Did the FinTech Loan Cause More Bank Credit?
The central question here is whether a junior unsecured FinTech loan caused small firms to borrow more from banks later, or whether both loans were simply responses to an investment opportunity that researchers could not observe. Answering it matters because the collateral story is causal only if the bank-credit increase follows from the first loan rather than from pre-existing demand. The French study uses timing, comparison groups, and variation in banks’ collateral requirements to make that distinction more credible.
The before-and-after trap
A simple chart could seem to settle the issue. Draw a line for a firm’s bank credit before and after it receives a FinTech loan. If the line rises afterward, one might conclude that the FinTech loan unlocked the bank credit. But the firm may have applied to several lenders at once. It may have needed money for a profitable expansion, bought the first loan available, and received a bank loan a few months later. In that case, the FinTech loan would be earlier in time but not the cause of the later bank borrowing.
This is the familiar problem with before-and-after comparisons: time order is necessary for a causal story but not sufficient. The firm’s demand for credit could be the hidden factor that produces both events. A firm whose sales are growing, whose equipment is aging, or whose owner has identified a new contract may seek funding from every plausible source. A FinTech platform may approve quickly while a bank takes longer. The later bank loan could then reflect the opportunity that prompted both applications, not a new willingness to lend created by the FinTech loan.
The study is designed around this concern. It follows firms over time using French administrative and commercial data. The researchers observe FinTech loans, bank-firm credit relationships, firm characteristics, individual bank loans, specific-asset collateral for a subset of loans, and later bankruptcy or liquidation procedures. The main FinTech sample contains 2,013 applications from 10 platforms, representing more than 80% of FinTech lending to French SMEs in the relevant data. The loans were originated before July 2019, which gives the researchers a meaningful post-origination window and keeps pandemic-era lending outside the main analysis.
The outcome is also defined carefully. When the study asks whether bank credit increased, it excludes the outside loan itself—the FinTech loan for a FinTech borrower or the new bank loan for a bank-borrower comparison. The result therefore concerns credit from the firm’s other bank relationships. It is not a mechanical statement that total debt rose because a new loan was added. That distinction is essential: the question is whether other banks changed their lending after the outside loan.
The timing pattern is consistent with a response rather than an already-visible trend. In the author-hosted version, researchers observe 12 months before and 24 months after the outside-loan origination. Bank credit remains broadly flat in the pre-period. After a FinTech loan, it rises gradually, with the detailed version describing an average increase of 8% and a plateau of about 20% six months after origination. The rise is not a one-day jump caused by counting the FinTech loan; it appears in the credit relationships that the researchers continue to measure separately.
That pattern still does not prove causation. A firm can have an unobserved investment plan that begins to affect bank borrowing precisely after the FinTech loan. The study therefore needs firms that resemble the FinTech borrowers but did not receive the same kind of loan.
What the comparison groups add
The first comparison group consists of firms that obtained a new bank loan. This is not a comparison between FinTech borrowers and an arbitrary set of firms with no borrowing. The bank borrowers also sought an outside loan from a new lender, and the study matches the outside loans on features such as size and year. The idea is to compare two firms with a similar reason to approach a new creditor, then ask whether their subsequent relationships with other banks evolve differently depending on whether the outside loan came from a FinTech platform or a bank.
The researchers use propensity-score matching within industry and size categories. The matching incorporates recent bank-credit dynamics, the size and year of the outside loan, whether the firm had a credit line, and characteristics such as age, rating, liabilities, tangible assets, and earnings before interest and taxes. The details are not important because matching magically makes firms identical; they matter because they show which observable dimensions the study treats as relevant to credit demand and lender choice.
The second comparison group consists of firms that applied for FinTech credit but were rejected. One platform supplied the researchers with the list of rejected applicants, totaling 30,539 firms between January 2014 and July 2019. These firms help address a different selection problem. A bank borrower may differ from a FinTech borrower because choosing a bank rather than a platform reflects the firm’s preferences, relationships, or expectations. A rejected applicant, by contrast, made the decision to seek FinTech credit but did not receive an approved loan. Comparing accepted and rejected applicants can therefore hold more of the demand for a FinTech product constant, although the groups are not identical in every unobserved respect.
The matching results make the post-loan pattern harder to explain as a simple difference in borrower quality. In both matched samples, FinTech borrowers and their benchmarks have broadly parallel bank-credit dynamics during the 12 months before the outside loan. After origination, FinTech borrowers’ bank credit grows faster and the gap persists. The study also examines credit-line utilization, a measure related to a firm’s ongoing demand for credit. The trends remain parallel before and after the outside loan; if anything, FinTech borrowers use their lines slightly less than rejected applicants.
