Maximum-entropy Spectrum

Maximum-entropy Spectrum measures which cycle currently dominates a market — and whether the reading deserves trust. Each bar the build high-passes the source, fits an autoregressive model with Burg's recursion, and evaluates its power spectrum across a period grid: the maximum-entropy spectrum, prized for sharp peaks on short windows. The smoothed dominant period plots as a line in bars, the spectrum as a heat profile beside the last bar.

How to Trade the Maximum-entropy Spectrum?

• Sharp rhythm: peak power at or above the Sharpness Threshold times the spectral average means one rhythm dominates; below it the dashboard reads flat or drifting and the line deserves less weight.
• Cycle shifts: a jump in the smoothed dominant period beyond the shift threshold fires its own alert.
• Half cycle: the optional half-period plot is the traditional choice when setting oscillator lengths from a measured cycle.

Maximum-entropy Spectrum Settings

• Source (default close) with Conditioning (default First Difference) and Detrend Length (default 48): high-pass preparation so trend cannot pose as a very long cycle.
• Sample Window (default 64): shorter tracks drifting cycles faster, longer stabilizes the spectrum.
• AR Order (default 12): too low blurs neighboring peaks, too high can split a real cycle; capped below half the window.
• Minimum Period (default 8) and Maximum Period (default 48): the evaluated grid.
• Cycle Smoothing (default 3): median-smooths the readouts.
• Sharpness Threshold (default 4.0) and Cycle Shift Threshold (default 5.0): the alert levels.
• Calculation Bars (default 2000), Show Spectrum Profile (on) with Profile Width (default 30), Show Half Cycle (off), plus dashboard toggles.

Frequently Asked Questions

How does this compare to the Hilbert Transform for cycle work?

The Hilbert Transform derives one instantaneous period from the analytic signal's phase — responsive, but a single estimate. The maximum-entropy spectrum lays out the whole period landscape, including the dashboard's second peak, so competing rhythms stay visible.

Why fit a model instead of transforming the data directly?

A direct transform needs long samples to separate nearby periods and treats its window as if it looped forever. The autoregressive route extrapolates the measured structure instead, producing sharp peaks from a few dozen bars.

What does the profile beside the last bar show?

The current spectrum as horizontal heat bars on the shared period axis, the strongest peak claiming the full width. It is a snapshot of the latest fit and redraws as new bars update the model.