be estimated more complicated.7 We use the change in the number of foreign banks rather than the change in foreign banks’ assets since asset data are not available for all banks at all points in time. We do, however, conduct robustness tests using the available bank asset data.8
Empirical framework In order to explain bilateral FDI in banking, we use a first-difference model relating the change in the number of foreign banks from country j located in country i (over a period) to changes in our institutional competitive advantage measure and other variables. A first difference model is preferred for two reasons, econometrically and given the behavior underlying the entry decisions. Econometrically, first-difference is a preferred model since it controls directly for all country-pair, host-country and source-country fixed effects.9 It thus already controls for those bilateral, time-invariant factors that have proven to have explanatory power for foreign investment, including foreign bank entry, like distance, a common border or past colonial links between host and source countries.10 Besides any other bilateral, time-invariant effects (e.g., the existence of free trade arrangement), host- and source-country specific effects, such as the general risk of investing in a country, are also controlled for.
Behaviorally, first difference is also a preferred model because of the way banks can be expected to make their investment decisions. Banks will enter a particular country on the basis of, among others, the institutional competitive advantage at the time. Entry, however, comes with many fixed costs and is a discreet choice, made at one point in time and is (in general) not repeated or reversed easily. Banks are therefore not likely to review their decisions as circumstances, including institutional competitive advantage, vary from one year to the next, but rather consider their overall presence from time to time in light of changing circumstances. Consequently, it is more logical to explain the (change in) overall presence of foreign banks as a function of the (change in the)
7 We also estimated, however, the model including both entry and exits and found that this does not affect our main result (results are available upon request). We also used the number of foreign banks from source country j present in host country i relative to the total number of foreign banks in host country i and found similar results (available upon request). However, the results also hold when using a panel model approach (results are available upon request). Buch (2005), for example, find that banks hold significantly lower assets in distant markets, but that the importance of distance for the foreign asset holdings of banks has not changed over time. 8 9 10