CoDebris Spectre, a software application written for the Windows platform, encodes a novel method for extremely high-precision spectral analysis of simultaneous data from multiple sources. CoDebris, the company which developed Spectre, has been awarded significant grants and contracts from NIH and NASA under the federal Small Business Innovation Research program for automated interpretation of EEG "brain wave" and other physiological data. During the last two years, Spectre has been adapted to analyze market price data.
In addition to very pure spectral analysis of time series or price data to extract precise frequency domain measurements, Spectre uses a "family" of fast minus slow oscillators to extrapolate from past behavior. Ordinarily, only one such oscillator is employed using available technical analysis tools. But Spectre typically generates three to five significant (trending, long period) oscillators and a plethora of shorter cycles. In a family of such oscillators, the short periods are usually "shaping" factors we ordinarily visually identify as confined to a channel, a pennant formation, or other recognizable pattern.
Spectre treats the oscillators as being coupled, damped and driven, just as in many physics models of collective behavior. Coupled means interacting: price data affects and is affected by not only adjacent prices, but more significantly by prices from distant “patterns” which recur during the price history. Damped means dissipative: eventually all patterns change, or even disappear. Driven means an external force is applied: in Spectre, the external force is derived from “sentiment” derivations, see below. Basically, the oscillators are semi-coherent collections of past prices.
This means that price is not only dependent on past price, but also affects future prices, to the extent that coupling is identified. However, coupling can be disrupted by unforeseeable events in the external world. Such a disruption affects the shortest period oscillators the most. Their effect therefore weakens substantially the further an extrapolation proceeds.
Far from a random walk, or even a log-normal distribution, an extrapolation using Spectre is not just a wavy line into the future, but uncannily resembles the past data in both the statistical and the visual sense. This is possible because the purity of retained "phase" information (a term used in sinusoidal wave math to indicate “timing” or position within the price history) from the analysis allows the family of coupled oscillators to operate into the future.
The oscillators remember their entire interaction history on multiple time scales simultaneously, and merely continue to behave that way, obeying the math required for a valid analysis of statistical "non-stationary" data where the moments (mean, variance, skew etc.) can vary with time. Stationarity can be violated from tick to tick in financial time series due to the "human element" of market sentiment. In recent years sentiment has become numerically definable and usable in predictive systems.
No other spectral analysis technique commercially available retains the essential phase data satisfactorily. Common Fourier Transform methods exhibit substantial corruption of identified phase as well as amplitude and period information (spectral “leakage”), so although they are used routinely their suitability for our purposes is destroyed.
If a price chart exhibits more than one occurrence of a particular pattern, like a slow rise to a plateau followed by a steeper drop, the retained phase info enables Spectre to position the pattern in the extrapolation at a statistically meaningful place where the pattern’s spectral components re-align.
Just as in Ichimoku (a Japanese currency trader's "One Glance Equilibrium Chart"), what is extrapolated into the future is not merely a price estimate, but a sophisticated estimate of "sentiment", which is really the prime market mover. Sentiment is gleaned from price extremes, volume, implied and historical volatility, and put/call ratio histories, considering both normal and stressed behaviors. These are derived similarly to the way Average True Range is defined, but again on multiple time scales simultaneously. In context, we use the term "price history" to refer to this multiple-source data set.
The enhanced capability to do this in Spectre arises from the vastly increased level of complexity employed by the family of coupled fast minus slow oscillators and retained as phase information in the program for positioning.
Long-period oscillators form the backbone of trending behavior, but require care because the extrapolation proceeds into a widening cone of uncertainty as required by formal prediction theory. Operationally this translates as meaning that volatility extremes can return seemingly out of the blue, but their return may be foreseeable from an analysis of price histories only if they have occurred before within that history. Similarly, calm periods called regions of “congestion” or “consolidation”, or in Ichimoku "confusion", can develop only if they have been seen before.
A pattern such as a breakout developing at the "hard right edge" may simply fail to exhibit statistically rational behavior, and thus not extrapolate as such. In other words, patterns change all the time: trends end when they please, or break down soon after they begin due to the collective (but independent) reactions of seasoned traders to newly-evolving patterns.