Principles of Managerial Statistics and Data Science

17Forecasting

17.1 Introduction

There is a lot of data out there with an index of time. These type of observations are known as a time series. For example, monthly sales of winter clothing from a local company for the past 15 years. These data sets have important features over time that are worthwhile taking into consideration. Returning to the winter clothing example, the company may want to determine if the winter clothing sales have been increasing overall. Additionally, the company may wonder if there is a seasonal pattern in the sales of their winter clothing (Do you think they should expect a seasonal pattern?). This type of information can help guide the company's short‐ and long‐term plans.

A Regression Model?

Some of the regression techniques discussed before may come to mind as a way to model the association between a response variable and time. Although possible, regression techniques may fall short due to differing components of temporal dependency and inefficiency in estimating regression coefficients when observations are not independent. If there is no strong temporal dependence in the data, then regression methods and more generally, predictive analytics can be applied to obtain forecasts.

Typically, the time index is at a monthly, quarterly, or yearly scale. It may be discrete and regular (i.e. ) or discrete and irregular (i.e. ). The index may be continuous ...