Secular vs Cyclical Growth: Telling Them Apart
Secular growth comes from a structural change and does not reverse when activity falls. Cyclical growth comes from the level of activity and does reverse. On the way up the two produce identical numbers. The only evidence that separates them is behaviour through a bad year. Almost all real growth contains both, and the work worth doing is estimating the mix and stating how wide it is.
Two fields can grow at exactly the same rate, for exactly the same number of years, for completely different reasons, and no amount of staring at the growth figures will separate them. The separation is real and it matters enormously, and the evidence for it lives in the years everybody skips over. Each kind is therefore defined on its own terms first, with no reference at all to the other, and only then are the two set side by side.
What is secular growth, on its own terms?
Secular growth is demand rising because something about the world changed and stayed changed. Not because buyers happen to feel flush this year. Because the underlying arrangement that produces the buying is different from what it was, and it is not going back.
Think about a household that moves out of a rented single room into a flat of its own. From that year onward it buys more of almost everything: more furniture, more electricity, more paint on more walls. If somebody in that household loses a month of work, the buying slows. The household does not go back to the single room. The flat is still the flat. The change that lifted the spending already happened, and it happened once.
The thing driving secular growth has already occurred, so weak activity slows the growth rather than reversing it. The word earns its keep on that property alone. Secular is not a claim that growth will be fast, or steady, or long lasting. Secular is a claim about where the growth came from, and therefore about what has to happen for the growth to stop.
Two things do eventually stop it, and neither of them is a bad year. The first is saturationThe point at which almost everyone who is going to buy the thing already has it, so the structural change driving demand has nothing left to add. The distance left to saturation is taken up separately.: everybody who was going to move into the flat has moved into it. The second is a second structural change running the other way. Both are slow, both are visible in advance to anyone looking, and neither has anything to do with how activity felt last quarter.
What is cyclical growth, on its own terms?
Cyclical growth is demand rising because the level of activity is high, and falling again when the level of activity is low. The driver is a position rather than a change. Activity is somewhere on a range that it moves around on, and demand moves with it.
The street vendor outside a construction site is the whole idea in one picture. While the site is busy there are two hundred workers buying lunch, and the vendor's takings rise every month. When the site finishes, the workers leave, and the takings fall. Then another site opens two streets away and they rise again. Nothing structural changed about how people eat. The number of people standing there changed.
The condition producing cyclical growth can go away and then come back, so the same demand can rise, fall and rise again without anything having been gained or lost in between. This is not a statement about how long a cycle takes or how deep it goes. Cycles do not arrive with a fixed length, a fixed depth or a schedule. The absence of a schedule is precisely why cycles are hard to work with, and why no length is stated for one.
One consequence is worth holding on to now. Because a cyclical component comes back to where it was, it averages out to roughly nothing over a stretch that contains a full up and down. A cyclical component contributes swing, not level. Over a long enough record it lifts nothing at all.
Cyclical growth defined without using the word secular anywhere. Which of these does it?
Why does a rising number reveal nothing at all?
Set the two definitions side by side and ask what each produces in a year when activity is strong. Secular growth produces a rising number. Cyclical growth produces a rising number. The two are not similar in a good year. The only thing visible is the size of the increase, and both mechanisms can produce any size at all, so the two numbers are identical.
A bad year is the only place where the separation between the two becomes visible. The inconvenience of that does not make it any less true. Everything else people reach for is weaker than it looks. A cycle can stay up for years, so a long run of growth is not evidence. A cycle can be violent, so a large increase is not evidence either. A convincing story about why the growth is structural is not evidence at all; it is a hypothesis with good presentation.
The same rule is already applied elsewhere without being called anything. Anyone can look reliable while things are going well. A bad month is the only month where the two possible explanations of earlier behaviour finally do different things. Behaviour in a bad month is therefore the only behaviour worth learning from.
A ten year record arrives for a field about which nothing is known. Which period in it says most about what kind of growth it holds?
What does a bad year actually do to each of them?
