Modeling and forecasting intraday VaR of an exchange rate portfolio

• Date: 01 November 2018
• Author: Omar Abbara,Mauricio Zevallos
• DOI: http://doi.org/10.1002/for.2540
• Published date: 01 November 2018

Received: 21 December 2017 Revised: 20 June 2018 Accepted: 30 June 2018

DOI: 10.1002/for.2540

RESEARCH ARTICLE

Modeling and forecasting intraday VaR of an exchange rate

portfolio

Omar Abbara1,2 Mauricio Zevallos2

1Canvas Capital SA, Rua ProfessorAtilio

Innocenti, 165, 15th floor, zip code

04538-000, Sao Paulo, SP,Brazil.

2Department of Statistics, University of

Campinas, Rua Sergio Buarque de

Holanda, 651, Cidade Universitaria, Barao

Geraldo, zip code 3083-859, Campinas,

Brazil

Correspondence

Omar Abbara, Canvas Capital SA, Rua

Professor Atilio Innocenti, 165, 15th floor,

zip code 04538-000, Sao Paulo, SP,Brazil.

Email: muhieddine@gmail.com

Funding information

CNPq; CAPES ; FAPESP,Grant/Award

Number: 2013/00506-1

Abstract

The main task of this work was to predict, for the next 15 minutes, the

value-at-risk (VaR) of an equally weighted portfolio composed of four exchange

rates against the American dollar: Japanese yen, euro, Australian dollar and

Swiss franc. The dataset consists of transaction prices of each asset recorded

every 15 minutes, from January 7, 2013 to December 31, 2013. For each time

series, the multiplicative-component generalized autoregressive conditional

heteroskedasticity model of Engle and Sokalska (Journal of Financial Economet-

rics, 2012, 10, 54–83) is fitted, and the dependence among the series is modeled

by a D-vine pair-copula. VaRpredictions are estimated based on simulated obser-

vations of the fitted model following the proposal of Berg and Aas (European

JournalofFinance, 2009, 15, 639–659). The proposed method presents good

results in terms of out-of-sample intraday VaR forecasting.

KEYWORDS

backtesting, high-frequency, IVaR, MCGARCH,pair-copula

1INTRODUCTION

Forecasting risk measures is one of the most impor-

tant problems from the academic and applied stand-

points in financial econometrics. The primary measure

for risk management established in the Basel Accords is

value-at-risk (VaR), created by the investment bank JP

Morgan. VaRis the minimum loss over a horizon (usually 1

or 10 days for daily measurement) given a confidence level.

This measure can also be applied to intraday risk manage-

ment, and in this case it is called Intraday value at risk

(IVaR).

Tocalculate IVaR forecasts practitioners use the so called

high-frequency data, and to operate in financial markets

practitioners of high-frequency trading want forecasting

procedures that work fast. When dealing with large time

series, forecasting IVaR is a challenging problem given the

large amount of information.

The main purpose of this work is to predict, for the

next 15 minutes, the VaR of an equally weighted portfo-

lio composed of four exchange rates against the American

dollar: Japanese yen (JPY–USD), euro (EUR–USD), Aus-

tralian dollar (AUD–USD), and Swiss franc (CHF–USD).

This portfolio is composed of some of the most traded

exchange rates, including EUR–USD, which gained spe-

cial attention with the European debtcrisis starting in 2010

and continuing today.

The dataset consists of transaction prices of each asset

recorded every 15 minutes, fromJanuary 7, 2013 to Decem-

ber 31, 2013. This was a period with strong speculation

about possible changes in monetary policy in the world's

most important economies, especially the USA.1This spec-

ulation became stronger after the second quarter of 2013,

1The American case is important in this context because of successive

policies, known as quantitative easing, adopted by the American govern-

ment to overcome the financial crisis which started in October 2008.

Journal of Forecasting. 2018;37 729–738. wileyonlinelibrary.com/journal/for © 2018 John Wiley & Sons, Ltd. 729: