What the Evidence Actually Shows
The subject remains the banking group that can move corporate lending between a regulated bank subsidiary and an affiliated NBFI. The focus here is narrower: what does the IMF study actually compare, how does it identify a tightening shock, and how far can its estimates take us? Knowing the design lets readers judge the lending shift as evidence rather than as a plausible story that has been mistaken for a measured fact.
What the comparison isolates
A simple before-and-after comparison would not be enough. Loan volumes can change because firms want more or less credit, because the economy enters a downturn, or because unrelated lenders change their behavior. If bank lending falls after a regulatory announcement, that fall might reflect weaker borrower demand rather than a constraint on banks. The IMF study addresses this problem by working at the lender–borrower–quarter level and comparing bank and NBFI subsidiaries within the same banking group.
The group comparison supplies the first control. A parent may own both a regulated bank and a nonbank affiliate, so the two entities share an organizational relationship while facing different direct regulatory conditions. The study asks whether their lending responses diverge after a macroprudential tightening. It also includes firm-by-quarter fixed effects. In practical terms, the key comparison is between different subsidiary types that belong to the same parent and lend to the same nonfinancial firm in the same quarter. Borrower demand that affects both lenders is therefore held much more nearly constant than it would be in a comparison of two unrelated institutions.
That design does not make the lenders identical. A bank and an NBFI can differ in business model, funding, size and market presence. Those differences are part of the mechanism being studied: the question is whether the same parent uses those different entities differently when regulation tightens. The group fixed effects center the analysis on substitution within the group, while the borrower controls reduce the chance that a change in the borrower’s appetite for credit is being mislabeled as a lender response.
The scope is substantial but specific. The loan data cover activity from the first quarter of 2005 through the fourth quarter of 2023 across 27 lender countries: 21 advanced economies and six emerging-market and developing economies. The sample includes 963 banking groups, 4,871 subsidiaries and 52,380 nonfinancial firms. A separate ownership exercise reconstructs changes in ultimate control from 2000 through 2024. The analysis focuses on the leading bank and NBFI lending entities in the syndicated-loan data, which capture most of the lending by each type in that market.
The ownership map is important because legal names can change while control remains, or control can change through mergers, acquisitions and restructurings. A time-varying map helps the study attach a lender to the parent that controlled it during the relevant period rather than relying only on a static label. That is a demanding data task, but it still has a boundary: the map is built around entities active in syndicated lending. It is not a complete inventory of every company a financial conglomerate may own or influence outside this market.
The outcome is also narrower than the word “credit” can suggest. The study examines newly issued syndicated loans to nonfinancial corporations. Syndicated lending is an important market in which multiple lenders participate, but it is not the whole corporate-finance system. Private credit, bilateral loans and other forms of finance may behave differently. An estimate in this market can establish a relevant lending response without establishing what happened to every source of funding available to firms.
The second challenge is identifying the tightening itself. Authorities may tighten macroprudential policy because credit is already expanding too quickly or because financial conditions look dangerous. If so, a policy index could be correlated with the same economic forces that affect lending. The IMF’s baseline approach first constructs a country-level macroprudential stringency measure and then removes the part explained by lagged macroeconomic and financial conditions and country effects. The remaining movement is treated as the policy shock used in the lending analysis.
This residualization is not a magic word for causality. It is an attempt to separate the policy action from observable conditions that may have prompted it. The credibility of the interpretation still depends on the specification, the quality of the controls and the assumption that important remaining forces are not driving both the residual shock and the lending difference. That is why the paper supplements the baseline with a second approach rather than asking one index to carry the entire argument.
For six large lender countries, the authors assemble a high-frequency, announcement-based measure. The design uses the response of bank stocks around macroprudential announcements relative to the broader market, then brings those surprises into the lending analysis. A market reaction provides a different way to identify news about bank regulation: instead of inferring a shock from a country-level policy series alone, it asks how investors immediately reassessed the banking sector when the announcement arrived. The two approaches are not identical, but agreement between them would make it less likely that the baseline pattern is an artifact of one construction.
The evidence chain makes the logic visible. A policy shock is connected to a parent group. The study compares the group’s bank and NBFI subsidiaries while holding the borrower and quarter context as constant as possible. It observes a lending response and tests whether the response survives alternative specifications and a separate shock measure. The final node is deliberately bounded: internal transfers, entities outside the market and wider financial-stability effects are not directly measured by this chain.
From policy shock to bounded result
The baseline estimate supports a clear pattern. After a one-standard-deviation tightening in the study’s macroprudential shock, bank subsidiaries reduce syndicated lending by 1.0 percent. Affiliated NBFI subsidiaries increase lending by 2.0 percent relative to bank subsidiaries and by 1.0 percent in absolute terms. The relative number answers a substitution question: how does the NBFI subsidiary’s response compare with the bank subsidiary’s response within the group? The absolute number answers a different question: does NBFI lending itself rise, rather than merely fall less?
Both perspectives matter. If NBFI lending rose only relative to a much larger bank decline, the result could describe a relative shift without an expansion by the nonbank. Here the paper reports an absolute increase as well. That combination is consistent with the parent group using an affiliated NBFI to preserve or expand part of its lending capacity while the regulated bank pulls back.
The group-level calculation then applies the estimated responses to the average composition of lending in the sample. NBFI subsidiaries account for 28.4 percent of the group’s lending on average. With that weighting, the paper estimates a group-level decline of about 0.43 percent, compared with a 1.0 percent decline for bank entities alone. The implied mitigation is approximately 57 percent, or more than half of the contraction that would be observed if the bank response applied without the NBFI offset.
