Impacts of Digital Transformation and Basel III Implementation on the Credit Risk Level of Vietnamese Commercial Banks

Abstract

For a bank-based economy like Vietnam, the commercial banking sector’s conduct greatly influences Vietnamese economic and social prosperity. In Vietnam, net income from credit activities hold the largest portion of the total revenue of Vietnamese commercial banks. Therefore, in the context of Vietnam, credit risk obviously also plays a pivotal important role in the banking sector. Hence, the risk of credit failure can lead to a bank’s collapse and have a profound effect on a country’s societal structure. As seen in the previous literature, there are many macroeconomic and bank-level factors that have commonly affected the level of credit risk; however, these factors may change in the recent development era of the banking industry, especially the new impacts of digital transformation and the transition to full Basel III adoption. The overall aim of this study is to analyze the impacts of digital transformation and Basel III implementation on the credit risk level of Vietnamese commercial banks during the period from 2017 to 2023, with a sample of 21 Vietnamese listed commercial banks. This study employs the pooled OLS, fixed effect model (FEM), and random effect model (REM) methods to reach the finding that investing in technology for the readiness of digital transformation and implementing Basel III could adversely affect credit risk. Based on this finding, the authors give some recommendations for commercial banks to enhance the sustainability, safety, and better management of credit risk.

1. Introduction

For a bank-based economy like Vietnam, the commercial banking sector’s conduct greatly influences Vietnamese economic and social prosperity. In most cases, banks provide credit to sustain firms in the manufacturing, agricultural, commercial, and service sectors, providing jobs and enhancing purchasing power, consumption, and savings. Hence, the risk of credit failure can lead to a bank’s collapse and have a profound effect on a country’s societal structure, and can rapidly resonate globally (Caruso et al. 2020). A recent study, with a sample consisting of 28 out of the 35 Vietnamese commercial banks in 2009–2018, conducted by Pham and Daly (2020), showed that credit risk is the main cause of bad debts in the Vietnamese banking system and coexists with banking operations. The first capital requirement framework of the Basel Committee, which was also called “the BASEL I Accord” in 1988, identified that “credit risk is the major risk of banks among others”, and determined “to set a framework mainly for assessing capital in relation to credit risk”. In Vietnam, net income from credit activities held the largest portion of the total revenue of Vietnamese commercial banks (State Bank of Vietnam (SBV) 2018). Therefore, in the context of Vietnam, obviously credit risk also plays a pivotal and important role in the banking sector. As seen in the previous literature, there are many macroeconomic and bank-level factors that have been commonly affected the level of credit risk; therefore, this should be considered part of a much more complex relationship beside these factors.

An obvious fact is that the credit risk level of a commercial bank is not only affected by external factors, but also from those inside banks. For the perspective of a bank, the need for analyzing the way that intrinsic factors interact with credit risk is necessary so that they can alter their strategies and supervise non-performing loans in order to maintain their operations and efficiency. In the context of the recent development era of the banking industry, a novel trend in digital transformation and technology adaptation can now be seen in most commercial banks all over the world. This trend might bring a positive future and lead to a revolution that would transform the age-old traditional basis of banking services to a new form that we have never seen before, or they could be a not-yet fully supervised, unsecured door for various threats and risks to enter and danger the pivotal pillars of banking—one of which is the credit risk level of a commercial bank. The impact of digital transformation on credit risk management is also affected by the characteristics of banks. During the digital transition era, many major financial institutions, including state-owned commercial banks, typically have the ability to make substantial and comprehensive investments, leveraging their strengths in capital, technology, and skilled personnel (Jin et al. 2020). On the other hand, smaller and medium-sized banks benefit from their ability to make quicker decisions and adapt more swiftly, giving themselves a competitive edge. Therefore, to which side the use of technology will bring more positive outcomes is a question worth discussing.

Furthermore, in the context of economic integration with international standard requirements in credit risk management, implementing Basel III in Vietnamese commercial banks has become more urgent than ever. Basel III, with its stringent principles on credit risk management, has been adopted by several Vietnamese commercial banks (see Appendix A,

Table A1. List of Vietnamese Commercial Banks in the Sample.

No.	Bank’s Name	BASEL Implementation Status	State-Owned Status
1	BIDV	No detailed information	Yes
2	Vietcombank	Basel III trials will start in Q1/2025	Yes
3	VietinBank	Preliminary assessment is being conducted	Yes
4	Agribank	No detailed information	Yes
5	VPBank	Basel III—partly implemented	No
6	Sacombank	Basel III—completed: 2023	No
7	Techcombank	Basel III—partly implemented	No
8	ACB	Basel III—completed: 2022	No
9	MB	Basel II, not yet reforming to Basel III	No
10	VIB	Basel III—partly implemented	No
11	TPBank	Basel III—completed: 2021	No
12	OCB	Basel III—completed: 2022	No
13	SeABank	Basel III—fully implemented, not yet completed	No
14	BacABank	No detailed information	No
15	HDBank	Basel III—completed: 2023	No
16	SHB	Basel III—partly implemented	No
17	BVBank	No detailed information	No
18	MSB	Basel III—fully implemented, not yet completed	No
19	PGBank	No detailed information	No
20	LPBank	Basel III—completed: 2022	No
21	NamABank	Basel III—completed: 2022	No

Source: the authors.

