Abstract
Building on the literature on regularization and dimension reduction methods, the paper presents a quarterly forecasting model for euro-area GDP. The pseudo-real-time nature of the information set is accounted for as the pattern of publication lags is explicitly considered. Forecast evaluation exercises show that predictions obtained through various dimension reduction methods outperform both the benchmark AR and the diffusion index model without preselected indicators. Moreover, forecast combination significantly reduces forecast error.
Similar content being viewed by others
Notes
Bayesian model averaging and selection (Koop and Potter 2004) could be an alternative to BM based on statistical procedures and not on ex ante selection. However, Mol et al. (2008) show that Bayesian shrinkage forecasts are highly correlated to the FM forecasts, and Stock and Watson (2012) find that shrinkage forecasts are inferior to the FM forecasts (more encouraging results using Bayesian methods have been recently reported in Bencivelli et al. 2012 and Carriero et al. (2015). Therefore, in this paper we will concentrate on the well established BM and its FM polar approach.
The pseudo-real time experiments mimic the actual situation faced by forecasters in terms of the schedule of data releases and therefore the availability of monthly indicators. Angelini et al. (2011), Banbura and Rünstler (2011), Drechsel and Maurin (2011), Giannone et al. (2009), Rünstler et al. (2009), Gayer et al. (2015) show that the depth in accounting for the timing of data releases plays a crucial role in forecast performance.
Although FM do not require as BM the ex ante selection of specific indicators, the researcher must—among the others—address the issues of estimating factors with either static or dynamic principal components (see Stock and Watson 2002a and Stock and Watson 2002b, henceforth SW, and Forni et al. 2005, FHLR). In any case, a key point is that all these approaches estimate the factors without any reference to the target variable.
For this, we will devote particular care to excluding composite leading and coincident indexes from our indicator data set, as they are affected by a spurious forecasting performance when using revised data (Diebold and Rudebusch 1991; Heij et al. 2011). In addition, we will also account for revisions related to the indicators statistical treatment before their use (check for outliers, forecast to balance the panel, quarterly aggregation and seasonal adjustment) in order to exactly mimic the steps adopted in the genuine out-of-sample forecasting practice.
Bulligan et al. (2010) and D’Agostino and Giannone (2012) found that SW and FHLR perform similarly and produce highly collinear forecasts, corroborating our choice of focusing on SW alone as the “representative method” of FM factor extraction in order to simplify result reporting. Boivin and Ng (2005) have shown that SW (i.e. our choice here) performs systematically better when more complicated, but realistic error structures are considered. Alvarez et al. (2012) have shown that FHLR results in similar problems as SW when the number of indicators in the data set is large.
Details about the panel of indicators are in Appendix A.1.
The inclusion of US figures aims at capturing the linkages related to the transmission of international business cycles; see, for example, Bodo et al. (2000).
Appendix A.1 describes the composition of the indicators’ data set and lists them together with the relevant metadata.
Appendix A.2 details the composition and calendar of the three vintages.
Another strategy with FM (but not with BM) would be the use of more specialized techniques rooted in the FM approach; see, for example, the shifting operator (by which all indicators with missing observations for the latest month are shifted in time so as to have a balanced panel) and the expectation-maximization algorithm of Stock and Watson (2002a). Other specialized techniques involve use of approximate Kalman filter models, of Markov switching dynamic factors, of nonparametric methods, of mixed-frequency VARs and of MIDAS regressions; see Camacho and Perez-Quiros (2010), Camacho et al. (2012), Ferrara et al. (2010), Giannone et al. (2008, (2009), Kuzin et al. (2011) and Clements and Galvão (2008).
Details are given in Appendix A.3. Financial variables are released daily and are always available for all the quarters to be forecast (i.e. they never need to be predicted). Although we refer to missing data extrapolation as if all indicators were at monthly intervals, the same steps are followed to predict quarterly indicators, when they are missing for the quarter to be forecast.
In detail, the total number of 148 rounds is obtained by accounting for the 3 different monthly vintages available for each quarter of the 49 rolling windows (the former is from 1990q1 to 2000q4 and the latter from 2002q1 to 2012q4) plus one additional vintage for the period 2002q2–2013q1.
Given a sample of \(N = 259\) indicators, we expanded it in order to account for the leading properties of indicators up to the fourth order. This choice delivers a panel of \(259 \times 5 = 1295\) variables which both enter the factor extraction in FM and the prescreening procedure in PFM.
