Confidence Surveys: What Business and Household Sentiment Capture Before the Data Does

A confidence survey asks people what they think conditions are and what they expect, then scores the answers. A decision comes before the transaction it produces, and the survey catches the decision, so the reading moves before output does. The same fact is the survey's limit: an intention is not a purchase, and plenty of intentions are never acted on.

Two ideas already established are doing the work underneath that. The phases of a cycle give contraction and recovery their meaning. Some spending can be put off while other spending cannot, and that is what decides whose decision moves first. One idea is still missing, and it is smaller and more awkward: a survey reading is manufactured by a design, and the design travels with the number wherever it goes.

What does a confidence survey actually collect?

Start with what it is not. A statistical office builds the national accountsThe organised set of totals that add an economy up: output, spending, income and the pieces of each. Compiled from records of things that already happened, and covered in its own right under growth and output. out of records of things that happened. Goods left a factory. A bill was raised. Salaries were paid. Every one of those is a transaction with a date on it, and the compiler's job is to find them all, avoid counting any of them twice, and add them up.

A confidence survey does none of that and never goes near a transaction. The survey takes a sampleThe subset of people actually asked, chosen so that their answers stand in for a much larger group. Who is in it, and how they were picked, is a design decision made by whoever runs the survey. of households or firms, puts a fixed set of questions to them, and writes down what they say. A typical question offers three doors: better, about the same, or worse. The answers are then counted and squeezed into a single number, most often by taking the share saying better, subtracting the share saying worse, and pinning that difference to some agreed starting point. Some surveys report the raw difference and call it a diffusion indexA score built by subtracting the share of negative answers from the share of positive ones. The score reads above its neutral point when optimists outnumber pessimists, and the label describes the arithmetic rather than the subject.. The label describes the arithmetic and nothing else.

Everything good and everything bad about a confidence survey comes from one fact: it collects answers, not transactions. That sentence has two halves, and both are paid out in what follows.

Think of a neighbour who mentions, over the gate, that the scooter is on its last legs and they are thinking about replacing it this year. Nothing has happened. No money has moved, no showroom has a record of them, no statistic anywhere in the country is a rupee different because of that sentence. And yet something has been learned that the showroom's own sales ledger could never tell. The ledger only knows about the people who already walked in. The listener is ahead of the ledger, and is also holding something that might come to nothing. The scooter may last another two years, and the money may go on a wedding instead.

What is Consumer Confidence, and why can its two halves disagree?

Consumer Confidence is the name given to a survey of ordinary households about their own economic situation. The question is not about the country in the abstract. Most people have no way of knowing that. The question is about their own situation: whether money feels tighter than it did, whether the job feels secure, whether a large purchase feels affordable. The narrowing is deliberate. A person is genuinely the best available authority on their own kitchen, and that is what makes the answers worth having.

Nearly every Consumer Confidence reading published anywhere is built from two separate groups of questions, and reading the headline without knowing that is the first mistake available on this subject. One group asks about now. Is the respondent's situation better or worse than it was, is this a sensible time to buy something large, has income kept up. The other group asks about later. Will the respondent's situation be better or worse a year from now, will there be more work about or less. The headline is an average of two readings that can point in opposite directions, and averaging them destroys the very information a reader most wants.

A household on a single salary shows why. The salary has not risen for two years and the rent has, so the answer about now is bleak and honest. But the elder daughter finishes her course in eight months and expects to start earning, so the answer about later is genuinely hopeful. Neither answer is a mistake. Averaging them gives a middling number that describes nobody.

Take the invented Republic of Sankhya, whose confidence index is used throughout this guide. In its year 3 the index read 93, built from a reading of 100 about current conditions and 86 about what came next: people were doing fine and expected trouble. In its year 5 the index read 94, one point higher, built from 78 about current conditions and 110 about what came next: people were struggling and expected relief. Two headline readings one point apart, describing two economies that have almost nothing in common. Check the arithmetic: 100 and 86 average to 93, and 78 and 110 average to 94.

How is a firm's confidence different from a household's?

Business confidence surveys run on the same machinery and ask a different kind of question, and the difference matters more than it first appears. A household is asked what it feels and what it expects. A firm is asked about its order bookThe orders a business has accepted but not yet delivered. A record of commitments already made by customers, kept in the ordinary course of running the business rather than created for a survey., whether it plans to add people, and whether it intends to spend on capacity. None of those is a feeling. An order book is a written record of commitments other people have already made, and a hiring plan has usually survived at least one argument about money before anybody outside the firm hears about it.

