ECB lending survey: check responses, country weights and timing

01

Read the new ECB result through its aggregation

The European Central Bank’s October 27 release of its third-quarter 2026 bank lending survey (BLS) reported net tightening of corporate credit standards of 9% and a net decline in corporate loan demand of 6%.[C1] This combination provides a starting point for examining lenders and borrowers together. Adding the two figures does not produce a rate of decline in lending. What the answers count determines how far the result can support an explanation about the world economy or financial markets.

SG Group focuses on the calculations between individual answers and the published total. For financing conditions, a broad movement and an offset between opposing movements call for different follow-up evidence, even if they produce the same number. One aggregate cannot reconstruct each bank’s stance, lending amount and customers’ plans simultaneously. Three windows help examine it: the distribution of answers, country weights and the alignment of time.

The numerical exercises use fictional responses to test which information aggregation preserves and which it loses; they do not estimate the actual banks’ answers. Readers can copy the values to paper or their own spreadsheet and recompute the differences. Changing country weights then reveals where the aggregate changes sign even when national answers stay fixed. No account information or trading history is required, and the article provides no data-entry, submission or storage service.

This approach helps when a strong headline tempts readers to make a blanket judgement about counterparties or business conditions. A weaker aggregate requires different supporting documents depending on its country, lending category and period. An unchanged figure may also conceal greater disagreement. Working through the arithmetic first makes it possible to specify what to look for in the next survey or a bank’s explanation. This article develops that measurement question separately from the general credit-cycle argument.

For the wider credit-cycle framework, Read lending standards and the credit cycle.

02

The same plus ten can hide different response patterns

Consider a fictional survey of 100 banks eligible for the same question. In case A, 30 tighten, 20 ease and 50 leave standards unchanged. In case B, 15 tighten, five ease and 80 remain unchanged. Subtracting the easing share from the tightening share gives a simple net balance of plus ten in both cases. Yet 50 banks change something in A, compared with 20 in B. An identical difference need not describe a change of identical breadth.

A contains more banks moving in each direction. The scope for borrowers to seek another lender, and the need to compare conditions, could therefore differ from B. In B, most banks do not change their stance and a smaller group creates the balance. Rather than assuming that all banks move together, it is more useful to examine what changing banks have in common. Information about customer industries and contract types can narrow the distance between the aggregate and a reader’s circumstances.

For the check, first make sure the three counts sum to 100. Divide tightening and easing counts by that same 100, then subtract. This sequence helps detect percentages with incompatible denominators. Using all respondents for tightening but only banks that changed for easing would mix bases inside one table. Writing down the population used by each column is a small, useful check before comparing the result.

The 100-bank population is fictional and chosen for convenient arithmetic; it is not a rescaling of the current ECB sample. Multiplying a euro-area net balance by the sample size cannot identify how many banks tightened. The easing count also matters, and country aggregation adds another step. Comparing A with B demonstrates why several distributions can produce the same difference, without inventing an integer count for the actual survey.

03

At zero, distinguish offsetting changes from no change

In fictional case C, all 100 banks report no change. In D, 40 tighten, 40 ease and 20 remain unchanged. Both net balances are zero, but borrowers need not encounter similar conditions. C is consistent with earlier standards being maintained. In D, many banks move, leaving borrowers to ask which side their lender occupies. Reading zero as a motionless financing environment would lose this distinction.

A net balance also does not directly measure the level of standards. A lender that was already difficult to borrow from and one that was accommodating can both report no change. Similar current movements can leave different conditions because their histories differ. Comparing a current contract proposal with an earlier quote and a renewal notice is more concrete than simply lining up rates of change.

The possibility of offsetting changes does not make aggregation useless. Repeated answers collected consistently provide a common language for tracking direction. The difficulty comes from asking the retained information to explain too much. As a temperature change does not determine a diagnosis, a balance points towards the next investigation. Use distributional evidence when available; otherwise preserve the range of possible distributions while examining other relevant measures.

Before replacing zero with the word stable in an internal briefing, note whether the unchanged share is known, whether opposing groups could offset each other and how the earlier level can be checked. If another department later reports different conditions, this avoids prematurely treating either the statistic or local evidence as wrong. Using the aggregate to establish direction, then handing the question to contract-level checks, connects statistics with practical decisions.

04

A net balance and an intensity-weighted index

The ECB glossary defines the net percentage as the tightening share minus the easing share. Its diffusion index also weights intensity, assigning one to considerable changes and 0.5 to somewhat changed answers.[C3] Similar-looking measures preserve different information. Before copying a figure from one table to another, check the column label to establish which measure it is.

