Risk and Return Spillovers in a Global Model of the Foreign Exchange Network

South African Reserve Bank

Working Paper Series

WP/21/15

Risk and Return Spillovers in a Global Model of the Foreign Exchange Network

Matthew Greenwood-Nimmo, Daan Steenkamp and Rossouw van Jaarsveld

Authorised for distribution by Konstantin Makrelov

4 August 2021

© South African Reserve Bank

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means without fully acknowledging the author(s) and this Working Paper as the source.

South African Reserve Bank Working Papers are written by staff members of the South African Reserve Bank and, on occasion, by consultants under the auspices of the South African Reserve Bank. The papers deal with topical issues and describe preliminary research findings, and develop new analytical or empirical approaches in their analyses. They are solely intended to elicit comments and stimulate debate.

The views expressed in this Working Paper are those of the author(s) and do not necessarily represent those of the South African Reserve Bank or South African Reserve Bank policy. While every precaution is taken to ensure the accuracy of information, the South African Reserve Bank shall not be liable to any person for inaccurate information, omissions or opinions contained herein.

South African Reserve Bank Working Papers are externally refereed.

Information on South African Reserve Bank Working Papers can be found at https://www.resbank.co.za/en/home/publications/Papers/working-papers

Enquiries relating to the Working Paper Series can be addressed to:

Head: Economic Research Department

South African Reserve Bank

P O Box 427

Pretoria 0001

Tel. [phone]

Risk and Return Spillovers in a Global Model of the Foreign

Exchange Network

Matthew Greenwood-Nimmo*, Daan Steenkamp† and Rossouw van Jaarsveld‡

4 August 2021

Abstract

We developed a network model to capture the dynamic interactions among foreign exchange returns and realised risk measures among 20 developed and emerging market currencies, including the rand (ZAR). We demonstrate how this framework can be used to assess the sensitivity of a given currency to shocks from other currencies and to provide narratives con- textualising currency movements, focusing on the ZAR. We show that variations in the risk- return profile of the USDZAR correlate with variations in the risk-return profile of many other currencies, and that this is especially notable with respect to emerging market currencies. We interpret this as evidence of the ZAR's role as a bellwether emerging market currency. We show that the model is able to highlight risk transmission channels in a timely manner during foreign exchange flash crashes and periods of heightened financial market uncertainty.

JEL classification: F31, G01, G15

Keywords: Foreign exchange markets; Higher-order moment risk; Realised moments; Network modelling; Spillovers.

• Department of Economics, University of Melbourne and Centre for Applied Macroeconomic Analysis,

Australian National University. Email: [e-mail].

• Corresponding author. South African Reserve Bank, PO Box 427, Pretoria, South Africa, 0001. Email:

[e-mail].

‡SARB. Email: Rossouw.VanJaarsveld.co.za.

1

1. Introduction1

There has long been concern among policymakers about the relatively high volatility of the South African rand (ZAR) (see Loewald 2021). This reflects concern that exchange rate volatility imposes costs on the economy by interfering with investment and consumption decisions, raising hedging costs and potentially adversely affecting the development of South Africa's exporting sector.

Despite the extensive policy debate about the relative volatility of the ZAR, there has been relatively little research into the nature and drivers of the ZAR's volatility.2 Likewise, the depreciating trend of the ZAR has drawn attention to directional risks, yet there is an absence of research into the skewness of the ZAR (i.e. the implied risk that the ZAR could move in a particular direction) or its kurtosis (i.e. the implied risk of very large ZAR changes). Given the interconnectedness of global foreign exchange (FX) markets and the unconventional monetary policy measures (such as quantitative easing) implemented by major economy central banks recently, there has also been a lot of policy interest in measuring spillovers between different markets, but few papers consider samples that include many emerging market (EM) currencies.

In this paper, we characterise the dynamics of the ZAR, as well as those of a large number of EM and developed market (DM) currency pairs and then assess the relationship between ZAR dynamics and those of other currencies. To this end, we construct a database of log-returns and higher-order realised moments for the ZAR, 7 other EM currencies and 12 DM currencies over the period 12 July 2006 to 10 February 2021, inclusive. Next, following the precedent in the literature (e.g. Greenwood-Nimmo et al., 2016, 2019b), we develop and estimate an empirical network model to measure the intensity of bilateral FX risk and return spillovers among currencies and to capture the dynamics of these spillovers. We adopt a modified version of the framework for network analysis put forth by Diebold and Yilmaz (2009, 2014), in which the structure of the network of bilateral spillover effects is estimated via a decomposition of the variance of the forecast errors obtained from a reduced form vector autoregression (VAR). By analysing the structure of the network on a rolling-sample basis, it is possible to track the evolution of spillovers over time.

