What the Credit Average Actually Means
In the evidence on formal credit and SME performance, the word average carries more information—and less certainty—than it first appears to. This chapter reads the Inter-American Development Bank’s review of impact evaluations as a population-level summary, then follows the conditions that make the summary change. The practical value is a more disciplined judgment: understand what the mean says about formal lending, and what it cannot say about the next firm in line for a loan.
An average answers a population question
The IDB report Formal Credit and SME Performance: Reviewing Evidence from Impact Evaluations reports meta-analytic mean effects that are economically meaningful. Across the evidence assembled in the review, employment rises by about 12% on average, while sales and profits each rise by about 18% on average. These are important findings. They indicate that formal loans can do more than move money between accounts: relaxing a financial constraint can be associated with larger businesses and better measured performance.
Yet each number has a built-in boundary. It is a mean across an evidence base, not a forecast attached to a particular firm. The relevant comparison is between outcomes under a credit intervention and outcomes under the corresponding comparison condition, aggregated across the firms, programs, settings, and measurement horizons represented in the review. The number says something about the direction and average magnitude of the assembled evidence. It does not say that every firm’s employment rises by 12%, or that every borrower’s profits rise by 18%.
This distinction is easy to lose when a percentage is presented without its population. A mean is calculated after unlike responses have been combined. Imagine, as a conceptual example, a group in which some firms use additional finance to take a profitable opportunity, some use it merely to smooth a difficult month, and some assume more risk than their business can carry. The group can show a positive mean even though the loan changes each firm’s trajectory in a different way. The mean is not contradicted by the variation. It is the arithmetic summary of it.
The word meta-analytic adds another layer of scope. The report is not describing one bank’s borrowers under one contract. It reviews impact-evaluation evidence, focusing particularly on developing countries and Latin America while using evidence from advanced economies to sharpen mechanisms and external validity. Its average is therefore a synthesis across studies rather than a result from a single standardized lending product. The comparison base, the firms included, and the timing of outcome measurement can differ from study to study.
That breadth makes the mean valuable for one question: whether formal credit, considered across the reviewed evidence, tends to improve SME outcomes. It makes the mean insufficient for another: whether a lender should increase the size of the next loan for a particular applicant. The first question is about a program or population. The second is about conditional response and selection.
The same issue applies to the three outcomes. Employment, sales, and profits are related but not interchangeable. A firm can increase sales while its costs rise faster. It can increase employment while profit remains unchanged. It can report higher profits for a period without sustaining that performance. The review’s summary says that all three means are positive by the reported approximate percentages; it does not turn them into one guaranteed chain of events for every business or every time horizon.
The figures should therefore be read with two habits. First, retain the comparison: the effect is relative to what would have happened under the relevant non-credit or alternative condition, not a description of business performance in the abstract. Second, retain the qualifier: “about” and “on average” are not decorative language. They mark rounding, aggregation, and uncertainty about how the result travels to a new firm.
There is also a difference between variation across programs and variation within a program. If one targeted intervention has a larger mean effect than one broad intervention, that comparison can tell us that the settings differ in a decision-relevant way. It cannot, from the summary alone, tell us how much of the difference came from targeting, the firms who enrolled, the way the program was implemented, the comparison group, or the time at which outcomes were measured. The review identifies these dimensions as sources of heterogeneity; it does not turn them into a universal formula for predicting a new program.
This is where the report’s use of evidence from advanced economies is useful but easy to misuse. The purpose, as described in the report summary, is to sharpen understanding of mechanisms and external validity. Evidence from a different economy may help clarify how a financial constraint could affect business behavior or how a result might travel. It does not make the institutions, markets, or borrower populations interchangeable. A mechanism can be portable while its measured effect is not. The reader must ask whether the condition that generated the result is present in the new setting, rather than treating geographic distance as either irrelevant or decisive.
The same discipline applies to a lender comparing two applicants. A review-level mean can establish a reasonable prior that formal finance may improve SME outcomes under some conditions. It cannot supply the applicant-level counterfactual: what this firm would have done with a larger loan, what it would have done without one, and how repayment risk would have changed. That missing comparison is not a minor technicality. It is the reason the allocation question requires a separate design and evidence base.
Why the mean changes when the program changes
The review’s most consequential point is that the average is not a property of the word credit alone. The report says effects are larger in some large-scale public interventions and in programs explicitly targeting financially constrained firms. In other programs, effects are smaller or statistically indistinguishable from zero. The contrast suggests that access is only one part of the intervention. Who receives finance, how the program is organized, and what surrounds the borrower can alter the economic result.
