Basic Concepts in Time Series Analysis (Part-3)

Basic Concepts in Time Series Analysis (Part-3)

This post is the third part of a series of tutorials regarding time series analysis. Here is the first part discussing about the definition of times series, and here is the second part discussing on time series components.

Forecasting models

The idea behind a forecasting model can be used for the representation of any numeric procedure aiming for a simple yet reliable approach of its underlining mechanisms. A forecasting model represents the process that is followed in order to produce predictions. Obviously, every such model has its own process of analysis and therefore, there is a variety of prediction models. These models are divided into two main categories; time series models and exploratory models:

  • Time series models: This category is the most popular type of forecasting models. They can be utilized in the case when historic data of the dependent variable are available. This approach is based in the assumption that past behavior of the data can be used to describe future values of the dependent variable. As an input to such models, past observations of the variable of interest is used and the relationship between inputs and outputs is linear. An example of such models is ARIMA and Exponential Smoothing methods.
  • Exploratory models: This model is based on the assumption that there is some constant relationship between the variable of interest and a number of predictors (independent variables). While in the time series models the function f that describes the relationship between input and output is derived from historic data, in the exploratory models this function is derived measuring the correlations between dependent and independent variables and then modeling the relationship between inputs and output with the best way possible.

A basic advantage of the time series models is that they don’t need exploratory variables to model the behavior of the dependent variable. However, the assumption that past data can fully describe the evolution of the variable of interest in the future is by no means an accurate approach since special events or other external factors that influence the output of the model, cannot be modeled. On the other hand, in exploratory models the researcher has a clearer idea of what factors have an effect on the variable of interest and therefore these models can be embedded in the model. However, for this approach more data are needed since it is a prerequisite that future values of the predictors are also know in order to be useful for the prediction of the variable of interest.

No Comments

Sorry, the comment form is closed at this time.

× How can we help you?