Split A’s 30 tightening banks into ten reporting considerable tightening and 20 reporting some tightening. Let all 20 easing banks report some easing. The intensity-weighted calculation is 10×1 + 20×0.5 − 20×0.5 = 10. Dividing by 100 eligible banks and multiplying by 100 gives an index of ten. If all 30 tightening banks instead report some tightening, it becomes 15−10 = five, despite the same net balance.

Numerical agreement does not establish that two measures are identical. Both produce ten in the first setting, making their different construction easy to overlook. Changing response intensity reveals the difference. When checking a transcription, consider whether the relationship survives a small input change, rather than accepting one matching number as proof. Keep fictional calculated values in separate columns from published observations.

An intensity-weighted index still cannot be read directly as the severity of losses or the effect on lending amounts. The one and 0.5 in this exercise are computational weights assigned to response categories; they do not mean one contract loses twice as much as another. Use the measure to summarise stronger and weaker categories, then obtain separate evidence for amounts or repayment burdens. Greater precision about the statistic should not create false precision about economic effects.

For the basics of expectations, outcomes and revisions, Match expectations, releases and revisions.

05

Equal country weights can change the aggregate’s sign

The ECB user guide describes two-stage euro-area aggregation: unweighted national results are weighted by each country’s share of outstanding loans. Banks normally receive equal weights within a country; Bulgaria, Malta and Slovakia also compile nationally weighted results.[C2] Counting banks across countries is therefore different from the euro-area weighting step. Keep respondent counts and country aggregation weights in separate fields.

Take two fictional countries. A has ten eligible banks: four tighten, two ease and four remain unchanged, yielding plus 20. B has one tightening, five easing and four unchanged, yielding minus 40. Their equal-weight average is (20−40)÷2 = minus ten. This gives each country one vote and includes no information about the relative size of their lending markets.

Give A a fictional country weight of 80% and B 20%. The aggregate becomes 20×0.8 − 40×0.2 = plus eight: A contributes plus 16 percentage points and B minus eight. The same bank answers now produce the opposite sign from the equal-weight average. The two calculations answer different questions: treating countries equally or reflecting specified market weights. Naming that choice explains the discrepancy.

These two countries do not represent the current euro-area breakdown or an assessment of particular countries. Reproducing an actual aggregate requires national results and weights for the same lending category and data vintage. The exercise shows where differences can enter when figures use different aggregation methods. It also helps check whether a collection of national headlines has silently substituted the number of countries for the intended weights.

06

Solve for the weight at which the sign changes

In the fictional example, let A’s weight be w and B’s be 1−w. The total is 20w−40(1−w), or 60w−40. It reaches zero at w = 40÷60, approximately 66.7%. Above that boundary the aggregate is positive; below it, negative. One equation shows how weighting changes the direction without altering national answers. Knowing the distance from the boundary is more useful than remembering only the 80% setting.

In a spreadsheet, try 60%, approximately 66.7% and 80%, always setting the other weight to the remainder of 100%. The aggregates are minus four at 60%, zero at exactly two thirds and plus eight at 80%. Rounding the boundary to 66.7% leaves a small positive result. Keep the fraction for calculation and round only the display, especially when a small balance makes the rounding rule consequential.

Varying weights here does not mean official weights are freely selectable. Reproduction follows the prescribed method; sensitivity analysis changes an assumption. Separate sheets can keep those purposes distinct. A firm whose trading geography differs sharply from the aggregate might construct a separate reference measure for its own exposure, but should not label that result as the ECB’s published statistic.

The boundary does not forecast which country’s banks will move next. It identifies the inputs that control the aggregate. Changing A’s balance alters the equation’s coefficients; changing B’s balance shifts the boundary too. Recompute it from the same input fields at each update rather than carrying the old threshold forward. Recording whether responses or weights changed also reduces the temptation to attribute the entire move to one country’s stance.

07

A response balance does not identify an amount growth rate

Two numbers may use similar notation while measuring different things. Suppose one of two fictional banks reports increased demand and the other decreased demand. Giving each answer equal weight produces a zero balance. The answers alone do not specify the initial amounts or the size of either change. Moving from that zero balance to unchanged financing needs or unchanged lending requires separate amount data.

As a separate amount exercise, move fictional A from 100 to 120 and B from 900 to 720. The total falls from 1,000 to 840, or 16%. Alternatively, move A from 900 to 1,080 and B from 100 to 80: the total rises to 1,160, or 16%. Each setting has one rising and one falling amount. Swapping initial sizes reverses total growth. This arithmetic demonstrates missing information; it does not model the mapping from actual BLS answers to lending amounts.

Do not draw a causal line between these calculations. A demand assessment, application, approval, disbursement and balance after repayments occupy different stages. The amount exercise shows that direction counts cannot determine totals; it does not assume that those stages change at the same rate. When adding company documents, identify whether a figure is a facility limit, actual utilisation or a closing balance before comparing it.

