First, Find the Shock: When Credit Supply Changes
This chapter focuses on the first causal link in the credit-supply-to-productivity relationship: how to tell that a firm received less or costlier finance because its lenders pulled back, rather than because the firm itself wanted to borrow less. That distinction supports a practical judgment. If the goal is to understand whether a credit crunch can damage productive capacity, the relevant comparison is not simply firms that borrowed more with firms that borrowed less. It is firms facing different lender-side conditions while their own demand, market, and pre-existing characteristics are held as constant as the data allow.
Why borrowing data do not identify credit supply
Credit observed on a balance sheet is an equilibrium result. It reflects what a firm wants, what a lender is willing to provide, and the conditions under which the two negotiate. A firm may draw less on a credit line because it expects weak sales. A bank may reduce that same line because its capital has deteriorated. A recession can cause both events at once. If a researcher treats the change in borrowing as the shock, the resulting productivity estimate may simply show that firms expecting trouble both borrowed less and performed worse.
The distinction is between a borrower-side demand shock and a lender-side supply shock. Demand concerns the firm’s desired quantity of finance at prevailing terms: its investment opportunities, expected sales, leverage, and need for working capital. Supply concerns the amount and price of finance a lender is prepared to offer, given its own capital, funding costs, asset risks, and lending strategy. In practice, the two are not observed separately in a single loan balance. The studies therefore use the network of existing lender relationships to find variation that comes from banks rather than from the borrower’s contemporaneous decision.
This strategy works because relationships are not frictionless connections that can be replaced instantly. A bank that has lent to a firm knows something about its payment history, operations, and creditworthiness. The firm, in turn, already has the contractual and informational infrastructure to borrow from that bank. If that lender contracts, the firm may be able to approach another bank, but the alternative may be slower, more expensive, or unavailable. A lender-side change can therefore have a disproportionate effect on its existing borrowers. The relationship is the transmission mechanism; it is not, by itself, proof that the transmission came from the lender.
The Italian study makes this logic explicit in its bank-firm data. It observes the growth of credit granted by each financial intermediary to each firm and decomposes that growth into a common time component, firm-year factors, bank-year factors, and a remaining match-specific component. The bank-year component captures how a particular lender changes its supply relative to other lenders, after accounting for changes shared by its borrower and the wider economy. The authors then aggregate the shocks of a firm’s pre-existing lenders using each lender’s share of the firm’s previous-period credit. A firm borrowing mostly from a contracting lender receives a more negative firm-level supply shock than a firm borrowing mostly from lenders that are expanding.
That weighting is economically meaningful. It does not assume that the firm has no alternatives. It assumes instead that the pre-existing relationship determines how strongly a lender’s change reaches the firm. The Italian paper also uses credit granted rather than credit used, because usage is more directly affected by a firm’s current demand. The resulting measure is defined for firms with credit relationships, while the bank-level shocks are estimated from borrowers with multiple lenders so that the researcher can compare how banks treat the same borrower at the same time.
The New York Fed study uses a different crisis setting but the same underlying logic. It examines Belgian manufacturing firms around the April 2010 request for an EU/IMF bailout by Greece, an event that sharply changed the perceived risk of sovereign bonds issued by peripheral European countries. Banks entered the episode with different holdings of these distressed GIPSI securities—those issued by Greece, Italy, Portugal, Spain, and Ireland. A firm’s exposure is the credit-weighted average of the GIPSI exposure of the banks lending to it in 2010:Q1, the quarter before the bailout request. Because the holdings are measured before the shock, they are not the banks’ reaction to the crisis itself.
In this design, two otherwise similar firms can face different lender-side pressure because their existing banks held different amounts of vulnerable sovereign debt. The event is common, but exposure is heterogeneous. The mechanism begins with the bank’s balance sheet: the repricing of sovereign securities reduces the value or collateral usefulness of assets and makes it harder for exposed banks to sustain lending. It reaches the firm as a contraction in credit availability. A one-standard-deviation difference in exposure corresponds to about an 18 percent cumulative reduction in total bank credit over the three years after the shock. The financing-cost proxy—financial charges relative to prior-period financial debt—eventually rises by about 3.5 percent for the more exposed firm. Because quantity and cost move in opposite directions, the pattern is consistent with a supply tightening rather than merely lower borrower demand.
The financing-cost result needs a qualification. The Belgian data do not contain the exact interest rate on every bank-firm loan, so the study uses a firm-level accounting ratio as a noisy proxy for the average cost of finance. That proxy is informative about direction and timing, but it should not be read as a precise quoted rate. This is a useful example of why identification and measurement must be evaluated separately: a credible lender-side shock can still be paired with an imperfect outcome measure.
The diagram separates the trigger from the outcome. The crisis is the trigger; the bank’s balance-sheet condition is the immediate mediator; the firm’s credit availability is the treatment that can influence prices, investment, employment, and productivity. The researcher’s task is to show that the treatment varies for reasons plausibly independent of the firm’s unobserved performance trajectory.
How lender exposure becomes a testable shock
The first diagnostic is a pre-trend check. Suppose firms attached to more exposed banks were already on a different productivity path before the crisis. Then a post-crisis difference could reflect that pre-existing trajectory rather than the lender shock. In the Belgian study, firms borrowing from more and less exposed lenders appear balanced on several pre-shock characteristics, including size, leverage, productivity, prices, inventories, and financial distress. The dynamic credit estimates show no differential pre-shock trend. In the Italian interbank experiment, firms more exposed to the 2007–08 freeze likewise show no statistically different productivity growth before the collapse, and their pre-crisis sensitivity to business-cycle fluctuations is not significantly different.
