PDF Gallery...
This section is for links to our PDF files created by CPDF Participants, Team Leaders, and IIF Researchers involved in experienced-based forecasting applications.

Why Demand Forecasting is So Important to Supply Chain Professionals and Managers Hans Levenbach, Executive Director, CPDF Training and Certification Curriculum Many planning organizations still operate under the notion that a forecast is "just a number". Moreover, they often fail to recognize the essential difference between an unbiased demand forecast and a balanced supply/demand plan for a consumer-centric supply chain. This Chapter examines and challenges some of the common myths surrounding best-in-class practices for achieving improved forecasting performance. It will describe a pathway for making quantitatively- challenging processes more accessible and useful to supply chain practitioners. It outlines a four-tiered program involving (1) the 'big data' issues of data quality and analysis, (2) the 'predictive analytic' process for selecting statistical forecasting solutions, (3) the approach to forecast evaluation and performance measurement and (4) the reconciliation of models and demand forecasts to support an effective integrated business planning process.   Smarter Forecasting is Mostly about Data: Improving Data Quality through Data Exploration and Visualization Hans Levenbach, Executive Director, CPDF Training and Certification Curriculum Many demand planning and forecasting forecasting organizations summarize performance with accuracy measurements based on the oversimplified use of the arithmetic mean (as with the MAPE), not recognizing that such data rarely follow conventional statistical assumptions about the data generating process. Much data analysis in demand forecasting is informal and exploratory. However, it is important to realize that (1) An understanding of historical data in forecasting demand will be enhanced when we can identify key patterns in a time series, (2) Analyzing historical data is a key part of the demand forecasting process. For example, when data contain trends, contain seasonal patterns, or have unusual, hard-to-explain outliers, it may be inappropriate to use some of the more traditional and familiar forecasting methods, rather than statistical forecasting models. As in any statistical modeling process, you will find that (1) Exploratory Data Analysis (EDA) is open-ended and iterative in nature, (2) The steps may not always be clearly defined, (3) The nature of the process depends on what information is revealed at various stages. At any given stage, various possibilities may arise, some of which will need to be explored separately.   Predictive Analytics: Selecting Useful Forecasting Techniques Hans Levenbach, Executive Director, CPDF Training and Certification Curriculum One of the first things you will need when you start putting together a forecasting model is a listing of projection techniques. Whatever the technique, you need to start the selection process with * A statement of the forecasting problem in terms of the stages products and services are in their respective life cycles * A market theory stating what changes will affect demand for a company’s products and services * A gathering of data and market intelligence from field sales forecasters, market research studies, competitive analyses, and legacy data repositories (ERP) * A listing of plans for new products and special events or promotions throughout a life cycle. * Reconciling supplementary and collaborative approaches for combining forecasts into credible and defensible final demand forecasts In this White Paper, you will learn that the most common application of a forecasting technique involves some form of smoothing to reveal or highlight an important aspect of the data. Familiarity with the moving average helps to motivate the basic ideas behind exponential smoothing for agile forecasting. In addition, a moving average produces a simple projection that may be seen for its practical, intuitive appeal to motivate forecasting with State Space (exponential smoothing and ARIMA time series) forecasting models.   Taming Uncertainty: What You Need to Know About Measuring Forecast Accuracy Hans Levenbach, Executive Director, CPDF Training and Certification Curriculum This White Paper describes • Why it is a necessary to define first what a forecast error is • What bias and precision means for accuracy measurement • How, when and why to make accuracy measurements • The systematic steps in a forecast evaluation process. After reading this paper, you should be able to • Understand the difference between fitting errors and forecast errors • Recognize that there is no one best measure of accuracy for products, customers and hierarchies, industry forecasts and time horizons • Realize that simple averaging is not a best practice for summarizing accuracy measurements • Engage with potential users of demand forecasts to clearly define their forecast accuracy requirements.   The Myth of the Mape - and how to avoid it . . . Hans Levenbach, Executive Director, CPDF Training and Certification Curriculum Outliers in forecast errors and other sources of unusual data values should never be ignored in the accuracy measurement process. With the simplest measure of bias, for example, the calculation of the mean forecast error ME (the arithmetic mean of Actual (A) minus Forecast (F)) will drive the estimate towards the outlier. An otherwise unbiased pattern of performance can be distorted by just a single unusual value. The outlier-resistant measures introduced here operate to reduce their impact on the calculation of measures involving the arithmetic mean. This includes the Mean Absolute Error (MAE), the Mean Absolute Deviation from the Mean (MAD), and the Mean Absolute Percentage Error (MAPE), which is a commonly used measure of precision for reporting forecast accuracy. When we deal with forecast accuracy in practice, a demand forecaster typically reports averages of quantities based on forecast errors (squared errors, absolute errors, percentage errors, etc.). To properly interpret a measure of forecast accuracy, we must also be sensitive to the role of unusual values in these calculations.   Characterizing Demand Variability: Seasonality, Trend and the Uncertainty Factor Hans Levenbach, Executive Director, CPDF Training and Certification Curriculum Exploratory data analysis is basic to the demand forecasting process. In this White Paper, we focus on data that are ordered sequentially in time, called time series. We show you how to work with historical data in a variety of graphical ways that are useful in allowing you to see that: " Not only do data summaries and data visualizations help to explain historical patterns, but the requirements of an appropriate modeling strategy can also be visualized. " Deseasonalized, detrended, smoothed, transformed data (e.g., logarithms and square roots), fitted values, and residuals are most effectively visualized in graphical displays. " Nonstationarity, a basic concept of ARIMA time series modeling, is analyzed by plotting correlograms (empirical autocorrelation functions) of the original and differenced time series. .   Dealing With Seasonal Fluctuations Hans Levenbach, Executive Director, CPDF Training and Certification Curriculum This White Paper describes how seasonal effects can be removed and adjusted for in historical data. You will learn that " Many business time series show seasonal fluctuations " Businesses need to know when a change in a time series is due to more than the typical seasonal variation. In business and financial applications, many time series data may have been adjusted for seasonal fluctuations already " Government agencies and national banks adjust statistical indicators for seasonality before publishing economic data for the public, and developing econometric modeling for forecasting and policy analysis. There are generally three distinct uses of seasonal adjustment: the historical adjustment of available past data, the current adjustment of each new observation, and the predicted seasonal factors for future adjustment. The widely used X-12-ARIMA and more recent X-13ARIMA-SEATS programs from the U.S. Census Bureau are the industry standard for the large-scale analysis of publicly reported seasonal adjustments of monthly and quarterly data. These hybrid model-based and data-driven seasonal-adjustment procedures involve smoothing data to eliminate unwanted irregular variation from patterns that are meaningful to demand forecasters and planners.   Trend-Cycle Forecasting with Turning Points Hans Levenbach, Executive Director, CPDF Training and Certification Curriculum In previous White Papers, we described techniques that show how business time series reveal time-dependent patterns, primarily in terms of trend, seasonality and economic cycles. This White Paper describes " How serial correlation manifests itself in trend-cycle data patterns " How to detect and remove non-stationary trending behavior in time series " How to characterize the variability in trending patterns " How to create trend-cycle forecasts with a turning point forecasting (TPF) method We consider those characteristics of business time series that lack time dependence in typical behavior and variability. Random data are an example. Hence, it is important to analyze data that are said to be stationary.   Big Data: Baseline Forecasting With Exponential Smoothing Models Hans Levenbach, Executive Director, CPDF Training and Certification Curriculum This White Paper deals with the description, visualization and evaluation of statistical models that: " Are widely used in the areas of sales, inventory, logistics, and production planning as well as in quality control, process control, financial planning and marketing planning " Are especially suitable for large-scale, automated forecasting applications, because they require little forecaster intervention or parameter adjustments, thereby releasing the time of the demand forecaster to concentrate on the few problem cases " Are based on the extrapolation of past patterns with forecasting equations that are simple to update and maintain in a database " Can be described in terms of a modern modeling framework that provides uncertainty with prediction intervals and procedures for model selection " Capture level (a starting point for agile forecasting), trend (a factor for growth or decline) and seasonal factors (for adjustment of seasonal variation) in data patterns.   Short-Term Forecasting with ARIMA Models Hans Levenbach, Executive Director, CPDF Training and Certification Curriculum This White Paper is an exposition of how to use (mostly) graphical means for creating trend/seasonal demand forecasts to circumvent the cumbersome mathematical derivations in the Auto-Regressive Integrated Moving Average (ARIMA) modeling approach. The topics covered include: " What an ARIMA forecasting model is used for " Why stationarity is an important concept for ARIMA processes " How to select ARIMA models through an iterative three-stage procedure " How, when, and why we should use the Box Jenkins modeling methodology for ARIMA forecasting models " Its relationship to the new State Space Forecasting methodology The ARIMA models form the theoretical framework for " Expressing various forms of stationary (level) and nonstationary (mostly trending and seasonal) behavior in time series " Producing optimal forecasts for a time series from its own current and past values " Developing a practical and useful modeling process.   Demand Forecasting with Regression Models Hans Levenbach, Executive Director, CPDF Training and Certification Curriculum This White Paper describes (1) The concept of a linear regression model, (2) The assumptions needed for analyzing regression models for demand forecasting purposes, (3) The model building steps for ARIMA models, but used as well for regression modeling, (a)Identification - using the data to tentatively identify a model, (b)Estimation - fitting algorithms and inferences about the parameters, and (c)Diagnostic checking - adequacy requirements for evaluating the model, (4) The use of transformations to improve the validity of the regression modeling assumptions, and (5)Regression analysis as the principal method of causal demand forecasting.   