Downturn and Recovery: The Early Signals, the Definitions That Disagree, and the Shapes Coming Out

A slowdown is growth getting weaker while output still rises. A contraction is output falling. A recovery starts at the turn, the moment output stops falling, not at the moment it regains its old level, and the long stretch between those two moments is why recoveries feel late and get announced late. The definitions disagree because each measures a different thing at a different speed.

Underneath that answer sits one habit of reading. Every word in the previous paragraph is about either a direction or a level, and the two are almost never talking about the same thing. Direction words name which way the economy moved last period: up, down, faster, slower. Level words name where it stands: high, low, back to normal, still short. A headline mixes them freely. A reader who does not separate them will hear a direction word and reach for a level conclusion, or the other way round, and will be wrong in a way that feels perfectly reasonable at the time.

Here is the everyday version, and it does most of the work below. Picture a small printing press that does wedding cards and shop stationery. In one season the diary fills with twenty per cent more work than the season before. Next season it fills with only four per cent more. Four per cent more work is still more work, so the owner is busier than she has ever been. Something has clearly changed all the same. The pace at which the diary is filling has dropped hard. Now imagine a different season, where the diary comes in ten per cent lighter than last time. A lighter diary is not a slower fill. A lighter diary is less work. The first season is a slowdown and the second is a contraction, and the press owner feels them as completely different events even though a careless summary would call both of them bad news.

Every figure below belongs to the Republic of Sankhya, an invented country built for macro arithmetic. Sankhya has a published output path, a re-estimated path for what it could produce, and a downturn and a turn inside that path. The rates are recomputed from the rupee levels each time, so the arithmetic can be checked against itself with a calculator rather than taken on trust from a stated percentage.

Economic Slowdown: is output rising or falling while growth weakens?

Output is rising. Output rising is the whole of the distinction, and it is worth saying flatly before anything else. An economic slowdown is a fall in the rate at which output is growing. The level of output keeps going up. A contraction is a fall in the level itself. One is a change in pace and the other is a change in direction, and no amount of severity turns the first into the second.

On its published annual path Sankhya has no slowdown year at all. The absence is awkward and useful at once. Growth ran 6.00 per cent in year 4, then 6.50 per cent in year 5, then minus 1.00 per cent in year 6. Growth did not weaken first and then go negative. Growth accelerated, and then went straight through zero into a fall. A slowdown on this path therefore has to be constructed out of the published levels. The weakening can begin and end inside a single reporting year and never appear as its own row, so an annual series can hide a slowdown completely.

The invented variant works like this. Hold year 5 output at its published Rs 19,72,414 crore and suppose year 6 had come in at plus 2.00 per cent instead of the published minus 1.00 per cent. Output would then be Rs 20,11,862 crore, a rise of Rs 39,448 crore. Growth would have dropped 4.50 points, from 6.50 to 2.00. A fall of that size is heavy by any standard, and the economy would still have produced more in year 6 than it did in year 5. Set that beside what actually happens on the published path, where year 6 output of Rs 19,52,690 crore is Rs 19,724 crore below year 5. Both are bad readings. Only one of them is output shrinking.

Why does the confusion cost anything? Because the two states call for completely different expectations about what happens next inside a business. In a slowdown a firm is still filling more capacity than last year and is deciding whether to add less rather than whether to cut. In a contraction the same firm has spare hands and spare machines and is deciding what to stop. A supplier who hears the word slowdown and prepares for the second of those has taken decisions that were not needed, and a supplier who hears contraction and prepares for the first has taken none of the decisions that were.

Which signals turn before output does, and what is each one worth?

Output for a whole economy is counted after the fact. The number describing a period only exists once the period is over and the counting is done. So anyone who wants an earlier view has to watch things that move before production does. Three are worth naming here, and all three are things a business person can see without any statistics at all.

The first is the order bookA list of jobs customers have committed to and nobody has delivered yet, counted in value or in units. An order book is a diary of work owed to the world.. An order is placed before the thing is made, so a thinning book is visible to the maker weeks or months before the making shrinks. The second is hiring intent, meaning vacancies opened, offers made, contracts renewed or quietly not renewed. Recruitment is the cheapest thing to pause, so a firm that expects less work stops recruiting long before it produces less. The third is the inventory buildFinished goods piling up in a warehouse because they were produced but have not sold. A build here only means the pile grows. Nobody chose to grow it., and it is the most treacherous of the three. A rising stock pile can mean a firm expects a boom and is stocking up for it, or that the firm expected a boom that did not arrive.

