Fintech and Commercial Banking

Data Collection and Selection of Sample

From 2011 to 2020, data were collected from the Wind database, the annual reports of listed commercial banks, the website of the National Bureau of Statistics, and the website of China. There are 37 listed commercial banks (Table 1). Among them, five are state-owned commercial banks, eight are joint-stock commercial banks, 14 are city commercial banks, and 10 are rural commercial banks.

The five state-owned commercial banks are the largest group of commercial banks in China. Moreover, eight joint-stock banks are the commercial banks established in accordance with the joint-stock system after the reform and opening in 1979. Similarly, 14 city commercial banks are regional commercial banks established in developed cities in China. In the last decade, 10 rural commercial banks are restructured from rural credit cooperatives.

Table 1. Selection of sample.

Selection of Variables

Explained Variable

The explained variable is commercial banks' risk-taking (RISK). In the previous studies, the variable has been used to measure commercial banks' risk-taking with z-value, risk-weighted asset ratio, non-performing loan ratio, capital adequacy ratio, loan loss reserve ratio, capital asset ratio, equity-to-liability ratio, expected default probability, stock market volatility, and stock price volatility. According to the availability of data, we used the z-value (Z) to measure the overall risk level of commercial banks.

$z = \dfrac{ROA + CAR}{ \sigma (ROA)}$ (1)

where ROA is the return on assets, CAR is the capital-to-asset ratio (CAR = E/A), and σ(ROA) is the standard deviation of the return on assets. The Z value measures the overall stability of the banks. The changes in the Z value are consistent with the change in the stability of the banks. A larger Z value means stronger stability of banks. The change in the strength of stability is the opposite of the increase or decrease in risk. A bank with enhanced stability will reduce its risk, and vice versa. The Z value has its unique characteristics, showing a tail after the peak, so the logarithm of the z value must be taken during regression. There are special cases where the z-value is zero, and log (1+Z) is used instead of log (Z). In addition, we used the nonperforming loan ratio (NPL) to replace the z-value for robustness analysis.

Main Explanatory Variables

In the literature, there are many indicators to measure the degree of financial technology development. For example, Qiu et al. used China's Digital Financial Inclusion Index. Guo and Shen used the "text mining method" to build their Internet financial index. Zhu and Hua used the "Internet Finance Theme Index" compiled by the Shanghai Stock Exchange to replace the development scale of Internet finance.

By analyzing the measurement methods above, we used the "text mining method" of Guo and Shen to construct our fintech index and used the method of Wu X. F. to measure the degree of independent use of financial technology by banks, namely, the fintech level of commercial banks. Specific steps can be written as follows:

The first step is to establish an initial lexicon of fintech based on the functions of fintech in commercial bank applications. At the same time, starting from the format of business, we divided financial technology into four dimensions, such as payment and settlement, business development, investment management, and resource allocation.

The second step is to match each keyword with the bank name, search for the bank name and keyword in Baidu Information, and then use Python web crawler technology to crawl all the news search results of each bank from 2011 to 2020. To ensure the accuracy of search results, we used double quotation marks to lock keyword groups when searching to filter out irrelevant or wrong information.

The third step is to filter effective keywords. In particular, the word frequency is standardized, and then the Pearson correlation analysis method was used to calculate the correlation coefficient between the word frequency and the annual average value of the nonperforming loan ratio of commercial banks. Following Larson and Farber, the critical value of weak correlation is set at 0.3. Finally, a total of 15 keywords were reserved, such as peer-to-peer transmission, digital currency, value transfer network, foreign exchange wholesale, third-party payment, online payment, online financing, Internet financial management, Internet of Things, artificial intelligence, blockchain, large-scale data, cloud computing, and digital exchange platform.

The fourth step is to synthesize the annual fintech index of each bank (FIN). The number of news search results for all keywords at the annual level of each bank is summed up to obtain the annual total news volume of the sample bank, and then logarithmically processed to serve as an indicator to measure the bank's annual level of fintech use. It should be pointed out that to accurately identify the business orientation of the commercial bank fintech index, we decomposed the fintech index of each bank into four dimensions, namely, payment settlement (PAYS), business development (BUS), investment management (INM), and resource allocation (RESA), and measured the sub-indicators of each dimension.

Mediating Variables

Fintech mainly affects the overall risk level of commercial banks through three channels such as operational efficiency, financial innovation, and risk management. To count the impact of financial technology on these aspects, we transformed them into measurable indicators. We used the cost-to-income ratio (CIR), noninterest income ratio (NII), and loan impairment to total loan ratio (ILR) to represent operating efficiency, financial innovation, and risk management to analyze the intermediary effect.

Control Variables

To analyze the changes in the explained variables and minimize the multicollinearity problem more efficiently, we added some control variables, including macro-level and microlevel. The value of variance inflation factor (VIF) for multicollinearity is around 5, which reveals the non-existence of the multicollinearity.

Macroeconomic Variables

Economic Development (GGDP)

The higher the GDP growth rate, the better the economic development, which affects the business and operations of commercial banks. Generally, there is an inverse relationship between the GDP growth rate and the risk-taking of commercial banks.

Inflation

There are three possibilities for the impact of inflation (INF) on bank risk. First, inflation will increase bank costs, which is adverse for banks. Second, the central bank's currency is over-issued, and inflation is beneficial for the bank, which is the debtors. Third, when the economy is prosperous, inflation will make the country use tight monetary policy to curb inflation, which is adverse for the banking system.

