# Techniques of forecasting - Marketing Management

There are two basic techniques:

1. Objective methods that are of a mathematical or statistical nature;
2. Subjective methods based on experience, judgement and intuition rather than on quantitative analysis.

The wide acceptance of objective techniques in recent years is primarily because objective methods have developed a record for accuracy and thus have inspired confidence in managers who use them as an aid to decision making. the development of better forecasting software has greatly improved its accuracy. Subjective methods still rely largely on intuition, but the practice of objective forecasting is more advanced. Marketers recognize that the pace of change in the marketing environment, and the increased uncertainty which this gives rise to, is making the use of such intuitive techniques more appropriate.

The discussion that follows relates to specific practical and managerial problems that can be encountered when using such techniques.

Objective methods

Moving averages This method of time series analysis involves compilation of the arithmetic average of a number of previous consecutive points in a time series. It is best employed in a situation where an extrapolation of a trend that is gradually increasing or decreasing is present. It has a low cost with ease of manual computation. Problems occur in the choice of the number of points to average, and the effects of a non-typical item in the time series. Seasonality and cyclical trends can be catered for by the application of relevant indices, provided they are known.

The major disadvantage of this technique is that it is purely quantitative in its approach and thus extremely introspective. It does not take into consideration any salient factors in the environment that may affect future sales.

Exponential smoothing Using a moving average has the problem that it gives equal weight, or significance, to all the items in a time series. More recent points in a time series will represent the present situation more accurately than older items, and it is therefore, only logical to attach more significance to more recent items by using a weighting method. The different weight attached to an item in a time series can be calculated either simply by using an arithmetic progression or, more sophisticatedly, by using a geometric progression. When a geometric progression is used and a graph is drawn, raw sales data are smoothed into an exponential curve; hence the name ‘exponential smoothing’. In the case where an arithmetic progression is used, this is simply known as a weighted moving average.

Exponential smoothing provides a forecast which is equal to the old one, plus or minus some proportion of the past forecasting error(s). There are many variations of exponential smoothing, ranging from the very simplistic to the more complex methods involving a greater number of data points and proportions of forecast errors. These techniques, because of their statistical nature, lend themselves particularly to purely quantitative data, thus neglecting other important market factors. A more realistic prediction is gained through the use of this technique than moving averages because it allows for new factors and influences that have emerged in the most recent sales period.

Trend projections By fitting a trend line to a mathematical equation it is possible to make forecasts about future sales using the equation. four typical growth curves that a firm may experience. The danger of using the trend approach alone is that when the analyst extrapolates, the assumption is that what affected sales in the past will continue to affect sales in the same way over future periods. An adequate number of past measurements or observations are also required for adequate statistical significance, but care must be taken not to include too many past observations or history will be too heavily weighted. Trend projections, like moving averages and exponential smoothing, are not ideal for short or medium-term forecasts. They are more fitting for predicting a ‘broad sweep’ trend over the long term.

The Box-Jenkins forecasting method is a special case of exponential smoothing in which the time series is fitted with an optimizing mathematical model that attributes minimal error to historical data. Once the model has been identified and constructed, the parameters must then be estimated. Of the available statistical routines, this is one of the most accurate and flexible in that it can cope with almost any type of data pattern. However, accuracy involves complexity, which, along with flexibility, results in a relatively high cost and the need for a skilful operator with plenty of time to reap the full benefit s of this technique. As this method’s accuracy is limited to the short term it is not very often used in practice, as there are many other cheaper and easier techniques that can be employed, and although they do not give as much accuracy, these are often adequate for short-term decisions.

Spectral analysis Incorporated in the classification of spectral analysis is the technique of Fourier analysis, where a time series is mathematically decomposed into its constituent sine wave forms. Thus, from one time series, a spectrum of time series are produced having the name ‘power spectrum’. The mathematical complexities of this method put it beyond the use of all but the most competent analyst, whose skill and understanding of the technique are imperative for its successful implementation.

X-11 technique is similar to spectral analysis in that it decomposes the original time series into a spectrum of time series. However, it only separates out the seasonal and cyclical trends, and then fits a time series to the remainder. It takes the best of spectral analysis and Box-Jenkins and combines them in one technique. Used by a skilled analyst, it rates as one of the most effective short to mediumterm forecasting methods, with its ability to identify turning points being a major asset. Causal methods are still objective techniques, but they all involve some degree of subjectivity. One of the best known causal methods is that of regression analysis, which attempts to assess the relationship between at least two variables: one (or more) independent and one dependent, the purpose being to predict the value of the dependent variable from the specific value of the independent variable. The basis of this prediction generally is historical data.

This method starts from the assumption that a basic relationship exists between two variables, and the least squares method of estimation is used to formulate the mathematical relationship which exists. Various forms of regression analysis exist, one being multiple regression analysis, where any number of variablescan be considered at one time.

Another form is stepwise regression analysis, where only one independent variable is considered at one time. The value of this technique is difficult to assess other than in individual cases, as the accuracy is dependent upon the degree to which the independent variables explain characteristics of the dependent variable. This relationship may vary considerably, but improved computing techniques have meant that the value of regression analysis is increased dramatically on a cost/benefit basis.

Company growth curves

Through techniques such as regression analysis, the Newspaper Society has established clear links between their different life stage categories with levels of demand for a range of products. Unsurprisingly, Mothercare shoppers are almost four times more likely to have pre-school-aged children than the national average.

