Seminar to be given at the University of Wollongong, 8 September 2010

Abstract: Functional time series are curves that are observed sequentially in time. In demography, such data arise as the curves formed by annual death rates as a function of age or annual fertility rates as a function of age. I will discuss methods for describing, modelling and forecasting such functional time series data. Challenges include:

• developing useful graphical tools (I will illustrate a functional version of the boxplot);
• dealing with outliers (e.g., death rates have outliers in years of wars or epidemics);
• cohort effects (how can we identify and allow for these in the forecasts);
• synergy between groups (e.g, we expect male and female mortality rates to evolve in a similar way in the future);
• deriving prediction intervals for forecasts;
• how to combine the mortality and fertility forecasts to obtain forecasts of the total population.

I will illustrate the ideas using data from Australia and France.

Slides (12Mb).

Talk to be given at the International Symposium on Forecasting, San Diego, 20–23 June 2010.

Slides

Presentation to College of Management, University of Fuzhou, China. 25 June 2009.

Slides

Keynote address, International Symposium on Forecasting, June 2009.

Abstract

Extremely bad data, extremely poor methods and extremely difficult problems will be used as the basis of some extremely useful lessons. I will describe three cases from my consulting experience and draw some general lessons that are widely applicable.

The first case involved forecasting passenger traffic on an Australian airline. The data showed variations due to school holidays, major sporting events, competitor activity, industrial disputes, changes in fare structures, and other factors. I will discuss the types of models that can be used to successfully forecast such data.

The Australian government subsidizes some pharmaceutical products, and requires forecasts of the likely expenditure on such products. After two consecutive years in which the expenditure was under-​​estimated by half a billion dollars, I was asked to review their forecasting procedures and recommend how to do it better. I will describe the results.

My third example involves forecasting the maximum electricity demand in any half hour period up to ten years in advance using only ten years of historical data. This seemingly impossible task was resolved so successfully, that the methods developed are now used as the basis of official forecasts for three states of Australia.

These three diverse examples will be used to draw some general conclusions about model complexity, structural change and forecast uncertainty.

Slides(18Mb)

Presentation to a meeting of Australian Deans and Directors of Graduate Studies, 1 May 2009.

Slides

Indian Institute of Management Calcutta. Melbourne, 18 July 2008.

Slides

R workshop. Melbourne, 29 June 2008.

There was an R workshop on 28–29 June, just before the Australian Statistical Conference. I put in an appearance on the second day.

Building R packages for Windows

• handout
• slides

R workshop. Melbourne, 29 June 2008.

There was an R workshop on 28–29 June, just before the Australian Statistical Conference. I put in an appearance on the second day.

Time series and forecasting in R

• handout
• slides

Australian Statistics Conference. Melbourne, July 2008.

• When: June 19–21, 2008
• Where: First International Workshop on Functional and Operatorial Statistics, Toulouse
• When : July 2008
• Where : Australian Statistical Conference Melbourne, Australia
• Speakers: Professor Rob J Hyndman, Monash University & Han Lin Shang, Monash University

Abstract: We propose some new tools for visualizing functional data and for identifying functional outliers. The proposed tools make use of robust principal component analysis, data depth and highest density regions. We compare the proposed outlier detection methods with the existing “functional depth” method, and show that our methods have better performance on identifying outliers in French male age-​​specific mortality data.

• Slides
• Paper

• When: 22–25 June 2008
• Where: International Symposium on Forecasting, Nice, France

Abstract: The most common forecasting methods in business are based on exponential smoothing and the most common time series in business are inherently non-​​negative. Therefore it is of interest to consider the properties of the potential stochastic models underlying exponential smoothing when applied to non-​​negative data. We explore nonlinear exponential smoothing state space models for non-​​negative data under various assumptions about the innovations, or error, process.

We discuss three problems with exponential smoothing state space models for non-​​negative data:

1. The forecasts and prediction intervals can be negative
2. The forecast distributions can have infinite variance
3. The processes can converge to zero almost surely.

We propose a new model with similar properties to exponential smoothing, but which does not have these problems, and we develop some distributional properties for our new model.

• Paper on which the talk is based.
• Slides