The whole method sits here, so work it through carefully. In a downturnA stretch when overall activity is falling rather than rising. The length of a downturn, what sets it off and how it ends is taken up separately, under the economic cycle., the cyclical part of demand falls, because the condition that was producing it has weakened. Then, when activity recovers, it comes back. Nothing was destroyed and nothing was created; the swing went down and then up.
The secular part behaves completely differently, and the reason is almost boring. The structural change is still there. The household is still in the flat. The household may spend less this year, so the secular part slows. Reversing it would require the structural change to unwind, and structural changes do not unwind over a dip in activity.
A field going through a bad year hands over the single most informative period its record contains. Bad years are usually treated as the exact opposite. The instinct is to call a contraction a gap in the data, to skip it, to normalise it out, or to describe the record as growth apart from the one bad year. Every one of those moves throws away the only observation that separates the two explanations.
What does the record actually say, once the bad year is looked at first?
The invented record used here runs across five years of volume growthThe change in the number of units sold, with price left out of it entirely. Volume growth is a different figure from revenue growth, and the two are separated under volume and price. for the coatings field: 2.1 per cent, 6.8 per cent, minus 1.4 per cent, 7.2 per cent and 4.5 per cent. The five figures average 3.84 per cent. Measured against that average the best year sits 3.36 points above it and the contraction year 5.24 points below it, and that spread is the whole of the swing the record contains. The third year is the only one carrying information, so take it first.
The field contracted by 1.40 per cent. The fact that it fell establishes least, so do not stop there. Ask how far it fell. A field with no structural component at all would be pure swing, and a pure swing averages nothing over a record that contains a full up and down. The five observations average 3.84 per cent. For that average to be nothing with no structural part carrying it, the contraction year would have had to absorb the whole 19.20 points the record adds up to. A worst year of about minus 20.60 per cent, then, instead of minus 1.40 per cent.
The size of the contraction is the evidence, and the mere existence of a contraction is very nearly worthless. A fall of 1.40 per cent in the worst year of a record that averaged 3.84 per cent is a shallow fall, and something was holding it up. Equally, a field with no cyclical component at all would print the same figure every year, and this record swings 8.60 points from its best year to its worst. So both pure readings are ruled out by the record and everything in between is not.
The field fell 1.40 per cent in its worst year. What does the size of that fall suggest?
For this record to contain no structural component whatsoever, what would have had to be different about it?
What does each reading imply about what comes next?
The distinction stops being vocabulary here and starts costing money. Take one current figure, say the 4.5 per cent the field grew in its most recent year, and hand it to two readers who disagree about what produced it.
The reader who thinks it is mostly swing expects reversionThe tendency of a series that has moved away from its own average to come back towards it. How fast that happens, and whether it can be leaned on, is taken up separately.. A peak is not a level to them, it is a position on a range, and positions on ranges come back. The change that produced the growth has not been undone and will keep producing it until saturation arrives, so the reader who thinks it is mostly structural expects persistence.
The same current number supports two opposite expectations depending on which reading is held, and that is the entire reason the distinction is worth the trouble of making. If both readings pointed the same way the distinction could be skipped and nothing lost. They do not. One of them says the number in front of the analyst is a high water mark and the other says it is a run rate.
Can growth be both at once?
Almost always, yes, and pretending otherwise is where most of the damage on this subject gets done. The coatings field described here has households forming and repainting more often, a structural driver. New construction and industrial capital spending sit in the same figures and move with activity. Both are running at the same time in the same figures. Nobody reports them separately because nobody can.
The realistic question is never which one, but in what proportion, and the proportion is estimated rather than measured. An estimate is a legitimate output. An estimate becomes illegitimate the moment it is dressed up as a measurement, and the tell is always the same: a split stated to a precision the record cannot carry.
So here is what this record can and cannot support. The record can rule out the two extremes, as the drawing above showed. A negative structural part makes no sense, and a structural part above the best year would make the swing negative in every single year including the boom. Within those extremes the structural part has to be somewhere between 0.00 and 7.20 per cent, and that possible band is uselessly wide.
