Abstract
A substantial amount of studies have estimated market risk by employing multivariate GARCH models but none of these studies according to be best of our knowledge has applied this technique on BRICS data . The aim of this paper is to compare the performance of three multivariate risk models (DCC-GARCH, ADCC-GARCH and CCC-GARCH) in estimating portfolio Value-at-Risk (VaR). Unlike the previous literature, we employ in our study the BRICS data and different weights to assess how changes in these weights affect the performance of the different multivariate risk models. The equity market indexes from the five countries that the paper employs are; the Brazilian Ibovespa Brasil Sao Paulo Stock Exchange Index (IBOV), Russian MICEX index, Indian S&P BSE SENSEX Index (SENSEX), Chinese Shanghai Stock Exchange Composite Index (SHCOMP) and the South African Johannesburg All Share Index (ALSI). In addition the Brazilian real/USD (brl/usd), Russian ruble/usd, Indian rupee/USD (inr/usd), renminbi/USD (cyn/usd) and the rand/USD (zar/usd) exchange rates are also employed in the study. The Average Deviations, Quadratic Probability Function Score and the Root Mean Square Error are used to backtest the performance of the models at 90%. The results indicate that multivariate GARCH models of dynamic correlation, in particular the DCC and ADCC-GARCH perform better than the CCC. In addition, giving more weight to currencies and less to equities proves to be the best way of minimizing risk in BRICS when holding a portfolio made of foreign exchanges and equities.
M.Com.