Cantonal Bank Panel Data Set
Description
Self-compiled, balanced panel dataset of 21 Swiss cantons and their respective cantonal banks over 61 years (1960-2020). The dataset contains bank variables like return on equity, capital ratio, interest rate spread, mortgage ratio and branch density as well as reform variables denoting which reforms cantonal banks underwent during the period. The purpose of the data is to test whether cantons have changed their conception of public ownership by testing the effect of reforms on performance (ROE) and bank metrics, which capture the social welfare function of banks. Cantonal banks were originally founded to serve an underbanked cantonal public by providing stable and inexpensive banking services for borrowers and savers. The data shows that reforms, summarized in a commercialization index, boosted the return on equity, lowered capital ratios, increased the net interest rate spread and decreased branch density. The results imply that a trade-off exists between reform-induced profitability increases and social welfare, as commercialization increased risk by lowering capital ratios, increased the costs of financial intermediation by widening spreads, and reduced bank service coverage in the respective canton.
Files
Steps to reproduce
Static and dynamic OLS models were estimated with unit and time fixed effects with GDP growth as an additional control. Mediation of the commercialization index effect on ROE through the four bank variables indicating social welfare was achieved by applying the method of Jonah B. Gelbach, described in his paper "When Do Covariates Matter? And Which Ones, and How Much?", published in 2016 in Journal of Labor Economics. The idea is to partition the attribution of each social welfare variable to the commercialization effect by comparing the base model to the extended model (including the four social welfare variables). As a robustness check, a stacked event study was conducted as proposed by Wing, Freedman, and Hollingworth ("Stacked Difference-in-Differences", published 2024 in NBER) which requires changing the dataset. First, you define the subexperiments, indicated by all possible reform years. Then you assign for every treatment group a control group, imposing a minimum time difference of 10 years without any reform. Clustering on the cantonal level and using the appropriate weights (see Wing, Freedman, and Hollingworth), reform effects and possible pre-trends can be investigated by applying the event study with time and unit fixed effects.
Institutions
- Universitat Zurich