Narrowing it requires adding a judgement, and requires saying out loud that one was added. The judgement carried here is that a swing component should average close to nothing across a record that contains a rise, a fall and a rise again, and the tolerance chosen is one full point either way. The judgement puts the structural part between 2.84 and 4.84 per cent. Half a point instead narrows the band to 3.34 through 4.34; one and a half points widens it to 2.34 through 5.34. The band moves when the tolerance moves. The width of the answer is a decision made by the analyst, not something the data handed over.
Somebody asks straight out: is this field's growth secular or cyclical? What is the right answer?
What does the decomposition look like when the setting is moved?
Reading about a band is not the same as watching one refuse to close. In the calculator below, the five year record is fixed and cannot be edited. The control sets the structural component, held flat across all five years, and everything else is the residualWhatever is left over once the modelled part is subtracted. A residual carries everything unaccounted for, including the analyst's own mistakes. A large residual is a warning rather than a result.: the swing component, year by year, computed as the record less the chosen setting. At every setting the two halves rebuild the record exactly, and that is the point.
Set the structural component and watch the swing take whatever is left
The record never moves. The flat green line is the chosen setting, the bars are the residual running from that line up or down to the actual figure, and the dark markers are the record itself. The swing readout does not change at any setting. The record therefore cannot choose one.
With the structural part held flat at 3.40 per cent, the swing component runs minus 1.30, 3.40, minus 4.80, 3.80 and 1.10, averaging 0.44 points a year, and the record cannot tell this setting apart from any other between 2.84 and 4.84.
Drag the setting from 3.40 per cent up to 4.84 per cent. What happens to the swing component measured from its best year to its worst?
What does separating them change about a forecast?
Separating them changes exactly one thing, and that one thing is the whole payoff: it changes what is carried forward. The structural part continues, on its own logic, until saturation. Continuing is the one thing swing does not do, so the swing part does not continue. So a forecast built without the separation carries forward a component that is defined by its refusal to persist.
Make it concrete on the case entity. Sarvani Coatings Limited, an invented maker, grew volume 6.0 per cent in its most recent year against a field that grew 4.5 per cent. Now watch what an analyst who has not separated anything is actually doing when she carries 6.0 per cent forward. At the setting worked above, that 6.00 points contains a field structural part of 3.40, a field swing part of 1.10, and a company part of 1.50 that is the difference between Sarvani Coatings and the field it sells into. Three components, added together into one number, none of them separated, and only one of them with any claim to repeat.
| The record, decomposed at a structural setting of 3.40 per cent, invented throughout | Recorded | Structural | Swing, as residual |
|---|---|---|---|
| Year one | 2.10 | 3.40 | minus 1.30 |
| Year two | 6.80 | 3.40 | 3.40 |
| Year three, the contraction | minus 1.40 | 3.40 | minus 4.80 |
| Year four | 7.20 | 3.40 | 3.80 |
| Year five | 4.50 | 3.40 | 1.10 |
| The five years added together, in percentage points | 19.20 | 17.00 | 2.20 |
Read the bottom row across and the arithmetic is unarguable: 17.00 plus 2.20 is 19.20, so nothing has been lost or invented in the decomposition. Read the last column down and the teaching is unarguable too: the swing component adds to 2.20 points across five years, or 0.44 a year, close enough to nothing for the setting to survive the tolerance test. An analyst who has not separated the two is extrapolating the sum. The sum is wrong by the swing component every single time, and the amount she is wrong by is exactly the thing she never computed.
None of this claims the forecast will be wrong. Swing components are not obliged to fall in the year after they were ignored, so a number carried forward blind can land perfectly well. The number may turn out right and the reasoning cannot be, and everything downstream that leans on the reasoning inherits the fault rather than the luck.
An analyst carries 6.00 per cent volume growth forward into next year without separating anything. What is wrong with the forecast?
The split that was right to attempt and wrong to state
Meghna Iyer does the work properly and then ruins it in the last sentence. She looks at the five year record, sees correctly that a shallow contraction rules out pure swing, sees correctly that a real contraction rules out pure structure, and concludes that the growth is a mixture. All of that is sound and most analysts never get that far.
