Full Terms & Conditions of access and use can be found at
http://www.tandfonline.com/action/journalInformation?journalCode=ubes20
Download by: [Universitas Maritim Raja Ali Haji], [UNIVERSITAS MARITIM RAJA ALI HAJI
TANJUNGPINANG, KEPULAUAN RIAU] Date: 11 January 2016, At: 20:21
Journal of Business & Economic Statistics
ISSN: 0735-0015 (Print) 1537-2707 (Online) Journal homepage: http://www.tandfonline.com/loi/ubes20
Rejoinder
Lucia Alessi, Eric Ghysels, Luca Onorante, Richard Peach & Simon Potter
To cite this article: Lucia Alessi, Eric Ghysels, Luca Onorante, Richard Peach & Simon Potter (2014) Rejoinder, Journal of Business & Economic Statistics, 32:4, 514-515, DOI: 10.1080/07350015.2014.958920
To link to this article: http://dx.doi.org/10.1080/07350015.2014.958920
Published online: 28 Oct 2014.
Submit your article to this journal
Article views: 91
View related articles
514 Journal of Business & Economic Statistics, October 2014
Romer, C. D., and Romer, D. H. (2000), “Federal Reserve Information and the Behavior of Interest Rates,”The American Economic Review, 90, 429–457. [510]
Rossi, B. (2013), “Advances in Forecasting Under Model Instability,” in Hand-book of Economic Forecasting, eds. G. Elliottand A. Timmermann, (Vol. 2B), pp. 1203–1324, Amsterdam: Elsevier Publications. [513]
Rossi, B., and Sekhposyan, T. (2011), “Forecast Optimality Tests in the Presence of Instabilities,” mimeo, Duke University. [510]
——— (2013a), “Alternative Tests for Correct Specification of Conditional Forecast Densities,” mimeo, Universitat Pompeu Fabra and Texas A&M. [512,513]
——— (2013b), “Conditional Predictive Density Evaluation in the Presence of Instabilities,”Journal of Econometrics, 177, 199–212. [513]
West, K. D., and McCracken, M. W. (1998), “Regression-Based Tests of Predictive Ability,” International Economic Review, 39, 817– 840. [510]
Rejoinder
Lucia A
LESSIEuropean Central Bank, 60311 Frankfurt am Main, Germany (lucia.alessi@ecb.europa.eu)
Eric G
HYSELSDepartment of Finance, Kenan–Flagler Business School, and Department of Economics, University of North Carolina at Chapel Hill, Chapel Hill, NC 27599 (eghysels@unc.edu)
Luca O
NORANTEJoint Modeling Project, Central Bank of Ireland, Dublin, Ireland, and formerly, European Central Bank, 60311 Frankfurt am Main, Germany (Luca.Onorante@centralbank.ie.)
Richard P
EACHMacroeconomic and Monetary Studies Function, Federal Reserve Bank of New York, New York, NY 10045 (richard.peach@ny.frb.org)
Simon P
OTTERMarkets Group, Federal Reserve Bank of New York, New York, NY 10045 (simon.potter@ny.frb.org)
1. THE USEFULNESS OF HIGH-FREQUENCY FINANCIAL DATA
Almost all the discussants comment on our findings regarding the usefulness of high-frequency financial market indicators as possible sources of forecast improvements. Hubrich and Man-ganelli as well as Kenny refer to Banbura et al. (2013) who reported that daily financial data do not significantly improve the nowcasting performance of factor-based models for U.S. GDP growth. Beyond the methodological differences between MIDAS regression and the dynamic factor state-space model discussed in detail notably by Andreou, Ghysels, and Kourtel-los (2013), it is worth noting that Banbura et al. only used a handful of daily financial assets, namely the S&P 500 stock price index, short- and long-term interest rates, the effective exchange rate, and the price of oil. These series are typically not very informative when trying to retrieve information from financial markets. The discussion by Scotti indeed highlights the fact that series which are incorporated in stress indices, such as credit default swaps, LIBOR-OIS spreads, corporate bond credit spreads, CDS data, VIX, etc. (series also included in the analysis by Andreou, Ghysels, and Kourtellos (2013)) tell a different story as illustrated by the analysis of a small Bayesian VAR model she reports in her discussion. Along these lines, Scotti notes that it might be easier to construct informa-tive high-frequency indices and use these directly in a MIDAS regression, as opposed to the forecast combination approached
used in our article. This point is well taken, and in fact one of the interesting series, besides the stress index, is the ADS index con-structed by Boragan Aruoba, Frank Diebold, and Chiara Scotti (see Aruoba, Diebold, and Scotti (2009)). Andreou, Ghysels, and Kourtellos (2013) reported empirical findings which shed some light on this, namely they compare a setting where prin-cipal components are extracted from a panel of high-frequency data and subsequently used in a MIDAS regression versus a set-ting where single high-frequency ADL-MIDAS regressions are used with the same high-frequency data which are then subse-quently aggregated through forecast combinations. The results indicate that the differences are minor. Practically, of course it may be very useful to rely on some well-chosen index series, such as the stress index or the ADS series.
