High Dimensional Forecasting and its Pitfalls – M. Hashem Pesaran

Forecasting is central to macroeconomic policy making and risk management, and it is now routine to exploit data on a large number of variables to produce forecasts. In doing so, you can use a number of techniques, such as penalized regressions, principal components, neural networks, ensemble, Bayesian and other forecast averaging techniques. Yet, each method faces its own set of challenges.

This lecture focuses on some of the pitfalls of the two most popular forecasting methods - penalized regressions and factor augmented VARs – and explains how these can be overcome by exploiting the multi-dimensional interdependencies that exist across markets and countries. Using UK inflation and output growth figures to illustrate this, Dr. Pesaran expands on some of his previous work to explore why it is important to take a panel approach to forecasting and include all major world economies. Ultimately, by including these global factors proxied by weighted cross-section averages, we return to an approach that bears similarities to the GVAR modelling approach.

About the Speaker

Professor M. Hashem Pesaran is the John Elliot Distinguished Chair in Economics and Professor of Economics at USC Dornsife, and Director of the Center for Applied Financial Economics.  He is also an Emeritus Professor of Economics at Cambridge University, a Lifetime Fellow of Trinity College, Cambridge, a Fellow of the British Academy, the Econometric Society and the Journal of Econometrics. He was awarded Honorary Doctorates by the University of Salford in 1993, the University of Goethe, Frankfurt in 2008, the University of Maastricht in 2013 and the University of Economics in Prague in 2016. In September 2013 he was named as Thomson Reuters Citation Laureate in Economics. More recently he was named by Thomson Reuters as one of the World’s Most Influential Scientific Minds for 2014 and 2015.

He has served as consultant to the United Nations, the World Bank and the International Monetary Fund. Hashem is the founding editor of the Journal of Applied Econometrics, and a founding Board Member of the International Association for Applied Econometrics. His work has been published around 250 articles in academic journals and edited volumes, and 19 books and edited volumes. He received the Royal Economic Society Prize (1992), the Best Paper Award of Econometric Reviews (2004-2005), Best Paper Award of the International Journal of Forecasting (2007), and the Econometric Theory Award (2008)

About the Deane-Stone Lecture

The Deane-Stone lecture was established to commemorate the work that Sir Richard Stone (1913–1991) and Professor Phyllis Deane undertook during the 1940s into the development of national income accounting (NIA) systems of measurement.  Much of this work was published jointly by ourselves and the Department of Applied Economics in Cambridge. Professor Deane later left NIESR to join the Colonial Office where she remained until 1949.  She was a former president of the Royal Economic Society (1980-1982) and Professor of Economic History at Cambridge University.  Sir Richard Stone  was awarded the Nobel Prize in Economics in 1984 for this work.