) but not all. For Vietnamese commercial banks, complying with Basel III is also an opportunity to improve asset quality and manage credit risk more effectively.

In order to theoretically contribute to the literature on factors affecting the credit risk level of banks and practically contribute to suggestions for banks in the sounder credit risk management in the digital era and international standards approach context, this study aims to analyze the influences of digital transformation and Basel III implementation on the credit risk level of Vietnamese commercial banks and then propose recommendations with which to improve the current situation of their non-performing loan levels. To achieve these targets, two proposed research questions need to be addressed: (i) How has digital transformation affected the credit risk level of Vietnamese commercial banks? (ii) How has Basel III implementation impacted the credit risk level of Vietnamese commercial banks? In order to clarify the answers, this paper uses an empirical data analysis of 21 Vietnamese listed commercial banks (see Appendix A, Table A1) during the period from 2017 to 2023 with static models, including pooled OLS, a fixed effect model (FEM), a random effect model (REM), and FGLS regression, and reaches the following important findings: (i) investing in technology for the readiness of digital transformation adversely affects the credit risk level; (ii) the continuous implementation of Basel III has helped to reduce the credit risk levels of banks; and (iii) some other influences are investigated, i.e., the growth of the economy and inflation have a positive correlation with the credit risk level, while profitability and customer deposit growth can help banks to decrease credit risk levels. The remainder of this paper, excluding the References and Appendices, is structured as follows: Literature Review, Data and Methodology, Regression Results, Discussions and Implications, and Conclusions.

2. Literature Review

2.1. Indicator Addressing the Credit Risk Level

There are different indicators in the existing literature that are related to the credit risk management of a commercial bank. However, in terms of reflecting the loan quality of banks, the non-performing loan (NPL) ratio is one of the most monitored critical indicators (Adekunle et al. 2015; Bayar 2019; Jianfu et al. 2023). In the same line of thought, Goyal et al. (2023) concluded that, in different countries and regions, the quality of the loan portfolio maintained by the banking sector is reflected by the level of non-performing loans (NPLs). Additionally, NPL minimization is especially a priority for regulators and policymakers (Bayar 2019).

In Vietnam, bank loans are classified into five groups, numbered “Group 1” to “Group 5”, following Circular 11/2021/TT-NHNN. Additionally, according to this regulatory framework, a non-performing loan (NPL) refers to a bad debt on a balance sheet, which is classified into Group 3, 4, and 5, as described above. The NPL ratio is determined by the ratio of the amount of NPLs to the total amount of outstanding loans (all five groups in total). In this study, the authors choose the NPL ratio for measuring the credit risk level of Vietnamese commercial banks.

2.2. Impacts of Digital Transformation on a Bank’s Credit Risk Level

The need for digital transformation has been recognized in the Vietnamese banking sector for a long time. However, this term has been used as a common catchword but not in the correct way, mostly due to confusion surrounding related terms. This perplexity could be attributable to the characteristics of digital transformation, as it is a much more complex strategy and action. According to Mark (2020), the process of digital transformation can be explained in a pyramid figure built from three foundational parts: digitization, digitalization, and digital transformation.

An example of academic research in China, conducted by Yang and Masron (2023), showed that digital transformation might help in reducing the ratio non-performing loans, though its effectiveness remains uncertain. Furthermore, as the era of technology has just started its beginning and blooming phase, it might be too early for the role of novel technology to become apparent in the banking sector. The results for researching these relationships have been mixed and inconclusive, with some claiming digital transformation could positively reduce the NPL ratio while others claim the contrary. The impact of digital transformation on risk in banks currently needs a solid consensus.

In Vietnam, the digitizing phase began in the late 1990s, with several notable systems that most Vietnamese joint-stock commercial banks favored at that time. The impact of these changes was considered mostly in terms of computerization and digitization, not digitalization, and the impact of these internal transformations on the whole banking sector was not obvious. It took several years for these novel platforms to have a significant impact on the banking sector as a whole. According to Ha and Nguyen (2022), only in the three years before their study took place had digital transformation in the Vietnamese banking system taken place “as a strong wave”. In that time period, most Vietnamese banks “have either implemented or are in the process of developing their digital transformation strategies” (Ha and Nguyen 2022). There is also no research on the effect of digital transformation on banks’ credit risk, just limited studies on banks’ performance in general. For example, according to Vinh et al. (2023), when studying the impact of ICT on the profitability of 25 Vietnamese commercial banks in the period 2010–2020, the authors found a possibility that the impact of digital transformation on commercial banks’ profits could be non-linear, or a “U-shaped” relationship, to be exact. They considered this result to be caused by Vietnamese commercial banks developing slowly and currently implementing ICT in a limited and incomplete manner (Vinh et al. 2023).

H1. 

Digital transformation had a negative effect on the credit risk level of banks.

2.3. Impacts of Basel III Implementation on Banks’ Credit Risk Level

Since the first version was introduced in 1988, the Basel Accords have been considered the main standard and framework for commercial banks around the world. The implementation of Basel III has significant implications for banks’ credit risk levels, primarily through enhanced capital requirements, liquidity standards, and risk management practices. Basel III was introduced in response to the financial crises that exposed the vulnerabilities of the banking sector, aiming to strengthen the resilience of banks and reduce systemic risk (Fidrmuc and Lind 2020). One of the core components of Basel III is the increase in capital requirements, which mandate banks to hold a higher level of high-quality capital. This change is intended to enhance banks’ ability to absorb losses, thereby reducing the credit risk associated with their lending activities (Le et al. 2023). The impact of Basel III on credit risk is also reflected in the operational adjustments that banks must undertake to comply with the new regulations, as banks are encouraged to adopt more proactive portfolio management and forward-looking risk management strategies, which can lead to improved credit risk assessment and management (Boora and Jangra 2019).