There is no universal ranking in which amount weighting is correct and counts are crude. Changing conditions across many small counterparties and a volume move in a few large loans can both matter. They inform different questions about breadth and quantity. Establish the question first, then use each measure for its purpose. When displaying both, separate their units rather than forcing them into a single scale.

08

Place standards, terms and demand along a process

In the ECB glossary, standards are internal lending policies or approval criteria; terms are those of approved loans. Demand concerns financing needs whether or not a loan results, as assessed by banks.[C3] The three fields are not alternative names for one contract’s rate. For a practical comparison, arrange evidence about eligibility, the offer and the funding need to locate where explanations diverge.

Imagine a company needing 100 to buy equipment. A bank may discuss the loan but request more collateral, increasing the documentation needed while the financing requirement stays fixed. Postponing the purchase could reduce enquiries or applications. Conversely, delayed customer receipts could increase short-term borrowing needs without stronger investment. Each situation changes different inputs, making one response category an incomplete assessment of business confidence.

The process changes the question when documents appear to point in opposite directions. Tighter bank approval criteria and a company’s rising borrowing balance need not conflict if they concern different periods or transactions. Balances can include existing contracts, while current discussions may lead to later disbursement. Rather than averaging the statements into one conclusion, locate their stage, population and timing to identify what is comparable.

For a business review, keep bank application documents distinct from the company’s cash plan. One supports an application and its conditions; the other schedules receipts and payments. A survey can prompt a review of both without predicting an individual approval. Naming the missing document in a worksheet turns a general concern into a specific question about what to request and from whom.

For the timing of profit and cash, Separate revenue, profit and cash.

09

Publication quarter and the quarter being assessed

One survey release combines assessments of past changes with forecasts. Read the collection window, reference periods and publication date in the diagram as separate fields. Aligning everything only by release date can make different horizons look like one observation. Keep separate fields for the period assessed and the forecast horizon.

The ECB user guide says the data’s quarterly label denotes the publication quarter.[C2] Placing an October release in a fourth-quarter row does not move its backward-looking answers into that quarter. Read the time label together with the question’s direction. Shifting an axis by one quarter merely to align charts could manufacture a lead or lag against another series.

Create three columns in private notes: publication label, assessed period and expected period, and copy the diagram’s mapping. Carry all three columns into another document rather than detaching the label. A later reader can identify the period assessed and the period expected without needing the original discussion. Check that the three fields stay together before making a time-series chart.

A collection window crossing quarter-end also matters when comparing daily events. Without individual response dates, juxtaposing a particular news event with survey answers cannot establish a reaction to that event. Events after the response deadline have an even clearer timing problem. The timeline does not prove predictive power; it helps eliminate explanations that the chronology cannot support.

10

Fix the comparison pair before assessing forecasts

To assess a forecast, do not simply subtract the backward- and forward-looking values in the same release: they refer to different periods. Pair an expectation for a quarter with a later survey assessing that quarter, then check the category, question and aggregation method. Without a fixed comparison pair, searching for nearby numbers after seeing the outcome can encourage a convenient but invalid match.

In a fictional record, one survey expects plus five for the next quarter and the following survey assesses that quarter at plus eight. The difference is three percentage points, not a 3% error in lending amounts. Agreement on direction and the size of the numerical difference are separate evaluations. Separate fields for directional agreement and the gap avoid reducing usefulness to a simple hit-or-miss judgement.

Also distinguish the value available at the time from a value downloaded today. Reviewing a past decision requires the information then available; assessing current conditions benefits from current definitions and revisions. Label files with retrieval date and vintage and preserve the original column. If a discrepancy appears later, this helps separate a calculation error from a data update. Readers perform that recordkeeping in their own environment.

A few matched releases cannot establish that the survey always leads other developments. Evaluation depends on the periods included, missing-data treatment and definition changes. Keep the same rule across favourable and unfavourable periods. The article’s worksheet fixes the correspondence between periods and columns; it does not report a forecasting test. Leave unfilled periods blank instead of turning missing observations into zeros.

11

Use the checks to narrow the economic interpretation

SG Group sees the corporate combination as a reason to prioritise investigation of financing conditions. Rather than assigning a production or investment decline from this survey alone, first establish which answers support the direction and which period they address. Businesses with European customers or suppliers can examine counterparties’ locations and financing purposes before translating the headline into an order forecast. Understanding aggregation improves the choice of that additional evidence.

For markets, treating negative demand answers as a single recession score compresses different questions. Directional responses, actual funding amounts and expectations embedded in prices have different comparators. The exercises show that direction aggregation does not retain enough information about size. Before explaining a market price, ask whether the proposed mechanism requires a quantity change or only a directional pattern. That exposes channels lacking the evidence they need.

