Demand Forecasting for Managers

Stephan Kolassa, Enno Siemsen

• Publisher: Business Expert Press
• Year: 2016
• ISBN:
  9781606495032
• Topic: Business
• Subtopic: Forecasting
• Index: Business

Most decisions and plans in a firm require a forecast. Not matching supply with demand can make or break any business, and that's why forecasting is so invaluable. Forecasting can appear as a frightening topic with many arcane equations to master. For this reason, the authors start out from the very basics and provide a non-technical overview of common forecasting techniques as well as organizational aspects of creating a robust forecasting process. The book also discusses how to measure forecast accuracy to hold people accountable and guide continuous improvement. This book does not require prior knowledge of higher mathematics, statistics, or operations research. It is designed to serve as a first introduction to the non-expert, such as a manager overseeing a forecasting group, or an MBA student who needs to be familiar with the broad outlines of forecasting without specializing in it.

PART I

Introduction

CHAPTER 1

Introduction

1.1. The Value of a Good Forecasting Process

It is common to become frustrated about forecasting. The necessary data is often dispersed throughout the organization. The algorithms used to analyze this data are often opaque. Those within the organization trained to understand the algorithms often do not understand the business, and those who breathe the business do not understand the algorithms. The actual forecast is then discussed in long and unproductive consensus meetings between diverse stakeholders with often conflicting incentives; in between, the forecast is often confused with goals, targets, and plans. The resulting consensus can be a political compromise that is far removed from any optimal use of information. These forecasts are in turn often ignored by decision makers, who instead come up with their own “guess” since they do not trust the forecast and the process that created it. Even if the forecasting process appears to work well, the actual, inherent demand uncertainty often creates numbers that are far away from the forecast. It is hard to maintain clarity in such a setting and not become frustrated by how hard it is to rely on forecasts.

Yet, what alternative do we have to preparing a forecast? The absence of a good forecasting process within an organization will only create worse parallel shadow processes. Every plan, after all, needs a forecast, whether that forecast is an actual number based on facts or just the gut feeling of a planner. Some companies can change their business model to a make-to-order system, eliminating the need to forecast demand and manufacture their products to stock, but this alternative model still requires ordering components and raw materials based on a forecast, as well as planning capacity and training the workforce according to an estimate of future demand. A central metric for every supply chain is how long it would take for all partners in the supply chain to move one unit—from the beginning to the end—into the market. This metric shows the total lead time in the supply chain. As long as customers are not willing to wait that long for a product, a supply chain cannot change to a complete make-to-order system. Someone in the supply chain will need to forecast and hold inventory. If that forecasting system does not work well, the resulting costs and disruptions will be felt throughout the supply chain.

One central tenet every manager involved in forecasting needs to accept is that there are no good or bad forecasts. There are only good or bad ways of creating or using forecasts. Forecasts should contain all the relevant information that is available to the organization and its supply chain about the market. Information is everything that reduces uncertainty. If a forecast is far away from the actual demand, but the process that generated the forecast made effective use of all available information, the organization simply had bad luck. Conversely, if a forecast is spot on, but the process that created it neglected important information, the organization was lucky but should consider improving their forecasting process. Bad forecasts in this sense can only be the result of bad forecasting processes. As with decision making under uncertainty in general, one should not question the quality of the decision or forecast itself given the actual outcome; one should only question the process that led to this decision or forecast. Betting money in roulette on the number 20 does not become a bad choice just because a different number is rolled—and neither does it become a better choice if the ball happens to actually land on the 20!

Different time series are more or less predictable, and if a series has a lot of unexplainable variation, there is a limit to how well it can be forecast. Figure 1.1 offers an example of two time series that are very different in terms of their forecastability. Importantly, while a good forecasting process will make time series more predictable by explaining some variation in the series, there are limits to the inherent predictability of such series. Repeated inaccurate forecasts can be a sign of a bad forecasting process, but they may also simply be a result of excessive noise in the underlying demand. The inherent forecastability of the series should thus be taken into account when judging the quality of a forecasting process. In this sense, a good forecasting process is not necessarily a process that makes a time series perfectly predictable but a process that improves the predictability of a series compared to simple forecasting methods.¹ We will discuss the concept of forecastability in more detail in Chapter 4.

Figure 1.1 Easy- and hard-to–forecast time series

From this perspective, one may be surprised to see how many organizations still exclusively rely on the use of point forecasts. A point forecast is a single number—an estimate of what an unknown quantity will most likely be. It is, however, highly unlikely that the actual number will be exactly equal to the point forecast. Thus, one always needs to think about and deal with the remaining uncertainty in the forecast. Ideally, one should conceptualize a forecast as a probability distribution. That distribution can have a center, which is usually equivalent to the point forecast. Yet that distribution also has a spread, representing the remaining uncertainty around the point forecast. Good forecasting processes will communicate this spread effectively; bad forecasting processes will remain silent on this issue, projecting unrealistic confidence in a single number. Further, not making explicit the inherent forecast uncertainty can lead to decision makers using both highly uncertain and highly certain forecasts in a similar way. It is not uncommon, for example, for firms to require equivalent safety stocks across different products, even though the uncertainty inherent in these products may vary vastly. The root cause of this problem often lies in insufficient reporting of uncertainty. We will further explore the idea of probabilistic forecasting in Chapters 2 and 3.

