• Requisites for effective forecasting memory, error learning and belief in the data.
• Types of forecasting methods
  • Extrapolative methods
  • Explanatory methods
  • Judgmental methods
• Forecasting publications
• Forecasting conferences
• Forecasting software
• Useful websites
  

Index | Top

Module 2: Time Series Decomposition

Defines and illustrates the basic terms of time series forecasting: time plots, trends, seasonality, cyclical factors, special events and noise. Participant learns how to tell global from local trends, how to distinguish types of seasonality, how to calculate seasonal indexes and use them to deseasonalize data, and how to avoid extrapolating noise.

• The principle of decomposition
• Assessing trends
• Seasonal indexes and seasonal adjustment
• Leading indicators of the business cycle
• Tracking special events
• Recognizing noise in the data

Index | Top

Module 3: Forecasting Accuracy

The many do's and don'ts in measuring forecasting accuracy. The participant learns that measures of goodness of fit are not a reliable guide to forecasting accuracy, learns how to set up out-of-sample tests of forecasting accuracy and how to interpret statistical measures of forecast accuracy.

• Goodness of fit vs. forecast accuracy: Fit to the past is a risky guide to forecasting performance
• Within-sample vs. out-of-sample tests
• Rolling out of sample evaluations: A highly efficient procedure for assessing forecasting accuracy
• 3 Principle statistical measures of forecast accuracy: MAD, MAPE, and RAE
• Experiential vs. theoretical prediction intervals: which to trust?
• Designing an out-of-sample test: How much data to hold out, and how many test periods to use
• Forecasting competitions: M-Comp to the M3-Comp

Index | Top

Module 4: Introduction to Exponential Smoothing

Exponential smoothing is today the most widely practiced method of extrapolative forecasting. The participants learn how it works, why it is so versatile, whether automatic implementation of these models can be trusted, and how to deal with problem data.

• Why smoothing is so widely applied
• The family of smoothing model
• Weighting the data: weighting gives more emphasis to the recent than to the distant past
• Modes of implementation: manual, standard and automatic
• Choosing a smoothing model
• The big three: Simple, Holt and Winters' smoothing models
• Understanding and explaining the forecasts
• Damped and exponential trends
• Automatic model selection
• Dealing with intermittent series - Croston's approach
• Strengths and weaknesses of exponential smoothing

Index | Top

Module 5: Modeling Special Events

Many time series have been jolted by special events. Unless these are recognized and adjusted for, the forecasts may go awry. In this module, the participants learn how to recognize, code and account for special events.

• Types of special events: disruptions, holidays, promotions
• Consequences of ignoring a special event
• Identifying the timing of a special event
• Special-event adjustments to exponential smoothing models
• Event index: measuring the effect of a special event
• Multiple events and event interactions
• Using event adjustments to represent explanatory variables

Index | Top

Module 6: Forecasting for Product Hierarchies

Organizations frequently must forecast at different levels - company wide, by product or brand, and by item produced (sku). Multilevel (or hierarchical) forecasts must be made to reconcile. In this module, the participant sees many examples of product hierarchies, compares three major approaches to reconciliation - top down, bottom up and middle out - and learns how to make effective use of group level data to forecast individual sku's.

• Hierarchical forecasts and need for reconciliation
• SKU (stock-keeping unit) data: often nasty, brutish and short
• Three reconciliation strategies: bottom-up, top down and middle out
• Multilevel exponential smoothing
• Using good group data to forecast poor item data
• Special event adjustments in a product hierarchy

Index | Top

Module 7: Refresher on Statistical Testing

Over time, we tend to forget the logic behind and meaning of statistical significance tests; but these make an important contribution to the selection and evaluation of forecasting models. This brief module covers the essence - skipping technical details - of statistical testing, preparing the participant for the application of statistical test in ARIMA and Regression.

• Testing for Statistical Significance
• The Prob. Value
• The "t" ratio and "t" test
• Meaning of statistically significant differences

Index | Top

Module 8: Overview of Box - Jenkins Models (Arima)

Box and Jenkins models try to improve upon the method of exponential smoothing by considering a new information source. Called autocorrelation, and representing the tendency that series "in motion tend to remain in motion", its measurement opens the doors to a method that can be very useful for data that lack strong trends and seasonal components. This module is an overview, illustration the potential benefits of the Box-Jenkins approach as wll as the difficulties in its implementation and interpretation.

• What does ARIMA stand for?
• How does ARIMA differ from Exponential Smoothing?
• Creating lagged variables
• Autocorrelations: the key to understanding ARIMA
• Types of ARIMA models:
  • Autoregressive Models
  • Moving Average Models
  • Mixed Models
• Three step approach to model building:
  • Identification
  • Estimation
  • Validation
• Explaining the forecasts
  • Series 1. ARIMA at its best. Smoothing at its worst
  • Series 2. ARIMA and Smoothing: Too close to call
  • Series 3. ARIMA breaks down for short time series
• Strengths and weaknesses of ARIMA

Index | Top

Module 9: ARIMA Modeling

For those wishing to extend their knowledge of ARIMA to be able to make custom model identifications, more comprehensive validations and to better understand how ARIMA forecasts are made. This module relies heavily on graphical tools but introduces a limited amount of algebra.

• Autoregressive (AR) and Moving Average (MA) Terms
• Autocorrelations and Partial Autocorrelations
• Model Identification
• Non-stationary processes and differencing
• Automatic Model Identification
• Model Validation
• Deriving the forecasts
• ARIMA equivalents to Exponential Smoothing
• Extensions of ARIMA

Index | Top

Module 10: Regression Models for Forecasting

Regression is the basic tool for measuring the relationship between variables. If developed carefully, a regression model can also be a useful forecasting tool, but cannot be readily automated in the manner of extrapolative forecasting procedures. In this module, the participant will learn what a regression model attempts to accomplish, the challenges it must overcome, some procedures for determining elasticities, seasonal effects, and delayed impacts of explanatory variables, and how to interpret key regression results.

• Classical regression model
• Difficulties with regression models
• Interpreting results
• Regression coefficients
• The R-sq statistic
• "t" ratios and statistical significance
• Multicollinearity
• Log transformations
• Dummy variables and seasonality
• Dynamic (lagged) terms:
• Lagged explanatory variables
• Lagged dependent variable
• Lagged error terms
• Model building strategy

Index | Top

Module 11: Dynamic Regression Case Study

This module presents a guided tour from design to implementation of a regression model. The participant will learn the various tests a model must pass, options for forecasting the model's explanatory variables, and issues in evaluation the forecasting accuracy of the regression model. The module emphasizes what can and what cannot be trusted in the information supplied by forecasting software.

• Classical model is the starting point
• Checking for error autocorrelations
• Checking for lagged variables
• Introducing lagged variables
• Checking transformations
• Finding an acceptable model
• Interpreting coefficients and elasticities
• Checking goodness of fit (within sample)
• Checking forecasting accuracy (out-of-sample)
• Forecasting the explanatory variables
• Prediction intervals (PIs): standard
• PIs when explanatory variables are forecast

Index | Top

Module 12: Forecasting with Econometric Models

Many dozens of econometric models attempt to forecast the macro economy: income, price inflation, unemployment, etc. This module helps the business analyst to understand the nature of econometric forecasts and better judge the value of econometric forecasts to the organization.

• Who makes econometric forecasts?
• Characteristics of econometric models:
  • Multi-equations
  • Feedback effects
  • Simultaneous equations estimation
• Applications of econometric models:
  • Unconditional forecasting
  • Policy simulations
  • Linkage equations
• Use of judgment in econometric forecasts
• Accuracy of econometric forecasts

Index | Top

Module 13: Combining and Consensus Forecasts

Hundreds of published articles show how and when the forecaster can benefit from combining forecasts from different methods. This brief module highlights the principles of combining forecasts. It also looks at consensus forecasts and their track record.

• Motivations for combining forecasts
• How to combine forecasts: unweighted and weighted averages
• Does combining improve accuracy?
• Consensus forecasts
• Groupthink

Index | Top