In Vietnam, the high inflation rate after the global financial crisis in the 2010s led to a high NPL ratio in the banking sector, and the impact was considered devastating. The SBV had to initiate the process of adopting the Basel Accords in 2005 to reform banking practices during economic turmoil, by adopting NPL and CAR calculation methods following the Basel I agreement. Only in 2018 did the SBV start updating the regulating documents to be in line with the Basel II Accord for better calculations and transparent data of the whole banking system. From the research by Phi et al. (2019), at the same time as the announcement of Circular 41/2016/TT-NHNN, some banks in Vietnam had already started implementing the Basel III standards. However, until now there has not been much research on the impact of Basel III adoption as a factor in the credit risk level of Vietnamese banks. The author hoped that this study would contribute a small step for filling in the research gaps about the impact of Basel III application on the credit risk level in the Vietnamese banking sector.

After the implementation of the Basel framework in regulations, as a result, the CAR in the Vietnamese banking sector on average has been improving recently as banks focus on obtaining more capital buffers, and the NPL ratio showed signs of a downfall to a lower and safer threshold (Phi et al. 2019). In this study, the authors expected a reverse relationship between the Basel III application and the NPL ratio of Vietnamese commercial banks.

H2. 

Basel III implementation had a negative effect on the credit risk level of banks.

Based on these hypotheses, the process of the research framework is designed as below:

Step 1: Classical Linear Regression Model (CLRM)

Pre-eliminating unwanted variables and fixing phenomena often occur in the pooled OLS model. The three crucial errors and phenomena are heteroskedastic errors, first-order autocorrelation errors, and multicollinearity errors. For each of these errors, an author would apply the White, Wooldridge and VIF test in that order to check whether the suggested model bears these errors.

Step 2: Regression Analysis

After checking for errors, an author would carried out a regression analysis with three main methods of estimation, which are pooled OLS, an FEM, and an REM. Select the most acceptable model, then the authors recheck the error with Breusch and Pagan—Lagrangian multiplier test for Heteroskedastic error and Wooldridge test for First-order autocorrelation error. If errors exist, an author would edit the model to eliminate and remove an unwanted variable out of the model.

Step 3: Model Correction

An author would use the feasible generalized least squares (FGLS) regression method to overcome the phenomena of heteroskedastic and first-order autocorrelation errors.

3. Data and Methodology

3.1. Methodology

For the regression analysis, the static models that were incorporated in this study include pooled OLS, the fixed effect model (FEM), and the Random Effect Model (REM). Moreover, to overcome errors, the authors also chose to apply the FGLS regression model for correction. In this study, the authors gathered and calculated the data with Microsoft Excel, Office 16. The technical tool and application used for the regression model is StataMP 17 software.

Besides two explanatory variables, including digital transformation (ICT) and Basel III implementation (BASEL3), this study examines the effects of control variables, including both macroeconomic and internal bank factors, on the credit risk level of a commercial bank, which are GDP growth (GDPG), inflation (INF), a bank’s size (LN_SIZE), a bank’s profitability (ROAA), a bank’s liquidity (P_LDR), a bank’s capital adequacy ratio (CAR), and a bank’s ownership structure (STATE). The authors chose to employ panel data from a sample of 21 Vietnamese commercial banks from 2017 to 2023. All the variables in this model would be calculated in the natural logarithm form, as these transformations ensure that the variables span over the interval [−∞, +∞] and are distributed symmetrically, as was the case in past research by Ghosh (2015) and Klein (2013).

Applying the variables based on the conceptual framework above, the author can describe the specific model for this paper as follows:

$$\begin{gathered} NP{L}_{i,t}=\alpha +{\beta }_{1}IC{T}_{i,t}+{\beta }_{2}BASEL{3}_{i,t}+{\beta }_{3}LN\_SIZ{E}_{i,t}+{\beta }_{4}ROA{A}_{i,t} \\ +{\beta }_{5}P\_LD{R}_{i,t}+{\beta }_{6}CA{R}_{i,t}+{\beta }_{7}GDP{G}_{i,t}+{\beta }_{8}IN{F}_{i,t}+{\beta }_{9}STAT{E}_{i,t}+{\epsilon }_{i,t} \end{gathered}$$

3.2. Data Collection

3.2.1. Data Sources

The authors extracted data for the period from 2017 to 2023 from various official documents, reports, and publications from the selected Vietnamese commercial banks (see Appendix A, Table A1) and regulators, including the following:

• Annual consolidated financial statements;
• Annual reports;
• Capital adequacy ratio (CAR) disclosures, in accordance with Circular 41/2016/TT-NHNN by the State Bank;
• Vietnam ICT index report—banking sector, publicized by the Vietnam Ministry of Information and Communications (MIC);
• Other macroeconomic data are provided by DataBank at https://databank.worldbank.org (accessed on 24 July 2024)

3.2.2. Data Measurements

• Credit Risk

For credit risk level, the authors choose Non-performing loan (NPL) as an indicator. Currently in Vietnam, the methodology for calculating the NPL ratio is based on Circular 11/2021/TT-NHNN, issued by the State Bank, and is described as follows:

$$NPL={\displaystyle \frac{Non-PerformingLoansinvalue}{TotalGrossLoans}}\times 100\%$$

The non-performing loans consist of all loans overdue on principal and interest payments.

b.

