As you will see, there are many techniques available for forecasting purposes, which makes it difficult for people to select the most appropriate technique. In fact, there is rarely one best technique for any given forecasting situation and manytimes it is advisable to use a combination of quantitative and judgmental techniques (see the section on Combining Forecasting Techniques).
In general, selection of an appropriate technique can be guided by the focal product's stage in its life cycle.
For example, forecasting sales of emerging products which have little or no sales history must rely on more judgmental techniques. As the product becomes more mature and more data is available, simple time series models become more useful. Causal models can ultimately be used with a rich data history. To see the relationship between choice of technique and product life-cycle stage, see Table A1 and A2.
In general, selection of an appropriate technique can be guided by considering the following key factors about the forecasting situation.
1. Forecast Horizon: Basically, you want to make sure that the technique allows you to pick up changes that might occur during the forecast time interval. For example,
Short-term: < 3 months
In the short-term, seasonal fluctuations and randomness are the major influences on sales. Because forecasts at many firms are typically for periods greater than 3 months, short-term forecasting methods are not emphasized in this guide.
Medium-term: 3 months to 2 years
These medium-term forecasts require that fluctuations of a medium-term nature (e.g., economic and competitive conditions) are accounted for by the technique. Since cyclical change and trend are important factors in this time frame, techniques such as regression analysis and time-series methods are useful.
Long-term: > 2 years
Here, the major consideration is with expected trends, as well as economic, competitive, and technological conditions which can only be estimated subjectively. Judgmental methods are usually best employed here.
2. Data requirements. Techniques differ by virtue of how much data is required to successfully employ the technique. For example, Box-Jenkins models require many data points while judgmental techniques require little or none.
3. Pattern of past data. The pattern of a product's previous sale history is an important factor to consider. While the major pattern is the trend, there are also cyclic and seasonal patterns to consider. Certain techniques are best suited for capturing the different patterns in the data. In Table A1 and A2, it is shown how well each technique captures various pattern elements in the data.
4. Explanatory requirements. Whereas some techniques are based purely on the pattern of past data and may do quite well at forecasting, manytimes these are not useful by themselves since it is difficult to explain the forecast to others who wish to understand the causal factors underlying the forecast. Certain techniques (e.g. regression, leading indicators, judgemental methods) are particularly well suited to incorporating causal relationships.
Table A1 and A2 also cross-lists the various techniques described in this guide with the above factors. In addition to the stage in the product's life cycle, use these factors to fine-tune your selection of an appropriate set of techniques.