The effective design of forecasting processes seems difficult, but the benefits of getting the forecasting process right are tremendous. Fixing the forecasting process is a managerial challenge that usually does not require major financial investments. The challenge of improving the forecasting process often does not lie in the risks of investing into advanced machines or information technology or the costs of hiring more people and expanding the organization. Rather, the challenge is to manage cross-functional communication and push through change despite a multitude of stakeholders (Smith 2009). Yet, if these challenges are overcome, the returns can be huge. For example, Oliva and Watson (2009) document that the improvement of a forecasting process at an electronics manufacturing company led to a doubling of inventory turns and a decrease of 50 percent in on-hand inventory. Similarly, Clarke (2006) documents how the major overhaul of the forecasting process at Coca Cola Inc. led to a 25 percent reduction in days of inventory. These are supply chain improvements that would otherwise require significant investments into technology to achieve; if an organizational change (though challenging and time-consuming) of the forecasting process can achieve similar objectives, every manager should take the opportunity to improve forecasting processes seriously.

1.2. Software

While we often highlight the managerial aspects of forecasting in this book, we also delve into the statistics of forecasting. Our goal in doing so is to provide a basic intuition to managers as to how forecasting algorithms work—to shine some light into this black box. In this context, we emphasize that this book does not assume the use of any particular forecasting software. There is a large set to choose from when selecting a forecasting software, and a comprehensive review of the features, strengths, and weaknesses of all commercially available products is beyond the scope of this book. For an overview, interested readers may visit the OR/MS biannual survey of forecasting software (www.orms-today.org/surveys/FSS/fss-fr.html).

Throughout the book, we often provide a reference to functions in Microsoft Excel to help readers implement some ideas from the book. This spreadsheet modeling software is widely available, and most managers will have a copy installed on their laptops or tablets. However, Excel is known to suffer from inaccuracies, both in its statistical and optimization functions (Mélard 2014). Further, the standard functionality of Excel only allows for very limited time series analysis, and therefore the use of Excel for forecasting inevitably requires some coding and manual entry of formulas. It is very hard to maintain a consistent forecasting process in Excel, particularly when a company is growing. Spreadsheets start accumulating errors and undocumented changes over time (Singh 2013). When implemented correctly, spreadsheets have the advantage of being very transparent. Commercially available forecasting software, on the contrary, can often have a black-box character. As such, Excel is a good complementary tool for forecasting—to learn, to communicate, and to test out new ideas—but it should not become a standard tool for forecasting in an organization in the long run.

An important alternative is the free statistical software R (www.r-project.org/). While R is more difficult to learn and use than Excel, its functionality is much broader, and through user-written content, many existing forecasting methods are available for free in R (Kolassa and Hyndman 2010). Furthermore, interface add-ons like R-Studio (www.rstudio.com/) make the software more accessible, and excellent introductory books to R from a forecasting perspective are available (e.g., Shmueli and Lichtendahl 2015).

1.3. Key Takeaways

• Almost every business decision is about the future and is thus based on forecasts. We need forecasts. We cannot eliminate forecasts, but we can question whether we have an effective forecasting process that makes use of all available information within our organization and supply chain.

• Different time series will differ in terms of how hard they are to predict. Inaccurate forecasts may be a result of an ineffective forecasting process or may simply be due to the unpredictable nature of a particular business.

• No forecast is perfect. We need to directly confront, quantify, and manage the uncertainty surrounding our forecasts. Failure to communicate this uncertainty makes risk management associated with the forecast ineffective.

• Fixing the forecasting process can lead to huge improvements in the supply chain without major investments into technology. The challenge is to manage cross-functional communications and to overcome organizational silos and conflicting incentives.

______________

¹The simplest forecasting method is naïve forecasting, which means using the most recently observed demand to predict the future (also called demand chasing). Another simple method is using a long-run average of demand to predict the future (also called demand averaging). These methods can perform well. The terms simple or naïve are not meant to describe their accuracy but only relate to their simplicity.

CHAPTER 2

Choice Under Uncertainty

2.1. Forecasting Methods

A forecast is an input to support decision making under uncertainty. Forecasts are created by a statistical model and/or by human judgment. A statistical model is nothing but an algorithm, often embedded into a spreadsheet model or other software, which converts data into a forecast. Of course, choosing which algorithm an organization uses for forecasting and how this algorithm is implemented often requires the use of human judgment as well. However, when we examine the role of human judgment in forecasting, we mean that judgment is used in lieu of or in combination with a statistical forecast. Human judgment is thus the intuition and cognition that decision makers can employ to convert all available data and tacit information into a forecast. Both statistical models and human judgment will be discussed in much more detail in the rest of this book. In reality, most forecasting processes contain elements of both. A statistical forecast may serve as the basis of discussion, but this forecast is then revised in some form or other through human judgment to arrive at a consensus forecast, that is, a combination of different existing forecasts (statistical or judgmental) within the organization. A recent survey of professional forecasters found that while 16 percent of forecasters relied exclusively on human judgment and 29 percent depended exclusively on statistical methods, the remaining 55 percent used either a combination of judgment and statistical forecast or a judgmentally adjusted statistical forecast (Fildes and Petropoulos 2015). Another study of a major pharmaceutical company found that more than 50 percent of the forecasting experts answering the survey did not rely on the company’s statistical models when preparing their forecast (Boulaksil and Franses 2009). Forecasting in practice is thus not an automated process but is strongly influenced by the people involved in the process.¹

Throughout this book, we will refer to a forecasting method very generally as the process through which an organization generates a forecast. This forecast does not need to be the final consensus forecast, although the consensus forecast is generated by some method. The beauty of any forecasting method is that its accuracy can be judged ex post and compared against other methods. There is always an objective realization of demand that can and should be compared to create a picture of forecast accuracy over time. Of course, this comparison should never be based on small samples. However, if a forecasting method repeatedly creates forecasts that are far off from the actual demand realizations, this observation is evidence that the method does not work well, particularly if there also is evidence that another method works better. In other words, the question of whether a forecas...