Cycles don’t predict prices
9 misconceptions of market cycle analysis
I have spent more than two decades measuring cycles in financial markets, and the same objection still lands in my inbox every week: cycle analysis is witchcraft and doesnt work. I understand where that comes from. Most of what circulates under the label of cycle analysis deserves the criticism. However, the criticism almost always targets one of nine misconceptions rather than the method itself.
The core statement comes first because everything else follows from it: cycles never predict prices. A cycle projection points to a turning window on the time axis. It tells you when the market is likely to turn. It stays silent about where price will stand at that moment and about how far the move carries afterwards.
What cycles actually are
1. Cycles predict prices
Everything downstream breaks when this part is wrong. A dominant cycle of 120 bars tells you that a trough window opens roughly 120 bars after the last confirmed trough. That is a statement about time, and nothing in it concerns the price axis. Even if you plot a sine-wave, you can not read the max. or min. turns of the plotted sine to read a price target from.
2. Price is the cycle
What you see on the chart is trend, cycles and noise in superposition, and the cyclic part is just one out of the three. I have watched analysts read raw price as if it were a clean wave. They find cycles that are nothing but the trend bending, and they miss real ones buried under noise. In practice this means detrending comes first, always. Once the trend is removed, the vibration of the market becomes measurable. Same otherway around: You can not plot a cycle on top of the raw price series to predict or forcast. You always need to remodel the trend and noise components if you want to get a somewhat related price curve-prediction.
3. One cycle explains the market
There is never just one. Several cycles of different lengths are active at any moment, and their combined wave shapes the swings you trade. As a result, a single-cycle overlay fits one stretch of history and falls apart in the next, because the additional dominant cycles moved the turning points. The dominant cycle operates alongside the others.
The detection trap
4. A peak in the spectrum proves a cycle
Run a spectrum analysis on pure random noise and you will get peaks, some of them impressive. A peak is a candidate. The spectrum tells you where to look, and the validation stage decides what you actually found. Trading the tallest bar in a scanner output means trading noise with conviction. Noise can generate peaks in a spectrum. So you never know if a peak in a spectrum is a valid, real cycle or just due to mathematical noise in the dataseries.
5. Detection is validation
Genuine cycles are separated from noise by testing, and two features do this work. The Bartels score measures the statistical significance of a candidate (the probability that the wave is a real periodic component rather than a chance pattern).
The Stability Score checks whether the cycle held its phase and length across the analysis window. A candidate that fails these tests remains noise, however tall its peak looked in the spectrum. My workflow has read the same way for years: detrend, detect, validate, then project. Skip the validate step and the projection is worthless.
6. More data means better detection
This one feels rigorous and fails in practice. Cycles morph. A detection window that reaches back too far averages the current cycle with its ancestors. The result is a length that fits the whole history while describing the average historical fit. At at the right side of the chart, the average fit over a long history is useless for prediction. The relevant question is which window captures the vibration in terms of phase and length that is active now, and the amount of available data says little about that. Even worse, you more data you throw at the analysis the more you average and loose precision for the current point in time where everything matters.
Cycles rhyme rather than repeat
7. Cycles are fixed clocks
The most common message I get after a turning window reads: “Lars, the cycle projection did not work.” In many of these cases the projection worked and the expectation was wrong, because the reader measured the market against a perfectly symmetric sine wave. Cycles in financial markets carry a skew. The downward phase runs shorter than the upswing. This asymmetry has been documented in the business-cycle literature for decades. From Keynes’s remark that downturns arrive suddenly and violently while recoveries build slowly, to the formal statistical work on asymmetric expansions and contractions. Hold a simplified sine next to a skewed real-world cycle and the projection looks broken; account for the skew and the same projection lines up. Skew is one example among several: length, amplitude and phase all morph as the market digests new conditions. Therefore cycle analysis is the continuous re-measurement of a living rhythm. The fixed-date folklore (anniversary dates, rigid multi-year counts applied without re-detection) fails in live markets for exactly this reason, while adaptive detection keeps working. Have you every applied a skew to you analysis? If not, start doing.
8. A cycle that worked will keep working
Cycles fade in and fade out. A dominant cycle can lose its dominance within a few swings when a longer or shorter vibration takes over, and a validation score from last year protects nothing today. As a result, the full pipeline runs on every update, and I have retired more than one cycle I was fond of.
The analyst owns the forecast
9. The tool makes the forecast
This is the misconception I push back on hardest, because it is the comfortable one. A fool with a tool still stays a fool. The scanner delivers validated turning windows and their statistical quality. The analyst weighs those windows against trend, liquidity and risk, and the analyst owns the forecast decision. A cycle projection is one measured input into a human judgement. Treat it as an oracle and you have converted a measuring instrument into an alibi. ;-)
The takeaway
Strip these nine away and cycle analysis becomes a sober craft. Decompose price into trend, cycles and noise. Detect candidates in the detrended series. Validate with the Bartels score and the Stability Score. Project turning windows on the time axis, and re-measure continuously, because cycles rhyme rather than repeat. The prediction at the end of that chain belongs to the analyst, and the tools for the earlier steps have never been better.
Always remember, a fool with a tool still stays a fool. I have been fooled myself on this subject many times. So I will keep writing to help others to overcome these issues.
Yours,
Lars
Companion article to the Market Cycles Report video of July 20, 2026. Watch the full episode here: