Z = 1,2X1 + 1,4X2 + 3,3X3 + 0,64X4 + 0,99X5 Z-score is the foundation of classifying default and non-default companies. - If the company has Z-score index being smaller than 1,81, it is classified as a default company. - If the company has Z-score index being higher than 1,81, it is classified as a non-default company. + Applying Z-score model (1993) Z= 6,567X1 + 3,26X2 + 6,72X3 + 1,05X4 If the company has Z-score index being smaller than 1,10, it is classified as a default company - If the company has Z-score index being higher than 1,10, it is classified as a non-default company. * Application of market-based approach via KMV model The research identified default point through estimation of short-term debts and half of long-term debts: DPT = STD + 0.5*LTD (19) In which: 13 + DPT: default point + STD: Short-term debts + LTD: Long-term debts In addition, before calculating probability of default, KMV firstly calculates one more quantity called Distance to Default (DD). Distance to Default is the expected difference between the asset value of the firm relative to the default barrier, after correcting and normalizing for the volatility of assets. DD=  Calculation of distance to default includes two steps: 1. Calculation of Absolute Distance to Default- DD’ 2. Calculation of Relative Distance to Default - DD) Absolute Distance to Default (DD’) is expressed as a distance between expected assets and Default Point (DPT) or it can be displayed as a sum of initial distance and the growth of that distance within the period T  =     +  − 1 2    21 According to the model’s definition of default, the company’s assets in the normally distributed time T have value lower than default point and therefore, the research can identify probability of default via this following formula: PD = 1 - N(d2) = N (-d2) In KMV model, probability of default (PD) is replaced with Expected Default Frequency (EDF). • Default point Lopez (2002) selected probability of default P for customers with D ranking in The Empirical Relationship between Average Asset Correlation, Firm Probability of Default In case of P >= 18.25, the company does not default. In case of P <18.25, the company defaults. Step 3: Results and pratical observations of corporate defaults The study created a table to compare results collected from Z-Score model (1968); Z-Score model (1993) and KMV model with pratical observations Step 4: Using MCC and ROC methods to test Z-score and KMV models ● Through the usage of Z-score and KMV models, the research compared and classified status quo of Vietnamese firms: 14 - Type I error (Rejection of its correctness and assumption of it being incorrect) and type II error (approval of of its correctness while it is incorrect) Regarding a forecast with positive and negative possibilities, results are classified into these four following groups: TP (True positive) and TN (true negative) state correct forecast while FP (false positive) and FN (false negative) state incorrect forecast.FN is equivalent with type I error and FP is equivalent with type II error. Table 3.1. Confusion matrix Content Default Non default Rejection Correct prediction (TN) Type I error (FN) Non-rejection Type II error (FP) Correct prediction (TP) * Using MCC (Matheews correlation coeficient) to test the accuracy of two models: The formula is as follows: MCC =  ∗*+ ∗+, - .+  .+**.+ *.+* MCC is in the range from -1 to 1; the closer it is to 1, the better the model is. Using absolute FN and FP, the research calculate error rate: FNR (false negative rate) and FPR (false positive rate). Additionally, the study considered sensitivity and probability of type II error to test the accuracy of models. False negative rate (FNR) demonstrates the degree of incorrect prediction for an event when it is positive but the conclusion is negative: FPR=FN/(FN+TP). False negative rate (FNR) illustrates the degree of incorrect prediction for an event when it is negative but the conclusion is positive: A good MCC model is essential to have low FNR and FPR. In credit ranking, the false classification of capital loaning application from good to bad will lead less serious consquences than the other case. ● Using ROC (Receiver Operating Characteristic) to test the accuracy of two models. 15 Figure 3.1. ROC Sobehart and Keenan (2001) offered the usage of ROC to confirm internal credit ranking for companies and area under ROC (AU ROC). Figure 3.2. Area under ROC ● Classification of AU ROC Figure 3.2: Classification of AU ROC AU ROC Significane (accuracy of ROC) AU > 0.9 Very good 0.8 <AU≤0.9 Good 0.7 <AU≤0.8 Average 16 0.6 <AU≤0.7 Not good 0.5<AU≤0.6 Not be able to discriminate AU ROC can be under 0.5 but at that time, it is necessary to reconsider criteria with large deviation that lead to probability of correct negative customers being smaller that probability of incorrect positive customers (Y<X). 3.2. Research data and in-depth interviews 3.2.1. Research data * Data from 2014 to 2016 on Ho Chi Minh Stock Exchange has 935 company entries 3.2.2. In-depth interviews - Implementation of in-depth interviews is to clarify research orientation, test the practical application and propose appropriate recommendations/ - Interviewees: leaders of credit institutions, securities companies and managers (State Bank, Securities Commission) - Contents of interview: collection of information, opnions and evaluation on (i) The usage of quantification models in default forecast in Vietnam and interviewees’ units. (ii) The influence degree of factors such as capital structure, business efficiency, scope of assets, state managing mechanism on corporate defaults. (iii) Solutions and recommendations - In-depth interview contents were prepared and designed according to questions’ categories. Chapter 3 demonstrated the research design according to four steps in using models via accounting-based and market-based approaches to predict corporate defaults and compare with status quo of companies. The chapter also suggested the usage of MCC and ROC to compare the accuracy when applying forecast models. CHAPTER 4 APPLICATION OF ACCOUNTING-BASED AND MARKET-BASED APPROACHES IN CORPORATE DEFAULT FORECAST 4.1. Business performance of companies listed on HOSE from 2014 to 2016 Until the end of 2017, there were 344 enterprises active on HOSE. In 2017, there were 31 newly listed companies with considerable first listed value and fastly-growing transaction volume through many years. 