The closer a survey question sits to something already committed, the more reliably the answer turns into activity. That is why business confidence tends to translate into output more dependably than household expectation does, and it is also why business confidence gives less warning: a commitment is further along than a feeling, so by the time it exists some of the deciding is already behind. The trade is lead time for reliability, and there is no setting on that dial where both arrive together.

A street food stall makes the trade visible. Asked how he feels about the year, the man running it gives an answer that could change with the weather. Asked how many kilograms of potatoes he has ordered for next week, he gives a number he has already paid a deposit on. The first answer arrives sooner. The second one is much harder to walk away from.

Why does sentiment move before the output data does?

Two separate reasons, and they are worth keeping apart because they fail in different ways.

The first is about sequence. A transaction is the last step in a chain that started with somebody deciding. A household decides the year is going to be tight, and only later does it not buy the fridge. A firm decides the order pipeline looks thin, and only later does it not place the steel order. The transaction that never happened is what the statistics eventually notice, but the deciding happened months earlier, and a survey that asks people about their situation walks straight into the middle of that gap. The lead is a consequence of what the survey asks about, not a special property that surveys have. Ask about a decision and the answer arrives in front of the transaction. Ask about a transaction and it arrives exactly as late as everything else.

Which spending gets cut first is a question settled earlier in this sequence, and the answer is that postponableSpending that can be delayed without much immediate cost: a new vehicle, a renovation, a machine that would have replaced a working one. The opposite is spending that recurs whether conditions are good or bad, such as food and electricity. spending goes first. The purchases a household can put off are almost exactly the purchases a confidence survey asks about. The match is what makes a confidence survey useful rather than merely early. Nobody is surveyed on whether they intend to keep buying rice.

The second reason is duller and gets forgotten. A household knows its own situation the day it changes. The compiler of a national statistic knows it only after collection, checking and publication, and there is no version of that process that is instant. So even where a survey adds nothing at all about the future, it can still report a deterioration that is already real before the statistical machinery has finished counting it. The second part of the lead is a reporting-speed advantage rather than a foresight advantage, and it disappears entirely the moment the statistic catches up.

Can the lead be seen on one path from end to end?

The Sankhya confidence index runs beside the output path this sequence has carried throughout. The index scale is arbitrary. There is no reason the neutral point sits near 100 rather than near 50 or near zero, and no point on it means anything at all except relative to other points on the same index.

The Sankhya output record runs like this. Output stood at Rs 16,80,000 crore in year 2 and Rs 17,47,200 crore in year 3, growing 4.00 per cent. Output then grew 6.00 per cent to Rs 18,52,032 crore in year 4 and 6.50 per cent to Rs 19,72,414 crore in year 5. Year 5 is where output stood furthest above potential outputThe level of output an economy could sustain with what it has, estimated rather than measured. Different reasonable methods give materially different answers, and the estimate gets revised. and where the output gapThe distance between what an economy actually produced and the estimate of what it could have produced, stated as a percentage of the estimate. Covered in its own right under growth and output. stopped rising. Output then fell 1.00 per cent to Rs 19,52,690 crore in year 6 and 2.50 per cent to Rs 19,03,873 crore in year 7, the troughThe lowest point output reaches before it starts rising again. A trough is only identifiable once output has already turned, and every low point in a cycle carries that same awkwardness., before recovering 5.00 per cent to Rs 19,99,066 crore in year 8.

Now put the index beside it. The index read 106 in year 2, then 93, 108, 94, 79, 90 and 102 across years 3 to 8. Read the turns rather than the levels. The index peaked in year 4 at 108 and fell 14 points in year 5, a full year before output fell at all. The index bottomed at 79 in year 6 and rose 11 points in year 7, a full year before output turned up. At both ends of the cycle the answers moved first and the transactions followed.

What happens the year the index turns and output does not?

Year 3 deserves a second look, and it is the year that keeps the record honest. The index fell 13 points, from 106 to 93. The fall is a serious one, the second largest in the whole record. Output that year grew 4.00 per cent. The next year output grew 6.00 per cent, and the year after that 6.50 per cent, the fastest reading anywhere on the Sankhya path. Between the fall in confidence and the eventual contraction, output rose Rs 2,25,214 crore. A confidence reading can move on news that never reaches behaviour at all, and this is a false signal rather than a broken survey.

