How We Know What Changed in Lending
The Banco Popular resolution raises an empirical problem that the legal transfer alone cannot answer: after Santander acquired the failed bank, did firms receive less credit because banks reduced supply, or because firms changed what they wanted to borrow? The study addresses that problem by following the same firms across their bank relationships, then comparing Santander’s inherited borrowers with Santander’s own pre-existing customers. This design makes the main lending estimates more informative than a before-and-after change in Banco Popular’s loan book, while also making clear why the evidence about firm outcomes requires greater caution.
The comparison is the control
A bank-firm loan is the outcome of two decisions. The firm asks for financing, and the bank decides whether and how much to provide. If a loan balance falls after a resolution, the observation alone does not reveal which side changed. The firm may have less demand because sales weakened, or it may want credit but face a bank that has become more cautious. A credible study of Banco Popular therefore needs comparisons that hold as much of the borrower’s circumstances constant as possible.
The authors use three sources of Bank of Spain information. The Central Credit Register records lending relationships between Spanish firms and banks. Firm financial statements add measures such as sales, assets, cash, and employment. The Bank of Spain’s in-house credit assessment system supplies a consistent estimate of each non-financial firm’s probability of default. Together, these sources allow the study to observe not just whether a bank lent, but the borrower characteristics and risk categories associated with the relationship.
The baseline covers approximately 215,100 bank-firm lending relationships granted by 72 banks to 80,284 firms over nine quarters, from the second quarter of 2016 through the second quarter of 2018. The time window contains four quarters before and four quarters after the 7 June 2017 resolution. The unit of observation is important: the analysis is not asking only whether Santander’s total loan book rose or fell. It is asking how a particular bank’s committed credit to a particular firm changed relative to comparable relationships around the same event.
The strongest comparison comes from firms that borrowed from more than one bank. Consider a firm that had Banco Popular as one lender and another bank as a second lender before the resolution. If the firm’s demand for credit changes, that change is likely to affect its several relationships in some proportion. If the Banco Popular relationship is transferred to Santander and Santander’s lending moves differently from the other bank’s lending to the same firm, the difference is evidence about the bank side of the market. It is not perfect isolation—firms may not rotate demand proportionally across lenders in every crisis—but it is far more informative than comparing different firms with different needs.
The study creates two key borrower categories for the Santander comparison. An inherited borrower had a pre-resolution relationship with Banco Popular but not with Santander. A pre-existing borrower had a relationship with Santander but not with Banco Popular. This classification lets the analysis ask two related questions. Relative to other banks serving the same inherited firm, did Santander change its supply? And within Santander’s own portfolio, did inherited borrowers receive different treatment from customers Santander already knew?
The second question matters because a bank-wide lending number can conceal a reallocation. Suppose Santander keeps credit stable for newly inherited customers while reducing credit to some existing customers. A comparison only with other banks might show that inherited lending was protected, but it would not show where Santander found the capacity to do it. The inherited-versus-pre-existing comparison reveals whether the acquisition changed the composition of Santander’s lending, not merely its total volume.
Risk is another part of the comparison rather than a footnote. The study uses the Bank of Spain’s supervisory-based risk assessment to distinguish firms that were riskier before the resolution from safer firms. This timing matters. Risk is measured before the event, rather than defined by a later fall in lending. The question is therefore whether banks responded differently to borrowers already viewed as risky, not whether a borrower is called risky simply because the resolution harmed it.
The design can be read as two nested comparisons. The first holds the firm and time broadly constant while comparing the acquiring relationship with other lenders. The second holds Santander and the time period constant while comparing customers inherited through the transaction with Santander’s existing customers. Each comparison has a different interpretation, and neither should be replaced by a single headline about the bank’s aggregate lending.
What the fixed effects remove—and what they do not
The phrase “fixed effects” can obscure a simple idea. A firm-time fixed effect asks the analysis to compare the banks serving the same firm at the same time. Any factor shared by that firm in that period—its current demand for working capital, its general business prospects, and much of its borrower-specific risk—does not explain the difference between its lenders. The estimate then comes from how one bank relationship changes relative to another relationship for that same borrower.
The bank-firm fixed effect adds a different safeguard. Some relationships are naturally stronger than others. A bank may know a borrower better, have a longer history with it, or specialize in its sector. Those time-invariant features could otherwise look like a treatment effect. Holding the bank-firm relationship constant over time means that the estimate focuses on the post-resolution change rather than the fact that one relationship was stronger at the outset.
A second specification adds bank-time fixed effects. These absorb lending policies that affect all of Santander’s relationships in a given period, such as a general change in appetite for corporate credit or a bank-wide response to capital pressure. The remaining comparison is between the borrowers Santander inherited from Banco Popular and Santander’s pre-existing borrowers in the same bank-period. This is the specification that helps identify whether Santander prioritized the acquired relationships relative to its own established portfolio.