Our empirical network model enjoys several benefits relative to competing techniques for the analysis of spillovers. First, by virtue of our consideration of the higher-order moments of the exchange rate distribution, our model achieves greater generality than popular ARCH-type models of FX volatility transmission (e.g. Engle et al., 1990, and the many studies that have adopted similar methods). Second, unlike many existing studies that develop separate models for each variable group under consideration (e.g. Cai et al., 2008; Diebold and Yilmaz, 2009),

• The computer programs required to replicate the results in this paper using R version 4.0.2 are available from the authors on request.

• Exceptions include Farrell et al. (2012), who show that inflation surprises have high-frequency impacts on the USDZAR, Hassan (2015), who suggests that there are relationships between macroeconomic fundamentals and ZAR volatility, and Arezki et al. (2014), who find that gold price volatility can explain the ZAR's volatility.

2

our use of a shrinkage and selection estimator allows us to develop a single high-dimensional model for multiple variable groups simultaneously that accounts for spillovers between different variable groups (e.g. spillovers between skewness and returns). Furthermore, relative to alternative techniques for network analysis based on Granger causality (e.g. Billio et al., 2012), our technique provides a means to capture not just the direction but also the strength of pairwise spillovers and to account not just for lagged dependence but also for contemporaneous dependence across currencies.

The spillover statistics that we obtain can be used to shed light on how changes in a cur- rency's risk-return profile transmit through global FX markets and to assess the degree to which changes in the risk-return profile of a given currency reflect global and local conditions. Our analysis focuses primarily on spillovers to and from the ZAR and provides new insights into the sensitivity of the ZAR to shocks arising in FX markets around the world, as well as the informational role played by the ZAR in global FX markets. We show that our model is able to highlight risk transmission channels in a timely manner during FX flash crashes and periods of heightened financial market uncertainty. In addition, we show that variations in the risk- return profile of the USDZAR correlate with variations in the risk-return profile of many other currencies, and that this is especially notable with respect to EM currencies. We interpret this as evidence of the ZAR's role as a bellwether EM currency. Our results suggest that global conditions have an important impact on the ZAR, which would likely limit the effectiveness of interventions aimed at dampening currency volatility. Our results do, however, suggest that the ZAR also reacts strongly and rapidly to South Africa-specific shocks, such as bouts of political instability. This is consistent with the exchange rate playing the role of a shock absorber in the economy, as argued by Soobyah and Steenkamp (2019) and Loewald (2021).

2. The dataset

We construct risk and return measures at daily frequency for a group of M = 20 currency pairs, with each expressed in units of foreign currency per US dollar (USD).3 Our sample contains 12 DM currencies, as follows: the Australian dollar (USDAUD), the Canadian dollar (USDCAD), the Swiss franc (USDCHF), the euro (USDEUR), the British pound (USDGBP), the Hong Kong dollar (USDHKD), the Japanese yen (USDJPY), the Korean won (USDKRW), the New Zealand dollar (USDNZD), the Norwegian krone (USDNOK), the Swedish krona (USDSEK), and the Singaporean dollar (USDSGD). In addition, our sample includes the following 8 EM currencies: the Brazilian real (USDBRL), the Chinese renminbi (USDCNY)4, the Indian rupee (USDINR),

• Maintaining consistency in the quoting convention across currencies facilitates comparisons of higher-order risk measures across currencies.

• China maintains a managed floating currency framework. The renminbi is traded in two markets: while the offshore renminbi (USDCNH) is largely based on market trading, the onshore renminbi (USDCNY) is regulated and only allowed to trade within a 2% range from a given value that is set each day at 9.15 am. The framework has evolved over time. Between July 2005 and July 2008, the currency referenced a basket of currencies and a fixed trading band; between July 2008 and June 2010, the CNY was fixed to the USD; after July 2010 the managed float was reinstated and the trading band gradually expanded (from 0.3% between January 1994 and May 2007, then 0.5% to April 2012, then 1% to March 2014 and thereafter it was set at its current level of 2%).

3