The two main types and further sub-types of forecasting methods are as under:
Qualitative Methods
1. Market Survey
This is considered a direct and very old method to conduct a survey among the prospective users of the product. Depending on the respondent category, a survey can be in the form of a salesforce survey or customer survey. The main purpose of the organization to conduct this type of survey is to either forecast or predict the demand for products or determine the new market potential. The information on the survey can be obtained from customers or users through telephonic interviews, in-person interviews, or mail questionnaires. The survey method is mainly useful for new product development by estimating the product demand or identifying new potential markets to launch the product.
The market survey can be further categorized as sample or census surveys. A sample survey is used in the case of consumer products in which a small group of customers is selected through the method of statistical sampling.
A Census survey is mainly useful when the number of participants is less and all can be included in the survey. For example, industrial buyers can be selected for the census survey.
2. Salesforce Opinion
The opinion of the salesforce is considered another reliable source of gathering information for estimating future demand. This method includes taking input from sales staff on future estimated sales. This method works on the assumption that salespeople understand the needs of customers in a better way because they are in direct contact with most of the customers. The response of this survey is aggregated to determine future demand.
3. Executive or Expert Opinion
This method of qualitative forecasting includes gathering the advice or opinion of expert members who are part of a group of experts and their opinion is further utilized to determine the forecast. Key executives from different departments such as accounts, production, sales, finance are also a part of this expert team. The estimation of the demand is provided by these experts according to their experience and expertise.
An approach of the personal focus group can be used to obtain such expert opinion in which all experts gather at a common area and conclude a final estimate through an agreement.
4. Delphi Method
This method also includes a group of experts i.e. decision-makers, industry key people, etc. that provide their opinions to determine a forecast. But unlike the expert opinion method, experts don’t gather physically in this method to provide their opinion. In this method, each individual member of the expert group provides their responses to independent representatives who are responsible for making a summary of these forecasts and any supportive statements. They forward the summary back to the experts along with any further queries. This process continues until a final agreement is obtained.
5. Nominal Group Testing
This method includes providing a product or service for trial purposes to a particular group of people such as employees, students, etc. Their responses based on the experience of using products or services are collected and examined further to forecast.
6. Life Cycle Analysis
In this method, opinion is formed after assessing the stage of the life cycle of a product into which it currently lies.
Quantitative Methods
This method includes activities such as sufficient data collection and use of various statistical techniques to determine some sort of patterns that will act as a forecast. Quantitative methods are categorized into two methods i.e. Time series analysis and Causal methods.
1. Time Series Analysis
In the time series analysis method, analysis of data of one type or several types is done to determine the forecast. In this method, patterns of the prior or historical data are examined and future demand is forecasted based on those patterns.
Before understanding times series analysis in detail, let’s first have a look at what time series is all about.
– Time Series
This is considered a group of data that is collected or recorded at regular intervals i.e. on a weekly, quarterly, or yearly basis. Few examples can be collecting sales data of an organization on a quarterly basis since 2015, recording temperature on an hourly basis, agricultural output on annual basis, etc. Time series can be used for predicting forecast for the long-term i.e. for more than 5 years because it is based on the assumption that there is a repetition of past patterns in the future. Different purposes can be there for long-term projections such as making decisions on manufacturing, purchasing, planning of new manufacturing units, new product development. There are four main elements of time series i.e.
- Trend
- Cyclic Variations
- Seasonality or Seasonal Variations
- Random or Irregular Movements
Trend
The trend in time series indicates the data tendency on the basis of increasing trend or decreasing trend over a longer time period. An increase or decrease may or may not be in the same direction during the given time period. Or we can say, that though it is quite possible that tendencies may go high or low or remain stable in different fractions of time, still, the pattern of the overall trend is required to be upward or downward or stable.
Cyclic Variations
These variations include data repetition for more than one year at regular intervals by following a certain pattern. Sometimes, this is also termed as a business cycle which is commonly used to understand the market trend. Different phases of cyclic variations are recession, recovery, prosperity, and depression.
Random or Irregular Movements
This is considered an unpredictable element of time series and is irregular or random variations. These types of fluctuations or variations can’t be controlled or predicted. Also, these are unforeseen and erratic. For example, different disasters such as floods, earthquakes, famines, wars fall under this category.