Digital Transformation

In Vietnam, the level of technology application in businesses and corporations is typically evaluated using the information, communications, and technology application readiness index (ICT index) carried out by the Vietnam Ministry of Information and Communications (MIC). The MIC scores corporations and organizations on various determined categories, and sum those scores in total for the final ranking. This index is regularly featured in the yearly Vietnam ICT index reports.

c.

Basel III Implementation

Based on the announcements of sample banks on their public websites and in their reports, the authors define the status of whether Basel III was implemented or not in banks. This variable (BASEL3) is a dummy, with a value equal to 1 if a bank announced that they had finished the Basel III routine.

d.

GDP Growth

A typical method for measuring global economic growth involves choosing a weighting scheme for different regions of the world. The global real output growth rate is often calculated as a weighted average of real GDP growth across major world regions, with the weights indicating each region’s relative significance in the global economy (i.e., its proportion of the total world GDP value). To calculate this proportion, national data—initially in the national currency—are converted into a common currency using an exchange rate measure (ECB Monthly Bulletin 2006). From then, the formula to calculate GDP growth is described as follows:

$$GDPG=({\displaystyle \frac{RealGD{P}_{recentyear}}{RealGD{P}_{previousyear}}}-1)$$

Growth rates of GDP and its components are calculated using the least squares method and constant price data in the local currency. Constant prices in U.S. dollar series are used to calculate regional and income group growth rates. Local currency series are converted into constant U.S. dollars using an exchange rate in the common reference year. The data of Vietnam’s GDP growth rate is collected from DataBank by the World Bank at: https://databank.worldbank.org (accessed on 24 July 2024).

e.

Inflation

Inflation, as measured by the consumer price index, reflects the annual percentage change in the cost to the average consumer of acquiring a basket of goods and services (generally generated by the Laspeyres formula) that may be fixed or changed at specified intervals, such as yearly. The formula for calculating inflation applied in this paper is as follows:

$$INF={\displaystyle \frac{CP{I}_i-CP{I}_{i-1}}{CP{I}_{i-1}}}$$

f.

Banks’ Liquidity

As in the instructions of Circular 26/2022/TT-NHNN, Vietnamese commercial banks were required to start taking into account the balance of deposits of the State Treasury when calculating their LDR, with a roadmap consisting of one-year steps with certain percentages for each period until reaching 100% from 1 January 2026. This requirement by the SBV brings complexity to the calculation of the LDR (as a version of the stipulated LDR); within the framework of this study to ensure consistency in the methodology and also in accordance with international practices, the authors chose to apply the basic formula of the loan-to-deposits ratio (pure LDR) in this study with the formula given below:

$$P\text{\_}LDR={\displaystyle \frac{TotalLoans}{TotalDeposits}}$$

In this study, the authors calculated the total deposits from the “Deposits from Customers” account only, laying in the liabilities of banks’ balance sheets, in order to present the capital input in a more correct way without calculating the deposits of other financial institutions.

g.

Banks’ Capital Adequacy Ratio

According to Circular 41/2016/TT-NHNN by the State Bank of Vietnam (SBV), all commercial banks and financial institutions must report on their capital adequacy ratio (CAR) on a semiannual basis in accordance with regulations of the SBV on statistical reporting systems that are also applied. The CARs of Vietnamese banks from 2017 onwards are calculated based on Circular 41/2016/TT-NHNN. The general CAR formula is described as follows:

$$CAR={\displaystyle \frac{C}{RWA+12.5\ast \left(K_{OR}+K_{MR}\right)}}$$

h.

Banks’ Profitability

The authors chose to use ROAA as the profitability indicator for Vietnamese commercial banks. The methodology and formula used for the calculation are as below:

$$ROAA={\displaystyle \frac{NetIncome}{AverageTotalAssets}}$$

All components for calculating these ratios are collected from the consolidated financial reports published annually by selected Vietnamese commercial banks.

i.

Banks’ Size

To indicate the size of a commercial bank, the volume of the total assets is often used both academically and in practice. However, to reduce the differences between banks, a natural logarithm expression for total assets is used:

$$LN\text{\_}Size=ln\left(TotalAssets\right)$$

j.

Banks’ Ownership Structure

Based on the ownership structure, a bank is defined as a state-owned bank in case the state has the stake in the bank’s equity structure. Therefore, this variable (STATE) is a dummy and equals 1 if a bank is state-owned.

3.2.3. Variable Description

From

Table 1. Variable descriptive summary.