Every one of these three turns before output does, and every one of them also turns when nothing follows. A signal is therefore a reason to look harder rather than a conclusion to act on. A thin quarter in the order book can be a lost tender that a rival won. A lost tender says something about that firm and nothing about the economy. A hiring freeze can be a change of management. A stock pile can be a delayed shipment. The discipline is to ask what else would have to be true if this signal were real, and then go and look for that as well.

Economic Recovery: does it begin at the turn or at the return to normal?

At the turn. An economic recovery begins in the first period in which output stops falling and starts rising, and that is a statement about direction only. The turn says nothing about the level output has reached, nothing about whether the previous peak has been regained, and nothing about whether anybody feels better. Level, peak and relief are separate questions with separate and much later answers.

Work the Sankhya path. Output fell to Rs 19,03,873 crore in year 7. No year in the stretch is lower, so year 7 is the trough. In year 8 output is Rs 19,99,066 crore, a rise of Rs 95,193 crore. Take that rise against the Rs 19,03,873 crore it started from and the ratio is 0.05, so growth is 5.00 per cent. Output has turned. On the definition just given, the recovery begins in year 8, and nothing that follows changes that.

Now the uncomfortable part of the same year 8 reading. Sankhya's potential outputA yardstick figure for how much an economy could turn out with its people and equipment in ordinary use, neither idle nor stretched. It is arrived at by a method rather than counted, and the method is arguable. in year 8 is Rs 21,25,785 crore, on a path re-estimated at 4.00 per cent a year and settled under potential output. Subtract: Rs 21,25,785 crore less Rs 19,99,066 crore is Rs 1,26,719 crore. The shortfall divided by potential is 5.96 per cent, so the output gapHow far production sits from that yardstick, written as a share of the yardstick. A minus reading means the economy came in under it. in year 8 is minus 5.96 per cent. The same shortfall divided by actual output instead is 6.34 per cent. Year 8 would need that much on top of everything it already produced just to draw level with the potential estimate for the same year. The recovery is running, and the economy is still Rs 1,26,719 crore short of where the estimate says it could have been.

One more reading, and it cuts the other way, so it is worth stating plainly rather than hiding. Year 8 output of Rs 19,99,066 crore is above the year 5 peak of Rs 19,72,414 crore, by Rs 26,652 crore or 1.35 per cent. So on this path the old peak level is already back, in year 8, in the first year of the recovery. And the economy is still nearly six per cent below what the estimate says it could be producing. Potential did not stand still while output fell, but kept growing at 4.00 per cent a year straight through the contraction. Regaining the old peak therefore settles far less than it sounds like it should. The target moved while the economy was busy falling.

How long does the rest of it take? Suppose, purely as arithmetic on invented numbers, that Sankhya output kept growing at 5.00 per cent a year while potential kept growing at 4.00 per cent. The distance closes by roughly one point a year, so output draws level with potential in year 15, seven years after the turn. Growth rates do not stay fixed and the potential estimate itself gets revised, so year 15 is arithmetic rather than a date. A turn takes one period and a return takes years, and everything that makes a recovery feel like a lie is sitting in the space between those two facts.

How does the whole Sankhya stretch read as one downturn and one recovery?

Laid out in rows, the four years stop arguing with each other. Read the growth column for direction, the level column for where the economy stands, and the gap column for how the level compares with the estimate of what was possible. Each of the three columns is telling the truth, and each of the three would produce a different headline.

Three things in that table deserve a second look. Between year 5 and year 7 output falls Rs 68,541 crore, the whole of the contraction measured in rupees. Between year 7 and year 8 output rises Rs 95,193 crore, more than the entire fall. The potential line rose Rs 81,761 crore over that same year, so the gap only improves from minus 6.86 to minus 5.96 per cent. And the jobless rate on the same Sankhya path was 6.90 per cent in year 7 and 7.40 per cent in year 8. The jobless rate went up in the first year of the recovery. A reader watching only the jobs number in year 8 would say the downturn was still deepening, and every figure they were looking at would be correct. Percentages are rounded to two decimals.

Why do two people announce the same downturn at different moments?

Because they are answering with different tools, and both tools work. One approach applies a mechanical rule to a single series: state a test in advance, count the periods, and the moment the count is met the answer is produced. The other convenes a considered judgement across many series at once, output and jobs and incomes and sales together, and declares only when the whole picture agrees. The two tests, and the trouble each buys, are covered separately.

Having both produces one consequence. A single downturn gets several start dates, and the disagreement is not a sign that anybody is wrong. The mechanical rule answers fast, on one measure, and it will sometimes fire on a stretch that never turns into anything. The judgement answers slowly, on many measures, and it will sometimes be silent through months that everybody living in the economy already knows were bad. Speed and confidence trade off against each other directly, so a definition that answers early is buying that speed with false alarms and a definition that answers well is paying for accuracy with time. There is no third definition that has both, and a reader who understands this stops asking which announcement is right and starts asking which question each one answered.