Monetary Policy (M₂)

We used the growth rate of the broad money supply to represent monetary policy. The lower the cost of financing by enterprise, the looser the monetary policy. The larger-scale credit business of commercial banks and the relaxation of credit standards are detrimental to credit risk management.

Variables of Bank Characteristics

Operating Scale

Logarithm of total assets (LNTA) belongs to the operating scale of a bank, which is generally expressed by the bank's total assets or total income. According to the theory of economy of scale and scope, the expansion of scale can reduce the credit risk faced by banks. We have taken the bank's asset size as one of the control variables and taken the logarithm of the asset size.

Capital Structure

We used the owner's equity ratio (OER) to measure the bank's capital structure. The OER is equal to the ratio of equity capital to total assets. The higher the OER, the less debt the bank has, the stronger the repayment ability and the stable capital structure. In addition, we also used control of the micro-characteristics of commercial banks, such as return on assets (ROA), net interest margin (NIM), and loan-to-deposit ratio (LOD).

Analytical Framework

To verify the impact of commercial bank financial technology development on its risk-taking, we used the given regression function.

$RISK_{it}=α_0+α_1FIN_{it}+βcontrols_{it}+μ_i+λ_t+ε_{it}$ (2)

where RISK is the explained variable and represents the bank's risk-taking, the main explanatory variable (FIN) represents the extent of the bank's financial technology, controls_it is a set of control variables, μ_i represents the individual fixed effect, the λ_t represents time fixed effect, and the ε_it is the random error, which is assumed to be normally distributed at zero mean value and constant variance and i = 1, 2, …, 37, and t = 2011, 2012, …,2020.

In addition, we used an intermediary effect model to examine whether fintech affects risk-taking through intermediary indicators. The regression functions can be written as follows:

$Med_{it}=C+_2+αFIN_{it}+βcontorls_{it}+μ_i+λ_t+ε_{it}$  (3)

$z_{it}=C+α_3x_{it}+δMed_{it}+βcontorls_{it}+μ_i+λ_t+ε_{it}$  (4)

The intermediary variables (_{Med_it}) are operational efficiency, financial innovation, and risk management.

Construction of a Risk-Warning Model for Commercial Banks

The KLR refers to Kaminsky, Lizondo, and Reinhart because it was introduced by Kaminsky, Lizondo, and Reinhart in 1998. The KLR model is a signal analysis method that is used to study bank risk and financial crisis early warning. The main idea of the KLR model is to build an indicator system and combine these indicators to determine the threshold range. When the predicted data exceed the threshold range, a danger warning signal will be issued.

The higher the indicator value, the greater the risk corresponding to the indicator. The prediction time window of the KLR model was set to 12 months and decomposed the signal into four categories. Suppose "A" represents the number of correct signal months, "B" is the number of false signal months, "C" is the number of months that should have been signaled but were not signaled, and "D" is the number of months that should not have been signaled but were not signaled. If the crisis signal occurs within the 12-month range, the signal is valid, and if the crisis signal occurs after 12 months, the signal is invalid.

According to the assumption, the ideal model of the early warning model is that "A" and "D" are greater than 0, and "B" and "C" are equal to 0. Based on the actual situation, if the ideal state cannot be achieved, Kaminsky and Zhang used the method of minimizing the noise-signal ratio to determine the indicator threshold. A/(A+C) represents the ratio of valid signals; B/(B+D) represents the ratio of invalid signals; and [B/(B+D)]/[A/(A+C)] represents the signal-noise ratio to the effective signal. The indicator is abbreviated as NSR, and the case with the smallest indicator is the best threshold. There are n indicators. X_it represents the value of indicator i in time period t, and S_it represents whether there is a danger signal for indicator i in time period t. S_it = 1 means that the indicator i sends a danger signal in the t time period, and S_it = 0 shows that the indicator i does not send out a danger signal in the t time period. Following Bisceglia and Scigliuto, we constructed the bank risk index (BRI).

$BRI_t=w_{Ldr}(\dfrac{Ldr_t−Ldr_{t−1}} {Ldr_{t−1}})+w_{Rir}(Rir_t−Rir_{t−1})+w_M(\dfrac{M_t−M_{t−1}}{M_{t−1}})$  (5)

where Ldr_t, Rir_t, and M_t represent the loan-to-deposit ratio, real interest rate, and money supply at time t (or time period), respectively. W_Ldr, W_Rir, and W_M are the weights of the deposit-loan ratio, the real interest rate, and the money supply, respectively. The weights are determined by the standard deviation that can be calculated using a given function.

$w_k=\dfrac{1}{Std_k} / (\dfrac{1}{Std_{Ldr}} + \dfrac{1}{Std_{Rir}}+\dfrac{1}{Std_M})$  (6)

The criteria for the early warning can be written as follows:

$Warning= \overline{BRI}+σt$  (7)

If $BRI_t > \overline{BRI}+σ_t$, then the bank has systemic risk at time t. If $BRI_t < \overline{BRI}+σ_t$, then the systemic risk of the bank at the time t is small.

The bank risk-warning index can be written as follows:

$RWI=\dfrac{∑_i(\dfrac{S_{it}}{NSR_i})}{∑_i(\dfrac{1}{NSR_i})}$   (8)

where NSR_i represents the noise-to-signal ratio corresponding to the indicator i. The calculation formula is [B/(B+D)]/[A/(A+C)], and Si_t represents whether the indicator i has a danger signal in the t time period. S_it = 1 means that the indicator i sends a danger signal in the t time period. When S_it = 0, the indicator i does not send a danger signal in the t time period.