Forecasting Models

Understanding Forecasting Models

Forecasting models utilize historical and current information to provide a range of probable outcomes. The objective of a forecasting model is to extrapolate past and current trends with the help of various statistical and analytical tools to predict a future scenario. The results of such forecasting models form the basis of strategic decision making. Following are some examples of forecasting model applications:

  • To ascertain the future movement in the price of a stock
  • To determine the manpower turnover based on past trends
  • Forecasting the demand for a particular product

A forecasting model takes into account all the variables and possibilities that are associated with the subject to be forecasted. Such models are based on a number of assumptions, aggregations, and probabilities. Risk and uncertainty will, therefore, always underlie any forecasting model. The goal is only to forecast an outcome which would be closest to the real picture in order to minimize deviations from management expectations.

Types of Forecasting Models

The entire range of forecasting models available today is vast and ever increasing. They vary from fundamental to extremely complex in form. While an understanding of advanced models can be developed only with extensive study, a few basic models have been illustrated here below:

Time Series Forecasting

It is a quantitative forecasting technique. This model seeks to uncover patterns hidden in the movement of data over specific intervals: hourly, weekly, monthly, yearly etc. This method, therefore, relies on sequential repetition of events to forecast a future outcome.

Time Series Forecasting With Excel

  • Insert data with time or duration in one column. A fixed interval, say a day, month or year should lapse in between. (Ensure the format of the concerned column is set to date, otherwise, the forecast sheet shall show an error).
  • Insert corresponding values sought to be forecasted in the next column.
  • Select relevant data
  • On the data tab > Forecast Group > Forecast Sheet.
  • In forecast worksheet select line or bar graph as per preference.
  • In forecast end box, set forecast end date > Create.

The historical values of S&P 500 for the last 24 years have been plotted below. A confidence interval of 95% is set.

Timeline Values Timeline Values
Jan-95 465.25 Jan-07 1,424.16
Jan-96 614.42 Jan-08 1,378.76
Jan-97 766.22 Jan-09 865.58
Jan-98 963.36 Jan-10 1,123.58
Jan-99 1,248.77 Jan-11 1,282.62
Jan-00 1,425.59 Jan-12 1,300.58
Jan-01 1,335.63 Jan-13 1,480.40
Jan-02 1,140.21 Jan-14 1,822.36
Jan-03 895.84 Jan-15 2,028.18
Jan-04 1,132.52 Jan-16 1,918.60
Jan-05 1,181.41 Jan-17 2,275.12
Jan-06 1,278.73 Jan-18 2,789.80

The consequent graph will look as follows :

Interpreting a Forecasted time series

Values: The line plotted as values is nothing but a graphical representation of the data under review. The X-axis bears the time with an intermission of one year. The Y-axis bears the range of values.

Upper Confidence Bound: Denotes that 95% of future values will be “less than” or within its range.

Lower Confidence Bound: Denotes that 95% of values will be “greater than” or beyond its range
Together, the upper and lower confidence bounds seek to explain that 95% of all future values will lie between the maximum and minimum limits that have been carved out by this forecast.

Forecast: Basically, an aggregation or average of the upper and lower confidence bounds. It considers the minimum and maximum extremes to come up with a line that may be a “best fit”. The future values may reasonably be expected to lie closest to this range. It absorbs historical patterns to manifest its future counterpart assuming to carry on the same trend.

Regression Analysis

Regression refers to the statistical analysis of various variables and scanning for any inter-relationship that may exist among them. This technique is the most useful when the variables under consideration display dependence on another such that the movement of one (dependent variable) is a function of the movement of another (independent variable). The analysis aids in uncovering deep-seated correlation that exists between factors. Thus, such knowledge can be exploited and the proportions of variables are controlled to gain desired results.
The relationship between rainfall in centimeters and crop harvested, gold prices and exchange rates, wage scale/working hours and labor productivity. Each of the aforementioned relationships can be subjected to regression to extract relevant patterns.

