Spectral analysis of pre-industrial English wheat price periodicity

The primary data consists of a 414-year time series of annual English wheat prices, spanning the years 1501 to 1914. This longitudinal record was subjected to targeted spectral analysis to test for the presence of a Kitchin-style 3.4-year inventory cycle.

The analysis identified a spectral peak at a period of 4.4 years, yielding a surrogate p-value of 0.0120. While this result allows for a verdict of confirmed for the targeted hypothesis, the finding must be contextualized within the broader spectral landscape of the dataset. The 4.4-year signal is not the dominant feature of the periodogram; rather, the top five spectral peaks all reside in the 5.4 - 7.4 year range, possessing greater power than the 4.4-year peak. This indicates that the identified signal is a secondary feature that required specific hypothesis testing to surface.

Several structural vulnerabilities in the data preclude a definitive assignment of this cycle to an inventory-driven mechanism. First, the 30% deviation from the canonical 3.4-year Kitchin cycle suggests that the observed 4.4-year period may represent a distinct agricultural cycle, such as those driven by weather or crop rotation, rather than an inventory cycle. Labeling this phenomenon as “Kitchin” may inadvertently impose a modern macro-economic framework onto pre-industrial grain prices.

the effective resolution of the annual data is insufficient to reliably distinguish between a 3.4-year and a 4.4-year cycle. Aliasing from sub-annual price variations, such as seasonal harvest effects, could shift the apparent peak period. The reliability of the signal is also challenged by the non-stationary nature of the underlying data. Pre-industrial wheat prices are subject to high-amplitude disruptions from weather shocks - including droughts, floods, and frosts - and discrete events such as wars, plagues, and enclosures. These events violate the stationarity assumption required for spectral analysis and can generate spurious periodic signals during surrogate testing.

Finally, the statistical significance of the finding is marginal. Given that the periodogram contains approximately 200 independent frequency bins, one would expect 2 to 3 bins to reach a p-value of 0.012 by chance alone. Without a formal multiple-comparisons correction, it remains uncertain whether this signal would survive a more stringent statistical threshold.

The following gaps remain unaddressed:

1. The degree to which the 4.4-year peak is an artifact of seasonal aliasing.
2. The stability of the signal when non-stationary disruptions (wars, plagues) are filtered from the series.
3. The independence of this signal from other potential drivers, such as multi-year climatic oscillations or crop rotation schedules, which lack corroborating evidence from independent proxy domains.