Bull market vs bear market

Ask where the threshold is written down and the trail goes cold quickly. It is not in any regulation. India's securities regulator publishes an investor education page on precisely this topic, and its definitions are entirely qualitative: a bull market is described as a phase when prices are rising and investors are optimistic, a bear market as a phase when prices are falling and the outlook is negative. There is no percentage in it at all. No exchange rulebook defines the terms either, because they are not operative terms in any rule. Nothing depends on them. No margin changes, no surveillance measure triggers, no disclosure obligation attaches when an index crosses some line from one word to the other.

The 20 per cent figure is a newsroom convention. It exists because financial journalism needs a repeatable, checkable trigger for a headline, and a round number computed from a peak is the cheapest such trigger available. It is a good convention for that purpose. It is unambiguous once the peak is known, it can be computed by anyone from public data, and it produces a consistent editorial standard across outlets. Those are real virtues, and none of them is a virtue about markets. A threshold designed so that reporters agree with one another is not thereby a threshold that tells you anything about what happens next.

Notice what the convention actually requires before it can fire. It needs a peak, and a peak is only knowable in retrospect, because a high is only a high once nothing has exceeded it since. It then needs a decline of a specified size from that retrospective point. Both inputs are backward-looking by construction. A rule assembled entirely out of backward-looking inputs cannot, even in principle, contain forward-looking information. This is not a criticism of the rule. It is a description of what kind of object it is: a label for something that has finished happening, in the same family as a coroner's finding or a match report.

That last row is where this page is going. The first four rows are vocabulary. The fifth is the only one with any measurable content, and it is not defined by a decline from a peak at all. Keeping the two apart is most of the work, because in ordinary usage they are hopelessly run together: people say bear market when they mean the tape has turned violent, which is a claim about volatility, and they say bull market when they mean the tape has turned placid and drifting, which is also a claim about volatility. The percentage rule is not measuring either of those things.

Before the argument goes any further, here is the arithmetic in a form you can run yourself. Enter a current index level, the recent peak, the recent trough and a long-term average, and the tool below computes the drawdown from the peak, the distance travelled up from the low, and where price sits against the average. It then reports which conventional label those numbers earn.

Use it as an instrument, not an oracle. Everything it reports is a statement about what has already happened, computed from four numbers you supplied. That is genuinely useful, in the way a thermometer is useful. It is not a forecast, and the tool says so in its own output, because the single most common error with regime language is to hear a description of the present as a claim about the future.

The claim that the label lags is easy to make and easy to nod along to. It is more useful to measure it. The figure below takes a single complete cycle, one rise, one fall, one recovery, and marks the exact sessions on which the conventional vocabulary would have changed. Everything quoted is computed from the drawn path rather than asserted about it.

The asymmetry embedded in that arithmetic is worth sitting with, because it is not a quirk of the simulation but a property of how percentages work. The deeper the eventual decline, the earlier in it the 20 per cent line is crossed, and so the more useful the label would have been. The shallower the eventual decline, the later in it the line is crossed, and the more likely it is that the line is never crossed at all. The rule is therefore most informative in exactly the cases where it matters least, the catastrophes that are obvious anyway, and least informative in the ambiguous middle where a trader would actually want help.

Worse, the rule has no memory and no confirmation requirement. A market that falls 19.6 per cent from a peak and turns is, by convention, not a bear at all, and a market that falls 20.1 per cent and turns the following week is one. Two paths that were the same market behaving the same way get different names, and the names then go into headlines, into fund commentary and into the stories people tell themselves about what they lived through. The false precision is not harmless. It manufactures a distinction where none existed and then invites people to reason from it.

If the labels are empty, it would be tidy to conclude that the whole idea is empty. That conclusion is wrong, and the distinction matters more than anything else on this page. There is something real underneath the vocabulary. It is just not what the vocabulary measures.

The real property is this: large moves cluster with large moves and small moves cluster with small moves. A violent session is far more likely to be followed by another violent session than a calm one is. Markets do not deliver their volatility evenly, sprinkled at random through the year; they deliver it in bursts, with long placid stretches in between. This is one of the most robust and most replicated findings in the study of financial prices, and it is the thing the word regime should be reserved for.

The cleanest demonstration is not a chart of a market. It is a chart of a market next to the same market with its days shuffled. Reshuffling preserves the average move, the distribution of moves, the best day and the worst day, and every summary statistic that ignores order. It destroys nothing but the sequence. If the calm-and-storm structure survives the shuffle, it was never in the sequence to begin with; if it vanishes, the sequence was carrying it.

