What the Data Reveal: Home Bias, Sparse Networks, and Two Kinds of Integration
For the incomplete European Banking Union, the first empirical question is not whether a cross-border loan has a competitive interest rate. It is what counts as a cross-border connection, who is connected to whom, and which banks are actually present in a market. This chapter maps those units and the patterns they produce so the reader can interpret Europe’s home bias without confusing a large foreign loan, a foreign-owned bank, and broad access to foreign lenders.
Three levels of the banking map
The paper combines three kinds of evidence because no single dataset captures the whole banking system. The first level is the loan relationship: a bank-firm pair with positive outstanding credit. This is the level at which a borrower and a lender meet, negotiate, monitor one another, and decide how much credit to maintain. A relationship is more than a transaction recorded in a ledger. It is an ongoing connection between a particular firm and a particular bank.
The second level is the banking-group network. A bank may be a standalone institution, a branch of a foreign parent, or a subsidiary belonging to a wider group. The question here is not whether one firm has borrowed from one foreign institution, but whether a banking group has established an operating presence in another country. That presence can enlarge the local supply of potential lenders before any individual firm chooses a bank.
The third level is the country pair. This is where the paper studies differences between national banking environments: rules, supervisory practices, deposit protection, resolution, entry conditions, and other features that can make one cross-border direction easier or harder than another. A country pair is not itself a loan or a bank. It is the setting in which many possible relationships and entry decisions take place.
The empirical design links these levels. AnaCredit provides detailed cross-country credit information, including loan amounts, maturities, interest rates, collateral, default probabilities, sectors, and locations. The paper’s final real-sector sample spans the fourth quarter of 2019 through the second quarter of 2025. RIAD supplies ownership, control, branch, and subsidiary links, allowing the authors to identify the ultimate parent of a bank and reconstruct banking-group expansion. Bank-level balance-sheet information comes from iBSI, which covers roughly 95% of euro-area banking-sector assets.
Each source answers a different question. AnaCredit can show whether a particular bank and firm have an active credit relationship. RIAD can show whether the bank belongs to a group that has entered a foreign market. iBSI can place those relationships inside a bank’s wider balance sheet of loans and deposits. Combining them prevents a country from appearing integrated merely because a foreign group owns a local entity, or appearing closed merely because a particular loan happens to be booked domestically.
The observation rule matters as much as the database name. A positive relationship is evidence that credit exists, not evidence that the firm compared every possible lender. A bank-group link is evidence of ownership or control, not proof that all of the group’s local decisions are made at headquarters. And a country-pair measure summarizes an environment; it cannot reveal which individual rule stopped a particular loan. Keeping those boundaries visible is what allows the descriptive map to motivate, rather than replace, the later analysis of barriers.
Two definitions of crossing a border
The paper uses two definitions of “cross-border,” and the distinction is substantive rather than cosmetic. Under operational geography, a relationship crosses a border when the immediate lender and the firm are physically located in different countries. This is the relevant definition for relationship formation and loan pricing because lending work is treated as local: bank staff gather information, monitor borrowers, and operate under the legal setting where the lender and firm are located.
Under ownership nationality, a relationship is treated as foreign when the bank’s ultimate parent is based in another country. This is better suited to studying group entry. A group deciding whether to establish a branch or subsidiary is deciding where to deploy an organization and balance sheet, not merely where a particular loan officer sits.
Consider a foreign banking group with a local branch. From the operational perspective, a loan made by that branch to a local firm is a domestic relationship: the branch and the borrower are in the same country. From the ownership perspective, the same loan is supplied by a foreign group. Both statements are true, but they describe different forms of integration. The first concerns the local production of lending; the second concerns who controls the institution supplying it.
This is why the paper does not choose one definition and treat the other as an error. The operational measure is appropriate when the question is whether a firm can connect with a bank across a physical and institutional boundary. The nationality measure is appropriate when the question is whether a banking group has crossed that boundary through organization and ownership. Using the wrong measure can answer a different question while still producing a plausible-looking percentage.
The denominator changes with the definition as well. A local branch owned by a foreign group can count as domestic under the operational measure and foreign under the nationality measure. Reassigning the same balance-sheet activity therefore changes the measured share without creating a new loan or a new branch. The result is not an inconsistency to be averaged away; it is information about the difference between local delivery and foreign control.
The alternative nationality-based remapping does raise measured integration. The cross-border share rises from roughly 5%–5.6% under the operational reading to 11% of loans and 14.6% of loan amounts. Smaller economies show some of the largest changes because local branches may be controlled by groups headquartered elsewhere. Yet the result does not erase home bias. The diagonal of the country-by-country lending matrix remains much larger than the off-diagonal positions, and the remapped matrix can become sparser because activity attributed to a local lender is reassigned to the parent’s country.
The distinction also clarifies why “foreign bank” is not a single economic condition. A local subsidiary may be foreign-owned but subject to more host-country control over governance, reporting, resolution, supervision, and capital or liquidity. A branch remains part of its parent bank and can preserve more group-wide balance-sheet mobility. For the descriptive map, both are foreign establishments. For understanding how regulation shapes entry and lending, their organizational form can matter.