It is perhaps worth reiterating that we believe that the virtues of our approach can be summarized in one word: parsimony. The MIDAS regression-based framework incorporates addi-tional and timely information, and hedge against major mis-takes by using model averaging techniques, thereby circum-venting the curse of dimensionality. Each MIDAS regression focuses on a single high-frequency indicator, and uses a parsi-monious parameterization. In the second step, model averaging
© 2014American Statistical Association Journal of Business & Economic Statistics October 2014, Vol. 32, No. 4 DOI:10.1080/07350015.2014.958920
Alessi et al.: Rejoinder 515
selects the relevant indicators on the basis of performance. In comparison with large-dimension VARs and factor models this technique uses up less degrees of freedom and selects away more information—a fact that has been shown to be advanta-geous when many but potentially noisy indicators are available; see Koop and Onorante (2012) for a general discussion of this topic.
2. HOW WERE FORECASTS BEING USED AND WHICH FORECASTS WERE USED?
Hubrich and Manganelli note that it would have been helpful to the reader to know how forecasts were used. They highlight the traditional use of forecasts, the one of informing the de-cisions of the policymakers, and wonder whether other uses related to the coordination of expectations could be detected. Related to this is the question by Kenny on whether we had access to the forecasts which were the real thing, compared to the study by Stockton who carried out a study mandated by the Bank of England with unlimited access to the internal resources. While it is true that the study by Stockton notably included one-on-one interviews with staff members, something we did not do, it is fair to say that our analysis is at par with Stockton’s in terms of data access. We were fortunate to have the unique opportu-nity to work with this type of input from both the ECB and the FRBNY. At an abstract level, we know that central bankers may have asymmetric loss functions, a point elegantly formalized by Kilian and Manganelli (2008). Therefore, the optimal forecast for decision makers may not even be the unbiased prediction. The standard response of central bankers is: we look at every-thing (an ill-defined concept as Maurice Allais once noted by saying:tout est dans tout et inversement). What we know for sure is that the forecasts we analyzed were key inputs to the de-cisions made by policymakers. In this regard our study parallels that of Stockton.
3. DENSITY FORECASTS AND THEIR EVALUATION
Rossi highlights an impressive set of new tools in her dis-cussion. She is absolutely right that we could have gone much further and deeper in our analysis of the point forecasts as well as density forecasts. What is particularly useful in her approach is the fact that her analysis is specifically designed to appraise forecast performance whether it is point- or density forecasts when structural breaks or instabilities are present. While we only document crudely that the forecast performance
deterio-rated during the crisis, her discussion provides a clear illustration on what the test developed by Giacomini and Rossi (2009) and (2010) and Rossi and Sekhposyan (2011) and (2013) are capa-ble of. Her analysis highlights that the main conclusions of our article should be augmented by the observation that researchers at central banks should take note of these recent advances in the literature on forecast evaluation. The more widespread use of (1) multihorizon point forecasts, (2) density forecasts, (3) scenario-based analysis, etc., by the research staff of central banks should be accompanied by the type of analysis she suggests.
4. NONLINEARITIES AND UNCERTAINTY
Many of the discussants touch on issues related to either un-certainty in various forms such as the use of VSTOXX and accounting for model uncertainty through the combination pool analysis of Geweke and Amisano (2011) or the importance of nonlinearities due to regime switches as highlighted in the work of Hubrich and coauthors. We very much agree with their ob-servations, suggestions, and comments. The fact that we high-lighted in our article density forecasts, driven by projected sce-narios, as one of the areas of great progress, and the use of forecast combination methods and high-frequency data, were very much in line with these comments.
REFERENCES
Andreou, E., Ghysels, E., and Kourtellos, A. (2013), “Should Macroeconomic Forecasters Use Daily Financial Data and How?,”Journal of Business and Economic Statistics, 31, 240–251. [514]
Aruoba, S. B., Diebold, F. X., and Scotti, C. (2009), “Real-Time Measurement of Business Conditions,”Journal of Business and Economic Statistics, 27, 417–427. [514]
Banbura, M., Giannone, D., Modugno, M., and Reichlin, L. (2013), “Now-Casting and the Real-Time Data Flow,” inHandbook of Economic Forecast-ing(Vol. 2A), eds. G. Elliott, and A. Timmermann, Amsterdam: Elsevier, pp. 195–236. [514]
Geweke, J., and Amisano, G. (2011), “Optimal Prediction Pools,”Journal of Econometrics, 164, 130–141. [515]
Giacomini, R., and Rossi, B. (2009), “Detecting and Predicting Forecast Break-downs,”Review of Economic Studies, 76, 669–705. [515]
——— (2010), “Forecast Comparisons in Unstable Environments,”Journal of Applied Econometrics, 25, 595–620. [515]
Kilian, L., and Manganelli, S. (2008), “The Central Banker as a Risk Manager: Estimating the Federal Reserve’s Preferences Under Greenspan,”Journal of Money, Credit and Banking, 40, 1103–1129. [515]
Koop, G., and Onorante, L. (2012), “Estimating Phillips Curves in Turbulent Times Using the ECB’s Survey of Professional Forecasters,” Working Paper No. 1422, European Central Bank. [515]
Rossi, B., and Sekhposyan, T. (2011), “Forecast Optimality Tests in the Presence of Instabilities,” Working Paper 109, ERID. [515]
——— (2013), “Conditional Predictive Density Evaluation in the Presence of Instabilities,”Journal of Econometrics, 177, 199–212. [515]