Central Bank Macroeconomic Forecasting during the Global Financial Crisis: The European Central Bank and Federal Reserve Bank of New York Experiences

Abstract

Standard-Nutzungsbedingungen:Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden.Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen.Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may AbstractThis paper documents macroeconomic forecasting during the global financial crisis by two key central banks: the European Central Bank and the Federal Reserve Bank of New York. The paper is the result of a collaborative effort between staff at the two institutions, allowing us to study the time-stamped forecasts as they were made throughout the crisis. The analysis does not exclusively focuses on point forecast performance. It also examines methodological contributions, including how financial market data could have been incorporated into the forecasting process.Keywords: Forecast evaluation; Mixed frequency data sampling. However, our analysis also goes beyond the traditional forecast evaluation exercises and discusses methodological advances which most likely will shape the future of the macroeconomic forecasting process at central banks. Since the main trigger of the crisis was a meltdown of the US subprime mortgage market, we pay special attention to whether financial market signals were fully accounted for in the central banks forecasting process. Considering that the crisis was first and foremost a crisis caused by financial market turmoil, we address the question of whether central banks could have done a better job at reading financial market signals.The issue requires one to think about mixed frequency data since financial market data are intrinsically high frequency (in our case daily) whereas the forecasts pertain to low frequency macroeconomic phenomena such as GDP growth. We propose to use MIDAS regression models. We find that simple MIDAS regressions using financial series could have improved the forecast accuracy both in the U.S. and in the euro area, and provide some evidence that the usefulness of financial series increased during the crisis.We also describe some new conceptual and methodological frameworks for prediction and ECB Working Paper 1688, July 2014 2 policy guidance which grew out of the challenges faced in particular at the FRBNY. We identify two key ingredients: (1) the emphasis on what might be called scenario-driven forecasting schemes, and (2) the recognition that one should pay attention to distributional features beyond point forecasts, in line with general notions of macroeconomic risk.Overall, we reach the following conclusions.• First, macroeconomic forecasting during the global financial crisis was a challenging task faced by the Federal Reserve Bank of New York and the...

UK Regional Nowcasting Using a Mixed Frequency Vector Auto-Regressive Model with Entropic Tilting

“…Increasingly, macroeconomic forecasters want to move beyond unconditional forecasts to incorporate extra information or restrictions on their forecasts (see, among many others, Alessi et al (2014) and Krüger et al (2017)) such as our cross-sectional restriction. Conditional forecasting and entropic tilting are two ways of doing this.…”

Financial nowcasts and their usefulness in macroeconomic forecasting

“…36 At each forecast origin, we iteratively generate 1-through 8-step-ahead forecasts for 35 When averaging across forecasts in the MIDAS case, the combination weights are determined by the discounted mean square forecast error (MSFE) method, as in Andreou et al (2013). We also ran our MIDAS exercises using forecast combination weights determined by a simple arithmetic average (as in Alessi et al 2014), BIC criteria, and AIC criteria; all three yielded slightly worse forecast accuracy compared with the MSFE method. 36 For comparability, both the bivariate BVAR model and the MIDAS model use matched real-time information sets; however, for simplicity within a given quarter, we use the real-time macroeconomic data that would have been available at the time of the SPF survey for that quarter, although the financial data differ depending on the case.…”

“…The estimation of empirical macroeconomic models at a quarterly frequency with financial variables has a long tradition going back at least to Mitchell and Burns (1938), and the 2008 financial crisis has rekindled interest in the topic. In particular, Alessi et al (2014) emphasize that the failure of forecasters to predict the financial crisis and their subsequent poor forecasting record was due in large part to inadequate modeling of the relationship between quarterly macroeconomic variables and daily financial variables, given that the latter are inherently forward-looking. They show that, by using a MIDAS framework that links quarterly real GDP to timely intra-quarterly daily financial data as inspired by Andreou et al (2013), one could improve upon the accuracy of real GDP growth forecasts and importantly could have seen the recession coming, albeit with limited advance warning.…”

Extracting information shocks from the Bank of England inflation density forecasts

“…A second area in which this paper contributes to the literature is in the evaluation of the forecasting practice of the Bank of England. As Alessi, Ghysels, Onorante, Peach, and Potter (2014) stress, it is important to assess the forecast accuracy of Central Banks' forecasting ability in the light of the recent financial crisis. There is already a vast literature on the evaluation of the Bank of England's point and density forecasts.…”