Dependent Variable

To control for the bank size effect of the dependent variables, the securitization transaction volume is scaled by the entity's total assets. The sample was collected from 35 securitizing banks, and their total transaction volume is around CNY 1.2 trillion.

## *3.3. Empirical Model*

This paper employs fixed effects and random effects estimation methods on panel data in order to compare the determinants of banks' engagement in loan securitizations pre- and post-2017 in China. Panel data (also called longitudinal data) embodying information across both time series and cross sections (entities) are multi-dimensional (Diggle et al. 2002). The sample of this study comprises panel data on 35 banks across 7 years and 42 banks across about 11 years for analysis. There are broadly two classes of panel estimator approaches, fixed effects and random effects models, that can be employed in this research. These two models are normally employed to obtain a function that predicts whether an observation belongs to a particular group or when trying to analyze the influence of a series of independent variables on the dependent variable (in our case, the three bank-specific determinants that may influence the amount of securitization). The unobserved variables can have any associations with the observed variables in the fixed effects model, while the unobserved variables are assumed to be uncorrelated or more strongly statistically independent than all of the observed variables in a random effects model. It is difficult to determine whether or not the unobserved variables in this case are statistically independent of the four bank characteristics. To determine the appropriate model, we used the Hausman test. If the probability in the Hausman test is larger than or equal to 0.95 and less than or equal to 1 (0.95 ≤ Prob. ≤ 1), it is suggested that the error term is not correlated with the independent variables, the hypothesis is not rejected, and the random effects model should be applied for an analysis. By contrast, if the probability is too low, the unobserved variables are related to the observed variables, and a fixed effects model will be acceptable.

The empirical models are as follows:
