Techniques of Demand Forecasting

Techniques of Demand Forecasting

The importance of selecting the right demand forecasting method cannot be overstated. However, the choice is complicated, because each situation might require a different method. Management should be aware of the factors favoring one method over another in a given demand-forecasting situation.

In some cases, managers are interested in the total demand for a product or service. In other circumstances, the projection may focus on the firm’s probable market share. Forecasts can also provide information on the product mix. Major decisions in large business houses are generally based on forecasts of some type.

In some cases, the forecast may be little more than an intuitive assessment of value judgment of the future by those involved in decision-making. Thus, no forecasting method is suitable for all circumstances.

The selection of a forecast has to be appropriate to the situation, i.e., objective, urgency, data availability, nature of the product, time horizon, cost the firm can afford, and accuracy level required. Further, various forecasts have to be linked properly for correct decision-making. It is always prudent to use more than one forecast for crosschecking and for improving the credence of the forecast.

None of these methods is perfect. These methods can be used to derive macro as well as micro forecasts. When requisite data or firm forecasts are not available, the firm’s demand can be forecasted by multiplying the forecasted firm’s share in industry demand (through some statistical method like trend method to the past data of variable) and forecasted industry’s demand of the product.

Similarly, likely regional demand can be derived from the aggregate demand in the industry as a whole. The following are important methods of forecasting:

Survey Methods

Under this approach, surveys are conducted about the intentions of consumers (individuals, firms, or industries), the opinion of experts, or of markets. Under the census survey, all consumers/experts/ markets are surveyed. When taking sample surveys, a selected subset is surveyed and through study, inferences are drawn.

These methods are usually suitable for short-term forecasts due to the nature of consumers’ intentions. New product demand forecast also makes use of a survey approach, as the data availability problem is overcome through surveys of consumers. A few important survey methods are discussed here:

Consumer Survey Method

Surveys of consumer plans can be one of the important methods of forecasting. The rationale for conducting such surveys is that plans generally form the basis for future actions. By using this method, a firm can ask consumers,

what and how much they are planning to buy at various prices of the product for the forthcoming time period, usually a year. If the product happens to be a consumer good, the consumers are firms or industries using that product. The survey may involve a complete enumeration of all consumers of the given product, whose demand is to be forecast.

If there are ‘n’ consumers in all, each demand Di, then the total demand forecast will be Di ∑i 1

This type of complete enumeration or census method is time-consuming, tedious, and cumbersome, particularly, for a product having a vast market. So, a relatively cost-effective, less tedious, less time-consuming sample or survey method may be used.

The forecaster selects a few consuming units out of the relevant population and then collects data on the likely demands for the product. If the probable demand of the selected consumers is Di, then the total demand forecast for the commodity will be:

n ∑i 1= Di

Where ni is the number of consumers in group ‘i’. Here, care needs to be taken that the forecaster’s bias does not creep into the sampling procedure. Further, the consumers may be uncertain about the quantity they intend to purchase from a particular firm.

The merits of this method are as follows:

  • It’s a direct method of assessing information from primary sources.

  • It is a simple method, as it is not based on past historical records.

  • It saves time and cost by conducting a survey on a representative sample through the issue of questionnaires or otherwise, depending upon the complexities of the problem.

  • It does not introduce any bias or value judgment, particularly in the ‘census method’, where the data is simply recorded and aggregated.

The demerits of this method are as follows:

  • There may be personal bias of the consumers in answering the questions of the questionnaire.

  • It becomes difficult for a firm to ascertain the number of consumers that intend to buy from that firm. These methods only give information about the demand for a product in the industry. The survey method may not be directly useful for estimating demand for a particular firm.

    Tools like the Markov chain process has to be employed to find the firm’s share in the total estimated demand for the product market.

  • The utility of these estimates is limited to a period of about one year.

  • There may be sampling errors if the sample is not properly chosen.

