Technical vs Economic Recession: A Rule and a Judgement

A technical recession is a rule: real output falls for two quarters in a row and the label applies. An economic recession is a judgement: activity is deeply, broadly and persistently weak across many measures at once. The rule is fast and mechanical, the judgement is slow and considered, and the two disagree often enough that knowing which one a headline means is most of reading it.

One fact sits underneath both definitions, and it is worth a minute before either of them arrives. An economy is not one thing that can be measured once. An economy is millions of separate decisions to buy, to make, to hire and to hold back, and there is no single dial anywhere that reads off how well all of it is going. Instead there is a set of measures, each of them partial, each published with a delay, and each revised afterwards as better information turns up. No dial settles whether an economy is in trouble. The question is settled either by agreeing on a rule in advance and then applying the rule, or by looking at everything available and forming a view.

A rule agreed in advance and a view formed from everything available are the only two routes there are. Consider how a household decides whether money is tight. One route is a rule the household agreed on earlier: if the balance in the account is lower at the end of two months running, money is tight. The rule answers the moment the second statement lands, and nobody can argue with the arithmetic. The other route is a look at everything: the balance, yes, but also whether anyone lost work, whether the shop-front takings held up, whether a bill got pushed to next month, whether the tightness is in one week or across the whole quarter. The second answer is better and it takes longer. The two definitions of recession are exactly those two routes, applied to a country instead of to a household.

Each definition can be stated on its own, applied to the same series of quarterly readings, and then set against the other across four possible outcomes, one of which almost nobody checks for.

What exactly does the technical rule test?

The technical rule is a single test applied to a single series. Take real outputActivity measured in volume terms rather than in rupees. A rise in prices on its own cannot make the number go up. Real means the price effect has been taken out. for a country, quarter by quarter. Compare each quarter with the one immediately before it. If output is lower in one quarter than in the quarter before, and lower again in the quarter after that, the test is met and the label applies. If that never happens, the test is not met and the label does not apply. There is no second condition and no room for interpretation.

The rule needs nothing but one series and the ability to subtract, and it is used for precisely that reason. It does not ask how far output fell. The test is about direction and not about size, so a fall of one tenth of an index point counts exactly as much as a fall of ten points. The rule does not ask what else was happening. Employment could be climbing all the way through and the answer would not change. The rule says nothing about how long the weakness lasted beyond the two quarters it looks at, and nothing about whether the fall was general across the economy or concentrated in one bad season for one crop. All of that lies outside the test, not through oversight but because the test asks about direction and nothing else.

Two words have to be nailed down or the test becomes slippery. Consecutive quartersTwo three month periods that sit next to each other in the calendar, with no gap and no quarter in between. Two bad quarters in the same year are not consecutive unless they are neighbours. means neighbours, not merely two bad quarters somewhere in the same year. The comparison is with the quarter immediately before, not with the same quarter a year earlier. A year earlier is a different comparison entirely and can point the other way. A crop that comes in every winter makes every winter quarter look big and every summer quarter look small. Quarterly series are therefore usually seasonally adjustedAdjusted so that a pattern that repeats at the same point every year, such as a harvest or a festival buying season, does not show up as a genuine rise or fall in activity. before anybody applies this test to them. Applying the rule to an unadjusted series measures the calendar, not the economy.

Work it once, on an invented series, so the mechanics are concrete. Take the Republic of Sankhya, an invented country, and a real output reading of 100 in its first quarter, then 98, then 99, then 97, then 95, then 96. Now slide a two quarter window along the five changes. Quarter two fell and quarter three rose, so no. Quarter three rose and quarter four fell, so no. Quarter four fell and quarter five fell, so yes, and the label applies from that point. Quarter five fell and quarter six rose, so no. The rule fired in exactly one window out of four, and finding that window took nothing but five subtractions anybody could redo.

What does a crude rule that anybody can check actually buy?

Treating the technical rule as a poor substitute for real analysis is tempting, and the reading misses what the rule is for. A rule that ignores almost everything has three properties that a careful judgement can never have, and each of them is worth something real.