It could be argued that the great recession of 2008–2009 favours more those models which exploit indicators compared to the AR benchmark as during deep recession episodes indicators embody such news in the nowcast. Therefore, forecasting ability over the AR benchmark may be overstated. However, our aim was to rank the forecasting performance of BM, PFM and FM, i.e. of models that exploit (albeit to a different extent) indicator information.
We have also computed other measures of forecasting accuracy, i.e. mean errors (ME) and mean absolute errors (MAE) that qualitatively give us the same outcomes as those with RMSE (these unreported results are available upon request).
Indicators composition and calendar are detailed in Appendix A.2.
Details are given in Appendix A.4.
These promising results from the use of soft-thresholding techniques are confirmed in Kapetanios et al. (2016) who present a similar application.
Note also that ortogonal regressors improve the ability of our automatic general-to-specific modelling procedure of discovering the better diffusion index model specification.
The procedure is sketched in Appendix A.5.
In fact, after the first training sample needed to estimate the quantiles of the distribution of picks’ frequency in the PFM approach, the available span of data here cannot allow for a second training sample to run ex ante the hierarchical forecast combination, in order to run fully ex ante the procedure to assess the forecasting ability.
Detailed outcomes are given in Appendix A.6.
References
Alvarez R, Camacho M, Pérez-Quirós G (2012) Finite sample performance of small versus large scale dynamic factor models. CEPR discussion paper, no 8867
Angelini E, Camba-Mendez G, Giannone D, Reichlin L, Rünstler G (2011) Short-term forecasts of euro-area GDP growth. Econom J 14:25–44
Antipa P, Barhoumi K, Brunhes-Lesage V, Darné O (2012) Nowcasting German GDP: a comparison of bridge and factor models. J Policy Model 34:864–878
Baffigi A, Golinelli R, Parigi G (2004) Bridge models to forecast the euro-area GDP. Int J Forecast 20(3):447–460
Bai J, Ng S (2008) Forecasting economic time series using targeted predictors. J Econom 146:304–317
Banbura M, Rünstler G (2011) A look into the factor model black box: publication lags and the role of hard and soft data in forecasting GDP. Int J Forecast 27:333–346
Banbura M, Giannone D, Modugno M, Reichlin L (2013) Now-casting and the real-time data flow. In: Elliott G, Timmermann A (eds) Handbook of economic forecasting, vol 2, Part A, Chapter 4, pp 195–237
Bencivelli L, Marcellino M, Moretti G (2012) Selecting predictors by Bayesian model averaging in bridge models. Bank of Italy, Temi di Discussione, no 872
Bernanke BS, Boivin J (2003) Monetary policy in a data-rich environment. J Monet Econ 50:525–546
Bessec M (2013) Short-term forecasting of French GDP: a dynamic factor model with targeted predictors. J Forecast 32:500–511
Bodo G, Golinelli R, Parigi G (2000) Forecasting industrial production in the euro area. Empir Econ 25:541–561
Boivin J, Ng S (2005) Understanding and comparing factor-based forecasts. Int J Cent Bank 1(3):117–151
Boivin J, Ng S (2006) Are more data always better for factor analysis? J Econom 132:169–194
Brunhes-Lesage V, Darné O (2012) Nowcasting the French index of industrial production: a comparison from bridge and factor models. Econ Model 29:2174–2182
Bulligan G, Golinelli R, Parigi G (2010) Forecasting monthly industrial production in real-time: from single equations to factor-based models. Empir Econ 39(2):303–336
Bulligan G, Marcellino M, Venditti F (2015) Forecasting economic activity with higher frequency targeted predictors. Int J Forecast 31:188–206
Caggiano G, Kapetanios G, Labhard V (2011) Are more data always better for factor analysis? Results for the Euro area, the six largest Euro area countries and the UK. J Forecast 30:736–752
Camacho M, Perez-Quiros G (2010) Introducing the euro-sting: short-term indicator of euro area growth. J Appl Econom 24:663–694
Camacho M, Perez-Quiros G, Poncela P (2012) Markov-switching dynamic factor models in real time. CEPR discussion paper, no 8866