That number is the whole case for taking regimes seriously. It also quietly tells you what a regime is not. Nothing in the shuffle test says anything about direction. The correlation measured is between the size of consecutive moves, with the sign thrown away. A market can be in a loud regime while going up, and this happens; it can be in a quiet regime while grinding down, and this happens too. Volatility and direction are separate axes, and the bull and bear vocabulary collapses them into one, which is the source of a great deal of confused thinking.

This is also why a volatility index is a genuinely different instrument from a price chart rather than a decoration on top of one. It reads the axis the price chart does not. If you want to know how the market is behaving rather than where it has been, that is the measurement to look at, and the volatility index published for the Indian market is the standard version of it.

Granting that regimes exist, the immediate question is what follows. Here it pays to be extremely precise, because this is the exact point at which a true statement about markets gets stretched into a false one. Take a long simulated history, sort every session by how volatile the previous twenty sessions were, put the sessions into five equal buckets, and then ask two separate questions of the very next session. How far did it move? Which way did it go?

Read the two panels together and the honest summary of regime knowledge falls out. Knowing the regime tells you a great deal about the amplitude of what is coming and close to nothing about its sign. This is not a limitation of the particular measurement; it is the consistent finding across the literature on financial prices, and it survives changes of market, instrument and period far better than almost anything else in technical work.

It is also the precise reason most regime-based trading advice is wrong in a specific and diagnosable way. The advice usually runs: identify the bull market, then buy; identify the bear market, then sell or stand aside. That takes a variable which conditions amplitude and uses it to make a call about direction. The left panel is doing the persuading and the right panel is doing the work, and the right panel says the direction call has no support.

The rows sort into two groups and the sorting is the lesson. Everything about magnitude, participation and co-movement is measurable, more or less promptly, with more or less noise. Everything about direction and duration is not. A trader who builds on the first group is building on something. A trader who builds on the second is building on a story about the first.

One qualification is owed here, because the table risks flattering the measurable rows. Measurable does not mean stable. Correlation between holdings rises in stress, which is well documented and inconvenient, since it means diversification thins out at the moment it is most wanted. But the rise is not instantaneous and not uniform, and a correlation estimated over the last sixty sessions is a description of those sixty sessions rather than a promise about the next ten. The same caution applies to every row in the yes column. These are measurements with error bars, not readings off a dial.

Everything above concerns a single series. The moment the series is an index rather than an instrument, a second problem appears, and in the Indian market it is a large one. A benchmark is a weighted average, and a weighted average can be doing something that most of its constituents are not.

The mechanism is arithmetic rather than mysterious. A broad benchmark is weighted by size, so the largest few members carry a disproportionate share of it. If those few are rising strongly while the rest are drifting lower, the headline goes up. Anyone watching only the headline sees a bull market. Anyone holding a portfolio drawn from the rest of the list is in something that does not feel like one at all, and is not wrong to feel that way. The measurement that separates the two is breadth: not where the index is, but how many of its members are participating.

This is not a hypothetical structural feature of Indian benchmarks; it is how they are built. A headline index is a capitalisation-weighted construction in which a modest number of the largest constituents dominate the arithmetic, which is a sensible way to measure the market's aggregate value and a poor way to measure the experience of a typical holding. If you want the details of how the main benchmark is put together and weighted, the guide to the headline index covers the construction. The practical consequence for regime reading is direct: an index-level regime label describes the index, and generalising it to your own positions is an inference you have not earned.

That second card is the one that gets ignored. Breadth deterioration is often described as a warning, and there is a persistent temptation to treat it as a timing tool. It is not one. Narrowing can persist for a long time and can resolve in either direction, and a reader who takes a falling participation count as an instruction to act has smuggled a forecast back in through a side door after the front door was closed. The value of breadth is that it stops you mistaking an index statement for a portfolio statement. That is worth a lot. It is not worth a trade on its own.

So what is left, once prediction is off the table? Something narrower than most treatments promise and considerably more usable than nothing. The honest use of a regime is conditioning: letting the measurable properties of the current environment set the parameters of what you do, without letting them generate an opinion about which way the market goes next.

The clearest instance is position size. Suppose you risk a fixed drawdown per position, and you set the stop at a fixed multiple of how far the instrument has lately been moving each session. Those two commitments together determine your position size, mechanically, with no view required. In a placid stretch the stop distance is small, so the same rupee risk buys a larger position. In a violent stretch the stop distance is wide, so the same rupee risk buys a smaller one. The regime has changed your exposure substantially, and you never formed a view.

Notice how much this differs from the usual advice, which is to be aggressive in a bull and defensive in a bear. That advice requires you to know which one you are in, which is the thing you cannot know until afterwards. Conditioning on volatility requires only that you measure something that is measurable now and demonstrably persists, which is precisely what the shuffle test established. It is the same instinct, stripped of the part that cannot be done.

