FFT/spectral Analysis

What is FFT/spectral Analysis?

Spectral analysis asks how much of a series' variance lives at each frequency. The discrete Fourier transform rewrites a window of data as a sum of sinusoids; the FFT (fast Fourier transform) is the efficient algorithm that computes it, and the resulting power spectrum shows which cycle periods carry the most energy. The Goertzel algorithm, a common alternative in trading tools, evaluates single frequency bins instead, which is cheaper when scanning a short list of candidate periods rather than the whole spectrum.

The mathematics is old and the algorithm is not: Fourier's decomposition dates to the early nineteenth century, while the FFT as computing knows it arrived with Cooley and Tukey's 1965 paper, which made spectral analysis cheap enough to be routine. Markets picked it up through the cycle-analysis tradition, from J.M. Hurst's 1970s work on price cycles to the digital-signal-processing indicators of later decades, which treat the chart as a noisy signal whose rhythms can be measured rather than guessed.

Raw price breaks the transform's assumptions: it is nonstationary and trend-dominated, so energy floods the lowest frequencies and spectral leakage smears the peaks. Practical use means detrending first (typically a high-pass or roofing filter), analyzing short windows, and reading the spectrum as an estimate for that window only. Practitioners in the Ehlers tradition often prefer the autocorrelation periodogram for exactly this reason, but Fourier analysis remains the reference frame for dominant cycle measurement.

The unavoidable trade-off is resolution against relevance. Frequency resolution improves with window length, but long windows average over regime changes, and markets rarely hold one rhythm for hundreds of bars; short windows stay relevant and blur the spectrum. Windowing functions manage leakage, spectrogram views stack windows over time to show rhythm stability, and every reading remains an estimate whose peaks shift as new bars arrive.

How to read a spectrum or spectrogram on a chart

Spectral tools plot power against cycle period, sometimes stacked over time; the reading is about peak quality, not peak existence.

1. 1Confirm the preprocessing: the tool should detrend or high-pass the series first, since raw price spectra are dominated by trend energy.
2. 2Find the dominant peak and read its period; that is the window's best estimate of the operative cycle length.
3. 3Judge the peak's sharpness: a narrow, tall peak marks a coherent rhythm, while a broad or flat spectrum says no single cycle is in charge.
4. 4Check stability over time on a spectrogram: a period band that persists across successive windows is worth adapting to; one that jumps every window is noise.
5. 5Cross-check against a time-domain read, such as autocorrelation or a Hilbert-transform period estimate, before feeding the period into adaptive indicators.

How traders use it

• To estimate the dominant cycle period in a lookback window, which then sets adaptive lengths for oscillators and moving averages instead of one fixed setting.
• As a regime read: a sharp, stable spectral peak suggests a tradable cyclic rhythm, while a flat or constantly shifting spectrum argues the market is trending or too noisy for cycle tools.
• As a bandpass building block: once the dominant period is measured, filters can isolate that component and oscillators can be tuned to swing at its rate.
• To audit seasonality claims: a proposed session or weekly rhythm should appear as spectral energy at the matching period; its absence is evidence the pattern is imagination or already gone.
• As a model-selection gate: cycle-based strategies are switched off when spectral coherence collapses, the same way trend systems gate on regime labels, so the tool trades only in the conditions it assumes.

FFT/spectral analysis vs other cycle tools