local employees and skills. In addition, mergers and acquisitions have to work, at least initially, with acquired bank’s existing technology, processes and procedures. If the bank is not familiar working in a business environment similar to that of the host country, it may be harder to operate a merged/acquired bank than a greenfield. At the same time, however, institutional competitive advantage can also be important for greenfields. Unless the foreign bank already has a competitive advantage to work in the host country’s institutional environment, it may find it hard to start a greenfield operation, a new bank, from scratch. As a consequence, for different reasons, institutional competitive advantage is important for both modes of entry.

Robustness tests We next conduct a number of robustness tests, reported in Table 4, where we only provide the coefficients of (the change in) our institutional competitive advantage variable (all other variables are not reported, but generally retain their statistically significance). We have so far conducted the main regressions using Poisson regressions. Tobit can, however, be an alternative technique when the change in the gross number of banks present is a positive number. The regression result using Tobit is reported in Table 4, second row (first row repeats the base regression of Table 3). The result is very similar to that using Poisson, with the sign and statistical significance of the institutional competitive advantage the same (the coefficient is of different size since it is a different technique and is not a semi-elasticity).

Instead of using either Poisson or Tobit techniques, we also estimate the model with simple OLS using all observations (Table 4, row 3). This is admittedly not the right econometric model, since only a limited number of countries experience a change in the number of foreign banks present, i.e., many observations are zero. Nevertheless, the regression results do show the same sign and significance for our institutional competitive advantage variable.

We next re-estimate the base regression using Poisson estimation techniques but using different sample periods over which we calculate the change in foreign bank presence. Specifically, we consider changes over the periods 1996-2000 (row 4) and 2001-2006 (row 5). We adjust all the explanatory variables to match these periods. In

19