A Prediction Is Not a Forecast: Temporal Structural Forecasting

A Prediction is unidimensional univariate: a single value along a single dimension, with no temporal coordinate and no specification of normal at any position, supporting neither planning nor decisions. A Forecast is multidimensional univariate: a {4,5,6} object carrying a Dimension 4 value, a Dimension 5 Calibrated Probability Band that defines the range of normal at each position, and a Dimension 6 calendar coordinate. A Forecast supports both planning and decisions.

Every method that forecasts a time series produces a Prediction because it operates on a single sequence of values along a single dimension. It does not matter whether the method is a simple moving average or a deep neural network with billions of parameters; the operation is a Univariate Unidimensional (UU) Operation. The UU Operation reduces multidimensional univariate data to unidimensional univariate form. Dimension 5 Probability collapses at observation. The UU Operation discards the Dimension 6 calendar coordinate, replacing it with a sequential index that carries no calendar meaning, and converts the temporal absolute into an atemporal absolute. The Prediction, an atemporal absolute, is the structural ceiling of what a UU Operation can produce.

Time series forecasting reads the historical record as a single line of prices across trading days and produces a Prediction: a single value at the next position (Figure 1). A Prediction tells the trader WHAT the price will be. It cannot tell the trader WHEN the price is abnormally low, because it carries no specification of what counts as normal at any position.

The horizontal axis is the sequential timeline; the vertical axis is price. Two orthogonal axes define a two-dimensional plane, and the data at each position is a Dimension 4 value at a Dimension 6 calendar coordinate. Yet the visual impression is not of paired coordinates in a plane; it is of a single line tracing a path. The line’s continuity invites the viewer to perceive unidimensional values rather than the multidimensional objects the chart actually plots. This perception is the cognitive trap that makes UU Operations appear sufficient: if the data looks one-dimensional, then an operation that produces a one-dimensional output appears to lose nothing.

Figure 2 displays the structure that was present before observation collapsed it. Each calendar date belongs to multiple historical seasons. The Distribution Mean (blue dashed line) is the average value those seasons have historically produced. The 85% Calibrated Probability Band (green band) is the range containing 85% of historical outcomes within those seasons. Each position is a {4,5,6} object: a Dimension 4 value, a Dimension 5 probability distribution, and a Dimension 6 calendar coordinate.

TSF resolves this by generating a Forecast Value: a temporal absolute, recovered through Orthogonal Projection. The UU Operation along the Dimension 4 sequential timeline produces an atemporal absolute: the inertial trend. The UU Operation along the Dimension 6 seasonal timeline produces a temporal relative: the temporal coordinate preserved at each seasonal position, the value expressed as a ratio, capturing how values at this seasonal position have historically departed from the inertial trend. The product of an atemporal absolute and a temporal relative is a temporal absolute, restoring the multidimensional structure that every single UU Operation discards.

The {4,5,6} Forecast locates this temporal absolute at a Dimension 6 calendar coordinate and surrounds it with a Dimension 5 Calibrated Probability Band recovered from the empirical distribution at the corresponding seasonal position. The band at each future position defines the range of normal at that position; taken across the forecast horizon, the bands form a Map of Normal. Planning operates against the Map: the range of normal at each future position specifies what to expect, and positions, resources, and commitments can be staged accordingly. Decisions operate against the Map in real time: as actual values arrive, they either fall inside the band (normal at that position, no action indicated) or outside it (an exceedance, the signal to act). The same Forecast supports both because the Calibrated Probability Band is present at every position, forward and current.

The Calibrated Probability Bands at each future position in Figure 3 were committed one week in advance. The lower band at each trading day is the price for a pre-placed Buy Limit Order. The orders are set before the trading window opens. When the intraday low touches the band price, the position is entered at a structurally specified abnormally low price.

The full forecast methodology is available here.