Variable	Obs.	Mean	Std. Dev.	Min	Max
NPL	144	0.0170394	0.0090927	0	0.0573465
GDPG	147	0.057516	0.0211421	0.0256156	0.0801979
INF	147	0.0304541	0.0054734	0.0183472	0.0353963
LN_SIZE	147	19.56331	1.058639	17.19303	21.55653
ROAA	147	0.0131434	0.0079122	0.0006777	0.0365264
P_LDR	147	0.9888231	0.1386309	0.6368026	1.469094
CAR	142	0.1140529	0.0220241	0	0.1948
ICT	100	0.507759	0.1407528	0	0.7762
BASEL3 (dummy)	147	0.0884354	0.2848977	0	1
STATE (dummy)	147	0.1904762	0.3940192	0	1

Source: the authors.

above, it can be seen that the 21 commercial banks selected in the sample had an average NPL ratio of around 1.7%, with a small dispersion of only 0.92% from the mean of the NPL ratio. The highest NPL ratio of all of the data collected was the NPL ratio of VPBank in 2022, which was 5.73%, higher than the 3% threshold that the SBV often chose as the criterion specified in Circulars and other related documents (such as Circular 18/2022/TT-NHNN, Circular 22/2019/TT-NHNN, etc.). However, from the overview that the banking sector in general is still staying in the safe zone, several Vietnamese banks are on alert for credit risk management.

4. Regression Results

4.1. Pooled OLS Regression

4.1.1. Correlation Coefficients of Variables

In the first step, the authors checked if there are any correlations between the coefficients of variables. The results are in the table that follows.

With 95% confidence, the results in

Table 2. Correlation coefficients among variables.

	NPL	GDPG	INF	LN_SIZE	ROAA	P_LDR	CAR	ICT	BASEL3	STATE
NPL	1.0000
GDPG	0.1135	1.0000
INF	0.0262	0.0844	1.0000
LN_SIZE	−0.2620	−0.0492	−0.1056	1.0000
ROAA	−0.2103	0.0154	−0.1027	0.2413	1.0000
P_LDR	0.2411	0.0776	−0.0686	−0.0104	0.4135	1.0000
CAR	0.2329	0.1709	0.2418	−0.3334	0.2784	0.1919	1.0000
ICT	−0.2331	−0.1499	−0.0583	0.2461	0.3333	0.1583	−0.0979	1.0000
BASEL3	−0.0998	0.1496	−0.0237	−0.0104	0.1654	0.0224	0.0498	0.2186	1.0000
STATE	−0.1907	−0.0476	−0.0715	0.7442	−0.1929	−0.1486	−0.4592	0.0414	−0.1083	1.0000

Source: the authors.

show that only the GDPG, INF, P_LDR, and CAR variables show a positive correlation with the NPL ratio, while the remaining independent variables have a negative correlation with the NPL variable in the model.

4.1.2. Heteroskedasticity and First-Order Autocorrelation Tests for Pooled OLS

To see if the regression model that is suggested in this paper is applicable and suitable for pooled OLS regression or not, the authors carried out heteroskedasticity and first-order autocorrelation tests to check for any heteroscedasticity and first-order autocorrelation problems.

The test results in

Table 3. Heteroskedasticity and first-order autocorrelation tests for pooled OLS.

Problem	Test	Result
Heteroskedasticity Pooled OLS	White	H0: Homoscedasticity (no heteroskedasticity) ${\chi }_{\left(47\right)}^2=55.57$ $Prob>{\chi }^2=0.1833$
First-order autocorrelation Pooled OLS	Wooldridge	H0: No first-order autocorrelation $F_{\left(16\right)}^1=10.826$ $Prob>F=0.0046$

Source: the authors.

show that the p-values from the White tests are much higher than 5% (as Prob > χ² = 0.1833), while the p-values from Wooldridge test are lower than the threshold of 0.05; therefore, H₀: homoscedasticity is accepted, and on the contrary, H_0: no first-order autocorrelation is rejected, meaning that pooled OLS regression has the phenomenon of first-order autocorrelation (or AR(1) problem) only.

4.1.3. FEM/REM Regression

• Multicollinearity Test (VIF)

For a classical linear regression model (CLRM), there are seven assumptions that are required to be applied so that the model can be statistically acceptable (Poole and O’Farrell 1971). One of these classical assumptions is that none of the explanatory variables in the regression model is a perfect linear function of any other explanatory variable, indicating no perfect multicollinearity. If this assumption is violated, it could be said that the multicollinearity problem is present. Therefore, to see if any variables have a correlation with the other independent variables, the authors used the multicollinearity test via the variance inflation factor (VIF) and its tolerance (equals 1/VIF).

The test results in

Table 4. Multicollinearity test results (VIF).

	VIF	1/VIF (Tolerance)
LN_SIZE	3.60	0.277440
STATE	3.47	0.288275
ROAA	2.10	0.476085
CAR	1.54	0.650125
ICT	1.29	0.776153
P_LDR	1.25	0.797080
BASEL3	1.12	0.894895
INF	1.11	0.902273
GDPG	1.10	0.909499
Mean VIF	1.84

Source: the authors.

show that the VIF coefficients of the remaining independent variables all have values within the threshold of less than 10.0. However, the VIFs of the LN_SIZE, STATE, and ROAA variables are higher than 2.0, meaning that these independent variables show multicollinearity phenomena. The authors chose to remove the variable that has the highest VIF, which is LN_SIZE, out of the regression model. To recheck whether the problem is solved or not, authors should run the multicollinearity test with the VIF again. The test results also show that the VIF coefficients of the remaining independent variables all have values within the threshold of less than 2.0. Thus, after removing LN_SIZE, these independent variables show that there is no multicollinearity phenomenon, and this means that after the removal of the unwanted variable the dataset is now acceptable for use.

b.

