Essays on macroeconomic nowcasting

The thesis contains three empirical essays focusing on macroeconomic nowcasting. Nowcasting has been used for a long time in meteorology. The term is a contraction between 'now' and 'forecasting' and in meteorology indicates weather forecasting on 'a very short term mesoscale period of up to 2 hours according to the World Meteorological Organization and up to six hours according to other authors in the eld'1 . It has become popular in Economics only recently, around a decade ago, when monitoring the current economic conditions in real time drew attention in the aftermath of the nancial crisis. This crucial activity of tracking in real time the behaviour of the macroeconomic variables at the beginning widespread among institutions and central banks in helping policy-making decisions, started to meet the academia when the nowcasting problem was rst formalized in a statistical model by Giannone et al. (2008). Then, Banbura et al. (2011) de ned the term 'nowcasting' as 'the prediction of the present, the very near future and the very recent past'. Nowadays, macroeconomists live in a more complex era, with a massive amount of data available essentially in real time. In this context, the release of the key measure of the economy activity that is the Gross Domestic Product (GDP) makes old even the very recent past. First, the economic agents would like to anticipate the economic growth as accurately as possible in a timely way. In addition, they would like to 'read' the signal among this ocean of data. Progress in technology helps in gathering and collecting vaster quantity of more granular data. On the other hand, the issue is to understand whether these novel data enhance the predicting ability of a model. Some of standard models has become outdated in the innovative framework so this call for new statistical tools to take the original challenge. The rst chapter shows a comparative analysis to assess the role of alternative indicators in nowcasting macroeconomic variables. We gather many series from other sources of data di erent from the o cial variables published by the National Institute of Statistics. We exploit some novel information such as credit cards payments, online data, administrative and scal gures building a comprehensive dataset for the Italian economy. We compare the nowcasting ability of a dynamic factor model (in the version of Banbura et al. (2011) and Banbura and Modugno (2014)), cast in a state-space form, estimated by EM algorithm via Kalman ltering and smoothing. The purpose of the rst chapter is to provide an additional contribution with respect to the body of literature assessing the role of the alternative indicators with respect to the o cial ones. The evaluation is conducted through a pseudo real-time forecast performances of the factor model against a simple autoregressive model. The second and third chapters enrich the debate on the proper strategy to address the issue of contemporaneous disaggregation in forecasting from a di erent angle. First, we focus on the very short-term prediction, or better, the 'very near future'. Second, we consider the issue in a data-rich environment. Third, we consider also di erent levels of disaggregation. These are arguments not so deeply investigated in the literature. The second chapter address the issue of nowcasting in a data-rich environment. We compare two methods (see Proietti et al. (2020)) that take the challenge of a big data analysis, a direct and indirect approach, to assess which one outperforms in nowcasting the Italian Gross Domestic Product. The contribution is an empirical comparison between a direct and a bottom-up approach in a high-dimensional framework. The focus of the third chapter is again the issue of contemporaneous disaggregation, but we address the question in a di erent way. We employ a di erent econometric strategy to setting up a bottom-up approach. We propose a bottom-up model using a block structure in a nowcasting model by Banbura et al. (2011) for the demand side components of the US GDP. Our model takes into account the hierarchical structure of the expenditure approach, building a multi-level factor model. We compare the proposed model with di erent speci cations providing that the one suggested turns out to be the best in the nowcast horizon. Moreover, we consider di erent level of disaggregation and at each level we compare the aggregation of the nowcasts and forecasts from the disaggregated models and the nowcasts and the forecasts of the aggregate. The applications of the rst two chapters exploit Italian data downloaded from Re nitiv Datastream. The dataset is a cleaned version of a database owned by the Directorate I of the Treasury Department of the Ministry of Economy and Finance, which I have contributed to populate and maintain. The alternative indicators are collected from di erent sources, as explained in details in Chapter 1. Data on credit card payments are kindly provided to the Directorate I by Nexi. Autostrade and Terna as well gentle provide the Directorate I with data on motorway ow of trucks and on electricity consumption and net production of energy. The US data used in the third chapter are downloaded from Haver Analytics during my visiting at Now-Casting Economics Ltd. The codes of this thesis are written in Matlab, Julia and Ox Metrics programming languages. For Chapter 1 and Chapter 3, the codes are based on the nowcasting model written in Julia language2 . The codes for the Chapter 2 are written in Ox metrics. These routines are based on the programs realized by Tommaso Proietti for the Eurostat project on "EuroMIND: the Monthly Indicator of Economic Activity in the Euro Area", that were rst coded in Ox 3.3 by Doornik (2001).