The first is that it is reproducible. Hand the same series to ten people and ten identical answers come back. There is nothing in the test for them to differ about. The second is that it is checkable by the reader. If a report says the test was met, the same series can be pulled and the subtraction redone in about a minute. No judgement can be rechecked that way. The third is that it cannot be argued away by somebody with an interest in the answer. Nobody can lean on a subtraction. A crude rule that everybody can verify has a real advantage over a subtle judgement that only some people are in a position to make, and that advantage is not about accuracy at all.

Rules of this kind are already trusted in ordinary life for exactly these reasons. A cricket team is all out at ten wickets, not when the batting looks hopeless. A train is late when it arrives after the scheduled minute, not when the delay felt significant. Neither rule captures what anybody actually cares about, and both survive because a rule that anybody can check settles an argument that a judgement would keep open. The cost is the same in every case: the rule fires in situations nobody would have called by hand, and stays silent in situations everybody would have.

What does an economic recession mean, on its own terms?

Set the rule aside completely for this section. An economic recession is not a test that gets applied to a series. An economic recession describes a condition. The condition is that activity across the economy is significantly weak, in more than one part of it, for more than a moment. Somebody has to look at the evidence and decide whether that description fits. Deciding whether the description fits is what makes an economic recession a judgement rather than a rule, and no amount of care removes the judgement from it.

Three questions are usually asked together, and the answers to all three are needed before anybody can say the description fits. The first is depth: how far did activity fall from its highest point to its lowest? A fall of a fraction of a per cent and a fall of eight per cent are different events, and the rule cannot tell them apart while the judgement must. The second is breadth: how many of the measures being watched fell along with output? Output, employment, incomes, household spending and industrial production can move together or they can come apart, and a fall that shows up in one of them while the rest hold is a different condition from a fall that shows up in all of them. The third is duration: how long did activity stay below its peak? A single soft quarter that is fully reversed in the next one has not left anybody worse off for long.

The three tests are applied at the same time rather than one after another, and a weak answer on any one of them weakens the whole judgement. That is the part people get wrong. The three tests are not a checklist where two out of three is a pass. Deep but narrow means one sector had a terrible quarter. Broad but shallow means everything wobbled slightly at once. Wobbles like that happen for reasons that have nothing to do with an economy being in trouble. Long but neither deep nor broad means a period of dull growth. Dull growth is a real condition and it is not this one. Only when all three read weak together is the description worth using.

Two further pieces of vocabulary matter here. The question is where the economy stands rather than where it may be heading, so a judgement leans hardest on coincident measuresMeasures that move at roughly the same time as overall activity, rather than ahead of it or behind it, so they describe where things stand now rather than where they may be heading.. And every one of those measures is subject to revisionThe updating of an already published figure once more complete information arrives. An early estimate is built from partial returns. The number for a past quarter can change after it was first put out., which means the evidence itself moves under the judgement. A quarter that looked flat when it was first estimated can read as a fall once fuller returns arrive, and the reverse happens too. The judgement is not slow because the people making it are cautious; it is slow because the evidence it needs is genuinely not all there yet.

What can a judgement see that a rule never will?

The judgement earns its cost in one specific situation, and it is worth naming precisely rather than in general. Two soft quarters inside an economy that is otherwise sound are not the same event as a broad and lasting collapse. The two events produce the same reading on the rule and completely different readings on everything else, and only the judgement can tell them apart.

Picture a street of ten shops. In the first case one shop shuts for two months because the owner was unwell, takings on the street dip slightly, and every other shop keeps trading, keeps its staff and pays the same wages. In the second case demand across the whole street falls away, every shop cuts hours, two let staff go, and takings are still below where they started a year later. Adding up the street's takings in both cases can produce the same small dip. The total hides the difference, and the difference is the only thing anybody actually cares about. Looking at employment and at each shop separately is what tells the two cases apart, and that is exactly what breadth and duration are doing at the level of a country.

There is a second thing the judgement sees. Because it looks at the level of activity rather than only at the change from one quarter to the next, it can register weakness that never shows up as two neighbouring falls at all. A path that gives up ground, takes a little back, gives up more, takes a little back again and ends the year well below where it started has never once produced two falls side by side. The rule has nothing to report. The level, meanwhile, has fallen the whole way. The ratcheting path is the case the whole comparison turns on, and it is worked in full below.

Now that both are defined, where do the two actually part?

