Time-Series Techniques
Time-series techniques are based on the interrelationship of four data patterns: level, trend, seasonality, and noise. Level is a horizontal sales history, or what sales patterns would be if there was no trend, seasonality, or noise. Trend is a continuing pattern of a sales increase or decrease, and that pattern can be a straight line or a curve. Seasonality is a repeating pattern of sales increases and decreases such as high sales every summer for air conditioners, high sales of agricultural chemicals in the spring, or high sales of toys in the fall. The point is that the pattern of high sales in certain periods and low sales in other periods repeats itself every year. Noise is ran- dom fluctuation—that part of the sales history that a time-series technique cannot explain. This does not mean the fluctuation could not be explained by regression analysis or judgment; it means the pattern has not happened consistently in the past, so the time-series technique cannot pick it up and forecast it.
Time-series techniques arrive at a forecast by assuming one or more of these pat- terns exist in a previous sales history and projecting these patterns into the future. Exponential smoothing is a common time-series technique.
Time-series techniques are often simple and inexpensive to use and require little data storage. Many of the techniques also adjust very quickly to changes in sales conditions and, thus, are appropriate for short-term forecasting. Time-series tech- niques, however, will probably be less accurate than correlation analysis if the fore- caster utilizes a time-series technique that assumes data patterns do not exist but are, in fact, in the sales history. Simple exponential smoothing assumes, for example, that the sales history consists of only level and noise. If trend and season- ality exist in the sales history, simple exponential smoothing will consistently err in its forecast.