Econometric models are an extension of the regression technique whereby a system of independent regression equations is evolved. These equations describe a particular sector of economic activity whose parameters are usually estimated simultaneously. Generally, these models are relatively expensive to develop, the precise cost being dependent on the amount of detail incorporated in the model. However, the inherent systems of equations in such models express the causalities involved far better than an ordinary regression equation, and thus will predict turning points more accurately.

Input-output models are particularly applicable in the field of industrial marketing since they are concerned with inter-industry or interdepartmental flows of goods or services in a company and its markets. The technique is based on the theory that the output of one industry comprises the basic inputs of products and materials of another, thereby providing an inter-industry/interdepartmental flow of goods and services within the economy/industry. Major inputs required for this type of model are not in the operation of the model, but in the collection and presentation of data. This is because tables and government statistics showing the extent to which one industry obtains its basic inputs from another are very broadly defined in standard industrial classifications.

The cost of these models in producing a sales forecast, especially when they are combined with econometric models to produce economic input-output models, is often high but as Lackman4 shows this should be offset against the relatively high accuracy produced by these more sophisticated models.

Diffusion indexes use the many economic indicators available that represent general economic activity, or the activity within a particular industry or product class. By formulating an index based on a certain combination of these indices, an indication of future trends can be compiled. The accuracy and applicability of the index can be tailored to fit specific requirements, such as predicting turning points in the short term by choosing appropriate economic indicators. A forecasting technique of this nature can be accurate and relatively economical to apply in certain industries and product classes.

Tied indicators are used where the sales of one product are closely related to sales of another product and the sales trend of one product precedes that of the other. The preceding product is the indicator, and in the case of leading indicator models, the sales of more than one product may be utilized along with indicators of general economic activity. The leading indicators generally increase or decrease prior to the pending increase or decrease in the dependent variable, and they usually take the form of a time series of economic activity. As a forecasting method the real value of this type of model is its ability to predict turning points rather than as a predictor of future trends in general.

Life cycle analysis is particularly applicable where there is no historical data. Sales of similar products are analysed over time and usually a particular ‘S’ curve is found to apply for a certain product class. The phases of product acceptance by various groups i.e. innovators, early adopters, early majority, late majority and laggards are essential to the analysis ( Roger’s ‘Diffusion of innovations’). Consideration of the concepts of life cycle analysis by individuals is often more valuable than a thorough detailed expert analysis, as the database for this type of model is conceptually weak. A manager can readily appreciate that the product has to pass through the various stages in its life cycle and subjective opinion might even be as accurate as any expert analysis.

Subjective methods

Market research involves studying a representative sample of a market involving a systematic formal procedure for evolving and testing hypotheses about a market. For any valid information to be obtained, two reports should be made over a period of time so one can be compared with the other and conclusions drawn. This necessitates the collection of market data from questionnaires, surveys, published statistics, marketing intelligence reports and time series analyses of market variables.

Due to the relatively high cost of this technique, it is only used in cases where there is a considerable financial risk, which generally means it is restricted to large companies. The major application of this technique for sales forecasting is in the area of predicting new product sales by investigating consumer reaction to a new product concept or prototype..

The Delphi method is a technique that involves the marshalling of expert opinion to cope with the problems of eradicating the ‘bandwagon effect’ of majority opinion.

Panel consensus is not unlike the Delphi method, except that the panel of experts are encouraged to communicate and discuss matters in relation to the future prospects of what is to be predicted. Developed primarily for long-term forecasting, this method is rarely used due to the problems of personal and social bias influencing the members of the panel. Methods of this nature often do not arrive at a true consensus of opinion because of the effects of such bias. Experts are not infallible. Predictions regarding the growth in access to the Internet in the UK proved to be too conservative. Growth rates in the diffusion of this technology into UK households have been much higher than the experts predicted.

## Visionary Forecasting

Visionary forecast is where ‘visionary’ forecasters or ‘futurologists’ attempt to prophesy through personal opinion and judgements. The method is characterized by subjective guesswork and imagination, and in the method is non-scientific. A set of possible scenarios about the future is prepared by a few experts in the light of past events. At one time, visionary forecasts were felt to be too subjective to be used in marketing decision making. However, as mentioned earlier, the pace of chance and dynamic nature of the marketing environment have begun to make companies appreciate some of the advantages of visionary forecasts even though they may sometimes be wrong.

Many believe that at least in part the success of Microsoft in being ahead of its competitors in many areas is down to visionary forecasts. The company has a system whereby senior managers are encouraged to think about the future in the widest possible sense, including, for example, social trends and developments, and how these developments, might potentially open up new opportunities for Microsoft for future product development.

Historical analogy is a comparative analysis of the introduction and growth of similar new products, and this bases the forecast on similarity patterns. By comparing a new product with a similar previous new product, forecasts of future sales performance can be made. This technique, however, is conceptually weak, as a true new product will not be similar to any previous product, and even a new version of a product will probably not be similar enough to make any comparison really valid.

Sales force opinion is where members of a sales force are in constant contact with customers, and are in a position to predict their buying plans, attitudes and needs. An obstacle to gaining true estimates is that salespeople often tend to be pessimistic, owing to their compensation system. It is common practice for salespeople to be remunerated according to the degree to which they attain sales quotas which, in turn, are based on sales forecasts.

Thus it is in their own interest to underestimate future sales, resulting in low quotas and possibilities of high compensation. However, Jobber and Lancaster5 provided evidence that being involved in the sales forecasting and hence quota setting process can actually increase salesforce motivation, therefore making the achievement of agreed sales quotas more likely. This method has the advantage of being relatively cheap and easy to introduce and administer through the existing sales organization.