Can macroeconomists get rich nowcasting economic turning points?
Thomas Ra not
Exclusive: York Capital to wind down European funds, spin out Asian funds
York Capital Management has decided to focus on longer-duration assets like private equity, private debt and collateralized loan obligations. The firm also plans to wind down its European hedge funds and spin out its Asian fund. Q3 2020 hedge fund letters, conferences and more York announces structural and operational changes York Chairman and CEO Jamie Read More
This paper aims at nowcasting economic cyclical turning points in real time to get useful signals for policymakers but and for investors. Although the literature has mostly focused on the business cycle,Anas and Ferrara (2004) emphasize the relevance for policymakers of the growth cycle, which seeks to represent the uctuations around the trend. Furthermore, Ra not (2014) highlights the greatest interest of the growth cycle for euro and dollar-based investors. To nowcast growth cycle turning points, probabilistic indicators are created from a simple machine-learning algorithm known as Learning Vector Quan- tization (LVQ), introduced in economics by Giusto and Piger (2014). The real-time ability of the indicators to quickly and accurately detect growth cycle turning points in the United States and in the euro area is gauged. To asses the value of the indicators, pro t maximization measures based on trading strategies are employed in addition to more standard criteria. A substantial improvement in pro t measures over the benchmark is found: macroeconomists can get rich nowcasting economic turning points.
Forecasting economic turning points in real time is a notorious di cult task (see Berge (2013) or Katayama (2010)), but ex-post identi cation of turn- ing points is also challenging: economists fail to detect if a new economic phase has already begun. For instance, in a survey in March 2001, 95% of American economists said there would not be a recession, even though one had already started. That being said, governments and central banks are naturally sensitive to indicators showing signs of deterioration in growth to allow them to adjust their policies su ciently in advance. In this respect, timing is important and the earlier the signal, the better.
A signi cant literature focuses on the business cycle detection (see Hamil- ton (2010)). The business cycle is meant to reproduce the cycle of the global level of activity of a country1. However, Anas and Ferrara (2004) point out the importance for policymakers of the growth cycle, de ned as the devia- tion of the real GDP to the long-term trend2. The growth cycle seeks thus to represent the uctuations around the trend. In particular, they highlight the progressive follow-upof the cyclical movement: all recessions involve slow- downs, but not all slowdowns involve recessions. Growth cycles are thus more numerous than business cycles. Furthermore, Ra not (2014) emphasizes the greatest interest of the growth cycle for euro and dollar-based investors.
One stylised fact of economic cycles is the non-linearity: the behaviour of the series describing the cycle depends on the phase in which it evolves. Real- time regime classi cation and turning points detection require thus methods capable of taking into account the non-linearity of the cycles. In this respect, manynon-linear parametric models have been proposed, such as smooth- transition autoregressive models (see Ferrara and Guegan (2005)), non-linear probit models (see Liu and Moench (2014)) or Markov switching models (see Guidolin (2011)). Recently, Giusto and Piger (2014) introduce a simple machine-learningalgorithm known as Learning Vector Quantization (LVQ), which appears very competitive with commonly used alternatives.
In this article, probabilistic indicators are created from Learning Vector Quantization to nowcast growth cycle turning points in real time. The aim is to get useful signals for policymakers and for investors. The rest of the pa- per proceeds as follows. Section 1 introduces Learning Vector Quantization. Section 2 describes the data, the model selection and the evaluation of the forecasts. Section 3 analysis the empirical results.