FinTech adoption and bank risk-taking: evidence from China

ABSTRACT

We examine the impact of FinTech adoption on bank risk-taking. Applying the instrumental variable panel quantile regression (IV-QRPD) approach to the data of 160 Chinese commercial banks over the period of 2011–2020, we provide fresh evidence on the quantile-varying relation between FinTech adoption and bank risk-taking. We show that FinTech adoption increases bank risk-taking in low and middle quantiles but reduces bank risk-taking in high quantiles. We further find that this quantile-varying impact works through the efficiency-enhancing, gamble for resurrection, and credit expansion channels.

KEYWORDS:

• FinTech
• bank risk-taking
• panel quantile regression
• China

JEL CLASSIFICATION:

• C23
• G21
• G28

Highlights

• The quantile-varying impact of FinTech adoption on bank risk-taking is observed by applying a novel panel quantile regression approach.

• The quantile-varying impact operates through the efficiency-enhancing, gamble for resurrection, and credit expansion channels.

• Bank-level FinTech adoption measures are constructed using a textual analysis method.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

¹ We greatly appreciate an anonymous referee for the suggestions regarding the analysis of gamble for resurrection.

² Z-score, a common measure of bank risk-taking, is defined as (ROA+CAR)/σ(ROA), where ROA denotes the return on assets ratio, CAR the equity-to-assets ratio, and σ(ROA) the standard deviation of ROA. Following the practice of Laeven and Levine (2009) and Beck, DeJonghe, and Schepens (2013), we adopt a three-year rolling time window to compute σ(ROA). We then apply the natural logarithm to (1+Z-score) because the Z-score is highly skewed. Finally, we denote the inverse of the natural logarithm of (1+Z-score) as Risk_Z. A higher value of Risk_Z suggests a higher level of bank risk-taking.

³ Note that the quantile regression (QR) method is useful to examine the effect of variable X (e.g. FinTech) on Y (e.g. Risk_Z) across quantiles of Y, but it is improper when we want to explore the effect of X on M (e.g. CIR or gLoan) across various levels of Y. Thus, here we adopt a regime-varying regression model instead of a QR model.:

⁴ The initial agreement between the two independent research assistants was strong. Where the assistants disagreed, one of the authors made a final judgement by reading the annual report.

Additional information

Funding

This work was supported by the National Social Science Foundation of China (Grant No. 21FJYB005); the Humanity and Social Science Foundation of Ministry of Education of China (Grant No. 20YJC790034); China Postdoctoral Science Foundation (Grant No. 2019M663758).