Collective Opinion Method

Under this method (also called sales-force polling), salesmen or experts are required to estimate the expected future demand for the product in their respective territories and sections. The rationale of this method is that salesmen, being the closest to the customers, are likely to have the most intimate feelings of the market, i.e. customer reaction to the products of the firm and their sales trends.

The estimates of individual salesmen are averaged or consolidated to find out the total estimated sales and then reviewed by the top executives to eliminate the bias of optimism on the part of some salesmen and pessimism on the part of others.

These revised estimates are further examined in light of factors like proposed changes in selling prices, product designs, advertisement programs, expected changes in competition, and changes in secular forces like purchasing power, income distribution, employment, population, etc.

The final sales forecast would emerge after these factors have been taken into account. This method is known as the ‘collective opinion method’, as it takes advantage of the collective wisdom of salesmen, departmental heads like production managers, sales managers, marketing managers, managerial economists, etc., and the top executives.

The merits of this method are as follows:

  • The method is simple and does not involve the use of statistical techniques.

  • The forecasts are based on firsthand knowledge of salesmen and others directly connected with sales.

  • The method may prove quite useful in forecasting sales of new products. Of course, here salesmen will have to depend more on their judgment than in the case of existing products.

The demerits of this method are as follows:

  • It is almost completely subjective, as personal opinions can possibly influence the forecast. Salesmen may even understate the forecast if their sales quotas are to be based on it.

  • The usefulness of this method is restricted to short-term forecasting, i.e., for a period of about one year. These forecasts may not be useful for long-term production planning.

  • Salesmen may be unaware of the broader economic changes likely to have an impact on future demand.

Reasoned Opinion (Delphi) Method

A variant of the opinion polls and survey methods is the Delphi method, developed by Rand Corporation of USA in the late 1940s for predicting technical changes. This method was earlier followed by the Greeks. It consists of an attempt to arrive at a consensus in an uncertain area by questioning a group of experts until the responses appear to converge along a single line or the issues causing disagreement are clearly defined.

The researcher or coordinator provides each expert with the previous responses, including the reasons thereof. He may also bridge the gap between the opinions of experts by arranging more meetings and bringing them together. He can also temper the ‘interval forecast’ (i.e., the lower and upper limits within which the demand forecast is to be) to the needed ‘point forecast’ through his conceived overall assessment and judgment.

Thus, like the collective opinion method, the Delphi method also uses the opinion of the experts to come to a logical conclusion regarding future demand. However, in this method, the group of experts employed should not be directly involved with that particular firm or industry.

Merits of this method are as follows:

  • It facilitates the maintenance of anonymity of the respondent’s identity throughout the discussion.

  • This method renders it possible to pose the problem to the expert at one time and have their response. The experts can even revise their responses.

  • This technique saves time and other resources in approaching a large number of experts for views.

The demerits of this method are as follows:

  • Since it is a tedious method, the panelists must be rich in their expertise and possess wide knowledge, and experience of the subject, along with an aptitude and earnest disposition towards the participants.

  • The Delphi method presupposes that its conductors are objective in their job, possess ample abilities to conceptualize the problems for discussion, generate considerable thinking, stimulate dialogue among panelists, and make an inferential analysis of the multitudinal view of the participants. However, this may not be so.

Market Experiment Method

This method is very popular in developed countries, but relatively new and less tried in India. Under this method, the main determinants of the demand for a product like a price, advertising, product design, packaging, quality, etc., are identified. These factors are then varied separately over different markets or over different time periods, holding other factors constant.

The effect of the experiment on consumer behavior is studied under actual or controlled market conditions, which is used for overall forecasting purposes. Here, the market divisions must be homogenous with regard to income, population, caste, religion, sex, age, tastes, preferences, etc.

For example, a controlled experiment was performed by the Parker Pen Company in the USA to gauge the effect of the price rise on the demand for Quink ink. In controlled experiments, token money may be provided to the consumers and they may be asked to shop around in a simulated market. This method is called the simulation method or laboratory experiment technique.