The denominator and comparator are essential. “More than half” does not mean that more than half of all corporate credit is supplied by NBFIs, nor that macroprudential policy loses more than half of its effect in every economy. It describes a weighted calculation for banking groups with the composition observed in this syndicated-loan sample, under the paper’s baseline shock. A reader who removes the lending share, the market and the shock definition has removed the information that gives the percentage meaning.
The evidence is also consistent with the mechanism rather than a direct movie of the mechanism. The authors do not observe every internal transfer between the parent bank and its NBFI affiliates. The estimates show that lending moves differently across subsidiary types after tightening. They do not show the exact contract through which funding was reallocated, the price of that funding or whether the parent formally instructed one entity to replace another. Internal capital markets and favorable affiliate funding are plausible explanations supported by the pattern and related evidence, but they remain partly inferred.
The results survive several forms of challenge described in the paper. Replacing firm-by-quarter fixed effects with a less restrictive borrower-demand control does not eliminate the reallocation. The response also appears along the extensive margin: NBFI subsidiaries are more likely to establish new lending relationships after tightening in the overall analysis, although the country-specific patterns are not uniform. The study tests alternative ways of imputing lender participation shares, includes further controls and fixed effects, changes the policy-shock construction, and examines different lags and clustering choices. These checks do not prove that no omitted factor exists, but they make a single fragile specification a less satisfactory explanation of the pattern.
The high-frequency analysis points in the same qualitative direction. Around lender-based macroprudential announcements, parent groups curtail credit through bank affiliates by about 0.9 percent and expand lending through NBFI subsidiaries by about 1.5 percent relative to bank subsidiaries. The response is again concentrated among US banking groups, while the selected UK and European estimates are not statistically distinguishable from zero in the reported analysis. The high-frequency exercise also finds a small additional increase in spreads charged by NBFI subsidiaries relative to banks. These results reinforce the direction of the baseline finding, while reminding us that the magnitude varies with the sample and identification method.
The country concentration is itself part of the interpretation. The study covers a broad international sample, but the strongest results are driven by US banking groups and, to a lesser extent, euro-area groups. A pooled estimate can therefore describe an average relationship that is much more relevant in some regulatory and organizational environments than in others. The result should not be transported mechanically to a country whose banking groups have different ownership structures, NBFI markets or consolidated-supervision rules.
The composition of the nonbank sample adds another qualification. Investment banks and broker-dealers account for most of the observed NBFI response. Investment funds and asset managers do not appear to expand syndicated lending after tightening. That does not establish that those funds never increase credit. They may be more active in private or bilateral markets that the data do not observe, or their reliance on investor capital may make them less able to expand in this particular market. What the evidence supports is a market-specific subgroup result: the affiliate route is most visible among NBFIs already positioned to participate in syndicated lending.
The risk tests are similarly precise. NBFIs lend, on average, to riskier borrowers than banks, but the incremental lending response after a macroprudential shock is not disproportionately concentrated in the risky-borrower categories tested by the study. Those categories include measures based on default probability, leverage, loan spreads, firm age, vulnerability and zombie status. In selected specifications, the additional response is more consistent with lending toward less risky borrowers. The evidence therefore weakens a simple story in which regulation automatically causes the group to chase the most fragile borrowers through its nonbank arm.
It does not support the opposite claim that the shift is safe. The risk variables are proxies for borrower characteristics, not a full measure of financial risk. They do not capture every form of leverage, the liquidity of the assets held by an NBFI, derivatives and market-making exposures, common collateral or the stability of the affiliate’s funding. The paper itself says that it documents lending regularities rather than directly assessing the financial-stability implications of the reallocation. Absence of a detected increase in the tested borrower-risk measures is not evidence that no other risk increased.
Pricing offers another useful but limited signal. Both bank and NBFI subsidiaries pass some higher regulatory cost through to borrowers, but the additional spread response for NBFIs is small relative to banks in the baseline and remains small in the high-frequency exercise. NBFIs already charge higher spreads on average. A higher spread can compensate a lender for risk or funding cost, but the observed adjustment does not tell us by itself whether the borrower, parent group or wider system ultimately bears the risk. It is a pricing observation, not a complete welfare or stability calculation.
The design also has selection limits that matter for interpretation. The ownership sample emphasizes the leading lenders in the syndicated market, and balance-sheet information is more available for listed institutions than for private ones. Lender participation amounts are partly imputed, though alternative imputation exercises preserve the main pattern. Entities active outside syndicated loans may be missing, and post-origination ownership can differ from origination shares. These choices do not erase the evidence, but they define the population to which it applies.
The strongest conclusion is therefore layered. The study provides evidence that, in its sample, tighter macroprudential policy is associated with a decline in lending by bank subsidiaries and an increase in lending by affiliated NBFIs, producing a sizable but incomplete group-level offset. The fixed-effects design and alternative shock construction make intra-group substitution a more credible interpretation than a raw correlation. The result is strongest for particular groups, instruments and NBFI business models. It does not reveal every internal transfer, cover every credit market or determine the wider financial-stability outcome.
That boundary changes the next question. Once the lending response is established, it is no longer enough to ask whether a parent can shift credit. We must ask where the shift occurs, which regulatory instrument permits it, and what other bank–NBFI relationships remain around the loan. The next comparison turns to those conditions, beginning with the fact that “tighter regulation” is not one uniform treatment.