ECB European Central Bank : Nowcasting business cycle turning points with stock networks and machine learning

Andres Azqueta-Gavaldon,

Dominik Hirschbühl, Luca Onorante,

Lorena Saiz

Working Paper Series

Nowcasting business cycle turning points with stock networks and machine learning

No 2494 / November 2020

Disclaimer: This paper should not be reported as representing the views of the European Central Bank (ECB). The views expressed are those of the authors and do not necessarily reflect those of the ECB.

Abstract

We propose a granular framework that makes use of advanced statistical methods to approximate developments in economy-wide expected corporate earnings. In particular, we evaluate the dynamic network structure of stock returns in the United States as a proxy for the transmission of shocks through the economy and identify node positions (rms) whose connectedness provides a signal for economic growth. The nowcasting exercise, with both the in-sample and the out-of-sample consistent feature selection, highlights which rms are contemporaneously exposed to aggregate downturns and provides a more complete narrative than is usually provided by more aggregate data. The two-state model for predicting periods of negative growth can remarkably well predict future states by using information derived from the node-positions of manufacturing, transportation and nancial (particularly insurance) rms. The three-states model, which identies high, low and negative growth, successfully predicts economic regimes by making use of information from the nancial, insurance, and retail sectors.

JEL Classication: C45; C51; D85; E32; N1.

Keywords: real-time; turning point prediction; Granger-causality networks; early warning signal.

ECB Working Paper Series No 2494 / November 2020

1

Non-technical summary

A real-time identication of recessions and economic regimes" (i.e. phases of the business cycle) proves dicult, but it is of great importance for policy-makers. We propose a novel approach for identifying changes in economic regimes in real time. Our approach utilises various measures of connectedness in a stock returns network as a proxy for the transmission of shocks through the economy. We show that the dynamic network structure contains forward-looking cyclical information and is a good predictor of booms and reces- sions.

We emphasise the role of granular, rm-level information, and its connectedness to unveil aggregate shocks. When consumer and business spending softens in some part of the economy, corporate earnings start to decline and businesses seek ways to cut costs. They might downsize their workforce, put a freeze on hiring and delay investments. As these factors become more widespread, unemployment rises, aggregate wages fall and demand weakens further, tipping the economy into recession. Once demand for goods and services increases again, earnings rise supporting higher stock prices. Although every recession is dierent, there is a clear link between the state of the economy, corporate earnings and stock price developments. Our dynamic stock network captures the connectedness of shocks to expected corporate earnings and hence allows us to detect the propagation of shocks within the economic system.

The main ndings can be summarised as follows. The baseline binary state model can predict upcoming recessions by using information from the node-positions of manufacturing, transportation and nancial, particularly insurance rms. The ternary state model (featur- ing high, low growth, and recessions) successfully predicts economic regimes highlighting the role of information stemming from the nancial, the insurance, and the retail sectors. Looking at the economic system as a whole, we highlight that during an expansion adverse shocks to rms are mostly idiosyncratic, while during contractions shocks become more widespread resulting in higher connectedness in certain parts of the network. Measures of centrality eciently summarise economy-wide developments and allow us to monitor the state of the economy in real time.

ECB Working Paper Series No 2494 / November 2020

2

• Introduction

A real-time identication of recessions and economic regimes" (i.e. phases of the business cycle) is of great importance for policy-makers. We propose a novel approach for identifying changes in economic regimes in real time. Our approach utilises measures of connectedness in a stock network as a proxy for the transmission of shocks through the economy and, hence, for widespread developments in corporate earnings. We show that the dynamic network structure contains forward-looking cyclical infor- mation, making it an optimal predictor of booms and recessions.

The identication of turning points or dierent phases of the business cycle in real time has proven rather dicult, despite a large literature on the subject and an ever increasing number of leading indicators. In this respect, the consensus in the literature is that the set of relevant indicators for predicting shifts in business cycle phases (e.g. recessions) changes over time. Therefore, indicators that are useful for predicting one recession do not necessarily serve to predict other recessions as every cycle is dierent". A general problem of forecasting economic regimes as shown by Ng (2014) is that only a few indicators are actually useful and the eectiveness of these indicators varies according to the forecast horizon. For example, nancial variables (e.g. term and corporate spreads) are appropriate for forecasting 6 to 12 months ahead in view of their forward-looking nature.1

One important reason for the delay in identifying shifts in business cycle phases is that the overall economy is normally reduced to a summation of its components and micro-level shocks are therefore assumed to oset each other at the aggregate level. By contrast, the approach adopted in this work emphasises the importance of granularity and the role of economy-wide return connectedness. When consumer and business spending softens in some parts of the economy, corporate earnings start to decline and businesses seek ways to cut costs. They might downsize their work- force, put a freeze on hiring and delay investments. As this situation becomes more widespread, unemployment rises, aggregate wages fall and demand weakens further,

• In our network approach, we focus on nowcasting. Investigating the forward-looking information content in the derived network measures is left for future work.

ECB Working Paper Series No 2494 / November 2020

3