Those controls do not make the study a randomized experiment. They make a narrower causal interpretation more plausible: conditional on the assumptions, a relative change in Santander lending to an inherited borrower is unlikely to be explained solely by a demand change shared with the firm’s other lenders or solely by a bank-wide policy that affected all Santander customers. The estimate still depends on whether the comparison groups would have evolved similarly in the absence of the resolution and whether firms’ credit demand behaved sufficiently proportionally across banks.
The paper addresses the first concern with event-time estimates. Before the resolution, credit supply to borrowers of Banco Popular and Santander followed similar trends in the relevant comparison. After the event, their paths diverged. Similar pre-resolution trends do not prove that no unobserved difference existed, but a sharp break at the resolution is more consistent with an event-specific change than with two groups that had already been moving apart for unrelated reasons.
The timing of the failure also supports, rather than proves, the design. Banco Popular’s collapse was associated with a sudden and severe liquidity deterioration rather than a broad macroeconomic downturn. Market participants did not fully anticipate the resolution, according to the paper’s account. These features reduce the chance that firms, competing banks, and Santander had already adjusted their lending policies in response to a known transaction. They do not remove every anticipation or information problem, but they make the 7 June event a more useful shock than a slow deterioration whose consequences were widely priced in advance.
Several robustness checks test whether the comparison is being driven by a particular corner of the sample. The estimates remain stable when industries are excluded one at a time, which weakens the explanation that a sector-specific demand shock made Banco Popular borrowers look different. They also remain stable when municipalities are excluded one at a time, reducing concern that regional shocks or uneven bank specialization created the pattern. The study additionally checks whether firms shifted demand toward their main bank during uncertainty; including a main-bank interaction does not materially change the main results.
These tests are evidence against specific alternative explanations, not a certification that every assumption holds. A firm may seek more credit from one lender than another because that bank has special information or because the firm expects a relationship to continue. Banks may also specialize in sectors or localities in ways that are difficult to observe completely. The value of the design is that the authors identify these failure modes and examine whether the estimates survive reasonable attempts to expose them.
Why firm outcomes deserve a different confidence level
The bank-firm analysis is best suited to the question of credit supply because it can compare lenders serving the same firm. The firm-level analysis asks a larger question: did the resolution change total credit, sales, employment, investment, or cash holdings for firms with greater exposure to Banco Popular? That question matters for the real economy, but the design necessarily changes when the unit of observation becomes the firm.
A firm cannot be compared with another bank serving the same firm once the outcome is the firm’s total credit or its employment. Firm-time fixed effects would absorb the variation needed to estimate a firm-level post-resolution effect. The authors instead use each firm’s pre-resolution share of bank debt supplied by Banco Popular as an exposure measure. They combine that exposure with firm fixed effects, controls grouping firms by province, industry, and size over time, risk-by-quarter interactions, and controls for the type of the firm’s main bank.
This is a reasonable exposure-based strategy, but it leaves more room for contemporaneous developments to matter. The firm financial statements used for sales, employment, investment, and cash are annual, covering 2015–2019, whereas the credit register is quarterly. Annual observations are less precise around a resolution that occurred on one date. A change in investment measured over a year may reflect other developments that coincide with the event, even when the bank-firm estimates are more tightly linked to the lending shock.
The distinction between evidence of absence and absence of evidence is important here. If the study finds no significant fall in employment for more exposed firms, that is evidence against a large employment contraction detectable with its sample and controls. It is not proof that no firm lost a job, faced a financing problem, or experienced a short-lived disruption. Likewise, the study’s inability to observe an unmeasured channel does not establish that the channel did not operate.
Selection also affects applicability. Multi-bank firms provide the comparisons needed to isolate supply, but firms with only one bank have fewer immediate alternatives and may face higher switching costs. The paper’s extension that adds single-bank borrowers changes the average inherited-borrower result somewhat: lending to the broader group declines modestly, while the relative support for inherited borrowers—especially riskier ones—remains. That pattern is useful precisely because it shows that the baseline is not the experience of every borrower. Access to alternative lenders is part of the mechanism.
The evidence should therefore be read in layers. The matched bank-firm estimates provide the clearest basis for saying that the resolution changed relative lending supply and that the change differed by borrower risk and relationship status. The firm-level estimates provide evidence about whether those changes translated into total credit and real outcomes, but with coarser timing and stronger assumptions. The single-bank extension shows where dependence may make continuity less reliable. Together, these layers support a disciplined conclusion: the Banco Popular resolution can be associated with preserved credit intermediation without implying uniform protection for every firm.
This way of reading the evidence also prevents a common analytical mistake. A stable total loan balance could arise because no bank changed its behavior, because one bank expanded while another contracted, or because firms changed demand in ways that offset supply changes. Only by keeping the firm, bank, borrower category, risk measure, comparator, and time period visible can the estimates tell those stories apart. The next task is to apply that discipline to the result itself: which lenders withdrew, which borrowers were most exposed, and how Santander’s response differed from the surrounding market.