The practical habit that follows is small and useful. A claim that a downturn has started or ended invites two questions before anything else. Which series was that decided on, and at what vintageWhich edition of a figure is in hand. The same past period carries a first estimate and one or more later ones, and a claim built on the earliest edition can be undone by a later one. was the series read? The two answers usually explain the whole of a disagreement that looked like a dispute about the economy and was actually a dispute about method.

What shapes can the path out of a downturn take?

Once output has turned, the path it traces afterwards can look very different from one episode to the next, and people describe those paths with letters because the letters are quick. A sharp fall followed by an equally sharp climb makes a V. A fall, a flat stretch where nothing much happens, then a climb, makes a U. A fall followed by a level that simply stays down makes an L. A fall, a partial climb, then a second fall, makes a W. Each of the four paths drawn below starts from the same Sankhya trough of Rs 19,03,873 crore, so the only difference between them is the shape itself.

Look at the third panel for a moment longer than the others. Output turns up there, by a fraction, and never climbs. Output is rising, so on the strict definition given above that is a recovery. The dashed potential line pulls away from it year after year, and by the fourth year the distance has grown to 19.82 per cent from 6.86 per cent at the trough. A recovery can be real, continuous and completely inadequate at the same time. Direction and level therefore have to be reported as two separate readings rather than blended into one verdict.

Now the sentence that matters more than the letters. A shape is only visible after the years that make it have happened, so naming one from inside it is a forecast wearing the clothes of a description. At the end of year 8, all four panels show exactly the same thing: a trough, then one year of rise. The four panels are identical up to that point in every way that a reader could observe. Everything that separates a V from a W happens later. When somebody says a recovery is V shaped and the recovery is one year old, they have made a prediction, and it deserves to be examined as one rather than accepted as a summary of the data.

Why is a recovery almost always declared late?

Two reasons stack on top of each other, and only the second one is a choice anybody made. The first is that the count itself is not ready. A period has to end before it can be measured, the measurement takes time, and the first version published is an estimate built on partial returns. Later, when more of the returns are in, that estimate is revised. A statistical revisionA published figure being replaced by a better one for the same past period, once more complete source data arrives. The period being described has not changed; only the estimate of it has. is routine and expected, and it is not a correction of an error in any ordinary sense.

The second reason is deliberate. A judgement that waits for several series to agree buys the ability to be sure at the cost of being early, and it does that on purpose. A turn called and then withdrawn is worse for everyone relying on it than a turn called four months late. A late declaration is the method doing exactly what it was designed to do, and the alternative on offer is not an early correct call but an early call that is sometimes wrong. Nobody has been given a way to be both fast and certain, and the honest position is that a turning point is not fixed in advance and can only be identified once the periods around it exist and have settled.

What does a lender watch at a turn, if not the announcement?

Its own book, and it watches it borrower by borrower. A lender with two hundred small business accounts is holding something no statistical office has: the current order position of two hundred real firms, updated as those firms draw on their limits, deposit receipts and ask for renewals. The aggregate is built partly out of what these same firms will eventually report, so the lender holds that information weeks or months before anyone else can read it.

Concretely, what does the credit officer actually look at? Utilisation of working capital limitsThe ceiling a lender sets on short term borrowing that a firm dips into for wages, stock and raw material, and repays as its own customers pay it. How such a limit is sized is a subject of its own. is the first of them, and a firm whose orders are recovering draws more to buy inputs before the sales appear. Cheque and transfer receipts matter next. A customer paying earlier is a customer whose own book has turned. Renewal requests for limits that were being allowed to lapse a year ago. A rise in requests to fund a new machine rather than to bridge a gap. No one of these readings measures the economy. Taken together they tell a lender whether the firms it has already lent to are turning, and that is the only question its own book asks.

The everyday version is a wholesaler in a cloth market who supplies forty retail shops. Long before anybody publishes anything about the state of retail, she knows which shops have started ordering their festival stock early again and which are still buying week to week. She is reading a turn from the inside, on a sample of forty, with no statistics whatsoever. The lender is doing the same thing on a sample of two hundred, and the statistical system is doing it on a sample designed to represent everybody. A sample built to represent everybody is both better and slower. Each of the three is answering a different question, and which of the three a reader needs is worth settling before the looking starts.

The Republic of Sankhya is invented.
Educational material. Not advice on any investment, tax, budget or market position.