Journal of Money, Credit, and Banking

During most of the 20th century, Latin American banking systems tended to be small, inefficient, and highly concentrated. Most were also characterized by regulatory barriers to entry. One component of these regulatory barriers was restrictions on the entry of foreign banks. Beginning in the early 1990s, most of the barriers to foreign investment in banking were lifted, and over time, foreign banks have come to play an increasingly larger role in most Latin American financial systems. Foreign banks now control roughly 45% of all Latin American bank assets. This change in ownership structure poses an interesting question for both academics and policy makers: what effect did the entry of foreign banks have on the spreads charged by banks? Did foreign banks increase competition and produce administrative innovations that reduced the gap between what banks paid for funds and charged for loans; or did the entry of foreign banks increase concentration, thereby allowing banks to behave less competitively?

Answering this question requires careful econometric analysis. At the same time that foreign banks were entering, both creating de novo operations and purchasing existing banks, domestic banks were also merging with one another. Thus, independent of foreign bank entry, there was also an increase in concentration. In order to measure the independent impact of foreign bank entry, we need to be able to separate [End Page 539] out the independent effects of domestic mergers and acquisitions, foreign bank mergers and acquisitions, and foreign bank de novo operations.

Martinez Peria and Mody (2004, this issue of JMCB) therefore carry out a very clever and very careful econometric analysis of data from five countries, Argentina, Chile, Colombia, Mexico, and Peru, during the late 1990s. Some of their findings are unsurprising: bank concentration was positively related to higher spreads and higher administrative costs; and foreign banks charged lower spreads than domestic banks. Some of their findings are curious: de novo foreign banks charged lower spreads than foreign banks that merged with domestic banks. In addition, foreign banks charged lower spreads because they had lower administrative costs.

This is a wonderful paper, which advances our knowledge in important ways. I do have some comments and questions, and I pose these in the spirit of making a very strong paper even stronger.

Let me start by posing a series of questions related to their result that de novo foreign banks had lower administrative costs and lower spreads than foreign banks that purchased already existing domestic operations. The choice of strategy (de novo, or merger and acquisition) is taken in the paper as exogenous. But, why would foreign banks purchase already existing banks (instead of opening a de novo operation) if the already existing banks were inefficient? What would be the advantage of this purchase over setting up a de novo operation?

One answer may be that if you want to enter particular segments of the market you need the expertise of local bankers. The authors mention that this may be a factor, when they note that de novo foreign banks may specialize in loans that are more transparent, where it is easier to obtain information from borrowers. I would suggest that they pursue this hypothesis systematically. In fact, differences in lending strategies may explain differences in spreads and administrative costs, all other things held constant. If that is the case, then the difference in the coefficients on foreign mergers and acquisitions and foreign ownership may be endogenous. There are several ways to check for this. First, at least in the Mexican case, there are data by bank and by quarter about the riskiness of bank loan portfolios. Do such data sets exist for other countries? Second, there is information in most bank superintendancy data sets about loans by type: consumer, housing, agriculture, commercial, etc. One could at a minimum add a variable to the spread and administrative cost regressions to control for either portfolio risk or market segment. I would also suggest estimating a Probit where the independent variable is...