Research on the Contagion Effect of Bank Systemic Risk Embedded with the Fintech Public Opinion Index

1. Sample Selection: Sixteen systemically important banks listed on China's A-share market were selected as research subjects, covering the period from January 2011 to December 2024. Unlisted and late-listed institutions were excluded. These include state-owned banks, joint-stock banks, and city commercial banks.
2. Core Data Sources:
(1) Fintech public opinion data: Obtained from Infobank’s "China Economic News Database", news texts containing "bank name + fintech keywords + sentiment words" were crawled through Python web scraping.
(2) Bank stock return data: Retrieved from the Wind Database, with weekly logarithmic returns calculated using average closing prices.
(3) Macro - micro variables: Seven macroeconomic and four bank - specific variables were sourced from the Wind Database; quarterly characteristic variables were converted to weekly frequency via cubic spline interpolation.
(4) Fintech keyword lexicon: Independently constructed with 359 terms (three dimensions).
(5) Financial sentiment dictionary: Integrated Jiang et al. (2019) and other Chinese financial sentiment dictionaries, including 11,487 positive and 18,014 negative words.
3. Data Processing Methods:
(1) Quantification of public opinion index: The frequency of positive and negative sentiment words in news texts was counted. The daily public opinion index was calculated through indexing formulas, and weekly average smoothing was applied to alleviate data gaps and random fluctuations.
(2) Rolling window analysis: A 52 - week rolling window was adopted, combined with CoVaR at the 1% quantile and the TENET model to construct a dynamic risk spillover network.
(3) Threshold filtering: Edge connections below the median in network analysis were removed to avoid information redundancy, and the Fruchterman - Reingold algorithm was used to present node clustering characteristics.