The same logic reaches beyond sizing. Whether a strategy's assumptions still hold is a conditioning question: a method that depends on moves running on will behave differently when moves stop running on, and measuring how strongly the market has lately been trending is a way of asking whether the conditions the method needs are present. A trend-strength indicator is one standard way to put a number on that, and the useful reading of it is as a description of the environment rather than as an entry trigger. Selection is a conditioning question too, in the sense that a narrow market and a broad one are different opportunity sets, whatever either is eventually called.

Horizon deserves its own mention, because regime language means different things at different holding periods. A stretch of violent sessions is a serious operational fact to a trader holding positions for days and close to irrelevant to someone contributing monthly over decades, for whom the same stretch is a purchase price rather than an event. Neither perspective is more sophisticated; they are answers to different questions. If the distinction is unfamiliar, the difference between trading and investing is worth settling first, because a regime discussion conducted without a fixed horizon generates arguments in which both sides are right.

None of the above says that regime labels are useless. They are excellent at what they are: a compact vocabulary for describing the past. That is a real service, and the Indian record is a good place to see both the service and its limits.

The last row is the instructive one, and it is instructive precisely because the label is unsatisfying. On a closing basis the 2026 decline ran roughly 15 per cent from the January high, which under the convention is a correction and not a bear market. Anyone holding through it experienced something that did not feel meaningfully different from the early stages of the 2020 fall, and the convention nonetheless assigns the two episodes different words. The vocabulary is doing what it was designed to do, which is to sort finished episodes into bins by depth. It is not doing, and was never designed to do, the thing readers want it to do.

You will notice this page does not tell you which regime the market is in today. That is deliberate, and it is the conclusion of the argument rather than an omission from it. Any such statement would be stale within days, and more importantly it would be the very move the page has spent five figures arguing against: converting a description of the past into an implied claim about the future. The tool above will compute the present reading from levels you check yourself, which is the correct division of labour.

Knowing the definitions does not protect you from the errors, because the errors are not definitional. Each of the following is fully compatible with using the words correctly.

The conclusion is narrower than the topic usually gets and it is worth stating without hedging. Regimes exist and are measurable. Volatility clusters, and the clustering is strong enough to be worth conditioning on. Breadth is countable and tells you something the index level does not. All of that is real, and none of it delivers a direction. The labels bull and bear describe the past reliably and the future not at all, and the 20 per cent line that separates them is a convenience for headline writers that has been mistaken, for decades, for a piece of market knowledge.

What that leaves is a modest and durable discipline: measure what is measurable now, let it set your size and your expectations of how far things will move, and decline to convert it into a view. It is less satisfying than a framework that tells you when the bull ends, and it has the compensating advantage of being true. The work that actually decides outcomes sits downstream of all of this anyway, in entries, exits, position sizing and the discipline to keep applying them when the tape is unpleasant. Reading the regime honestly is what clears the ground for that work. It is not a substitute for it.

• No regulatory or exchange definition, and no percentage threshold. The securities regulator's investor education material on bull and bear markets defines both phases qualitatively, in terms of rising or falling prices and optimistic or negative sentiment, and states no numeric threshold. Checked 18 July 2026; pages are revised, so verify at the source. investor.sebi.gov.in
• The 2008 and 2020 Indian declines. The 2008 crisis took the Sensex from about 21,000 in January 2008 to below 8,000 by October 2008, roughly a 60 per cent decline over about a year. The 2020 pandemic decline took the Nifty 50 from about 12,360 in January 2020 to 7,610 on 23 March 2020, about 38 per cent in roughly a month. Public market record; figures approximate and rounded. nseindia.com
• The 2026 decline, on a closing basis. Daily closing levels for the Nifty 50 over the twelve months to 17 July 2026 give a closing high near 26,329 on 2 January 2026, a closing low near 22,331 on 30 March 2026, a peak-to-trough decline of about 15 per cent, and a level on 17 July 2026 of about 24,334, which sat slightly below the mean of the trailing 200 closes. Taken from a market data feed rather than the exchange directly, and rounded; treat as indicative and verify at the exchange. nseindia.com
• Volatility clustering. That large price changes tend to be followed by large price changes, of either sign, and small by small, is among the most replicated empirical properties of financial price series, and is the property the family of models with time-varying conditional variance was built to capture. The figures on this page demonstrate it on simulated series so that the measurement can be checked against the drawing.
• Every figure on this page is illustrative and simulated. No figure plots a real index. Each one computes its quoted statistics from the series it draws, so the numbers in the captions can be verified against the paths shown rather than taken on trust. No figure should be read as a description of any particular market or period.