Now the uncomfortable part. Compare the year 3 fall of 13 points with the year 5 fall of 14 points. One point apart in size. The levels they arrived at, 93 and 94, are one point apart too. Nothing about the shape, the size or the level of those two moves separates them, and yet one was followed by two more years of strong growth and the other by two years of contraction. Anybody who claims they could have told the difference at the time is describing hindsight.

Something did move in year 3, and it is worth naming. A false signal does not mean nothing happened. Something frightened enough respondents to knock 13 points off a national reading. The cause might have been a price shock that faded, a run of alarming headlines, a bad monsoon forecast that did not come true. Any of those is a real event that genuinely changed what people expected. None of them changed enough of what people did.

So how much is a single confidence reading actually worth?

Less than it looks, and more than a sceptic allows. An intention is not a purchase. A person who tells a surveyor they will probably delay the fridge has committed nothing, spent nothing and can change their mind for a hundred reasons that have no relation to the economy at all, including the fridge finally breaking. Surveys also pick up news rather than experience, and news is louder than experience and travels faster.

A confidence reading is a reason to go and look at something else, never a finding on its own. That is the same status the early signals carried earlier in this sequence, and it is not a downgrade. Prompts are useful. Something that reliably shows where to point attention two quarters before the statistics do is worth having, provided it is never mistaken for the thing it points at. The mistake is not using confidence data. The mistake is stopping there.

Concretely: the honest sentence after a sharp fall in a confidence index is not "output is going to fall." It is "something changed enough to move a lot of answers, and the place to find out whether the answers turned into anything is orders, hiring and the things people can actually postpone."

Why is a change in a survey worth more than its level?

One habit changes how every confidence number is read, and it is not really about confidence at all. The habit is about what a survey level contains.

A level carries three things that have nothing to do with the economy. A level carries the wording. A question asking whether a respondent expects their situation to worsen and a question asking how they rate their prospects for the year ahead collect different answers from the same person. A level carries the sample. A survey of urban salaried households and a survey covering rural districts are asking two different populations. And it carries the scale convention, including the thing nobody checks: what a respondent silently takes the middle option to mean. In some cultures of surveying the middle is a polite refusal and in others it is a genuine "about the same".

All three of those are frozen into the level and none of them is comparable to another survey's version of them. Comparing two different surveys' levels is close to meaningless, and it is the commonest misuse of survey data there is.

A change behaves differently. In the change from one round of a survey to the next the wording is identical. The questionnaire is the same one. The sample design is identical, and so is the scale convention. Every one of the three contaminants sits on both readings and cancels in the subtraction. Movement is what is left, and movement is the only part that was ever about the economy.

The simple version overstates the point, so one refinement is needed. Subtracting removes the neutral point completely, so directions become genuinely comparable across surveys: if one is up and another is down, that disagreement is real. Subtracting does not remove the sensitivity, meaning how many index points a survey moves for a given shift in its underlying answers. So two surveys can both rise on the same shift in sentiment and rise by different amounts. Directions compare. Sizes still do not, unless somebody has done the work of standardising them.

What does an analyst actually do with a confidence reading?

In competent hands the routine is narrow and unglamorous. The level is never quoted on its own. Asked where the index stands, such an analyst gives the number and then immediately says that it does not mean anything by itself. The remark is not modesty but accuracy.

She quotes the change instead, and specifically the change against the same survey's own recent run. Three rounds of small moves and then one large one is what she looks for. A survey that has been quiet and suddenly is not has told her something about attention having shifted, even before she knows to what.

Then she does the thing the reading was only ever a prompt for. She goes to the measures that sit nearer a commitment: order books, hiring notices, and the categories of spending that are easiest to postpone. If the confidence move shows up there too, she now has two independent things pointing the same way and a case worth writing. If the move does not show up there, she has learned something equally useful. A lot of people got worried and carried on much as before, and that is a year 3 rather than a year 5.

The whole practitioner routine is one habit: the survey shows where to look, and the commitment data says what is happening. A credit team does the same with its own book, watching whether enquiries thin out before the drawdowns do. So does a household: three neighbours stop talking about buying a car before the showroom on the main road looks empty.

The Republic of Sankhya, its output record, the Sankhya confidence index, Survey A and Survey B are invented.
Educational material. Not advice on any investment, tax, budget or market position.