Seasonal Variations
These are the forces that occur on regular and periodic intervals and that too within a span of less than one year. The pattern is the same in these forces during the whole year. If the data is recorded on a daily, weekly, hourly, quarterly, or monthly basis, then the time series will contain this variation. Seasonal variations are the result of natural forces or conventions developed by humans. There is an important role of different seasons or weather conditions in seasonal variations such as the production of few crops like Rabi, Kharif, Zaid are season-based, sales of woolen wears are high in winters, conventions that are developed by humans are customs, festivals, occasions like marriage, birthdays, etc.
Different forecasting models or methods under time-series analysis are as under:
1.1) Naive Method
This is the simplest method among forecast methods of time-series analysis. In the naive method, the past period’s (the most recent one) actual demand is used as a forecast to predict demand for the next period. It considers the assumption of the repetition of the past data.
For example, a two-wheeler company that sells two-wheelers wants to make sure that it has enough sales staff and vehicles to meet the next demand. So, the company forecast the demand for offerings (two-wheelers) of next month, and for this, data of the previous 5 months is obtained as mentioned below:
In the naive method, the actual demand for the month of sep’20 will be considered as the forecast for the month of Oct ’20.
So, the demand of Sep’20= Forecast of Oct’20 = 300.
The same is shown in the below table and chart:
1.2) Moving Average Method
In this method, the moving average is calculated by doing the sum and average of the values mentioned in a time series over periods that are specified on a repetitive basis. In this, the old value is deleted and the new value is added each time. The next period’s forecast will be the same as the average calculated by summing previous or recent observations. Also, equal weightage is given to each observation.
For example, in the above example of two-wheelers, the forecast can be calculated through the moving average method.
To calculate the moving average of three months, an average of the demand for the previous three months will be calculated and the same will be considered in next month’s forecast. Let’s say to forecast the demand of Aug’20 using the moving average method, actual demand data of two-wheelers from May’20-Jul-20 will be considered which is 100, 150, and 200 respectively. So, the computation of the forecast for Aug’20 would be:
Forecast of two-wheeler demand for Aug’20: (100+150+200)/3= 150.
Similarly, the forecast for other months will be calculated and is shown in the below table:
Forecasts of five months will be calculated in the same manner, except for the previous five months’, an average of demand i.e. from May’20-Sep’20 will be taken.
So, the forecast for two-wheeler demand for Oct’20: (100+150+200+180+300)/5= 186.
1.3) Weighted Moving Average Method
In this method, more weightage is given to recent data or observations as compared to past data or observations. Unequal weights can be assigned to the past data. The weighted moving average is used to calculate the weighted average related to recent sales in order to estimate the demand for the short-term. The total of all weights in this method is required to be equal to 1.
For example, in the above illustration of two-wheeler demand forecast, to calculate the monthly demand forecast for May’20 through Oct’20 using a weighted moving average of three months, we need to assign weights to each month’s actual demand for two-wheelers by ensuring that heavier weight is assigned on the months that are more recent.
So, by keeping the above in view, to calculate the forecast of Aug’20, we have assigned weights 0.12, 0.38, and 0.50 to May’20, Jun’20, and Jul’20 respectively. So, the total of all assigned weights is equal to 1 and more weights are assigned to the most recent month.
Forecast of Aug’20= (weight of May’20 * actual demand of May’20) + (weight of Jun’20 * actual demand of Jun’20) + (weight of Jul’20 * actual demand of Jul’20)
= (0.12*100) + (0.38*150) + (0.50*200)= 169
Similarly, the forecast for other months are shown in below table and chart:
1.4) Exponential Smoothing Method
This method determines forecasts by using weighted averages that are based on past observation. More importance is given to recent data or values in a particular series. Moreover, in the exponential smoothing method, weights start declining or decreasing exponentially with past observations. In more simple words, we can say, that in the case of the observation that is the most recent one, the associated weight is higher.
In this method, each new forecast is determined by adding the past forecast and the percentage of the value which is the difference between the actual forecast and that past forecast. So, the new forecast will be:
New forecast = Past forecast value + α (Actual demand value – Past forecast value)
In the above formula, α is considered a Smoothing constant that varies from 0.01 to 0.50. If the value of α is higher, then changes in time series can be tracked more closely.
For example, a computer PC or Laptop assembling company assembles personal computers or laptops from different generic parts. For this, it purchases generic parts in bulk at a discounted price from different sources wherever there is a good deal in terms of quality and less price. So, the company wants to forecast the demand for their PCs or laptops in order to determine the required generic parts to be purchased and keep as an inventory.