Balance on observables does not prove balance on everything that matters. The stronger test in the New York Fed paper compares different banks lending to the same firm. If one firm simultaneously borrows from several banks, a firm-specific decline in expected sales or investment demand should affect its demand for all of them. Yet the study finds that banks with larger pre-crisis GIPSI holdings reduce lending to that same firm relative to less-exposed banks. Adding firm fixed effects greatly increases the explanatory power of the relationship-level regressions while leaving the exposure coefficients broadly similar. This pattern supports the interpretation that the difference originates with lenders, not with a borrower-wide change that happened to line up with bank exposure.
The Italian paper approaches the same concern through a broader panel decomposition and several validation exercises. If its estimated supply shock were actually contaminated by a firm’s desire to seek more or less credit, a positive “supply” shock might be associated with more applications to new lenders. The data show the opposite: firms receiving positive supply shocks are less likely to search for new lenders, both in whether they apply and in how many applications they make. That result is consistent with a borrower replacing less of its existing funding when current lenders are providing more.
The natural experiment around the Italian interbank market adds a different kind of protection. Interbank funding was a major source of bank liquidity before the crisis. Italian banks’ gross interbank transactions fell from about €24 billion in 2006 to €4.8 billion at the end of 2009, while the Euribor–Eurepo spread, close to zero until August 2007, rose above 50 basis points in the following year and widened further after Lehman’s failure. Banks differed in their pre-crisis reliance on this market. The study uses the credit-weighted average interbank dependence of a firm’s 2006 lenders as a time-invariant exposure measure and examines the 2007–09 credit-crunch window.
Firms whose lenders relied more heavily on interbank funding experienced lower revenue and value-added productivity growth during the crunch. The estimated effect is larger than the baseline relationship from the general panel, suggesting that reliable credit is especially valuable when the financial system itself is under stress. The identification is strengthened by placebos: exposure to the later freeze does not predict productivity growth in the 2004–06 pre-period, and a hypothetical exposure measure based on 2002 does not predict outcomes over 2003–05. A placebo cannot prove that every remaining confounder is absent, but a correctly timed effect with no comparable pre-event effect is difficult to reconcile with a simple pre-existing matching story.
The New York Fed design uses a similarly timed external event. The Greek bailout request did not arise from the Belgian firms’ own production decisions, and pre-bailout sovereign holdings were fixed before the event. The authors also include bank-level controls—such as size, funding structure, liquidity, capitalization, and lending-portfolio quality—because sovereign exposure may be correlated with a bank’s broader willingness to adjust lending. Firm-level controls capture the number and length of lending relationships. Industry and region fixed effects limit the possibility that exposed banks simply specialized in places or industries hit harder by a local demand shock.
These controls do not transform the estimate into a laboratory experiment. They make a specific comparison more credible: among firms operating in the same broad industry and region, with observed differences in lender relationships and bank characteristics accounted for, did exposure to the distressed-asset shock predict a change in credit and subsequent outcomes? The answer in both settings is yes. The causal interpretation is conditional on the design’s assumptions, but it is stronger than an unadjusted correlation between debt and productivity.
A further issue is what happens when a firm changes lenders. The New York Fed results show that credit supplied by the exposed pre-shock lender remains depressed at the firm-bank level, even though total firm credit is more transitory. This suggests that firms gradually establish new relationships that compensate for lost funding. That fact does not invalidate the original exposure measure. It clarifies its meaning: it captures the initial shock transmitted through legacy lenders and the firm’s ability to replace it over time, not a permanent assignment to one bank.
Both studies also test whether the additive treatment of lenders is too simple. A firm might respond to one bank’s contraction by drawing more from another bank, so the supply shock of the other lender could matter. The Italian authors add controls for substitute lenders and for match characteristics such as relationship length, loan shares, interest rates, credit instrument, and past non-performing loans. The resulting shocks are highly correlated with the baseline estimates, and the productivity results remain similar. This is reassuring in the data at hand, although the paper notes that substitution patterns could matter more in concentrated markets such as syndicated lending. The Belgian within-firm comparisons provide a complementary check rather than a general solution to every form of lender interaction.
What the coefficients mean—and what they do not
A standardized exposure coefficient is a comparison unit, not a forecast. In the Belgian study, the reported credit, price, and productivity coefficients describe the difference associated with a one-standard-deviation increase in the firm’s pre-shock exposure to lenders holding distressed sovereign securities. They do not say that every 1 percent fall in a firm’s credit produces the same response, nor that the same estimate applies to a firm with a different financing structure. In the Italian study, the firm-level shifter is scaled so that a 1 percent increase in the supply shock is the change in supply associated with roughly a 1 percent increase in credit granted; its productivity coefficient then describes the average response to that normalized variation.
The denominator and horizon matter as much as the coefficient. A cumulative three-year credit response is not the same object as a one-year flow. A productivity estimate based on revenue is not interchangeable with one based on physical quantities. A firm-level response is not an aggregate welfare loss. Reading the number without its exposure definition, comparison group, period, and outcome measure would discard the very design that makes the estimate meaningful.
The most defensible conclusion at this stage is therefore narrow but important. The evidence supports a lender-driven credit-supply shock in both settings, and it shows that the shock reaches firms through sticky financing relationships. The studies do not show that every fall in credit is supply-driven, that firms cannot find alternatives, or that bank exposure is unrelated to every unobserved characteristic. They show that several independent diagnostics—pre-trends, within-firm lender comparisons, application behavior, alternative controls, and crisis-timed placebos—make the supply interpretation substantially more credible.
That gives us a sound starting point for the next problem. Once the shock has been isolated, the remaining question is what a measured productivity response actually represents. If a firm’s revenue per input changes, is the change in physical efficiency, in the price it charges, or in the demand and quality characteristics folded into revenue? Without that distinction, even a well-identified credit shock can be assigned the wrong timing and the wrong economic meaning.