Both definitions are now built. The rule is a test on one series; the judgement is a description checked against many. Each definition can now be held in mind without borrowing from the other, so a comparison is finally honest. Four differences separate them, and they are really one difference seen from four sides.

The first is what gets measured: one series against many series. The second is how many measures have to agree: exactly one, against as many as the person judging can lay hands on. The third is how quickly the answer arrives. The rule answers the moment the second quarter of data lands. The judgement answers once there is enough evidence to be worth stating, later and sometimes much later. The fourth is who can dispute the answer. Nobody can dispute a subtraction and anybody can dispute a judgement. Being open to dispute is not a flaw in the judgement but the direct consequence of it being one.

The two are not rivals and neither is a worse version of the other; they are answers to two different questions, and the shorthand exists because the better answer is too slow to be useful at the moment somebody needs an answer. That is the whole reason a convention grew up at all. If the careful judgement could be delivered on the day the quarter closed, nobody would ever have needed a rule of thumb. The judgement cannot be delivered that fast, so a rule that trades accuracy for speed fills the gap, and everybody agrees to know that it is doing so. The trouble starts only when a reader forgets the trade and treats the fast answer as though it were the careful one.

Can an economy be in one and not the other?

Yes, in both directions. Two definitions applied to the same path give two answers, and those two answers make four combinations. All four occur. Each of the four is worked below on a Republic of Sankhya quarterly path, and the arithmetic produces the answer rather than the answer being taken on trust. Every path runs six quarters, real output is set to 100 in the first quarter, and employment and incomes are indexed the same way.

Read the four paths as four different economies, not as one economy at four moments. The rule column is computed by sliding the two quarter window along each path. The judgement column applies three stated tests: a deepest fall of 2.5 per cent or more, all three measures below their starting level in the same quarter, and output below its peakThe highest level a series reached before it turned down. The starting point from which a fall is measured. for three quarters or more. Those three lines are assumptions rather than anybody's official definition, and a different set of thresholds would move some paths across the boundary.

Work each row rather than reading the answers off. Path A falls in all five changes, so the rule fires at the first opportunity, the deepest fall is fifteen index points on a base of 100 and therefore 15.0 per cent, employment ends at 87 and incomes at 89, and output sits below its peak for all five remaining quarters. Every test reads weak, so both definitions say the same thing. Path D never falls at all, so nothing fires and nothing reads weak, and again the two agree. Path A and path D are the easy rows, and they are why people assume the two definitions are the same thing.

Path B is where they come apart in the first direction. Output falls one point, then one point again, so the rule fires on quarters two and three exactly as it did for path A. Then the deepest fall is two index points, or 2.0 per cent, under the 2.5 per cent line. Employment gives up a single point and takes it straight back. Incomes do not fall in any quarter of the six, ending at 106. The rule fires on path B and the judgement does not. The two are answering different questions about the same six numbers, so both answers are correct.

Path C is the other direction and it is the important one. Output goes 100, 96, 98, 94, 96, 92. Look at the five changes: down four, up two, down four, up two, down four. Three falls out of five, and not one of them has another fall next to it, so the two quarter window never closes on two falls and the rule stays silent through the whole path. Now apply the judgement. The deepest fall is from 100 to 92, a fall of 8.0 per cent. Employment goes 100, 98, 97, 96, 95, 94, falling in every single quarter. Incomes go 100, 99, 99, 98, 98, 98. Output is below its peak in all five quarters after the first. Path C is deeply, broadly and persistently weak by every test, and the label that most headlines rely on never once fires.

What does the corner without a label look like up close?

Path C is the square almost nobody checks for, and it is the one where a reader waiting for a label waits forever. The shape that produces it is ordinary rather than exotic. Activity falls back, some of that is made up in the following quarter, then it falls back further, some of that is made up again, and the level ratchets downward while every fall is separated from the next by a small recovery.

Nothing about that shape is unusual. A stall outside an office building sees takings drop when the building empties out, then pick up a little when a new tenant moves in on one floor, then drop again when a second floor empties, and so on. Month by month the takings never fall twice running, and after a year the stall is taking far less than it was. Ask the stallholder whether things are worse and the answer is obvious. Apply a two-in-a-row rule to the takings and it never fires once. The rule is looking at the change from one period to the next while the damage is accumulating in the level, and those are different quantities.

