The combination of tight credit conditions and fiscal-austerity measures that occurred in 2010 and 2011 put significant downward pressures on Eurozone GDP growth. Since the summer of 2011, the media started to talk about the possibility that the Eurozone could enter into a double-dip recession. Only a few months before the analysts were all looking for ‘green shoots’. According to Google trends, the term ‘green shoots’ reached a peak in the searches in the week of 10 May 2009, while the term ‘double-dip’ reached its peak in the week of 7 August 2011.
These Google searches reflect the need to capture turning points – a constant obsession for many economic analysts since the financial turbulence began in the summer of 2007. Obviously, the goal of forecasters is to be the first in calling these turning points while minimising the number of false signals, in order to make their announcements credible. In a recent paper by Hamilton (2010a) and his post on Vox (2010b) publications delays, data revisions, and parameter instability, among others, are singled out as the main factors that make it so difficult to detect the business cycle turning points in real time.
Incorporating solutions to all these difficulties, we develop in Camacho et al (2012a)1 an early-warning system for recessions to capture the ‘trembling’ of the Eurozone economy before anyone else notices it, as seismologists do with earthquakes. Notice that this is turning point (earthquake) detection, not turning point (earthquake) prediction.2
To develop a warning system, we need a set of indicators that, while capturing all the relevant information in the Eurozone economy, do not unnecessarily enlarge the information set by adding noise. In Camacho and Pérez-Quirós (2010), we show that the set of indicators used to compute the Eurozone Short Term Indicator of Growth (Euro-STING) captures the Eurozone economic developments in real time very accurately. The indicators include hard data such as GDP growth, employment, industrial production, industrial new orders, retail sales, and exports, and the timeliest soft data available in the Eurozone.
Using these key indicators, we aim at identifying turning points (the ‘trembling’ events in seismologist jargon) by using a statistical model based on a multivariate extension of original Markov-switching model proposed by Hamilton (1989), which addresses the typical problems that any analyst faces when analysing business conditions, such as publication delays, mixed frequencies, and data revisions. We show in Camacho et al (2012a) that this algorithm captures the Eurozone recessions in real time very precisely.
As Hamilton documents, this formal definition of trembling (recessions) provides a useful tool to the business cycle analyst for two reasons.
First, we abandon the impreciseness of the terms ‘green shoots’ and ‘double dips’ and we substitute them with recession probabilities. They are computed from the most recent data, which may include preliminary releases and a few figures made available early.
Second, the recession probabilities provide the users with a ‘sufficient statistic’ in the sense that they contain all the relevant information about the state of the economy. Thus, looking at the recession probabilities, the users have all the relevant information about the current state of the economy, with the subsequent saving of time and costs in monitoring the Eurozone business cycles.
Finally, the methodology is transparent, objective, and easily replicable.
Remarkably, our findings point out that the model exhibits a significant ability to track the CEPR Business Cycle Dating Committee chronology as captured in the state probabilities. In spite of its outstanding performance in dating the historical Eurozone business cycles, the primary usefulness of our model arises from its significant improvements over the Committee in the speed with which business cycle peaks and troughs are identified. This is not surprising since the Committee is more concerned with establishing the correct turning point dates than establishing them quickly and, therefore, peak and trough dates are often determined with a substantial time lag.
Using the model, we perform an actual real-time evaluation of the model which relies on data vintages that are constructed from the preliminary and partially revised data that were available at the time of the forecasts. Figure 1 shows the probabilities of recession that would have been inferred daily by an analyst who used the information available at the day of the forecast from 1 January 2008 to 31 December 2010. As we can see in the figure, we were already calling a recession in July 2008 (when the latest GDP growth number published was a healthy 0.73 for the first quarter of 2008), and we were already announcing the green shoots in late April 2009 (when a -2.55 growth rate for the first quarter of 2009 was going to be released on 15 May).