Why do parallel pre-trends matter? Imagine two runners whose speeds begin diverging before the event being studied. If one runner is already accelerating, a later difference cannot confidently be attributed to the starting gun. Parallel pre-trends do not show that the runners are identical, but they make it less likely that the treatment group was already on a different trajectory. Here, the absence of a differential pre-trend supports the interpretation that something changes after the FinTech loan rather than merely being revealed by it.
The comparison groups also impose limits. Matching can balance measured characteristics; it cannot directly balance a firm’s private information about an upcoming contract, an owner’s unrecorded plan, or an informal relationship with a lender. The identifying assumption is that observably similar firms have sufficiently similar credit demand and growth opportunities. This is a reason to treat the design as stronger than a raw before-and-after comparison, not as a substitute for random assignment.
The study strengthens the case with several robustness checks, including alternative matching procedures and different fixed effects. Those checks do not remove the underlying assumptions, but they reduce the chance that the central result is an artifact of one particular matching recipe. The convergence of two benchmark groups is especially useful because the groups fail in different ways: bank borrowers may resemble FinTech borrowers in their pursuit of a new lending relationship, while rejected applicants resemble them in seeking a FinTech product.
The substantive conclusion at this stage is therefore narrower than “FinTech lending improves access to credit.” In the French sample, accepted FinTech borrowers later obtain more bank credit than comparable firms, and the timing and benchmark patterns make credit demand alone an incomplete explanation. The next task is to find an outside source of variation that changes the chance of turning to FinTech credit without being driven by the individual firm’s immediate demand.
Using banks’ collateral requirements as a source of variation
The study’s second strategy begins with a behavioral prediction. Firms should be more likely to seek a junior unsecured FinTech loan when the banks they already use are more likely to require collateral for new loans. If a bank insists on a pledge that the firm cannot provide—or that would make future financing less flexible—the platform’s unsecured product becomes relatively more attractive.
To measure this pressure, the researchers first calculate, for each bank and quarter, the fraction of newly issued SME loans that are secured. They then combine that bank-level measure with each firm’s earlier credit exposure across banks. A firm whose existing borrowing is concentrated at banks that suddenly issue more secured SME loans receives a higher exposure score. In plain language, the “share” is where the firm already banks; the “shift” is how those banks’ collateral practices change at the relevant time.
This construction is useful because the firm’s own credit demand does not directly determine the contemporaneous secured-loan fraction at each bank. The strategy assumes that, absent the change in bank collateral requirements, firms with greater exposure to banks imposing stricter collateral terms would have followed similar credit paths to firms with lower exposure. That is a demanding assumption. Banks’ collateral policies could be related to their broader credit supply, local economic conditions, or the kinds of firms they serve. The study addresses one such concern by controlling for bank credit supply, constructed from the volume of new SME loans issued by the banks to which each firm was exposed.
The first-stage result follows the proposed behavior: firms exposed to a one-standard-deviation increase in the collateral-requirement measure are more likely to receive a FinTech loan relative to a bank loan. The reported increase in that probability is 0.024 percentage points. The reduced-form estimate then indicates an 8.3% increase in bank credit associated with the same one-standard-deviation collateral shift. The key logic is not that collateral requirements directly create bank credit. They make the unsecured FinTech route more likely, and the subsequent bank-credit response follows through that route.
Placebo tests make the timing argument more informative. When the study replaces the contemporaneous bank secured ratio with a lagged ratio or with the average ratio from earlier periods, the estimates become economically small and statistically insignificant. If the result were merely picking up the fact that certain firms bank with certain institutions, older collateral measures might predict the same outcome. Their lack of predictive power is consistent with the current collateral-requirement change, rather than fixed bank exposure alone, helping to drive the result.
Still, an instrument is not a magic word that turns an observational study into a randomized experiment. The strategy relies on the bank-level collateral shifts being plausibly unrelated to unobserved firm-level changes in credit demand. It also relies on the shift affecting later bank credit primarily through the increased likelihood of FinTech borrowing, rather than through another channel that independently changes the firm’s access to credit. The study’s controls and placebo tests support the design, but they do not make those assumptions disappear.
The headline magnitude also needs careful handling. The detailed author-hosted version reports a 20% increase in bank credit in its abstract and matched comparisons, while the Crossref record for the published Review of Financial Studies article reports 13%. This may reflect a version or specification change. The evidence supports a meaningful positive response in the studied setting, but it does not justify presenting 13% or 20% as a single uncontested estimate without identifying the version behind it.
What survives these qualifications is the result’s structure. Bank credit is flat before the outside loan, rises afterward for FinTech borrowers relative to two different benchmark groups, and responds to a bank-collateral shock that makes the FinTech route more attractive. That combination makes the causal complementarity claim more credible than the simple story that FinTech borrowers were merely more optimistic or more eager to borrow. It does not yet explain the physical bridge between the two loans. For that, the decisive question is whether the additional bank credit appears where collateral should matter—against assets that the first, unsecured loan helped the firm acquire.