The diagram’s important interface is between the two views. The bank-firm relationship is attached to the operating locations of the lender and borrower when the paper studies matching and prices. The banking-group entry is attached to the group’s origin and destination when the paper studies expansion. The same institution can participate in both descriptions without making them interchangeable.
Home bias is visible at both the aggregate and relationship levels
Once the units are separated, the descriptive pattern is clear. Cross-border lending to firms averaged 5.6% of total euro-area lending from 2019 onward under the operational definition. At the relationship level, cross-border pairs accounted for less than 1% of relationships. Those two facts are not contradictory. A small number of very large relationships can produce a meaningful amount of cross-border lending while leaving most firms connected to domestic banks.
The percentages answer different questions. The 5.6% share uses lending volume as its denominator: it asks how much credit is foreign relative to all credit. The less-than-1% figure uses relationships as its denominator: it asks how often a bank-firm link is cross-border. A market can have a low count of foreign links and a larger foreign volume if the few links that exist are unusually large. Neither percentage alone tells us the probability that a randomly selected firm can obtain foreign credit.
The paper reports that cross-border relationships were, on average, about ten times larger in outstanding amount than the full set of relationships. But that average is driven by the right tail. Cross-border loans were also more likely to be syndicated, which is consistent with large or complex financing needs being more capable of supporting the costs of coordinating across borders. It would be a mistake to read the average amount as a typical foreign loan received by a typical firm.
The composition of the connected firms and banks reinforces that point. Roughly 40% of banks had at least one cross-border relationship, while very few firms borrowed across borders. Firms that did so were older and considerably larger in turnover, balance-sheet size, and employment. Banks engaged in cross-border intermediation were also substantially larger in assets, lending, and deposit-taking. Cross-border activity therefore has a selection pattern: the visible foreign relationships are disproportionately those involving institutions and borrowers with the scale or maturity to clear a demanding access hurdle.
The raw contract terms can make this selection easy to miss. Cross-border relationships show slightly lower average interest rates in the descriptive data, and the paper reports lower observed default probabilities for them as well. Neither fact, by itself, establishes that foreign banks offer a general price advantage. If safer, older, larger firms are the ones most likely to form cross-border links, their observed loans can look cheaper partly because the sample has already selected different borrowers and lenders. The relevant comparison must control for those differences, and even then it will describe realized relationships rather than rejected ones.
The geography of lending is uneven too. Ireland, Luxembourg, and Malta rely heavily on credit from banks elsewhere in the euro area, while Luxembourg, Ireland, and the Netherlands have among the largest foreign lending shares on the creditor side. Larger economies tend to show smaller foreign shares in part because their domestic credit markets are large, but the country-pair matrix also contains empty or very thin cells. In the 2024 third-quarter snapshot, for example, some banks had no lending relationships outside nearby countries. Integration is therefore not a smooth gradient from closed to open; it is a network of strong, weak, and missing links.
The supply network is sparse before firms make a choice
The same structure appears in bank-group entry. Most off-diagonal cells in the origin-country/destination-country map are empty. Except in five cases, fewer than five banking heads in each origin country moved across borders in the September 2024 picture. Foreign expansion is concentrated among groups from the largest euro-area economies and the Benelux financial hub. Groups from smaller countries often enter only neighboring markets or do not expand abroad.
Entry and lending should not be collapsed into one statistic. A group can enter without immediately creating many foreign relationships, and a firm can borrow from an institution whose parent is foreign while the actual lending operation is local. Even so, entry changes the supply-side conditions for later matching. It determines whether a foreign bank has staff, information channels, compliance capacity, and a balance-sheet commitment in the destination market. The entry map is therefore a constraint on the relationship map, not a substitute for it.
This matters because a firm’s choice set is partly determined before the firm makes a choice. A borrower cannot compare offers from foreign banks that have no operating presence, and a bank that has not entered a country may lack the local information or organizational capacity to build relationships at a distance. A sparse entry network can therefore reproduce domestic lending patterns even if the legal framework permits foreign capital and even if the rates on the few successful foreign loans are competitive.
The ownership view adds information without overturning the operational one. Reassigning local branches to their ultimate parents shows how large banking groups can reshape the apparent geography of credit. It can make smaller countries look more dependent on foreign banking groups and make the lending of large parent countries look more international. But it still leaves strong domestic concentration and low cross-border branching. The network is not integrated simply because ownership crosses borders; the practical reach of banks and the formation of firm-level relationships remain limited.
That is the empirical foundation for the next question. The descriptive data establish that aggregate shares, realized relationships, and foreign establishments measure different layers of the system. They also show why the apparent price advantage of some cross-border loans cannot settle the integration question: the firms and banks observed in those loans are selected from a much larger set of potential connections.
The issue is now precise. We do not merely need to know how much cross-border credit exists. We need to estimate where the missing links arise—when a banking group decides not to enter, when a firm and bank fail to form a relationship, or only after a relationship exists when the loan’s price and quantity are set. That separation is what allows the next analysis to distinguish Europe’s access problem from its contract-pricing problem.