Hausman’s Test

In the next step, the authors would use the estimation method for both the fixed effect and random effect models to see which model would be acceptable for further analysis. The test used for model selection is Hausman’s test, with Chi-squared statistics, and the null hypothesis is a test of H₀: difference in coefficients not systematic. The results are presented in the table below:

The test results in

Table 5. Hausman’s test result.

Dependent Variable: NPL	Pooled OLS	FEM	REM
	Coefficient. (Std. Err.)	Coefficient. (Std. Err.)	Coefficient. (Std. Err.)
GDPG	0.0545405	0.1852854 **	0.1367668
	(0.1204808)	(0.0840394)	(0.0835321)
INF	−0.3118003	0.0170561	−0.0942976
	(0.4675254)	(0.3284748)	(0.3276537)
ROAA	−0.29679236 ***	−0.0648861	−0.1640183 *
	(0.0818104)	(0.1116606)	(0.0935703)
P_LDR	1.317977 ***	−0.8992597 *	−0.1091571
	(0.3727575)	(0.4851552)	(0.4132176)
CAR	0.6747372 *	0.2947571	0.405158
	(0.3420783)	(0.3180541)	(0.3008918)
ICT	−0.2286449	0.0290765	−0.1267114
	(0.2031458)	(0.2295854)	(0.2060373)
BASEL III	−0.1188771	−0.1707417	−0.1382092
	(0.2426882)	(0.1724097)	(0.1736408)
STATE	−0.1434664	0	−0.2275523
	(0.1287184)	(omitted)	(0.2264176)
_cons (Constant)	−5.0410569 ***	−3.23838 **	−3.996282 ***
	(1.643108)	(1.381167)	(1.281648)
N	95	95	95
R-squared	0.2806 0.2137 (adjusted)	0.1913	0.1552
Hausman Test
Chi-squared statistics	${\chi }_{\left(7\right)}^2=11.75$	Test of H0: difference in coefficients not systematic
Prob.	$Prob>{\chi }^2=0.1092$

Source: the authors. Note: *, **, and *** indicate significance levels of 10%, 5%, and 1%, respectively. (* p < 0.1; ** p < 0.05; *** p < 0.01). Errors are presented in parentheses ().

show that the p-value (Hausman) = 0.1092 > 0.05; therefore, the REM is considered better and will be selected. However, nearly all of the variables in the REM have statistical insignificance, with an R-squared value of only 0.1552, which means that the independent variables can only explain about only 15.52% of the dependent. As this indicates a low level of statistical meaning, the authors would check for errors of heteroskedasticity with the Breusch and Pagan—Lagrangia multiplier test and for first-order autocorrelation phenomena with Wooldridge’s test to see whether these errors exist in the model.

c.

Heteroskedasticity and First-Order Autocorrelation Tests for the REM

The authors would test if there were heteroskedastic problems in the REM.

The Prob > chibar2 = 0.0000 in

Table 6. Breusch and Pagan—Lagrangian multiplier test for random effects.

l_NPL[bank,t] = Xb + u[bank] + e[bank,t]
Estimated Result	Var.	SD = sqrt(Var)
l_NPL	0.2569186	0.5068714
e	0.0771423	0.277745
u	0.1288476	0.3589535
Test: Var(u) = 0	chibar2(01) = 45.24 Prob > chibar2 = 0.0000

Source: the authors.

indicates that heteroskedasticity exists in the REM.

The Wooldridge test results in

Table 7. Wooldridge test for autocorrelation in the REM.

Problem	Test	Result
First-order autocorrelation REM	Wooldridge	H0: No first-order autocorrelation $F_{\left(16\right)}^1=10.63$ $Prob>F=0.0049$

Source: the authors.

show that the REM also has the AR(1) problem. To overcome these phenomena of both heteroskedasticity and AR(1), the authors would conduct corrections by using the feasible generalized least squares (FGLS) method, based on the instructions by Greene (2012) and Baum (2001).

4.2. Feasible Generalized Least Squares (FGLS) Regression

With the aim of fixing both the heteroskedastic and AR(1) problems, FGLS regression would be used for the after-elimination model of REM. The results are summarized in

Table 8. FGLS regression results.

Coefficients	Generalized Least Squares			Number of Obs		=93
Panels	Heteroskedastic			Number of groups		=20
Correlation	Panel-specific AR(1)			Obs per group	Min	=3
					Avg	=4.65
					Max	=5
Estimated covariances		=20		Wald chi2(8)	=207.83
Estimated autocorrelations		=20		Prob > chi2	=0.0000
Estimated coefficients		=9
Dependent Variable: NPL	FGLS		Conclusion
	Coefficient. (Std. err.)		Statistical meaning		Expectation	Result
GDPG	0.163369 ***		Statistically significant		(-)	(+)
	(0.0251681)					Not as expected
INF	0.119432		Statistically insignificant		(+)	(+)
	(0.1260667)					As expected
ROAA	−0.2767464 ***		Statistically significant		(-)	(-)
	(0.0305498)					As expected
P_LDR	0.3168414 *		Statistically significant		(+)	(+)
	(0.1654745)					As expected
CAR	0.4655638 ***		Statistically significant		(-)	(+)
	(0.1124666)					Not as expected
ICT	−0.1172925 **		Statistically significant		(-)	(-)
	(0.0585906)					As expected
BASEL3	−0.2291679 ***		Statistically significant		(-)	(-)
	(0.0838582)					As expected
STATE	−0.2191768 ***		Statistically significant		(-)	(-)
	(0.0745192)
_cons (Constant)	−3.596422 ***
	(0.4108595)

Source: the authors. Note: *, **, and *** indicate significance levels of 10%, 5%, and 1%, respectively. (* p < 0.1; ** p < 0.05; and *** p < 0.01). Errors are presented in parentheses ().

below.