The future demand forecast from Jan’20-Sep’20 using exponential smoothing and smoothing parameter α= 0.2 is computed by considering demand data of the last 12 months as shown in the below table. Also, in the below example, as the data for Dec’19 is not available, so it is assumed that both actual and forecast for Jan’20 are the same. Now, the forecast is made for each subsequent month (starting from Jan’20) till Sep’20 using the exponential smoothing method.
The graphical representation of the above forecast is as under:
The above calculation of the exponential smoothing forecast of each month is as under:
Jan’20= same as actual demand= 30
Feb’20= 30 + 0.2 (30-30) = 30
Mar’20= 30+ 0.2 (34-30) = 30.8
Apr’20= 30.8+ 0.2 (35-30.8) = 31.64
May’20= 31.64+ 0.2 (30-31.64) = 31.31
Jun’20= 31.31+ 0.2 (40-31.31) = 33.05
Jul’20= 33.05+ 0.2 (50-33.05) = 36.44
Aug’20= 36.44+ 0.2 (42-36.44) = 37.55
Sep’20= 37.55+ 0.2 (48-37.55) = 39.64
1.5) Trend Projection Method
This method uses past available data related to different variables i.e. both dependent and independent for sales projection of the future. It works on the assumption that the behavior pattern of factors that are behind past trends will be the same in the future as well. Time series is obtained by arranging the data in chronological order. This time series estimate trends in the past based on that estimation of the future trend of the market. Once the future trend is predicted, then an organization is able to forecast the future demand for its product or service.
This method is a type of linear regression method and includes an equation of straight line i.e. Y=a + bX. In this equation, the values of time are represented through horizontal axis X, and values of demand are represented through vertical axis Y. (b) represents the trend line and (a) denotes the point where the line crosses the axis Y.
In the above equation, a and b can be calculated by solving the below equations:
In the above equations, n is equal to the total data items in a particular series.
For example, the below table shows the sales data of an organization from 2014 to 2019:
Trend projection through straight-line trend and estimation of total sales for the year 2022 would be as under:
In the above table, the following data is available:
2. Casual or Associative Forecasting Models
Associative forecasting models include identifying variables that can be useful in estimating another variable that has some type of association with each other. These are also termed casual forecasting models. The assumption, on which these models are based, is that validity of the link between independent and dependent variables from a historical point of view will remain in the future, and also, it is easy to estimate each independent variable.
It has been observed that in organizations many situations occur in which there is some linkage between the data values of different variables. It may not be a direct situation of cause and effect, but the depth of association can be predicted. This helps in estimating the value of another variable, once the value of one variable is available through mathematical equations.
Different common casual models of forecasting include:
2.1) Regression Method of Forecasting
The regression model is considered a common tool or method to define a relationship between two or more variables in a dataset. This includes different statistical techniques through which the link between a dependent variable and one or more than one independent variable can be estimated. Using the regression method, it is easy to assess how strong the relationship is between variables. Also, future relationships between variables can be determined.
For example, in a business, the profitability of the business which is a dependent variable can be forecasted by considering different factors like inventory stock, cost of products sold, etc. which are independent variables.
The regression model is mainly of two types i.e.:
Simple Regression or Linear Method
In this method, the dependent variable is affected by only one independent variable. The following equation represents the simple regression or simple linear method:
Y = a + bX
In the above equation, Y stands for the dependent variable, X is the independent variable, a represents intercept, and b indicates slope.
The value of a and b is calculated through the following formula:
Multiple Regression Method
When two or more than two independent variables are there that influence a dependent variable, then it is considered multiple regression. The equation of multiple regression method is mentioned below:
Y = a + bX1 + cX2 + dX3
In the above equation, Y stands for the dependent variable, X1, X2, X3are independent variables, a represents intercept, and b, c, d indicate slopes.
The process of forecasting under the regression method is as under:
- To plot both the dependent variable and independent variable (s) along with rectangular coordinates
- To develop an equation of the trend
- To use the equation for forecasting
- It is not mandatory that variables are related according to the time.
2.2) Econometric Method
In this method, economic variables are used in forecasting future-related developments. It is based on how economic variables and internal data of an organization such as internal sales data interact with each other. Employment rate, exchange ratio, inflammation, etc. are a few economic variables.