Demand Forecaster Seasonality Calculator

This is a multiplicative decomposition: the forecast is a trend level multiplied by a seasonal index, the standard structure used when seasonal swings scale with the size of the business rather than staying fixed in absolute units. The trend term grows the base demand at the stated annual rate prorated over elapsed time. The seasonal term uses a single cosine wave with a twelve month period, peaking in the nominated month and troughing six months later. Amplitude is a fraction of trend, so 30 percent means the index ranges from 0.70 to 1.30 and the peak to trough swing is 60 percent of trend, not 30. Observed amplitudes vary widely by sector: utilities and staple groceries often sit under 10 percent, general retail typically 15 to 35 percent, and toys, garden centres, ski resorts, and tax preparation services can exceed 60 percent, with some categories concentrating a third or more of annual volume into the fourth quarter. The single sinusoid is the model's principal simplification. A pure cosine is symmetric, so it forces the trough to fall exactly six months after the peak and assumes the rise and fall are mirror images, which misrepresents patterns with a sharp spike and slow decay such as the run up to late December or a single event driven week. Real series often need multiple harmonics, or a separate index per month estimated from history, to capture that shape. The model also assumes stable seasonality, ignores calendar effects such as varying month lengths, moving holidays like Easter and Ramadan, and trading day counts, and produces a point forecast with no uncertainty interval. For inventory decisions the spread matters more than the central estimate, since safety stock is driven by forecast error variance and target service level rather than by expected demand. Promotions, price changes, competitor entry, and one off disruptions are outside the model entirely and must be layered on judgmentally.

Decomposing an economic series into trend, cycle, seasonal, and irregular components was formalized at Harvard in the years around the First World War. Warren M. Persons, working on the Harvard Index of General Business Conditions, set out the four component framework in a 1919 paper in the Review of Economic Statistics and developed practical methods for measuring and removing seasonal variation. The problem mattered because policymakers and businesses could not tell whether a month to month change signalled a genuine turning point or merely the usual calendar rhythm. Mechanized seasonal adjustment followed with computing. Julius Shiskin led development of the X-11 variant of the Census Method II seasonal adjustment program at the US Bureau of the Census, released in 1965 and adopted by statistical agencies worldwide; it evolved into X-12-ARIMA in the 1990s and the current X-13ARIMA-SEATS, developed jointly with the Bank of Spain. In parallel, forecasting practice was reshaped by exponential smoothing: Charles Holt described trend corrected smoothing in a 1957 Office of Naval Research memorandum, and Peter Winters extended it to include a seasonal index in a 1960 paper in Management Science, producing the Holt-Winters method that remains a supply chain workhorse. George Box and Gwilym Jenkins published Time Series Analysis: Forecasting and Control in 1970, giving the field seasonal ARIMA and a systematic identification procedure, and Rob Hyndman and colleagues later unified exponential smoothing under a state space framework that supplies the prediction intervals a deterministic model like this one lacks.