What are the mechanics behind these good signals that mark the changes in probabilities? When the probabilities of recession were still high at the beginning of April, the values of some soft indicators such as the Economic Sentiment Indicator of the European Commission and the Purchasing Managers’ Index of the Manufacturing Sector were 64.6 and 33.9, respectively. However, the following realisations since that date were 67.3 and 36.9, which implied significant improvements after several months of consecutive falls. In addition, the good news was confirmed when the hard indicators became available. Industrial production and industrial new orders increased from -0.46 and 0.10 in April to 1.64 and 1.42 in May, and sales and exports rose from -0.52 and -1.37 in May to 0.01 and 1.08 in June.
Therefore, the Markov-switching dynamic factor model had unequivocally signalled in April 2009 that the trough in the Eurozone had occurred and the potential users of our model would have had timely statistical evidence of this fact. The CEPR Committee waited until 4 October 2010 to announce that a trough in economic activity occurred in the second quarter of 2009.
Figure 1. Real-time recession probabilities 2008–09
Figure 2, which plots the real-time recession probabilities for each day in 2011, helps provide a statistical analysis of the potential double-dip recession in the Eurozone. Although the hard indicators exhibited negative growth in several months of 2011, the sharp decreases took place in September, when the annual growth rates of industrial production, new orders, sales, and exports became -2.00%, -6.35%, -0.62% and -1.01%, respectively. The figure also reflects the steadily decline in confidence indicators, whose falls were sharper in August 2011. On signs that weakness had started, the model increased the recession probabilities to above 0.5 and has been foreshadowing a downturn since the summer.
Notably, at the time of finishing the Camacho et al (2012a) paper, 30 November, the day before it was presented at the 8th International Institute of Forecasters workshop, somewhat less negative news had come from the soft indicators referring to the last months of 2011 which reduced the recession probabilities to 0.44.
Figure 2. Real-time recession probabilities in 2011
Although it is not shown in the figure, the latest sample (20 March 2012) shows further reduction in this probability to levels close to 0. The three-year long-term refinancing operations, the success of the private-sector involvement with Greece, and the progress made in areas of fiscal governance and the crisis resolution mechanism could be behind this increase in confidence. Therefore, it seems that the Eurozone faced a short-lived double-dip recession.
The Committee will wait until some of the uncertainty about the severity of the incoming shocks hitting the Eurozone is resolved. However, using the daily updated recession probabilities, which are calculated using our computationally simple algorithm, definitely facilitates the day-to-day monitoring of the economic developments in the Eurozone.
Camacho, M and G Pérez-Quirós (2010), “Introducing the EURO-STING: Short Term Indicator of euro Area Growth”, Journal of Applied Econometrics 25: 663–94.
Camacho, M, G Pérez-Quirós, and P Poncela (2012a), “Green shoots and double dips in the Eurozone. A real time measure”, CEPR Discussion Paper 8896.
Camacho, M, G Pérez-Quirós, and P Poncela, (2012b), “Extracting nonlinear signals from several economic indicators”, CEPR Discussion Paper 8865.
Camacho, M, G Pérez-Quirós, and P Poncela (2012c), “Markov-switching dynamic factor models in real time”, CEPR Discussion Paper 8866.
Hamilton, J (1989) “A new approach to the economic analysis of nonstationary time series and the business cycles”, Econometrica 57: 357–84.
Hamilton, J (2010a), “Calling recessions in real time”, International Journal of Forecasting 27: 1006–26.
Hamilton, J (2010b), “Calling recessions in real time”, VoxEU.org, 18 July.
Keilis-Borok, V, J Stock, A Soloviev, and P Mikhalev (2000) “Pre-recession pattern of six economic indicators in the US”, Journal of Forecasting 19: 65–80.
1 Based on the theoretical results of Camacho et al (2012b) and Camacho et al (2012c)
2 The use of methods developed in the earthquake literature is not new in economics. For instance, Keilis-Borok et al (2000) applied a parameterised pattern recognition algorithm, similar to one used for earthquake prediction, to the problem of predicting recessions.