What is Time Series Forecasting?
Forecasting the future is one of the most exciting sciences there is. The impossibility of knowing is going to happen beyond what has happened so far, causing enormous uncertainty. Future sales, the weather in the coming days, the behavior of the stock markets, supply chain management, are clear examples of the importance of forecasting.
The tools that forecasters use to carry out this task are based on the use of time series. This is an accumulation of data taken over time, called observed data, which is analyzed and future projections are made.
Forecasting techniques using time series produce forecasts, but not prophecies!!
How to perform a forecast?
Prediction methods are very varied, although a rational distinction can be made between all of them: fundamental, statistical and artificial intelligence or computational methods. It cannot be said that none of them predominates over the rest, but for each application, the use of one or the other may be convenient.
The fundamental methods are based on relating different variables considered fundamental, with which, by analyzing each of them separately, a prediction can be obtained.
Statistical methods are the most used, without a doubt. They use proven techniques and allow the effectiveness of the prediction to be assessed. On many occasions, they also propose a value of uncertainty about the measurement. Exponential smoothing models, ARIMA models, regression models, and a long catalogue of applications of statistical techniques are widely used.
Computational methods, based on A.I. they are gaining traction as algorithms improve and the power of computers increases. Neural network-based models find much application in prediction, although other A.I. is also widely used. However, these methods are very specific, and it is necessary to use a different tool for each occasion.
Software
Many software is available. As a general rule, they are part of generic statistical packages, such as SAS, STATA, IBM SPSS, or as libraries in statistical software, see R, MATLAB, etc. The specific software exists, but its efficiency is not proven.
mshw is a forecasting software based on multiple seasonal Holt-Winters models. It has been developed under MATLAB environment, thus it manages a large time series.
Click here to obtain more details about this software.