Carriero A, Clark T, Marcellino M (2015) Real-time nowcasting with a Bayesian mixed frequency model with stochastic volatility. J R Stat Soc Ser A 178(4):837–862
Clements MP, Galvão AB (2008) Macroeconomic forecasting with mixed frequency data: forecasting US output growth. J Bus Econ Stat 26(4):546–554
Costantini M, Pappalardo C (2010) A hierarchical procedure for the combination of forecasts. Int J Forecast 26:725–743
Costantini M, Kunst RM (2011) Combining forecasts based on multiple encompassing tests in a macroeconomic core system. J Forecast 30(6):579–596
D’Agostino A, Giannone D (2012) Comparing alternative predictors based on large-panel factor models. Oxford Bull Econ Stat 74(2):306–326
De Mol C, Giannone D, Reichlin L (2008) Forecasting using a large number of predictors: is Bayesian shrinkage a valid alternative to principal components? J Econom 146:318–328
Diebold FX, Mariano RS (1995) Comparing predictive accuracy. J Bus Econ Stat 13:253–263
Diebold FX, Rudebusch GD (1991) Forecasting output with the composite leading index: a real time analysis. J Am Stat Assoc 86:603–610
Diron M (2008) Short-term forecasts of euro area real GDP growth: an assessment of real-time performance based on vintage data. J Forecast 27:371–390
Drechsel K, Maurin L (2011) Flow of conjunctural information and forecast of euro area economic activity. J Forecast 30:336–354
Ferrara L, Guegan D, Rakotomarolahy P (2010) GDP nowcasting with ragged-edge data: a semi-parametric modeling. J Forecast 29:18–199
Forni M, Hallin M, Lippi M, Reichlin L (2001) Coincident and leading indicators for the euro area. Econ J 111:62–85
Forni M, Hallin M, Lippi M, Reichlin L (2005) The generalized factor model: one sided estimation and forecasting. J Am Stat Assoc 100:830–840
Foroni C, Marcellino M (2014) A comparison of mixed frequency approaches for nowcasting Euro area macroeconomic aggregates. Int J Forecast 30:554–568
Foroni C, Marcellino M, Schumacher C (2015) Unrestricted mixed data sampling (MIDAS): MIDAS regressions with unrestricted lag polynomials. J R Stat Soc Ser A 178(1):57–82
Gayer C, Girardi A, Reuter A (2015) The role of survey data in nowcasting Euro area GDP growth. J Forecast. doi:10.1002/for.2383
Giannone D, Reichlin L, Small D (2008) Nowcasting: the real time informational content of macroeconomic data. J Monet Econ 55:665–676
Giannone D, Reichlin L, Simonelli S (2009) Nowcasting euro area economic activity in real time: the role of confidence indicators. Natl Inst Econ Rev 210(1):90–97
Golinelli R, Parigi G (2007) The use of monthly indicators to forecast quarterly GDP in the short run: an application to the G7 countries. J Forecast 26(2):77–94
Golinelli R, Parigi G (2008) Real-time squared: a real-time data-set for real-time GDP forecasting. Int J Forecast 24:368–385
Grassman P, Keereman F (2001) An indicator-based short-term forecast for quarterly GDP in the euro area. EC economic paper, no. 154, June
Harvey DI, Leybourne SJ, Newbold P (1997) Testing the equality of prediction mean squared errors. Int J Forecast 13:281–91
Heij C, van Dijk D, Groenen PJF (2011) Real-time macroeconomic forecasting with leading indicators: an empirical comparison. Int J Forecast 27:466–481
Hendry DF, Clements MP (2002) Pooling of forecasts. Econom J 5:1–26
Hendry DF, Hubrich K (2011) Combining disaggregate forecasts or combining disaggregate information to forecast an aggregate. J Bus Econ Stat 29(2):216–227
Huang H, Lee T-H (2010) To combine forecasts or to combine information? Econom Rev 29(5–6):534–570
Kapetanios G, Marcellino M, Papailias F (2016) Forecasting inflation and GDP growth using euristic optimisation of information criteria and variable reduction methods. Comput Stat Data Anal 100:369–382
Kaufmann S, Schumacher C (2013) Bayesian estimation of sparse dynamic factor models with order-independent identification. Study Center Gerzensee working paper, no. 13.04
Kim HH, Swanson NR (2014) Forecasting financial and macroeconomic variables using data reduction methods: new empirical evidence. J Econom 178(Part 2):352–367
Kinal T, Lahiri K (1983) Specification error analysis with stochastic regressors. Econometrica 51(4):1209–1219
Kinal T, Lahiri K (1984) A note on selection of regressors. Int Econ Rev 25(3):625–629
