Benchmarking forecasting models

Forecasting models are the heart of forecasting. The question is how to find the right model. Here we will discuss the models that are available and are used in different industries. There are basically three types of models, which are (1) Time Series, (2) Cause-And-Effect, and (3) Judgmental.

Time Series Models: In time series models, forecasts are prepared by extrapolating the past data using one technique or another. Here we assume that the past trend will continue into the future. Let us say that, in the past, sales increased at the rate of 5% a year. When we make a forecast into the future we assume that the same rate of increase will continue. Within Times Series, there are a number of models, the most important among them are: (i) Averages including Simple and Moving, (ii) Box Jenkins, (iii) Decomposition, (iv) Exponential Smoothing and (v) Simple Trend. Among all the forecasting models, Time Series models are, by and large, the simplest easy to understand and easy to use. They generally work well for shortterm forecasting.

Cause-and-Effect Models: In Cause-And-- Effect models, there is a cause (called driver or independent variable) and there is an effect (called dependent variable). If sale depends on the amount of money spent on advertisement, then sale is the effect (dependent variable) and advertisement is the cause (driver or independent variable). The Cause-And-- Effect models include (i) Regression, (ii) Econometric and (iii) Neural Network. These models are used where there is a strong relationship between the cause and effect, and the relationship between them does not change significantly at least during the forecast period. Among them, Regression models are most widely used.

Judgmental Models: Although in judgmental models judgment predominates, they are by no means seat-of-- pant models. There are set procedures that are used to arrive at forecasts. The judgmental models most commonly used in...