Then she writes it down as two thirds structural and one third swing. On a mean of 3.84 per cent that is 2.56 structural and 1.28 swing, stated as though it had been measured. The split was not measured. The split was inferred from five observations containing exactly one contraction, and it does not even sit inside the plausible band the same record supports: 2.56 sits 0.28 points below the 2.84 lower edge, so the record does not merely fail to confirm the split, it argues gently against it.
The cost is not in the separation. The separation was the right exercise. The cost is the false precisionStating a figure to more decimal places, or with more confidence, than the evidence behind it can carry. The number looks measured and is not, and that is precisely why it travels well. bolted on to the separation. A point estimate travels through a model as though it were a fact, and every figure computed downstream from it inherits a confidence nobody ever earned. The band would have travelled too, if anybody had carried it.
The fix takes one extra clause. State the mix as a range, state the judgement that produced the range, and carry the range forward into whatever the mix feeds. An analyst who cannot state the uncertainty in her own split has not finished the exercise, she has only finished the arithmetic.
An analyst states that the growth in a field is exactly two thirds structural. What is the first question to ask?
What does this change in the hands of somebody actually using it?
For an analyst the habit is small and it runs before the model rather than inside it. Before any growth figure is carried into a following year, she asks which part of it is entitled to repeat. The question takes a minute, produces a range rather than a number, and the range goes into the model as a range. Nothing about the question requires a view on the economy or a call on a turning point, and it therefore survives being wrong about both.
For a lender the same reading answers a different question. A borrower whose recent growth was mostly swing has a repayment capacity that will be tested in the next weak stretch, and the covenant that looked comfortable at the top of the range will be the one that binds. A borrower whose growth was mostly structural has a slower and steadier capacity. The lender is not forecasting the cycle, she is asking which of two borrowers has a capacity that survives one, and the decomposition answers exactly that.
For a household the version is smaller and identical in shape. A person whose income rose because of a promotion is in a different position from a person whose income rose because overtime was plentiful this year, even where the two increases are the same rupees. The first is structural and survives a slow quarter. The second is swing and does not. Anybody deciding how large a monthly commitment to take on is making exactly this separation, usually without a name for it, and the ones who get it wrong are the ones who treated a good year as a level.
One last practical note about extrapolationCarrying an observed rate forward into a period not yet observed, on the assumption that whatever produced it is still running. The assumption is the whole content of the exercise, and naming it matters more than the arithmetic.. The separation does not reveal the future. The separation identifies which part of the past may still be used, a much smaller claim and a far more defensible one. If somebody objects that the mix cannot be known, the objection is correct. It cannot. The whole job is to state the band, state what was assumed to narrow it, and stay checkable.
Where conduct sits around a number like this
The separation above is arithmetic and reasoning rather than compliance, so no regulator decides any part of it. One thing sitting beside it is decided elsewhere. Disclosure by a research analyst, covering how a figure was arrived at and the assumptions sitting behind a projection put in front of a reader, is a conduct matter that sits with the Securities and Exchange Board of India (SEBI).
Whatever SEBI requires today is at sebi.gov.in, and that is where to take it from. A real company's own volume disclosure, rather than an invented one, sits in the results filings at nseindia.com and bseindia.com, and the product mixThe blend of what a maker actually sold in a period. Selling more of the pricier line lifts average realisation without any price being raised, and how that is separated out is taken up separately. commentary that changes how a volume figure should be read is usually in the same document.
Where a volume disclosure and the conduct rules can be checked
The reasoning is arithmetic that can be reworked on paper in ten minutes. The sources below settle where a real volume disclosure is published and who sets the conduct expected of somebody putting a decomposition in front of a reader.
| Source | Site | Consulted |
|---|---|---|
| Securities and Exchange Board of India | sebi.gov.in | 28 August 2026 |
| National Stock Exchange of India | nseindia.com | 28 August 2026 |
| BSE Limited, formerly the Bombay Stock Exchange | bseindia.com | 28 August 2026 |
| The teaching record used in this sequence | Held inside this library. Not a published source and not checkable outside it. | 28 August 2026 |
Sarvani Coatings Limited, the coatings field around it and the analyst Meghna Iyer are invented.
Educational material. Not advice on any investment, tax, budget or market position.