The troughThe lowest level a series reached before it turned back up. The end point against which a fall is measured. on path C is 92 in the sixth quarter, eight index pointsThe unit of a series that has been set to 100 at a chosen starting point. A move from 100 to 96 is a fall of four index points, or four per cent of the starting level. below where the path started, and the employment line beneath it never once turns up. Set the two panels one above the other and the disagreement is visible in a glance: the top line zigzags and the bottom line slides. A reader watching only the top line and waiting for two falls side by side sees a series that keeps bouncing back and concludes, quite reasonably given what is in front of them, that nothing has happened.

Readings taken off the panel and written down so they survive without it. A straight run at a pull of 3 points gives 100, 97, 94, 91, 88, 85. The rule fires, all three judgement tests read weak, and both definitions agree. A dip and a recovery at a pull of 1 point gives 100, 99, 98, 99, 100, 101. The rule fires, the deepest fall is 2.0 per cent, and that is the rule without the reality. A bounce at a pull of 2 points gives 100, 96, 98, 94, 96, 92. The rule stays silent, the deepest fall is 8.0 per cent, and that is the reality without the rule. A straight run with the pull set to plus 1, meaning activity being pushed up rather than dragged down, gives 100, 101, 102, 103, 104, 105, and neither definition fires. The one setting worth noticing is the bounce at a pull of 5 points. The pull finally overwhelms the tick-up, both falls land side by side and the rule fires after all, so the top left square is a property of the shape rather than a permanent hiding place.

How should a headline that says a recession has begun be read?

Three questions, asked in order, and they take about a minute between them. The first is which definition is being used. If the report says two consecutive quarters of falling output, it is reporting the rule. If it describes weakness across several measures over a period, it is reporting a judgement. If it says neither, the reader has not been told what was tested, and every number that follows is resting on something that cannot be seen.

The second is who applied it. A rule applied by anybody is reproducible by anybody, so a reader can pull the same series and check it. A judgement made by a body that publishes its reasoning can be read and weighed. A judgement made by nobody in particular, appearing only as the writer's own summary, is a view and should be read as one. The third question is whether the other measures agree. If output, employment and incomes are all pointing the same way, the two definitions are going to give the same answer and the reading can stop there. If they are pointing different ways, the reader is standing in one of the two corners where the definitions part, and which corner decides what the news actually means.

A headline that does not say which definition it used has not told the reader whether anything happened. That is a strong claim and it holds up: without knowing what was tested, the same words could be describing path B, where a rule fired inside a sound economy, or path A, where everything is genuinely weak at once. Path A and path B are not remotely alike, and the words in print can be identical.

Why does a lender care which of the two fired?

Take a lender writing small household loans in one district. A borrower brings home Rs 45,000/- a month and pays an instalment of Rs 12,000/-. The lender is not interested in labels for their own sake. It is interested in whether that Rs 12,000/- keeps arriving, and that depends on whether the borrower keeps the job and whether the household's income holds. Employment and incomes decide whether the Rs 12,000/- keeps arriving, and both sit in the judgement rather than in the rule.

But the two definitions reach the lender at different moments, and that is the whole problem. A technical label arrives quickly, gets reported widely, and moves sentiment and the price of money on the day it lands. The economic condition arrives slowly, and is what actually decides whether borrowers can pay. A lender who tightens on the day a technical label fires is acting on the fast signal, and a lender who waits for a label that a sawtooth path will never produce is acting on nothing at all. The two errors sit in the two corners of the grid, one in each.

The reading that avoids both is unglamorous. Treat the technical label as news about sentiment and pricing, real and arriving early. Treat the employment and incomes series as news about whether the loan book will be repaid, and that second reading is what the judgement watches. An analyst covering a consumer business reads it the same way. The label may move the share price this week. The breadth of the weakness decides whether volumes are still falling in three quarters. An investor holding for years cares about the second and should expect to be repeatedly wrong-footed by the first. Neither signal says what anybody ought to do with money. Each one answers a different question, and the error is putting the wrong question to a signal.

The Republic of Sankhya, its output, employment and income series, the street of ten shops, the stall outside the office building and the borrower on Rs 45,000/- a month are invented.
Educational material. Not advice on any investment, tax, budget or market position.