Nuffield College Report 2008–2009 

David F. Hendry (Professorial Fellow) completed the research under his ESRC-financed award Automatic Tests for Super Exogeneity and Invariance with Jennifer Castle and Jurgen Doornik, and was on sabbatical leave throughout.  

The first, and main, aim was to advance a research program into automatic methods for empirical model discovery in econometrics, which jointly tackled all the key problems likely to be met in empirical modelling, since experience suggests that many features of empirical models cannot be derived from theoretical analyses. While the framework for any empirical analysis is determined by subject-matter theory, formulating a sufficiently general model to capture the salient data features requires an automatic algorithm to create lags, non-linear functions, indicators for breaks and data contamination, and determine exogeneity, then search for the relevant determinants despite the proliferation of candidate `explanations’. Empirical model discovery seeks to create, select, estimate and evaluate all the features jointly, and we have developed methods for doing so. Powerful developments in automatic methods have radically improved the success rates of such approaches, and although many of the key technical problems have been resolved, several crucial difficulties remained, so he focused on developing methods for handling multiple location shifts in economic time series using indicator saturation, which has underpinned a number of recent advances (jointly with Jennifer Castle and Jurgen Doornik).  

The second major aim of his research was to establish whether, and if so how, one could forecast successfully when economies are subject to sudden large unanticipated shifts, such as the recent financial crisis. One theme explored what the information requirements were for doing so; a second examined what impeded success in forecasting during breaks; a third investigated how to mitigate forecast  failure once an unanticipated break had happened, either by model transformations to improved robustness or by averaging across a range of models; a fourth sought to apply the implications of the research to nowcasting, namely `forecasting' the present data when there are delays in its release; and a fifth considered prediction markets, such as betting markets for the outcomes of sports competitions, elections etc. (jointly with Jennifer Castle, Nicholas Fawcett and James Reade).

A third aim concerned new tests for the correct specification of empirical econometric models, including (a) their functional form, developing two alternative ways of testing encompassing of logarithmic versus linear models when data are integrated (jointly with Aris Spanos and James Reade), and one more generally on testing non-linearity (jointly with Jennifer Castle); (b) a test of whether a model was changed when a new economic policy was implemented, by checking for a lack of invariance (jointly with Grayham Mizon); as well as (c) testing the specification of empirical models involving expectations of future variables (jointly with Jennifer Castle, Jurgen Doornik and Ragnar Nymoen).

A fourth aim was to participate in empirical research in cognate disciplines which face similar situations to time-series econometrics, including epidemiology and climate change, where the new tools of empirical model discovery promise to be very fruitful. He commenced joint research into modelling 400,000 years of ice core observations, and while ongoing, the preliminary results are promising (jointly with Katarina Juselius). 

Professor Hendry gave invited lecture series on ‘Econometric Modelling and Forecasting’ to the IMF, Washington, the Bank of Japan, and at Copenhagen University, as well as delivering invited papers at the National Bank of Denmark, the 50th Anniversary of the Ökonomisk Institut in Copenhagen, `Econometrics and Epidemiology’ Meeting at the Department of Health, CREATES Conference, Aarhus, `Forecasting and Prediction Markets’, Copenhagen, and `Econometrics and the World Economy', Fukuoka University, Japan. He presented a CREATES Distinguished Speaker Lecture, Aarhus, and gave a paper at the Svend Hylleberg Festschrift, Koldingfjord.

Forthcoming

Invited presentation, OxMetrics Conference, London, August, 2009

He was a member of the Advisory Panel on Foresight for the Chief Scientific Adviser to HM Government and of the International Advisory Boards of the National Centre for Econometric Research, Australia, and the School of Economics and Management, University of Lund. He was knighted in the Queen's Birthday Honours List in 2009 and received an Honorary Doctorate from Carlos III University, Madrid.

Publications 

(edited with M. Marcellino and G.E. Mizon) Encompassing, Oxford Bulletin of Economics and Statistics, 70, and `Editors Introduction', 715–719, 2008.

(with L. Ermini) `Log Income versus Linear Income: An Application of the Encompassing Principle', Oxford Bulletin of Economics and Statistics, 70, 807–827, 2008.  

(with A. Spanos and J.J. Reade) `Linear vs. Log-linear Unit Root Specification: An Application of Mis-specification Encompassing', Oxford Bulletin of Economics and Statistics, 70, 829–847, 2008. 

(with M.P. Clements) `Economic Forecasting in a Changing World' in Capitalism and Society, 2008. `The Methodology of Empirical Econometric Modeling: Applied Econometrics Through the Looking-Glass', 3–67 in T. Mills and K. Patterson (eds), Handbook of Econometrics, Vol. II, Applied Econometrics, Palgrave–Macmillan, 2009.  

(with M. Lu and G.E. Mizon) `Model Identification and Non-unique Structure', 343–364 in J.L. Castle and N. Shephard (eds), The Methodology and Practice of Econometrics, Oxford University Press, 2009.  

(with J.L. Castle) `The Long-Run Determinants of UK Wages, 1860–2004', Journal of Macroeconomics, 31, 5–28, 2009    

(with J.J. Reade)  `How should we make economic forecasts?', VoxEU, 11 June, 2009 http://www.voxeu.org/index.php?q=node/3647.