5. Discussions and Implications

5.1. Result Discussion

Based on the empirical results that are presented in Table 8, some highlights are listed below:

Digital transformation: The regression result of a negative relationship between ICT readiness and the NPL ratio is statistically significant, as expected, and is staying in line with previous research. The application of technology and information systems helps commercial banks in the credit risk management process, enhancing and boosting the process of lowering the NPL ratio, concurring with the study by Yang and Masron (2023). In an era of complexity for each case of lending, commercial banks are in need of technology enhancement to reduce processing time and increase efficiency when identifying signs of credit risk to prevent it from happening. Based on the result of the regression, for each 1% of ICT enhancement, the NPL ratio would drop by around 0.1173%.

Basel III implementation: The Basel Accord is an important framework for banks to follow and build strategies for better operation as well as avoid catastrophic failure in terms of credit risk. Therefore, the results of the negative relationship of Basel III implementation with the NPL ratio are acceptable. As the requirement to apply the Basel Accord is currently in the process of materializing in Vietnam, commercial banks are trying to implement Basel III in the hope of better managing credit risk.

Furthermore, the regression results also show the effect of other factors on banks’ NPL levels:

GDP growth: GDPG has a positive correlation with NPLs and is statistically significant at the 1% level with a coefficient of around 0.16337. This indicates that, at a time when economic growth increases by 1%, the NPL ratio of banks would also rise by 0.163%. This result contradicts that in other, foreign literature, as they concurred that an economy that thrives leads to higher social income and helps to deduct the burden of financial liabilities of customers, hence associating with lower NPL ratios. However, this above result is in line with several studies on Vietnam. Duong and Huong (2017) suggested that the same type of relationship fits the context of Vietnam, as this country’s economy has the characteristic of being bank-based; therefore, when the economy is on the rise, capital supply and credit demand also increase, leading to easier accepted lending, and therefore a higher potential of impaired loans.

Inflation: INF shows a positive influence on NPLs, but with an insignificant result. This fits with the general view of how unclear inflation interacts with NPLs. As mentioned in the literature review, Nkusu (2011) expected that the result of inflation could be positive or negative based on how the economy of a research sample uses the capital channels provided by commercial banks. In the context of high inflation, borrowers could use credit with low interest rates, but for individuals, their income in an inflationary context is also deducted, so the ability to repay the debts of customers is also worsened. Hang et al. (2018) suggested that the unclear influence of the context of high inflation in Vietnam could be due to novel players that come into the debt trading market and become the debt market maker, notably the VAMC, which would help commercial banks reduce impaired loans becoming a write-off balance sheet in a time of high inflation. Duong and Huong (2017) recorded a positive but still insignificant correlation, and supposed the reason for this was due to the intervention of authorities—in this case, the SBV. When the economy shows signs of inflation, inflation in a relative form would mostly be kept at a certain threshold as a result of regulation application. Meanwhile, NPL ratios in the inflation context also differ among large and other smaller banks, and the lending portfolio structure of banks prefers more corporate or personal banking, based on their operating strategies.

ROAA: The FGLS regression result also shows a negative relationship at the 1% significance level between banks’ profitability and the NPL ratio, as expected. This result is considered obvious as banks obtain higher profits than the assets side on a balance sheet, with reducing signs of customers’ loans leading to lower total assets and showing the status of efficient management, with the capital buffer they obtained from revenue adequate to cover customers’ loans; therefore, the NPL ratio is mitigated. Moreover, this result concurs with moral hazard theory, as when banks obtain low capital and are not capable of covering impaired loans they might not have the ability to diversify their lending portfolio, hence the higher credit risk.

Pure LDR: The positive result for the connection between pure LDR and the NPL ratio stays in line with expectations and other past research. With the statistical significance for a coefficient of around 0.317, in circumstances where pure LDR increases by 1% the NPL ratio would increase by 0.317%, as when capital outflows are higher than capital inflows from customers’ deposits, a bank would face a higher potential of liquidity risk.

CAR: While previous research also shows an unclear result and currently does not have a consensus about the interaction of the CAR with the NPL ratio, the result of the positive relationship between the CAR and the NPL ratio is contrary to expectations. This could be attributable to the behavior of Vietnamese banks when increasing the CAR, as it often also leads to an increase in risk-weighted assets. However, this relationship is often considered as a two-way relationship, with the direction of the effect of the NPL ratio on the CAR being clearer than the other direction, as raising the CAR is how a bank responds to an increasing credit risk. According to Ahmad et al. (2008), since, for commercial banks, “expected bankruptcy costs reflect an increasing probability of failure, banks would likely raise their capital-to-asset ratios when their asset portfolio risk increases”. An increase in the CAR may be a sign that banks are trying to prepare for a worse situation that may occur due to increased bad debt. This could also be an action based on recommendations from regulators to protect the bank from potential future volatility. In some cases, an increase in the CAR may be the result of market pressure or new regulations requiring banks to have higher capital adequacy as asset quality declines. In conclusion, the authors accept this result and consider this reasonable in the Vietnamese banking context.