The merits of this method are as follows:

  • This carefully carried out exercise can help the researcher to come out with a demand function, indicating quantities that the consumers would be ready to take from the market at various prices, levels of income, etc.

  • This method can be used to check the results of demand forecasting obtained from other methods.

The demerits of this method are as follows:

  • These experiments are expensive and time-consuming. A large number of buyers can never be involved in these experiments.

  • Potential buyers may treat these experiments as a game and may not behave in a natural manner.

  • These methods are risky, as they might send wrong signals to consumers, dealers, and competitors, which may lead to unfavorable situations for the concerned firm.

  • The conclusions and inferences derived from the experiments in some sections of the market cannot be applied to the whole market.

  • It is difficult to satisfy the condition of homogeneity.

Statistical Methods

These methods make use of historical data (time series or cross-section) as a basis for extrapolating quantitative relationships to arrive at future demand patterns and trends. The data may also be analyzed through econometric models.

These are useful for long-term forecasting, for old products, and for larger levels of aggregation. They are based on scientific ways of estimation, which are logical, unbiased and proven to be useful. However, the biggest disadvantage is that it is difficult to apply these methods. One needs a competent person to handle, interpret and manipulate the data for statistical purposes.

Further, these methods cannot be used for forecasting the demand for new products and products, which have, short existence due to the data problems. Furthermore, the past may not be repeated in the future. Statistical methods are broadly classified under the following two categories.

Time Series Analysis

It is an arrangement of statistical data in chronological order, i.e., in accordance with its time of occurrence. It reflects the dynamic pace of steady movements of a phenomenon, over a period of time. Most of the variables in business,

economics and commerce, be it a series related to price, production, consumption, national income, foreign trade, foreign exchange reserves, investment, sales, projects, dividends, etc., are all time series data, spread over a long period of time.

The data may be presented in the form of a table or a graph. Here, the time series data on the variable under forecast is used to fit a trend line or a curve graphically. The trend lines can be worked out by fitting a trend equation to time series data through the least squares method or some other estimation method.

Time series analysis can be used for demand forecasting by first evaluating, extracting and interpreting its various components, so as to make it understandable and explainable.

These four components are described below:

  • Trend: It shows the underlying, long-term tendency of the data, which may be the result of basic developments in population, capital, technology, etc. Any event of the future period can be forecasted using a trend line.

  • Seasonal Variation: These are short-term, cyclic fluctuations in the data about the trend, which is measured in an interval in a year. If the average monthly sales of a product for a particular month are 10 percent above the trend line, a seasonal adjustment factor of 1.10 can be applied to the trend projection to forecast sales in that month.

    However, here, the word ‘season’ can have different meanings. It may be related to weather, holidays, customs, festivals, fashions, etc. Here, a series fluctuates according to seasons of the year.

    Daily ‘seasons’ over a weekly cycle for sales in a supermarket, monthly ‘seasons’ over a yearly cycle for purchases of a company and quarterly ‘seasons’ over a yearly cycle for sales of electricity in the domestic sector are some examples. There is some regularity with regards to their occurrence.

  • Cyclical Variation: It can be thought of as an underlying economic cause outside the scope of the immediate environment. Here, the length of the cycle is generally longer than one year. Study of these variations is essential for predicting the turning points in business cycles.

    Cyclical variations are affected by swings in general economic activity, wherein recovery and boom are followed by recession and depression, and vice-versa. Capital goods industry shows such pattern of business cycles of constant amplitude and periodicity.

    Though, the trend and the seasonal factors can be forecasted, the prediction of cycles is difficult, as they do not always repeat at constant intervals of time.

  • Residual Variation: These include other factors not explained by (a), (b) and (c) above. These are disturbances due to unforeseen future events, such as weather conditions, illness, strikes, lockouts, riots, fires, wars, transport breakdowns and so on.

