The Barrier Comes Before the Interest Rate
The incomplete European Banking Union is not facing one undifferentiated “cross-border banking” problem. This chapter separates four points at which integration can succeed or fail: a banking group can enter a foreign market, a firm and bank can form a relationship, the linked bank can set a loan price, and the relationship can carry a particular quantity of credit. That separation lets the reader read the paper’s estimates correctly: the largest gaps appear before a loan has a price.
Four questions hidden inside one word
When people ask whether banking markets are integrated, they may be asking several different things at once. The first question is whether a banking group operates beyond its home country. This is the entry margin. It concerns the decision to establish a branch or subsidiary and to commit an organization, staff, compliance capacity, and balance sheet to a destination.
The second is whether a particular firm and bank have a lending relationship. This is the relationship-formation margin. A bank can be present in a country without lending to every firm there, and a firm can be eligible to borrow without ever becoming a client of a particular foreign bank. Information, trust, monitoring, legal arrangements, and the expected value of the relationship all matter before a loan is recorded.
The third question is what interest rate the bank charges once the relationship exists. This is the pricing margin. It is the comparison most likely to be called “price convergence,” but it is also the most conditional comparison: it excludes every bank and firm that did not reach a signed or active relationship.
The fourth question is how much of the firm’s borrowing comes from that bank. This is the quantity margin. A foreign bank may have a relationship with a firm and still supply a smaller share of its credit than a domestic bank would. Quantity can reflect demand, risk, bank capacity, and the terms of the relationship, so it should not be inferred from the interest rate alone.
These margins are related, but they are not substitutes. Entry changes the available supply of lenders. Relationship formation determines which of those lenders actually connect with a firm. Pricing and quantity describe the terms of the links that survive. A statistic from the last stage cannot reveal how much activity disappeared at the earlier stages.
The paper calls the country-pair effects that capture these differences “wedges.” The term is useful because it describes an implicit cost or distortion in the model, not a fee printed on a loan contract. A relationship wedge makes a cross-border match less likely; a pricing wedge changes the rate conditional on a match; an entry wedge makes foreign establishment less likely. Treating all three as the same “border cost” would conceal the mechanism the paper is trying to measure.
How a bilateral wedge is recovered
A raw comparison between domestic and foreign loans is not enough. Foreign borrowers may differ systematically from domestic borrowers, and banks that lend abroad may be larger, more experienced, or better funded than banks that remain local. If those differences are not separated, an observed price or relationship gap could be attributed to the border when it actually reflects the people and institutions on either side of it.
The paper addresses this problem with within-bank and within-firm comparisons. In plain language, it asks questions such as: does the same bank form relationships with firms abroad less often than with comparable domestic firms? Does the same firm connect with foreign banks less often than with domestic banks? For pricing, does the same bank charge different rates to domestic and foreign firms, and do the same firms receive different rates from banks in different countries?
Bank and firm fixed effects absorb persistent characteristics of the two sides. A bank’s funding model, risk appetite, scale, and lending technology may make it more likely to operate internationally. A firm’s productivity, size, risk, or financing demand may make it more likely to attract a foreign lender. Holding those persistent differences in the comparison makes the country-pair component more informative.
The identification is a network argument, not a single before-and-after comparison. The data contain banks that lend to several firms and firms that borrow from several banks. Those overlapping links create local comparisons that can be chained across the network. In the simplest case, two banks from different countries each lend to two firms from different countries. Comparing each bank’s foreign-minus-domestic gap and then comparing those gaps removes the bank and firm effects. Longer overlapping paths perform the same function when a neat four-link example is not available.
This also explains an important technical boundary. A country-pair effect is not identified as an absolute level on its own. Country-wide origin and destination conditions can be shifted between the pair effect and the fixed effects without changing the fitted relationships. The economically meaningful object is a normalized difference—such as a pair’s deviation from domestic lending—not an isolated number that can be read as the complete cost of operating in that country.
The relationship equation has an additional challenge: most possible bank-firm pairs are zeros. Keeping only observed relationships would make the problem look like pricing conditional on a match and would erase the very outcome the authors want to study. The paper retains observed links and samples non-links, then uses inverse-probability weights so the weighted comparison represents the larger set of possible matches. The purpose is computational feasibility without conditioning away relationship failure.
Entry requires a different version of the same logic. A banking group is observed to earn profits in markets where it operates, but its profits in markets it did not enter are not directly observed. The paper uses group and destination effects and also model-implied profits to compare foreign entry with home operation. The result is therefore conditional on assumptions about how unobserved foreign-market profitability can be represented. It is not a direct observation of a regulatory fee charged at the border.
