Aastveit, K. A., H. Bjørnland, and L. A. Thorsrud, 2014, “What Drives Oil Prices? Emerging Versus Developed Economies”, Journal of Applied Econometrics.
Alquist, R., L. Kilian, and R. J. Vigfusson, 2013, “Forecasting the Real Oil Price,” forthcoming in the Handbook of Economic Forecasting, 2, Amsterdam: North Holland.
Baumeister, C. and L. Kilian, 2012, “What Central Bankers Need to Know About Forecasting Oil Prices,” CEPR Discussion Papers 9118.
Baumeister, C. and L. Kilian,, 2013, “Forecasting the Real Oil Price in a Changing World: A Forecast Combination Approach,” CEPR Discussion Papers 9569.
Baumeister, C. and L. Kilian,, 2014, “Real-Time Forecasts of the Real Price of Oil,” IMF Economic Review, Palgrave Macmillan, vol. 62(1), pages 119–145, April.
Baumeister, C. and L. Kilian,, 2015, “Understanding the Decline in the price of Oil since June 2014”, forthcoming in the Journal of the Association of Environmental and Resource Economists.
Beidas-Strom, S., and A. Pescatori, 2014, “Oil Price Volatility and the Role of Speculation,” IMF Working Paper 14/218, International Monetary Fund.
Benes, J., M. Chauvet, O. Kamenik, M. Kumhof, D. Laxton, S. Mursula, and J. Selody, 2015, “The future of oil: Geology versus technology,” International Journal of Forecasting, Elsevier, vol. 31(1), pages 207–221.
Chen, Y-C., K. S. Rogoff, and B. Rossi, 2010, “Can Exchange Rates Forecast Commodity Prices?” Quarterly Journal of Economics 125(3), 1145–1194.
Diebold, F. X., and R. S. Mariano, 1995, “Comparing Predictive Accuracy,” Journal of Business & Economic Statistics, 13(3), 134–144.
Giacomini, R., and B. Rossi, 2010, “Forecast Comparisons in Unstable Environments,” Journal of Applied Econometrics, 25, 595–620.
Kilian, L., 2009, “Not all Oil Price Shocks Are Alike: Disentangling Demand and Supply Shocks in the Crude Oil Market,” American Economic Review, 99, 1053–1069.
Kilian, L. and D. P. Murphy, 2014, “The Role of Inventories and Speculative Trading in the Global Market for Crude Oil.” Journal of Applied Econometrics, 29: 454–478.
Mincer, J. A., and V. Zarnowitz, 1969, “The Evaluation of Economic Forecasts,” NBER Chapters in Economic Forecasts and Expectations: Analysis of Forecasting Behavior and Performance, 1–46.
Reichsfeld, D. A., and S. K. Roache, 2011, “Do Commodity Futures Help Forecast Spot Prices?” IMF Working Paper 11/254. International Monetary Fund.
Stock, J. H., and M. W. Watson, 2004, “Combination Forecasts of Output Growth in a Seven-Country Data Set,” Journal of Forecasting, 23, 405–430.
The Economist, 2015, “Dry-bulk cargo shipping: Hitting the bottom—Worse is still to come for many bulk carriers”, Oct 31st 2015, From the print edition http://www.economist.com/news/business/21677207-worse-still-come-many-bulk-carriers-hitting-bottom
We thank Christiane Baumeister for generously sharing the Baumeister and Kilian (2012) code, Daniel Rivera Greenwood for excellent research assistance, and Jörg Decressin, Hamid Faruqee, Thomas Helbling, Andrea Pescatori, Emil Stavrev, other colleagues at the IMF Research Department and participants at the George Washington University’s Seminar for Forecasting for helpful comments and advice.
U.S. refiners’ acquisition costs (RAC) is the average price paid by U.S. refiners for imported crude oil and includes transportation and other fees. See http://www.eia.gov/dnav/pet/PET_PRI_RAC2_DCU_NUS_M.htm
Throughout our paper, near- or medium-term refers to horizons up to 24 months. Short-term refers to horizons between 1 to 12 months. The long-term, i.e. forecasts in excess of 24 months, are not examined in our paper.
It is important to note that our paper relates to this class of forecasting models, i.e. short to medium term, defined as those horizons from 1 to 24 months only. For longer horizons, structural models are more common. See Benes et al. (2015) for one example of these longer horizon models.
No dummy is needed for December as the constant already captures that month, with other months’ constants adjusted relative to December’s.
Since the forecasting methodology is well-known, we refer the reader to standard time series textbooks such as Hamilton (1994) or Lütkepohl (2007) for further details on estimation and recursive forecasting.
Kilian and Murphy (2014) first introduced inventories into oil VAR models using U.S. oil inventories to extrapolate global inventories data. Kilian and Li (2014) obtain proprietary data to estimate OECD and non-OECD inventories, including oil in transit (i.e. floating storage and at sea). See www.IEA.org for more details.
When following the literature and specifying the real price of oil in logs, the largest estimated eigenvalue is 0.996, questioning stationarity of the model for our sample period. As a robustness check we run the model with the log-specification and find that the forecasting performance deteriorates.
Hotelling (1931) assumes a constant discount/interest rate, however, these are rarely constant over time.
For this reason (i.e., the stationarity of real oil prices), Baumeister and Kilian (2011) take log real oil prices rather than log-differences. However, we ran the models shown in Table 1 with log real oil prices but these performed worse than those with log differences and hence are excluded from the tables for parsimony. See also footnote 8.
In other words, we set the steady-state change in the equation of real oil prices to zero.
See Beidas-Strom and Pescatori (2014) and The Economist (2015) for a discussion of the performance of this index since the onset of global financial crisis during a period of overcapacity in the bulk shipping sector.
In this paper we report the findings for Brent spot price forecasts only, but the results hold equally for WTI.
Note that the choice of VAR model A(i) is illustrative since all specifications were checked and conformed to the reported findings.
These findings generally hold for all models shown A(i) to H(i). The futures forecast generally features the largest bias for all horizons, while the random walk and VAR bias are not insubstantial either. Model G(i) features a stronger bias than the random walk for short horizons up to six months. Model H(i) is strongly biased for medium term predictions beyond 18 months, with the bias exceeding that of futures. Detrending reduces the short-term forecast bias, yet induces a larger medium-term bias.
While for horizons between 7- 21 the results are not statistically significant at the 5 percent level, this is due to the large variance of the forecast errors. In these instances, the VAR does neither better nor worse than the random walk.
See footnote 10.
This is illustrated for model A(i). The same holds true for all ARMA-type models.
We also evaluate the performance of other samples splits of interest. For example, 1997M01-2002M12, 1997M01-2008M06, and 1997M01-2014M06. Results are available upon request.
See footnote 10.
For the actual Brent price collapse and its drivers see Box 1.1 of the April 2015 World Economic Outlook.