    The evaluation of these components is not usually required in examinations, but its interpretation should be known. They create disturbances, which are erratic in nature. Their operation and effects are not orderly.

    Depending on the nature, complexity and extent of the analysis required, there are various types of models that can be used to describe time series data. However, the additive and multiplicative models are the simplest standard time series models, which assume that the effects of component factors are independent of each other. The components that go to make up each value of a time series are described below:

    Time series additive model Y = t + s + c + r
    Time series multiplicative model Y = t × s × c × r

    where, Y is given time series value.
    t is the trend component.
    c is the cyclical component.
    s is the seasonal component.
    r is the residual component.

    In other words, given a set of time series data, every single given value (Y) can be expressed as the sum or product of four components. The decomposition of these components is necessary for understanding the nature of business fluctuations. It is important to note that the trend component will be the same, no matter which of the two models are used.

    In other words, given a set of data to which both models are being applied, both trend values would be identical, whereas the respective seasonal and residual components would be quite different.

    The demand can be forecasted through the identification of trends and estimation of extraneous factors.
  • Graphical Method: This method gives the basic tendency of a series to grow, decline or remain steady over a period of time. This method is useful in forecasting India’s population, and demand for cement, textiles, steel, paper, etc., where the future is not too much different from the average of the past.

    The period of time in the trend analysis is always a long time period. The concept of trend does not include short-time oscillations and fluctuations.

    Trends can be both linear and nonlinear. If the time series values are plotted on a graph, one can pass a straight line depicting the trend such that most of the values will fall on or near the line. This line may be drawn up to the present period or the period for which the data is available. It can then be extrapolated for the forecast period.

    Figure (a) shows that there is an upward linear trend, while Figure (b) shows that there is a downward linear trend. We observe that the straight trend line has been fitted around the fluctuations. The actual movement of the data is shown by the dotted curve.

    If the values of the variables are such that they cluster around a nonlinear path, we get a curvilinear or nonlinear trend. The nonlinear trend can be either quadratic or cubic. Figure 9.3 (a) and Figure (b) show a quadratic trend.

    Figure (a) represents an inverse parabola with a maximum point. Figure (b) also shows a parabola, but, with a minimum point. This is called a straight parabola. Figure (c) and Figure (d) represent a cubic trend.

    Study of trend enables managers and firms to forecast their business in the long run and to plan future operations, without formal knowledge of economic theory and the market. It only needs the time series data; e.g., we get the negative linear trend of demand for a particular commodity.

    The managers know that unless something fundamental changes, the production of this commodity will have to be reduced in a phased manner. Accordingly, inventory and investment have to be planned.

  • Semi Averages Method: According to this method, the data is divided into two parts, preferably with the same number of years. The averages of the first and second part are calculated separately. These averages are called semiaverages.

    Semi-averages are plotted as points against the middle point of the respective time periods covered by each part. The line joining these points gives the straight-line trend fitting the given data. In case of odd number of years, the two equal parts are obtained by omitting the values for the middle period.

    For example, for the data of 9 years from 1994 to 2002, the two parts will be the data for years 1994 to 1997 and 1999 to 2002; the value for 1998 will be omitted.

  • Moving Averages Method: When time series analysis does not reveal a significant trend of any kind, the moving averages method may be used to smoothen the series. This is a very simple and flexible method of measuring trend.

    It consists of obtaining a series of moving averages of successive overlapping groups of the time series. The averaging process smoothens out fluctuations as well as the ups and downs in the given data.

    The moving average is characterised by a constant known as the period of extent of the moving average. Thus, the moving averages of period ‘m’ is a series of successive averages of ‘m’ overlapping values at a time, starting with 1st, 2nd, 3rd value and so on.

  • Least Squares Method: The principle of least squares provides us an analytical tool to obtain an objective fit to the trend of the given time series. Most of the data relating to economic and business time series conform to definite laws of growth or decay.

