where d stands for the difference operator. A negative and significant sign for our institutional competitive advantage variable would confirm our main hypothesis.

To estimate this model, we use a Poisson specification.^{14 }The Poisson model is the most commonly used model when the dependent variable takes on nonnegative (integer) values (Wooldridge, 2002). Poisson models have been used in a number of finance and economic applications. For example, Fernando, Gatchev, and Spindt (2005) use it to explain the number of underwriting issues, Eisenberg, Sundgren and Wells (1998) the change in corporations’ board size, and Vafeas (1999) the distribution of board meetings frequency. Using a Poisson model has a number of advantages. The major advantage is that it can account for the fact that for many source-host combinations there is no change in presence, i.e., the dependent variable consists of many zeros. Silva and Tenreyro (2006) show that a Poisson model is preferred when dealing with many zeros in de dependent variable, as for example when analyzing trade flows in a gravity model context. The second advantage is that the Poisson model has been shown to produce consistent estimates under relatively weak assumptions, including that the dependent variable does not have to follow a Poisson distribution (Gourieroux, Monfort and Trognon, 1984; Wooldridge, 1997). Lastly, coefficients are easy to interpret as (semi- )elasticities.

We estimate all standard errors robustly. Furthermore, to take into account that in general more banks exist that can potentially engage in cross-border investment in larger source countries and that more opportunities exist to invest in larger host countries, we use a weighted estimation, with the weights equal to the inverse of the product of the average dollar GDP of source j and host i countries, both measured over 1996-2005.

14

The model assumes that y given x has a Poisson distribution and that the conditional variance and mean

are equal: f (dforpresence_{ij }| X_{ij }) =

) e x p ( i j i j μ dforpresence! dforpresence_{ij }μ −

So the density of y given x under the Poisson

distribution is completely determined by the conditional mean.

14