Abstract
We show that when outcomes are difficult to forecast in the sense that forecast errors have a large common component that (a) optimal weights are not affected by this common component, and may well be far from equal to each other but (b) the relative mean square error loss from averaging over optimal combination can be small. Hence, researchers could well estimate combining weights that indicate that correlations could be exploited for better forecasts only to find that the difference in terms of loss is negligible. The results then provide an additional explanation for the commonly encountered practical situation of the averaging of forecasts being difficult to improve upon.
Similar content being viewed by others
Notes
For a book length exposition of these methods as well as the basic forecast combination problem, see Elliott and Timmermann (2016).
Alternatively, we could consider that forecasters do in fact observe the relevant information but have forecasts that differ from each other as they add in an additional random error. This would leave the following mathematical results unchanged, although it is perhaps a less defendable story for the differences amongst forecasts.
This was verified visually. Because forecast errors from this method display large forecast errors before this period, these errors are very influential observations in the regressions used to construct the weights.
Whilst not formally justified here, it is expected that this approximation is likely to be reasonable. First, so long as the optimal weights are not the average weights, the two combined forecast models will not have a degenerate spectral density at frequency zero (so issues of nested models do not arise). Second, we have used the same loss function (squared loss) in the forecast combination as the evaluation.
References
Alquist R, Kilian L (2010) What do we learn from the price of crude oil futures. J Appl Econom 25:539–573
Bates JM, Granger CWJ (1969) The combination of forecasts. Oper Res Q 20:451–468
Clemen RT (1989) Combining forecasts: a review and annotated bibliography. Int J Forecast 5:559–581
Clements MP, Hendry DF (2004) Pooling of forecasts. Econom J 7:1–31
Dana J, Dawes RM (2004) The superiority of alternatives to regression for social science predictions. J Educ Behav Stat 29(3):317–331
Davies A, Lahiri K (1995) A new framework for analyzing survey forecasts using three-dimensional panel data. J Econom 68(1):205–227
Diebold FX, Mariano RS (1995) Comparing predictive accuracy. J Bus Econ Stat 20(1):134–144
Elliott G (2010) Averaging and the optimal combination of forecasts (unpublished manuscript)
Elliott G, Timmermann A (2016) Economic forecasting. Princeton Unversity Press, Princeton
Faust J, Wright JH (2013) Forecasting inflation. Handb Econ Forecast 2(Part A):3–56
Granger CWJ, Ramanathan R (1984) Improved methods of combining forecasts. J Forecast 3:197–204
Ito T (1990) Foreign exchange rate expectations: micro survey data. Am Econ Rev 80:434–439
Lahiri K, Yang L (2015) A non-linear forecast combination procedure for binary outcomes, CESinfo Working paper No 5175
Lahiri K, Peng H, Zhao Y (2013) Machine learning and forecast combination in incomplete panels. Unpublished Manuscript, University of Albany, SUNY
Lahiri K, Peng H, Zhao Y (2015) Testing the value of probability forecasts for calibrated combining. Int J Forecast 31:113–129
Palm FC, Zellner A (1992) To combine or not to combine? Issues of combining forecasts. J Forecast 11(8):687–701
Patton AJ, Timmermann A (2012) Forecast rationality tests based on multi-horizon bounds. J Bus Econ Stat 30(1):1–17
Smith J, Wallis KF (2009) A simple explanation of the forecast combination puzzle. Oxf Bull Econ Stat 71(3):331–355
Stock JH, Watson M (2001) A comparison of linear and nonlinear univariate models for forecasting macroeconomic time series. In: Engle RF, White H (eds) Festschrift in honour of Clive Granger. Cambridge University Press, Cambridge, pp 1–44
Timmermann A (2006) Forecast combinations. In: Elliott Graham, Granger Clive WJ, Timmermann Allan (eds) Handbook of economic forecasting. Elsevier, Amsterdam, pp 135–196
Trueman B (1994) Analyst forecasting and herding behavior. Rev Financ Stud 7:97–124
Author information
Authors and Affiliations
Corresponding author
Appendix
Appendix
Proof
(Proposition 1) We generalize slightly from the proposition, where above \( b=1.\) From matrix inverse laws
By direct calculation, we have
and
so
yielding the result. \(\square \)
Proof
(Proposition 2). When the optimal weights are equivalent to the averaging weights, i.e. \((\iota ^{\prime }\Sigma \iota )^{-1}\iota ^{\prime }\Sigma =m^{-1}\iota ^{\prime }\), then \(rl=0\) and is independent of \(\sigma _{\varepsilon }^{2}.\) If not, we have by taking derivatives with respect to \( \sigma _{\varepsilon }^{2}\) that
and the equality sign only holds in the case we are in the situation of (a). \(\square \)
Rights and permissions
About this article
Cite this article
Elliott, G. Forecast combination when outcomes are difficult to predict. Empir Econ 53, 7–20 (2017). https://doi.org/10.1007/s00181-017-1253-2
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00181-017-1253-2