Kisinbay T (2010) The use of encompassing tests for forecast combinations. J Forecast 29(8):715–727
Klein LR (1990) Cyclical indicators in econometric models. In: Klein PA (ed) Analysing modern business cycles. M.E. Sharpe Inc., Armonk, pp 97–106
Klein LR, Ozmucur S (2008) The University of Pennsylvania models for high frequency macroeconomic modeling. In: Mariano RS, Tse Y (eds) Econometric forecasting and high-frequency data analysis, lecture notes series, National University of Singapore, vol 13, World Scientific, Singapore, pp 53–91
Klein LR, Sojo E (1989) Combinations of high and low frequency data in macroeconometric models. In: Klein LR, Marquez J (eds) Economics in theory and practice: an eclectic approach. Kluwer, Amsterdam, pp 3–16
Koop G, Potter S (2004) Forecasting in dynamic factor models using Bayesian model averaging. Econom J 7:550–565
Kuzin V, Marcellino M, Schumacher C (2011) MIDAS vs. mixed-frequency VAR: nowcasting GDP in the euro area. Int J Forecast 27:529–542
Kuzin V, Marcellino M, Schumacher C (2013) Pooling versus model selection for nowcasting GDP with many predictors: empirical evidence for six industrialized countries. J Appl Econom 28:392–411
Lahiri K, Monokroussos G (2013) Nowcasting US GDP: the role of ISM business surveys. Int J Forecast 29:644–658
Lahiri K, Yang L (2015) A further analysis of the conference board’s new Leading Economic Index. Int J Forecast 31:446–453
Lahiri K, Monokroussos G, Zhao Y (2015) Forecasting consumption: the role of consumer confidence in real time with many predictors. J Appl Econom. doi:10.1002/jae.2494
Marcellino M, Stock JH, Watson MW (2003) Macroeconomic forecasting in the euro area: country specific versus area wide information. Eur Econ Rev 47:1–18
McGuckin RH, Ozyldirim A, Zarnowitz V (2007) A more timely and useful index of leading indicators. J Bus Econ Stat 25(1):110–120
Ng S (2013) Variable selection in predictive regressions. In: Elliott G, Timmermann A (eds) Handbook of economic forecasting, vol 2, Chapter 14, pp 752–789
Rathjens P, Robins RP (1993) Forecasting quarterly data using monthly information. J Forecast 12:321–330
Rünstler G, Barhoumi K, Benk S, Cristadoro R, den Reijer A, Jakaitiene A, Jelonek P, Rua A, Ruth K, van Nieuwenhuyze C (2009) Short-term forecasting of GDP using large datasets: a pseudo real time forecast evaluation exercise. J Forecast 28:595–611
Rünstler G, Sedillot F (2003) Short-term estimates of euro area real GDP by means of monthly data. ECB working paper, no, 276, September
Schumacher C (2010) Factor forecasting using international targeted predictors: the case of German GDP. Econ Lett 107:95–98
Schumacher C, Breitung J (2008) Real-time forecasting of German GDP based on large factor model with monthly and quarterly data. Int J Forecast 24:386–398
Stock JH, Watson MW (2002a) Macroeconomic forecasting using diffusion indexes. J Bus Econ Stat 20:147–162
Stock JH, Watson MW (2002b) Forecasting using principal components from a large number of predictors. J Am Stat Assoc 97:1167–1179
Stock JH, Watson MW (2004) Combination forecasts of output growth in a seven-country data set. J Forecast 23:405–430
Stock JH, Watson MW (2006) Forecasting with many predictors. In: Elliott G, Granger CWJ, Timmermann A (eds) Handbook of economic forecasting, vol 1, Chapter 10, pp 515–554
Stock JH, Watson MW (2012) Generalized shrinkage methods for forecasting using many predictors. J Bus Econ Stat 30(4):147–162
Trehan B (1989) Forecasting growth in current quarter real GNP. Econ Rev Winter:39–52
Author information
Authors and Affiliations
Corresponding author
Electronic supplementary material
Below is the link to the electronic supplementary material.
Rights and permissions
About this article
Cite this article
Girardi, A., Golinelli, R. & Pappalardo, C. The role of indicator selection in nowcasting euro-area GDP in pseudo-real time. Empir Econ 53, 79–99 (2017). https://doi.org/10.1007/s00181-016-1151-z
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00181-016-1151-z
Keywords
- Euro-area GDP forecasts
- Bridge and factor models
- Indicators’ selection and prescreening
- Forecasting ability