13 Time Series Forecasting

Time Series Forecasting

643

CHAPTER OUTLINE

13.1 Assessing Time Series Behavior
13.2 Random Walks
13.3 Modeling Trend and Seasonality Using Regression
13.4 Lag Regression Models
13.5 Moving-Average and Smoothing Models

Introduction

Many business decisions depend on data tracked over time. Quarterly sales figures, annual health benefits costs, weekly production, monthly product demand, daily stock prices, and changes in market share are all examples of time series data.

The U.S. Energy Information Agency and the International Energy Agency track global demand for oil over time to make forecasts of future demand. Such forecasts have direct influence on market prices for gasoline.
Hilton Worldwide is one of the largest hospitality groups with its 10 brands, including Hilton, Hampton Hotels, and Embassy Suites. Hilton Worldwide makes annual forecasts of occupancy rates and revenue per available room. Favorable forecasts lead to decisions to add new rooms worldwide and benefits the company in terms of investor relations.
Kimberly-Clark, whose leading brands include Kleenex and Huggies, utilizes sales data to generate forecasts that trigger shipments to stores. As a result of improved forecasting, Kimberly-Clark has reduced its cash conversion cycle, cut its total supply chain expenses, and increased gross margins.
On the other hand, poor forecasting can incur tremendous costs to a company. For example, in 2014, The Wall Street Journal reported the following: “A billion-dollar forecasting error in Walgreen Co.’s Medicare-related business has cost the jobs of two top executives and alarmed big investors.”¹

Overview of Time Series Forecasting

Time Series

Measurements of a variable taken at regular intervals over time form a time series.

644

Large and small companies alike depend on forecasts of numerous time series variables to guide their business decisions and plans. In the short term, forecasting is typically used to predict demand for products or services. Demand forecasting helps in operational decisions such as establishing daily or weekly production levels or staffing. For the longer term, businesses use forecasting to make investment decisions such as determining capacity or deciding where to locate facilities. It is not uncommon for companies to provide forecasts of key quantities in their annual reports; for example, a 2015 annual report may contain forecasts for how the company will perform in 2016. Whether the forecasts are short or long term, the first step is to gain an understanding of the time behavior of the key variables so as to make reasonable projections.

We handle time series data with the same approach used in earlier chapters:

Plot the data, then add numerical summaries.
Identify overall patterns and deviations from those patterns.
When the overall pattern is quite regular, use a compact mathematical model to describe it.

In this chapter, we focus on the plots and calculations that are most helpful when describing time series data. We learn to identify patterns common to time series data as well as the models that are commonly used to describe those patterns. To recognize when patterns exist in time series data, it is useful to understand the meaning of a time series that lacks pattern. We explore such a time series in the next section.