Yule's sunspot model and the origin of autoregression
In 1927 George Udny Yule replaced the idea of hidden periodicities with that of a damped pendulum disturbed by random shocks. This picture led to the autoregressive model. A simulation reproduces Yule's model of the sunspot cycle.
Consider a pendulum in a quiet room. Once pushed, it swings with decreasing amplitude until it stops.
Now suppose that the pendulum is struck at random by small projectiles; Yule's own illustration used boys with pea-shooters. The pendulum never comes to rest, its amplitude varies, and the time between peaks changes. A chart of its position might suggest several hidden rhythms, although only one rhythm is present, together with random disturbances.
This picture, used by the statistician George Udny Yule in 1927, changed the analysis of series that appear to oscillate: sunspots, business cycles, river levels, server load. It is the origin of the autoregressive model, the subject of the series of notes this article opens. The question it addresses is: is an irregular cycle a mix of hidden regular cycles, or one cycle that keeps getting knocked about?
The following work provides the context.
Hidden periodicities (1890s-1920s). The standard tool of the time was Arthur Schuster's periodogram (1898): break a series into sine waves of different frequencies and look for peaks. The underlying assumption was that the data were a sum of fixed sine waves plus independent measurement errors. Sunspots, recorded by astronomers for centuries and compiled by Rudolf Wolf and his successor Alfred Wolfer, were the classic test case. They rise and fall roughly every eleven years, but individual cycles last anywhere from about nine to thirteen years.
Yule (1927). In "On a method of investigating periodicities in disturbed series, with special reference to Wolfer's sunspot numbers", Yule proposed that the disturbances are not added to the curve afterwards; they drive it. He wrote each year's value as a linear function of the previous two plus a fresh random shock, and fitted that recursion to the sunspot record from 1749 to 1924.
Walker (1931). Gilbert Walker, who was studying atmospheric pressure records, generalised the idea to any number of lags and gave the link between the model's coefficients and the series' autocorrelations: the Yule-Walker equations, still a standard way to fit these models.
What came next. Mann and Wald (1943) put the estimation on a rigorous mathematical footing, and Box and Jenkins (1970) turned autoregressive models into the practical forecasting toolkit taught today.
The model
Yule's recursion for the second-order case is
where is the year's disturbance: an independent random shock. Without shocks, this is exactly the equation of a damped oscillator sampled once a year. Writing the damping per year as (the share of the swing that survives one year) and the natural period as , the coefficients are
The two coefficients therefore correspond to the period of the oscillation and the rate at which it decays.
Application to sunspot numbers
Using the classic yearly series that ships with the Python library statsmodels (the record before its 2015 recalibration), a least-squares fit over Yule's years gives
which explains about 80% of the year-to-year variance. Inverting the formulas above: and years. Statistically, the sunspot series behaves like a damped oscillator with a natural period of about ten and a half years, losing about a fifth of its amplitude each year and driven by random shocks.
Significance of the model
- Forecasting. If the cycle were a fixed sine wave, its phase could be predicted centuries ahead. In Yule's world every shock shifts the phase, so uncertainty grows with the horizon. This is consistent with experience: the next solar maximum is easier to predict than the one after it. The difference has practical weight. Geomagnetic storms are more frequent around solar maximum, and they disturb satellite operations, radio communication and, in severe cases, power grids; the strong storm of May 2024, near the maximum of the current cycle, produced aurorae visible across much of Europe. Satellite operators and grid companies plan around the predicted timing of the next maximum, and Yule's model explains why the maximum after that remains uncertain.
- Spurious cycles. A series driven by shocks produces periodogram peaks that correspond to no fixed cycle at all. A peak is therefore not evidence of a periodic component.
- The role of the shocks. In Yule's model the disturbances are part of the machine, so their size and their distribution shape everything that follows. This idea recurs throughout this series of notes.
The figure below simulates the model. One damped oscillation plus random shocks is enough to produce irregular, wandering cycles, with no hidden periodicities at all.
Suggested settings:
- No shocks. The model predicts a pure damped oscillation, dying away with a half-life of years at the sunspot setting. The oscillation decays accordingly.
- Shocks restored. Consider the readout "lengths of the last cycles". The natural period is fixed at 10.6 years, yet individual cycles come out as 8, 12, 9, 14… years. This irregularity matches the sunspot record, although here it is produced by a single rhythm.
- Weak damping. Increase towards 0.99. The oscillation now persists for a long time, the cycles become regular, and the model starts to look like Schuster's world of fixed sine waves. Yule's model contains the old one as a limiting case.
- Strong damping. Reduce to 0.5. Persistence is short, the oscillation is barely visible, and the series looks almost like noise.
Derivation. Without shocks the recursion is , whose characteristic equation has roots when and . The solution is therefore : a cosine with period whose amplitude shrinks by a factor every year. Each shock simply restarts this motion with a new amplitude and phase, which is why the cycle length varies while the natural period is constant.
Yule treated randomness as part of the generating mechanism rather than as noise added to a clean signal. This led to the autoregressive model, a two-parameter description of the sunspot cycle that remains adequate, and to an important caution: irregular cycles do not require irregular causes. A series that appears to contain several overlapping rhythms may be generated by a single damped oscillation driven by random shocks.
References
- A. Schuster, "On the investigation of hidden periodicities with application to a supposed 26 day period of meteorological phenomena", Terrestrial Magnetism, 3(1), 13-41, 1898.
- G. U. Yule, "On a method of investigating periodicities in disturbed series, with special reference to Wolfer's sunspot numbers", Philosophical Transactions of the Royal Society of London A, 226, 267-298, 1927.
- G. T. Walker, "On periodicity in series of related terms", Proceedings of the Royal Society of London A, 131(818), 518-532, 1931.
- H. B. Mann and A. Wald, "On the statistical treatment of linear stochastic difference equations", Econometrica, 11(3/4), 173-220, 1943.
- G. E. P. Box, G. M. Jenkins, G. C. Reinsel and G. M. Ljung, Time Series Analysis: Forecasting and Control, 5th ed., Wiley, 2015.