Gaining Credibility through Root-Cause Analysis and Exception Handling Hans Levenbach, Executive Director, CPDF Training and Certification Curriculum This White Paper describes how regression models can (1) Provide more reliable forecasts (once the accuracy of the forecasts of the independent variables is assured), (2) Provide a forecast user with a model that is encompassing and explanatory, because more than one factor is taken into account, and (3) Be used to develop plans for agile forecasting where the model serves to express various alternative assumptions. In doing so, we learn (1) What role residuals play in validating modeling assumptions, (2) How resistant measure of correlation are used to test the validity of linearity in regression modeling, (3) Why nonconventional methods are so important in improving the robustness of regression models in forecasting, (4) To apply transformations to improve the validity of the modeling assumptions, and (5) Why forecast error patterns can give valuable insights into improving forecasting performance   The Final Forecast Numbers: Reconciling Change and Chance Hans Levenbach, Executive Director, CPDF Training and Certification Curriculum This White Paper describes (1) The systematic steps in the demand forecasting process, (2) How to assess the reliability of forecasts generated from statistical forecasting models, (3) How to prepare forecast scenarios and establish credibility for the numbers, (4) How and why establish forecasting standards for the forecasting process   Creating a Data Framework for Smarter Forecasting and Demand Management Hans Levenbach, Executive Director, CPDF Training and Certification Curriculum This White Paper describes describes (1) What the role of demand management is in the supply chain, (2) Why a demand forecasting data framework is essential to the success of an agile demand forecasting function, (3) How to identify the essential components of a forecast decision support system, and (4) When and how automatic forecasting should be used.   Blending Agile Forecasting with an Integrated Business Planning Process Hans Levenbach, Executive Director, CPDF Training and Certification Curriculum This White Paper describes (1) How management practices and processes are essential for an effective demand forecasting discipline, (2) Why forecasting agility comes from using statistical forecasting approaches, forecast error measurement, monitoring and forecast integration into the firm’s planning processes. After reading this book, you should be able to (1) Understand the nature of a demand forecast in a consumer-driven supply chain environment, (2) Recognize the components of an effective and efficient forecasting work cycle, (3) Engage with potential users of the forecast to help define, formulate, execute, evaluate and support their forecast data requirements, and (4) Provide advice on recommendations for forecast reviews, forecast data standards, model checklists, performance measurement activities, and business planning integration.   A Collaborative, Structured Approach to Demand Forecasting in the Supply Chain Hans Levenbach, Delphus, Inc. Annotated presentation at Logistics Summit & Expo (LATAM), Mexico City, Mexico, 2 April 2014.   Measuring Forecaster Performance in a Collaborative Setting with Field Sales Forecasters, Customers or Supplier Partners Hans Levenbach, Delphus, Inc. Measuring performance of forecasters is a complex task, especially with field sales forecasters, customers or collaborative partners as stakeholders in the final forecast.   15-PEERing into the Future: More Accurate and Reliable Forecasts Improve Inventory Planning Hans Levenbach, Executive Director, CPDF Training and Certification Curriculum More and more companies are discovering that good forecasting can lead to lower inventories and enhanced customer service. Over the years, these companies take a more granular approach to forecasting demand. This allows them to forecast at the SKU (stock keeping level) as well as by customer segments, like area, accounts, plants and distribution centers. Supply chains have become more geographically dispersed leading to new challenges in maintaining high customer service levels while achieving profit margin goals. As a result, the need for sound forecasting practices and efficient software tools have given demand planners added reasons to sharpen their tools, get additional training and become more pro-actively involved in inventory planning issues. In this talk we will describe how a four-step agile forecasting process guides demand planners to become key contributors to improved inventory practices. White Paper 15 was published by Global Contact after my presentation at the 36th EFCLIN Congress and Exhibition, Glasgow, Scotland, October 11, 2008.   Predicting the Demand for New Products and Services Mohsen Hamoudia, International Institute of Forecasters It is all about predicting the demand for new products and services based on limited data. What are the most useful approaches?   Ten Worst (and some Best) Demand Forecasting Practices That Impact Forecasting Performance Hans Levenbach, Delphus, Inc. Isolate worst practices while endorsing best practices. Maintain standards and checklists as an essential ingredient to enhancing professionalism in forecaster development while achieving greater efficiencies in your forcasting work cycles.   Demand Management - "Where Practice Meets Theory Elliott S. Mandelman, CPDF Team Leader Demand Management involves proactively executing the following functions: (1) The creation of a new demand forecast, (2) Overriding/adjusting the “system” forecast, (3) Achieving consensus across multiple groups, (4) The evaluation of past performance (error not accuracy), (5) Proper metrics that reflect reality, and (6) The communication of the demand plan for upstream or downstream use