Consider financial constraint first. A constraint is relevant when a firm has an opportunity or productive use for capital but cannot pursue it under its existing financing options. Additional formal credit can then release activity that was previously blocked. But if the constraint is not the main bottleneck, more borrowing may have a weaker relationship with performance. The firm may lack demand, managerial capacity, reliable suppliers, or a sufficiently attractive opportunity. The review does not say that one of these explanations dominates everywhere. It says the effects differ, and its summary specifically identifies programs targeting constrained firms as a setting in which effects can be larger.
Targeting changes the composition of the borrowers as well as the likely use of the funds. A blanket expansion asks what happens when credit is made more available across a broad population. A targeted program asks what happens when access is directed toward firms more likely to be limited by finance. Those are not the same test. A positive result in the second setting cannot be used to claim that indiscriminate expansion will have the same return. Conversely, a weak result in a broad program does not prove that credit would be unproductive for firms facing a binding financial constraint.
Scale matters for related reasons. A large public intervention can reach a different number or mix of firms, operate through different institutional channels, and affect the surrounding market differently from a small program. The review says some large-scale public interventions produce larger effects; it does not make “large-scale” a universal recipe. The appropriate conclusion is conditional: the way finance is organized can change both the opportunities it reaches and the population in which its effect is measured.
Local financial architecture supplies another condition. A loan does not enter an empty economic space. It interacts with existing lenders, informal finance, business relationships, and the competitive structure of the market. The report’s reference to local financial architecture is a reminder that the same nominal expansion can be absorbed differently where firms already have alternatives than where formal credit is scarce. It also warns against treating a result from one financial system as a transportable estimate for another without checking the surrounding institutions.
The outcome horizon matters as well. A loan may fund an investment whose return arrives later, or it may relieve working-capital pressure quickly. Employment, sales, and profits can move on different schedules. An effect observed at one horizon may not describe what happens after the firm has repaid, expanded, adjusted its product, or faced new competition. The review explicitly places outcome horizon among the sources of heterogeneity. A mean without its time frame can hide whether it describes an initial response, a persistent change, or a temporary adjustment.
The risk is not only that a program fails to produce a large benefit. The report also warns that credit expansions are not automatically welfare improving. They can create distributional and competitive spillovers. A recipient may gain while a nonrecipient faces a changed competitive environment. A program may redistribute opportunity among firms, alter prices or market shares, or amplify differences between businesses that can use capital productively and those that cannot. The report summary does not provide a universal direction or size for these spillovers, so they should be treated as effects to investigate rather than as a reason to reject credit expansion in general.
This is why the average must remain attached to its design. “Formal credit raises profits by about 18%” is a compressed description of evidence assembled under varied conditions. “This loan will raise this firm’s profits by 18%” is a different and unsupported statement. The second sentence has silently removed the population, the program, the comparison, the time horizon, and the uncertainty.
The flow is deliberately not a simple line from lending to profit. The measured outcomes depend on the context through which credit reaches firms, and welfare interpretation also has to account for effects beyond the direct recipient. The review-level mean is an output of that arrangement, not a free-floating return on capital.
Why a broad review leads to a within-program test
A review can establish that formal credit matters on average and still leave the allocation problem open. Its cross-setting breadth is useful for identifying recurring conditions: constraint-targeted programs may look different from broad ones; public interventions may not resemble smaller programs; local financial systems and outcome horizons may change the result. But those comparisons do not tell us whether two firms inside the same program respond differently to the same loan-size change.
That unanswered question is central because allocation happens within programs. A lender usually is not choosing between “credit” and “no evidence.” The lender is deciding which eligible applicants receive a larger amount, which receive a smaller amount, and how much confidence to place in an assessment made before the outcome is known. A review-level mean cannot, by itself, rank those applicants.
Nor can a positive average settle the question of scale. If the return to credit is concentrated among firms for which finance is the binding constraint, expanding the program without preserving that targeting may dilute the effect. If downside risk is concentrated among firms whose decisions or opportunities do not support a larger obligation, a uniform increase may create costs that the mean conceals. The review’s evidence does not identify a single targeting rule, but it gives a reason to look for conditional response rather than assuming it away.
The next case holds more of the environment constant. In Egypt, an experiment studies relatively large enterprise loans among applicants who entered a lender’s approval process. That setting cannot represent every credit market, and it does not erase the review’s warnings about design and local architecture. Its value is different: it lets us ask whether, within one lending margin, a small average effect is compatible with meaningful differences across firms. The review tells us why context matters. The experiment lets us see why the identity of the recipient may matter too.
The correct lesson from the broad evidence is therefore neither “credit works” nor “credit does not work.” It is more precise. Formal credit can improve SME performance, and the reviewed means are large enough to matter. But the size and meaning of the improvement depend on who is reached, how the program is designed, what financial system surrounds it, when outcomes are measured, and what happens to firms and markets beyond the direct borrower. Once those conditions are taken seriously, the question of who benefits is not a refinement at the edge of the evidence. It is the next step required to use the evidence responsibly.