Regression Analysis with Excel

Consider the following data for thefts and fires in Chicago.

X Y X Y
6.2 29 18.4 32
9.5 44 36.2 41
10.5 36 39.7 147
7.7 37 18.5 22
8.6 53 23.3 29
34.1 68 12.2 46
11 75 5.6 23
6.9 18 21.8 4
7.3 31 21.6 31
15.1 25 9 39
29.1 34 3.6 15
2.2 14 5 32
5.7 11 28.6 27
2 11 17.4 32
2.5 22 11.3 34
4 16 3.4 17
5.4 27 11.9 46
2.2 9 10.5 42
7.2 29 10.7 43
15.1 30 10.8 34
16.5 40 4.8 19

X= Fires per 1000 housing units; Y= Thefts per 1000 in population

Step 1: Draft the data set in excel
Step 2: Go to Data on the ribbon > Data Analysis Tab> Select Regression> Click OK

The following Sheet appears:

Regression Stats Decoded

Multiple R: This is the correlation coefficient. It varies from -1(perfectly negative) to +1(perfectly positive). In the given case, the correlation coefficient of 0.55 represents a moderately positive relationship.

R Square: Coefficient of Determination. It is the square of Multiple R and ranges from) to 1. Denotes how many points lie on the regression line or the “goodness of fit”. The closer to 1 the better the fit. In above example, an R square of 0.30 denotes a below average exactness between variables X & Y.

Standard Error: It is the measure of the accuracy of predictions of the regression analysis. Smaller the standard error the higher the accuracy. A standard error of <0.05 is considered suitable since it agrees with the well-accepted 95% confidence interval range. In the example, a standard error of 19.46% is higher than desirable.

Qualitative Means of Forecasting Models

This technique departs from statistical analysis and number crunching and is based largely on expert opinion and judgment. Qualitative forecasting methods are mainly adopted when the historical trends are not expected to follow in the future. They are also resorted to when the data set is too narrow to be capable of being extrapolated into a numerical analysis. Experts may then find it viable to exercise judgment and experience to discern sense out of a seemingly incoherent data set. Additionally, qualitative methods are the most sensible option when trends and population habits are constantly evolving with time. For example, choice of music, fashion, the medium of entertainments etc. these variables are very dynamic. A fashionable clothing item in high demand 20 years ago holds no importance today. Since the tastes have drastically evolved and the trends 20 years ago are just a useless piece of information.

Delphi Method

A panel of experts is questioned and asked about their opinion of the future trends and outcomes. The analysis is conducted by a third party so as to ensure authenticity. Each of the panelists is reviewed separately so as to prevent manipulation or domination among one another. After a preliminary round of questions, the results are then collated and a second specific set of questions is pitched back to the panelists. This process continues until the scope of the outcome can be narrowed down with reasonable precision.

Consumer Survey

Corporations often indulge in customer and industry survey to obtain firsthand information of their tastes and preferences. The relevant data is collected through several mediums such as questionnaires, checklists, sampling and even by a salesperson making visits door to door. Based on the results of such a survey the company is able to judge the demand for its products, consumer patterns and habits and the changes it is required to make in order to boost its sales.

Scenario Analysis

It is a form of projection technique where the fate of a situation under consideration is analyzed by building “alternative scenarios or alternative worlds”. The probable outcomes are scrutinized making three sets of assumptions- optimistic, pessimistic and likely. An important characteristic of this form of analysis is that, unlike most forecasting techniques, this technique does not rely on historical data. Rather it takes into account the futuristic probabilities making the prediction as close to life as possible.

Forecasting Models

References

https://en.wikipedia.org/wiki/Forecasting#Judgmental_methods

https://en.wikipedia.org/wiki/Regression_analysis

http://archive.ics.uci.edu/ml/datasets.html?task=reg

Last updated on : July 23rd, 2018

** Disclaimer: This post may contain Affiliate Links marked as ** and we may earn a commission on sale.

What’s your view on this? Share it in comments below.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.