Market cycles

Accumulation

Accumulation is the period after a decline has exhausted itself, when the price trades in a range for an extended time and the attention that surrounded the previous peak has gone. Volume is thin, volatility compresses, and the asset is widely regarded as finished. The 2015 range, the second half of 2018 and the middle of 2022 are the clearest examples. What defines the phase is not a price level but a change in who is holding: coins move from participants who bought near the previous peak to participants who are willing to wait. That transfer is invisible in the price and visible only in the length of time the range persists.

Accumulation is the phase most easily identified in hindsight and least easily identified in progress, because the same conditions — flat price, low volume, public indifference — also describe an asset that is simply declining slowly. The 2014 to 2016 period and the 2018 to 2019 period look identical on a chart until one of them resolves upward.

Markup

Markup is the advance, and in bitcoin it has been fast. The 2017 episode carried the price from under a thousand dollars in January to a peak near twenty thousand in December. The 2020 and 2021 episode ran from roughly five thousand in March 2020 to a peak above sixty thousand a year later. The characteristic of the phase is not merely that the price rises but that the rate of rise accelerates: the final months of each advance have produced a larger share of the total move than the preceding year. That acceleration is what makes the phase feel permanent to participants inside it, and it is the same acceleration that makes the subsequent decline so damaging.

Markup is also when the supply schedule becomes a story. Each of the completed advances has followed a halving by twelve to eighteen months, and that coincidence is the origin of most cycle narratives. Whether the halving caused the advance or merely preceded it is a question the record cannot settle, and the section below returns to it.

Distribution

Distribution is the top, and it is a process rather than a point. The price stops making progress, then makes a lower high, then another, over a period of weeks or months. In December 2017 the peak was reached on the seventeenth and the decline did not become unmistakable until February. In November 2021 the price made its final high after a first high in April of the same year, a double-top that took seven months to complete. Distribution is the phase in which the coins accumulated cheaply are transferred to buyers who believe the advance is resuming, and it is the phase least visible to the participants doing the buying.

Markdown

Markdown is the decline, and its scale is the defining feature of bitcoin's record. The table below lists the major drawdowns with their depth and the time taken to recover. Every completed cycle has produced a decline of at least seventy per cent, and the two largest exceeded eighty. Recovery has taken between two and three years in each case, measured from the peak rather than from the trough, which is the convention that matters for anyone who bought near the top.

The most durable cycle narrative is that the halving drives the advance. The argument is straightforward: the subsidy reduction cuts the flow of new coins to miners, miners must sell fewer coins to cover costs, selling pressure falls, and the price rises. The timing fits the record — every completed advance has begun within roughly a year of a halving — and the mechanism is at least plausible.

The difficulty is that four observations cannot distinguish a cause from a coincidence, and the halving is not the only thing that has happened on a four-year rhythm. The 2017 advance coincided with the first wave of retail exchanges and the launch of regulated futures. The 2021 advance coincided with pandemic monetary expansion and the arrival of corporate balance sheets. The 2024 advance coincided with the approval of spot funds. Each of those is a plausible driver in its own right, and each happened once. Attributing the pattern to the halving requires assuming the other candidates were incidental, which the record does not establish.

There is also a supply-side objection. The subsidy reduction is large in proportional terms — the 2024 halving cut new issuance below one per cent of circulating supply — but new issuance is a small fraction of daily trading volume. For the halving to move the price through the mechanism described above, the reduction in miner selling would have to be large relative to the flow of coins already changing hands. That is an empirical claim, and it is not obviously true.

A cycle model fitted to four episodes has four data points and an unlimited number of parameters. That ratio is the source of almost every problem with cycle analysis. It is possible to draw a four-year rhythm through the peaks, to draw a lengthening rhythm through them, to draw a diminishing-returns curve through the successive peak multiples, or to draw a halving-to-peak lag that shortens each cycle. All of these fit the same four points. They cannot all be true, and the data cannot tell us which one is.

The practical consequence is that cycle claims are usually stated with more confidence than the sample supports. A model that says the next peak arrives eighteen months after the halving is not a prediction derived from a mechanism; it is a line drawn through four points, and the fifth point will either confirm it or not. The honest use of the cycle vocabulary is descriptive: it names the phase the market appears to be in, and it reminds the reader that every previous phase has ended. It is not a schedule.

There is a second, subtler problem. The cycle narrative has become widely held, and a widely held narrative about the future changes behaviour in the present. If enough participants expect a peak at a particular point, some will sell ahead of it, which moves the peak. The more precisely the pattern is specified, the more likely it is to be altered by the people acting on it. This is not a reason to dismiss the pattern; it is a reason to hold it loosely.

The most useful discipline is to ask what would falsify a claim before accepting it. A cycle model that cannot be wrong is not a model. The drawdown record and the yearly returns are the two series against which any such claim should be tested, and both are published here in full so that the test can be run rather than asserted.