    Thus, in such situations, trend fitting will be the most reliable way of forecasting. Here, the assumption is that past rate of change for the given variable would continue in the future. Least squares can fit both linear and nonlinear trends. However, trend projection breaks down, when a turning point occurs.

  • Fitting Non-linear Trend: Demand or sales having a linear function in time, is an over simplistic assumption. The demand function can be a parabola, exponential logarithmic curve, etc. Least squares method can be used to fit such cases of nonlinear trends too. Let us suppose a new product is launched in the market. The market itself is not mature.

    We may expect that in the initial phase, the demand will grow at a slow rate, followed by a fast rate, till it reaches a saturation point. We will get something like the curve shown in Figure .
    We have to fit a logarithmic trend here, which is very common in industrial growth.

Regression Analysis

Regression analysis is perhaps the most popular method of forecasting among economists. It is a mathematical analysis of the average relation between two or more variables, in terms of the original units of the data. Here, the data analysis should be based on the logic of economic theory.

Further, the explanatory variable whose value is influenced or is to be predicted is called the regressed or dependent variable; while the variable, which influences the dependent or predicted value is called the regress or predictor or explanatory variable.

When the regression analysis is confined to the study of two variables, it is called simple regression. But, quite often the studied regressed variable depends on multiple factors simultaneously. In that case, we have to study the effect of more than one predictor on the value of the predicted variable. It is called multiple regression.

The general form of multiple regression is Y=f (x1 , x2 , x3 …………) .

We have already studied the simple regression under the method of least squares, where we have seen how time can affect the demand or sale of a commodity. Multiple regressions are a more realistic and accurate way of predicting the future demand. Demand for money (Mt ) in an economy as a function of national income, rate of interest and time is a common example. We can write

Mt = a(Yt )b 1 (rt )b 2 (t)b 3

Here, Mt is stock of money in time period ‘t’
Yt is national income in time period ‘t’
rt is rate of interest in time period ‘t’

This is an exponential function in which ‘a’ is constant and b1 , b2 and b3 are the respective coefficients. Here, the dependent variable Mt is to be predicted and Yt , rt and ‘t’ are independent variables.

Now, take a completely different kind of demand, i.e., demand for roses during festival and marriage seasons. It will depend on price of roses, price of other flowers and family’s disposable income. Empirical study shows that the demand for roses during the peak season grows exponentially.

DR = a(PR)-b 1 (POF)b 2 (I)b 3

log DR = log a – b1 log PR + b2 log POF + b3 log I

Here, DR is the demand for roses, PR is the price of roses, POF is the price of other flowers and ‘I’ is the family’s disposable income. Further, ‘a’ is a constant, while –b1 tells us that if PR increases by 1% DR will decrease by b1 %.

Similarly, 1% increase in POF and ‘I’ will increase DR by b2 % and b3 % respectively. –b1 (own price elasticity), b2 (cross-price elasticity) and b3 (income elasticity) are called partial regression coefficients, which measure the responsiveness of regressed variables on the regresses.

The method of solving multiple regression analysis is similar to the simple regression analysis, i.e., solving least square equations for values of the parameters.

The merits of this method are as follows:

  • This method is prescriptive as well as descriptive. Besides generating demand forecast, it explains why the demand is, what it is. Thus, the technique has both explanatory and predictive value.

  • It is an objective method using time series as well as cross-section data.

The demerits of this method are as follows:

  • If some explanatory variables are not realistically chosen, they tend to be misleading.

  • The forecast will be wrong to the extent that future relationship deviates from the past average experience.

  • Regression results will be biased in the case of auto-correlation, i.e. when for one year, demand is highly correlated with the demand of preceding years.
  • Tapan K Panda, Marketing Management, Excel Books.

  • V S Ramaswami and S Namakumari, Marketing Management, Macmillan, 2003.

  • Ramphal and Gupta, Case and Simulations in Marketing, Galgotia.

  • Jayachandran, Marketing Management, Tata McGraw Hill, 2003.

Leave a Reply