17 Table 4.1.Market transaction throughout years (2014, 2015, 2016, 2017) Year Number of trading sessions Transaction volume (billion VND) Market capitalization (billion VND) 2014 247 536.484 985.258 2015 248 487.407 1.146.925 2016 251 748.608 1.491.778 2017 250 1.061.183 2.614.150 Source: HOSE annual reports It can be said that along with increasing number of companies listed on HOSE in terms of scale and quantity, many enterprises also cancel their listing due to losses from business operation. 4.2. Application of default forecast models with companies listed on HOSE 4.2.1. Application of models and results In terms of data collected during three years from 2014 to 2016, there were 303, 305 and 327 firms being surveyed in 2014, 2015 and 2016 respectively. With 935 entries of companies, 100% selected companies for sample are non-financial enterprises listed on HOSE during three years from 2014 and 2016. These firms are monitored in terms of default until 2017 to compare with empirical results from selected models. * The application of accounting-based approach via Z- Score model (1968) and Z-Score model (1993) is conducted to calculate the value of Z index for each company’s entry and compare with the reality. Results shown at Appendix 2: Results according to index of Z-score, KMV and the reality The research aggregrated results from Z- Score model (1968) and Z-Score model (1993) to calculate Confusion matrix. Table 4.2: Results aggregrated from Z- Score model (1968) and Z-Score model (1993) Entries of surveyed companies Results of Z – Score model (1968) Results of Z – Score model (1993) Results from reality Default Non- default Default Non-default Default Non- default 935 488 447 135 800 69 866 18 Source: Author’s calculation Table 4.3: Confusion matrix according to Z-Score model (1968) Content Default Non-default Total Rejection TN: 68 Type I error (FN): 420 488 Non-rejection Type II error (FP): 1 TP: 446 535 Total 69 866 935 Source: Author’s calculation Table 4.4: Confusion matrix according to Z-Score model (1993) Content Default Non-default Total Rejection TN:51 Type I error (FN): 84 135 Non-rejection Type II error (FP): 18 TP: 782 800 Total 69 866 935 Source: Author’s calculation • Evaluation of results achieved from the application of Z-score model: + Z-score model (1968): The number of type I errors was high which predicted companies having the possibility of default but in reality, they did not. This means if credit institutions and investors apply this model, many investment opportunities will be missed. However, the number of type II error was considerably low which means that if the model is used to eliminate risks, it can effectively operate. + Z-score model (1993): corporate default forecast rates were improved compare with Z-score model (1968), especially the number of type I errors notably decreased. * The application of market-based approach via KMV model is conducted and results are used to compare with the reality. Results shown at Appendix 1: Results according to index of Z-score, KMV and the reality Table 4.5: Results aggregrated from KMV model Entries of surveyed companies Results of KMV model Results from reality Default Non-default Default Non-default 935 91 844 69 866 Table 4.6: Confusion matrix according to KMV model Content Default Default Total Rejection TN: 60 Type I error (FN): 31 91 Non-rejection Type II error (FP): 9 TP: 835 844 Total 69 866 935 19 * Evaluation of results achieved from the application of KMV model: The model predicted there were 835 non-default companies which is nine ones lower than the reality – this is considered as relatively close with the reality. 4.2.2. Evaluation through testing results * Correlation coefficient: Table 5.7: Calculation of correlation coefficient Content Z-Score (1968) Z-Score (1993) KMV MCC 0,261 0.477 0,735 FNR ( type I error rate) 0,485 0.097 0,036 FPR (type II error rate) 0,015 0.26 0,13 Source: Author’s calculation + Results: - MCC of Z-score (1968) was 0,261(lowest) while MCC of Z- Score (1993) was higher being 0,477. This proved that the accuracy of Z- Score model (1993) was higher than Z- Score model (1968) when being applied in Vietnamese firms. MCC of KMV index was 0,735 (highest) which means KMV model having the highest accuracy compared with the two Z-score models. - Type I error rate (FNR): Z-score index (1968) had type I error rate being 0,485 which was relatively high. This showed that the model predicted defaults in many companies but in reality, it did not happen. Notably, type II error rate was 0 which means the model correctly predicted those companies having defaults. If Z-score model (1968) is applied, many good enterprises for investment will be eliminated and not selected. This also demonstrates the great impact of financial index including leverage coefficient on companies in developing countries like Vietnam is more significant than those in developed nations. There were some enterprises with Z-score index (1968) being relatively lower than standard still did not default as data transparency and accounting methods could not illustrate the complete value of these firms. - Type II error rate (FPR): in KMV model, FNR was 0,015 which is l