Dependência de portfólios: uma abordagem multiescala via cópulas multivariadas

Abstract

The dynamics of economic and financial variables is a recurring subject in scientific research. Specifically in the financial market, asset price movements reflect various structures of behavior, each occurring in a different time horizons. In addition, the time series generated by these data have peculiar characteristics that must be incorporated into the financial analysis. In this context, for a more accurate and realistic understanding of issues in finance, this work seeks to apply the wavelet methodology to analyze the structure of dependence between financial assets in the Brazilian stock market in the time-frequency domain. In this thesis, two essays were developed that explore the application of copula theory to obtain dependency structures in financial series wavelet with the objective of measuring the behavior in the frequency components of stock returns, with the impacts of cycles short, medium and long term. The present work was divided in the application of two distinct techniques of multivariate copulas in frequency components of the series of returns obtained with filters maximal overlap discrete wavelet transform. In the first essay the hierarchical construction of pair copula D-Vine of Bedford e Cooke (2002) was used in an intraday portfolio with six stocks decomposed using the Daubechie filter with two null moments, in order to measure the multivariate asymmetric dependence on shortterm frequencies referring to 15 min., 1 hour, 1 day and 1 week of trading. The results indicated a greater association of assets during market recoveries in the first months after the COVID-19 pandemic. Small increments in measures of tail dependence were evidenced especially at lower frequencies. As the composition of the portfolio is diversified with stocks from different sectors, the levels of dependence are reduced considerably, which reveals the importance of strategies composition/selection of portfolios in the short term. In the second essay, the technique applied was the factor copulas of Oh e Patton (2012) in a larger portfolio, totaling thirty daily stock returns reconstructed by decomposition with the Haar filter, in order to incorporate the effects of economic cycles in risk estimation using the Value at Risk metric. In this analysis, the factor loadings were specified based on segments of action of the actions and dynamically with dependence parameters conducted with the structure Generalized Autoregressive Scores (CREAL; KOOPMAN; LUCAS, 2013) revealing the behavior since the subprime crisis in 2008 with the copula Skew t–t. The results demonstrate that the VaRs estimates, obtained out of the sample, are consistent considering short and short to medium term components.