A hypothesis that demand for components sold into Europe will fall needs order, inventory or investment evidence outside the survey. A practical decision to check counterparties’ terms earlier can proceed without a quantity forecast. These decisions need not demand identical evidence, but their conclusions should be explicit. Sometimes the useful response to a release is one additional check rather than a large forecast.

The same reasoning applies when comparing a broad bank survey with one lender’s interest income. The survey aggregates responses; company results depend on that bank’s assets, liabilities and accounting period. Neither direction automatically explains the other. The existing HSBC article examines reinvestment timing and deposit rates. Leave that calculation there and preserve this article’s focus on aggregation, rather than forcing different documents into a single narrative.

For transmission from overseas demand into business, Trace overseas demand into orders and jobs.

12

Three follow-ups that would change the assessment

The first follow-up is the response distribution or national breakdown for the same category. It can show whether an aggregate move is broad or concentrated. Concentration outside a firm’s relevant geography would weaken a concern inferred from the headline. A similar direction spreading within its trading footprint would raise the priority of a focused check. The evidence changes where an explanation applies, not merely its sign.

The second is amount or activity evidence aligned to the same period. Agreement with applications or disbursements can sharpen a quantity hypothesis. First align corporate versus household coverage and stocks versus period flows, since mixed populations can create apparent agreement. Weak responses alongside growing amounts are a reason to investigate size, existing contracts and funding purposes. Do not discard contrary quantities to preserve a directional story.

The third is a later survey’s assessment of the quarter covered by today’s expectations. Use the three-column timeline to match the later observation to the earlier forecast, rather than merely appending another publication quarter. If the direction reverses, retain the discrepancy and reconsider how much certainty was assigned to expectations. There is no need to invent the next release date or its outcome; fix the matching rule in advance.

This is not a procedure for collecting only evidence of deterioration. Preserve larger-than-expected offsets, quantities moving the other way and forecast–observation mismatches in the same table. An explanation that survives every possible result is hard to test. Decide whether contrary evidence would narrow the population, change the quantity hypothesis or correct the timing. That makes subsequent changes in judgement traceable.

13

Eight fields to carry into your own evidence

The first two worksheet fields identify the document and vintage, then what is measured. Recording the release and question is more useful than writing only ECB. The next two fields specify the reference period and time direction. Category and backward- versus forward-looking coverage take priority over numerical similarity. The aim is not to transform every document into the same thing, but to reveal incompatible premises before comparison.

Fields five and six record the denominator and the weights or formula. If eligible counts are unavailable, name the needed document instead of substituting the total sample. For country results, record equal weighting or the specified weights. Keep a personal sensitivity result separate from the official figure. Even a small table then lets another reader trace calculated results and assumptions. Do not fill missing inputs with zeros for appearance.

Field seven states one economic question. Asking whether to check terms earlier in a firm’s trading geography requires different evidence from estimating a euro-area investment change. Field eight names the evidence and source that would change the assessment. A focused question makes relevant passages easier to find in the next document. Filling the table is not the goal; empty fields should point towards the next check.

The worksheet is free to read and the exercises can be reproduced locally. It does not retrieve data automatically, determine loan eligibility, review contracts, forecast or store inputs. Readers supply original and company documents and keep notes in their own environment; no confidential contract information needs to be sent to this site. Actual contract decisions require the counterparty’s formal terms and authorised explanation. The table connects understanding an aggregate with that separate decision process.

14

Which calculation to check after the headline

After a short news summary, the remaining task is to trace the route to the aggregate personally. Different response distributions produced the same balance; different country weights reversed its sign; identical direction counts allowed opposite amount growth. Working through these examples makes it easier to identify the additional information that would advance an interpretation of a published figure.

Start by choosing whether the question concerns aggregate direction, a particular geography or funding quantities. Check the balance and horizon for direction, the breakdown and weights for geography, and amount data for quantity. Avoiding one oversized answer preserves the survey’s usefulness while respecting its limits. As evidence arrives, it fills different fields in the same worksheet, making the reason for a revised interpretation visible.

The procedure can help with other surveys aggregating countries or industries, but their weights and signs must be checked afresh. A positive value may mean tightening in one measure and improvement in another. Return to the questionnaire and guide rather than carrying the convention across automatically. The transferable result is a way of reading answers, weights and time together, not a reflex judgement about a particular number.

SG Group’s current assessment is to use this survey as an entry point for a closer examination of financing conditions, while establishing the scale of economic and market effects with separate evidence. Matching breakdowns, quantity data and later observations for the corresponding quarter would change the next judgement. The diagrams and eight-field worksheet let readers identify which evidence has arrived, then make the explanation more specific rather than fixing its strength in advance.