The sequence shows why the data need separate equations. If the group stays home, the firm has no foreign operating presence to choose. If the bank enters but the relationship does not form, there is no foreign interest rate to compare. Only in the final branch do pricing and quantity become observable. The absence of a price is not a missing detail to be filled in with an average; it is part of the access problem.
A large gap before a small one
The relationship-formation estimates are the clearest evidence that the main barrier occurs early. In the preferred specification, cross-border relationships are about 3 log points, or 95%, less likely than domestic relationships. An alternative level specification expresses the gap as roughly 70 percentage points lower probability. These are different scales, not two effects to add together. The important conclusion is that the shortfall is large and broad.
The result also survives a meaningful composition check. Including bank and firm fixed effects does not materially change the wedge compared with a specification using only bank-country and firm-country effects. That stability suggests that the broad foreign-match gap is not merely the result of large firms choosing large banks. Large firms and banks are more likely to be visible in cross-border activity, but their selection does not explain away the country-pair shortfall. The pair estimates still vary: the reported standard deviation is about 24 percentage points around an unweighted mean gap of 65 percentage points, while the gaps remain overwhelmingly negative.
The pricing result is much smaller. Among existing bank-firm relationships, the preferred fixed-effect estimate places cross-border interest rates about 28 basis points below domestic rates, roughly 8.5% of the sample mean rate. Some country pairs are slightly positive and others slightly negative. The paper’s point is not that every foreign loan is cheaper. It is that the conditional pricing difference is modest and can even favor foreign links, in sharp contrast with the large probability gap that determines whether a link exists.
The quantity result sits between those two in direction but closer to pricing in economic size. Conditional on a relationship, the foreign bank’s share of a firm’s borrowing is estimated to be about 14% lower than a domestic bank’s share in the preferred specification. The gap is negative across most pairs, but it remains small relative to the barrier to forming the relationship. A firm may therefore receive somewhat less credit from a foreign bank after matching, while the much larger loss occurs when matching never happens.
Bank entry shows that the same early-stage pattern exists on the supply side. Across specifications, the cross-border entry gap is about 99 percentage points. Because a group necessarily operates in its home market, the estimate is interpreted as an average implicit tax on foreign relative to domestic banking. It says that foreign establishment is far less likely than home operation under the paper’s assumptions, not that every group faces an identical literal tax. Alternative estimators use different scales, so the percentage-point interpretation belongs to the stated linear-probability result.
Taken together, the estimates establish an ordering. Entry and relationship formation are the dominant extensive-margin barriers. Conditional pricing and loan quantity matter, but they are smaller once the bank and firm have already connected. This ordering resolves the apparent contradiction in the title: similar prices can be real for realized cross-border loans and still coexist with a fragmented market because the observed loans are the survivors of a much larger matching process.
What the estimates can—and cannot—tell us
The fixed-effect design improves on a raw average, but it does not turn the exercise into a randomized test. AnaCredit does not observe offers that a firm rejected or offers that a bank never made. If foreign banks quote different terms to firms that never become relationships, the conditional pricing estimates cannot capture that difference. The relationship estimate may also be affected by unobserved factors that influence both a match and the terms that would have been offered.
The authors test the intensive-margin results against several time-varying and granular controls, including bank-by-time shocks, industry-by-location-by-size-by-time conditions, bank-industry specialization, and bank-firm size bins. These checks make the interpretation more credible, but they do not remove every possible omitted variable. The pair effects remain estimates under a specified model of comparison, not a direct reading of an invisible border toll.
The entry estimate has its own boundary because non-entry does not reveal the profits a group would have earned abroad. Geographic distance, language, culture, information, expected demand, and group strategy can all influence the decision. The paper’s later regulatory analysis helps explain why the estimated wedges vary, but the wedge measurement itself does not declare one cause.
The right way to use these results is therefore diagnostic. If a policy discussion cites similar cross-border rates, ask whether it is describing the conditional pricing margin only. If it cites a low foreign-credit share, ask whether the denominator counts loan volume, relationships, or group nationality. And if it proposes regulatory harmonization, ask whether the mechanism should affect entry, relationship formation, pricing, quantity, or several margins at once.
The estimates locate Europe’s main integration problem before the interest rate is set. The next question is why the early-stage gaps line up with national regulatory differences—and how much confidence that association deserves before it is carried into a macroeconomic model.