Empirical results

4.3.1 Testing Hypothesis 1: Purpose-driven banks are more likely to establish alliances with fintechs. Estimating the marginal effects of the purpose on the probability of at least one partnership between fintechs and bank (i) in a year (y)

This study used Probit regression for panel data using STATA version 17 with the deployment of the xtprobit command. The researcher was trying to examine whether purpose-driven orientation within the bank has an influence on the bank–fintech partnership. This method was chosen because it provides the marginal effects of the independent variables, so it is possible to identify the statistical difference between groups.

Table 7 provides the results of the random-effects panel probit regression to test the probability of at least one interaction between the bank and a fintech happening based on the full sample. The coefficients show the average marginal effects with bootstrapped standard errors in parentheses. Model 1 includes only control variables. Model 2 provides information on both control and independent variables. The full model was found to be statistically significant with a chi-squared value equal to 36.89 (n = 747), p = 0.000, showing that the model was able to distinguish between banks that formed the collaboration with a fintech versus those which did not.

It was found that the coefficient of Purpose is positive and significant, proving that having a purpose in the strategic focus of the bank increases the probability of partnership with a fintech by 37.98% at p<0.01 level of significance, standard error of 0.098 and 95% confidence interval from 0.187 to 0.572.

The first model’s chi-squared was equal to 18.65 with the marginal effects’ coefficient for ln (total assets) of 0.051 at the significance level at p<0.01, equity ratio with the coefficient of 0.44 at p<0.01, and ln (age) with the coefficient of 0.11 at p<0.01. All these control variables were found statistically significant.

The final model with both control and independent variables is significant with chi-squared equal to 36.89 at p = 0.00 level of significance. Moreover, statistically significant variables include purpose, the natural logarithm of assets and equity ratio with marginal effects of 0.041 and -0.029 at p<0.01 respectively. Overall, the results show that the model was able to distinguish between purpose-driven and traditional banks that formed the partnership with the bank. Consequently, it can be claimed that my findings support my Hypothesis 1: Purpose-driven banks are more likely to establish alliances with fintechs. It means that if the bank is purpose-driven, then the probability of at least one partnership with fintech increases by 37.98%. The hypothesis is accepted.

Model 1(control variables) Model 2

Number of observations 747 747

Number of banks 101 101

Log likelihood -346.67 -336.63

Prob > chi2 0.002 0.000

Chi-squared 18.65 36.89

Table 7. Results of panel data analysis with the use of random effects Probit regression measuring the probability of whether at least 1 partnership between a bank and a fintech happens (partnership = 0) or not (partnership = 0). *** significance level of p<0.01, **

significance level of p<0.05, * significance level of p<0.1. The coefficients reflect the average marginal effects with bootstrapped standard error reflected in parentheses. d indicates the dummy variable.

4.3.2 Testing Hypothesis 2: There is a positive relationship between purpose-driven banks and the number of partnerships between those banks with fintechs. Estimating the Number of the partnerships between the fintechs and a bank (x) in the year (y)

This paper uses the Panel Count model since the dependent variable of this model is a count variable with non-negative integers and the data contained multiple instances per bank over 8 years.

After running Poisson regression, it was clear that this method cannot be used since the goodness-of-fit test statistics were at the level of p<0.05 with a large value of chi-square equal which are the indicators that the data might suffer from overdispersion. That is why the study deployed the random-effect negative binomial regression method for the panel dataset as it allowed overdispersion with the usage of STATA version 17 using the xtbreg function. The random effect was chosen after running the Hausman test to identify which is more appropriate.

Table 8 provides the results of the negative binomial regression to test whether purpose correlates with the number of fintechs with which a bank collaborates based on the full sample. The Incident Rate Ratios report the effect of the independent variables on the count variable. Again, as in the previous hypotheses, Model 1 includes only control variables. Model 2 provides information on both control and independent variables. The full model was found to be statistically significant with a chi-squared value equal to 47.02 (n = 747), p = 0.000, showing that the model was able to distinguish the number of collaborations between purpose-driven banks and not.

It was found that the coefficient of Purpose is positive and significant, proving that having a purpose in the strategic focus of the bank increases the number of partnerships with fintechs by 4.9 partnerships for purpose-driven banks in comparison with those without the purpose at p<0.01 level of significance, standard error of 1.83 and 95% confidence interval from 2.36 to 10.204.

The first model’s chi-square value was equal to 33.4 with an IRR for ln (age) of 1.92 at the significance level at p<0.01, ln (total assets) with the IRR of 1.387, equity ratio with the IRR of 13.726 at p<0.01, listed with the IRR of 0.119 at p<0.01, and ROAA with the IRR of 0.045 at p<0.01.

All of these control variables were found statistically significant.

The final model with both control and independent variables is significant with chi-squared equal to 47.02 at the level of significance at p = 0.00. Moreover, as in the model 1, all of the control variables used in this analysis were found to be statistically significant at a different level of

significance from p<0.01 for ln (age), ln (assets), and equity ratio to listed and ROAA at the p-value<0.05. Meaning that the big, publicly listed, old banks with a higher Equity Ratio and Return on Assets form partnerships with fintechs more often. If the bank is listed, then the count increases by approximately 0.23 partnerships. If the bank is big in terms of total assets, the count of partnerships with fintech increases by merely 1.3 partnerships. Overall, the results show that the model was able to distinguish between purpose-driven and traditional banks in terms of the number of formed partnerships with fintechs. Consequently, it can be claimed that these findings support Hypothesis 2:

there is a positive relationship between purpose-driven banks and the number of partnerships between those banks with fintechs. It means that the purpose has a significant positive impact on the count of partnerships. The hypothesis is accepted.

Model 1 (control variables) Model 2

ln(age) 1.919***(0.342) 1.52***(0.24)

Number of observations 747 747

Number of banks 101 101

Log likelihood -814.701 -806.706

Prob > chi2 0.000 0.000

Chi-squared 33.40 47.02

Table 8. Results of panel data analysis with the use of random-effects negative binomial regression measuring the number of partnerships of the bank in a particular year. *** significance level of p<0.01, ** significance level of p<0.05. The coefficients reflect the average marginal effects with bootstrapped standard error reflected in parentheses. d indicates dummy variable.

Both proposed hypotheses were accepted according to the above-mentioned tests. A further discussion of the results and the elaborated reflection on the existing literature is expected in the subsequent “Discussion” section.

Figure 3. Conceptual model of the variables used in this research. *** p<0.01.

Purpose-driven/Traditional bank

Probability of collaboration with Fin-tech

Number of Collaborations H1. 0.3798***

H2. 4.912***