STATE: The result indicates a negative influence of state ownership on the NPL ratio, which is in line with previous research by Mahdi et al. (2023), as they noted that state banks in Vietnam outperform private and joint-venture banks because they are relatively larger in terms of scale and resources. Furthermore, Vietnamese SOCBs are all long-standing banks, with great reputation, scale, human resources, capital resources, and technology, hence the advantages that they have in advancing the lending process and credit risk management. Therefore, the difference between the group of state banks and the non-SOCBs is understandable, as the SOCBs have fewer NPLs.

5.2. Implications

There are three recommendations for Vietnamese commercial banks from the empirical findings above:

First, continue to implement Basel III.

Until the first quarter of 2024, only seven commercial banks in Vietnam had announced the completion of Basel III implementation (see Appendix A, Table A1). Besides these, other banks have applied only some of these standards, or are currently at the stage of trials. This is considered a scarce and small number of banks implementing Basel III, while there are still many banks facing high amounts of impaired loans. Applying Basel III standards with a series of regulations acts as a protective shield against risks, helping the banking system not encounter similar incidents of global financial or regional crises.

However, to meet these standards of Basel III, banks are required to prepare more capital and accept larger reserve buffers to reduce operational risks. The new capital requirements of Basel III are now an actual obstacle and challenge for Vietnamese commercial banks, as they are the main reason that the scarce number of Vietnamese banks fully applying Basel III can be attributed to.

Second, enhance capital buffering for the new Basel III capital requirements.

Compared to Basel II, Basel III has many new requirements and has many impacts on bank operations. These updates of Basel III emphasized the importance of the quality and soundness of banks’ equity capital buffers, instead of just paying attention to the total capital value. However, implementing Basel III requires the large investment of resources as well as careful preparation by banks. It can be seen that the Vietnamese banking sector is still “thin” in capital, while the challenges from the external economy at this time are still very substantial. Thickening capital “buffers” will help banks have more resources with which to cope with challenges, and will also be the basis and conditions for continuing to support businesses and the economy to recover and develop.

Although the regression result shows a positive correlation between the CAR and the NPL ratio, increasing capital to improve financial capacity is still a necessary solution for banks. Banks may have been aware that, with high NPL levels, financial risks also increase, and, therefore, they need to increase the CAR to strengthen their financial position, giving a bank a larger capital “buffer” with which to resolve bad debts. To achieve this goal, banks are racing to make plans to submit to shareholders to increase charter capital through (i) dividends, (ii) issuing new shares, and (iii) attracting foreign capital.

Third, invest in technologies to push digital transformation.

The requirement of enhancing banks’ ICT performance for credit risk management is increasingly high. In this context of cashless payments and transactions being the novel trend, commercial banks need to update and strengthen their systems for better management and lending supervision, and let technology take part in constructing banks’ portfolios as well as scanning, evaluating, and credit scoring their customers, in order to mitigate the chance of loans becoming impaired. The Basel Committee on Banking Supervision (BCBS) even released Principles for the Sound Management of Operational Risk (PSMOR) as a guideline for enhancing banks’ ability to endure, one approach being to implement robust ICT performance of a bank, in order to bounce back from significant negative occurrences (Basel Committee on Banking Supervision (BCBS) 2021).

To secure and maintain the confidentiality of data and systems, banks’ directors should consistently supervise banks’ ICT risk management effectiveness. This necessitates the regular synchronization of banks’ operations, risk management, and ICT readiness strategies in order to align with banks’ risk appetites and risk tolerance statements. Banks should persistently monitor their ICT and routinely provide reports to higher levels of authorities about ICT risks, controls, and events.

6. Conclusions

Credit risk management has always been a banking activity of paramount importance, especially at the stage of stricter bank management requirements, dynamic technology, and digital transformation. To verify the impacts of digital transformation and Basel III implementation on the reduction in banks’ credit risk levels and suggest suitable as well as applicable recommendations for credit risk management in banks, the authors aimed to conduct research on the detection impacts of digital transformation and Basel III implementation on NPL levels in the 21 selected Vietnamese commercial banks from 2017 to 2023. A quantitative approach, including the pooled OLS regression model, the FEM, and the REM estimation methods, as well as FGLS regression, are used for completing this aim. The results are derived from expectations and proposed hypotheses, as this paper found that digital transformation, ICT readiness, and Basel III application, along with banks’ profitability, negatively affect credit risk level. Furthermore, the empirical results show that GDP growth, inflation, LDR, and the CAR all have a positive correlation with the NPL ratio. Based on these findings, the authors affirm the contribution of digital transformation and Basel III implementation, among other internal and external factors, to reducing the credit risk level of banks and suggest recommendations for Vietnamese commercial banks regarding the continuous progress of Basel III routine implementation and pushing forward the digital transformation, which then contribute to sounder credit risk management. This paper then becomes distinctive to the current literature as it is deeply concerned with the Vietnamese banking system only and uses a large number of variables covering both macroeconomics and internal bank factors. However, this study still has the limitations of there being limited Vietnamese banks in the sample and a limited time period, so future academic research would work to fill in this gap.