Extrapolative Forecasting

In extrapolative forecasting we predict the future by extrapolating a historical trend. What has happened in the past determines what is forecast for the future [with other forecasting methods, such as exploratory forecasting, this need not be so. For example, with exploratory forecasting we can explore revolutionary, as well as evolutionary, scenarios]. In some circumstances it is right to use extrapolative forecasting. In other cases different approaches might be more suitable. It is not an appropriate approach to use in a new product/ new business situation, or in situations where circumstances have radically changed, and the past is no guide to the future.

Any time series [a series of numbers recording past events] will have been produced by the interaction of a number of variables. For example, a time series of a company’s past profits will have been produced by a complex process, which involves an interaction between multiple revenue streams and numerous costs. Extrapolative forecasting would not require the forecaster to identify these individual variables, or the ways in which they interacted. The underlying process [variables and their interaction], which generated profits, would be treated as a black box [not analysed in detail].

To produce a profit forecast scenario using exploratory [what if] methods it would be necessary to analyse and model the total sales and costs process; including cost and revenue behavior and the relationship between all the cost and revenue variables involved in the process. To produce a forecast profit scenario by extrapolative forecasting it is only necessary to obtain a series of past profit figures and extend them to produce a forecast figure.

This usually makes the extrapolative forecasting approach quicker and cheaper; it also makes it more risky. Since the underlying processes producing the forecast are not being modeled there is a danger of overlooking some fundamental change in the process, which makes it unwise to assume any continuity of events.

There are a number of different extrapolative forecasting methods, ranging form the simple to the very complex. One of the simplest methods is eyeball extrapolation. This involves producing a chart of past values and then forecasting by line of sight. Other simple methods include weighted average and exponential smoothing. Research has shown that the more complex methods are not necessarily superior to the simpler methods in terms of forecasting accuracy. Often, the simpler methods are just as reliable as the more complex. There is no single best method. Because different time series will contain different kinds of trends [linear, exponential, logarithmic etc], and different forms of noise [see below] forecasters need to carry out statistical tests [see below] to identify the method that will work best with their data and suit their particular requirements [e.g. how far in to the future do they need to forecast – some methods only work for one period ahead].

When To Use

The extrapolative forecasting approach can be employed when there is a set of historical data, which is believed to contain a trend, and it is believed that this trend can be isolated and used to forecast future events. For example, if a company has accurate records of its sales for the past six periods, and believes that there has been a trend in its past sales, and, also believes that this trend will continue into the future. In these circumstances it may be able to use extrapolative forecasting methods to help it to forecast its sales for future periods. Before extrapolative forecasting methods can be used a number of conditions have to be satisfied.

Consistency In The Environment

There needs to be a reasonable expectation that the conditions that have prevailed in the past will continue to hold true for the period being forecast. This is not an absolute requirement. If there has been some minor change in the environment a judgmental adjustment can be made to the values produced by forecast calculations. However, whenever there have been major changes in the environment, such as a competitor introducing a radical new product, extrapolative forecasting methods become less appropriate. If a manager is intending to make a decision with radical consequences extrapolative forecasting methods are of little value since the effect of the decision would be to change the environment.

Data Availability

Reliable quantitative historical data has to be available. Extrapolative forecasting methods would not be appropriate for start up, or new product launches where relevant historical data is not available. Nor would they be appropriate where the historical data is unreliable, perhaps because it has not been recorded in a consistent manner over the relevant time period. Most accounting and business systems contain copious amounts of data organised on a time basis, but one has to be certain that information is both reliable and in a usable form.

Data Comparability

Data relating to several past periods has to be reduced to a common basis. This can be a problem when forecasting something like future sales. Usually what one is trying to do is forecast future sales volumes and the temptation is to do this using sales values as a surrogate for volumes since the accounting system will have recorded values, not physical volumes. The danger in doing that is that price inflation will have affected sales values and £5m of sales in 2010 does not represent the same volume of physical sales as £5m in 2000. In a situation like that, if sales values are to be used for forecasting purposes the values have to be inflation adjusted. If possible physical volumes should be used instead of monetary units but there are unfortunately many instances where this is not feasible. Another problem in data comparability arises when a company sells a range of products or services and that mix has changed over time. Here again it may be necessary to forecast each component in the mix separately.


The benefit expected from the forecast should exceed the costs of collecting and processing the data. A supermarket might well want to forecast the short term demand for each of its many thousands of product lines but could only do so if it could easily collect data on past sales at point of sale via some method such a bar code reading by lasers at the checkout, and process the data using a fairly simple and computationally undemanding extrapolative forecasting technique.


As a general rule management will use information it understands. Someone who is responsible for a decision will usually not base that decision on a technique, which they cannot understand. There is a wide range of forecasting techniques. Ranging from simple ones, which are readily intelligible to most people, to others, which require a PhD in mathematics to understand. If it becomes obvious that the only appropriate forecasting technique is mathematically complex you have to ask yourself firstly whether you can competently employ it, and secondly whether the client will understand and accept the results of the calculations.

Bookmark the permalink.