CLASSIC AND COPULA DEPENDENCY MODELS AND KENDALL'S TAU DEPENDENCY MEASURES FOR RISK ENERGY RETURN PREDICTION

• Main Author: Xaverius Aditya P, Fransiskus
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/76462
• Institution: Institut Teknologi Bandung

The existence of dependence in asset returns is a phenomenon that cannot be ignored in financial investments. Understanding the effect of dependence allows investors and policymakers to better comprehend the movement of risk. Statistically, the dependence between two random variables can be modeled using bivariate (classical) distributions or copulas. Copulas can be interpreted as the joint distribution of two uniformly distributed random variables resulting from a specific transformation. With copulas, modeling dependence becomes easier because they can capture the dependence structure in investment assets that have non-identical distributions. Kendall’s tau dependence measure is used as a tool to quantify the relationship between financial assets. Kendall’s tau can also be understood as a measure of the extent of non-linear dependence between two specified data, such as asset prices of two companies or two different commodity prices. Kendall’s tau has a range of values from negative one to positive one. The more negative (positive) the Kendall’s tau value, the stronger the inverse (direct) relationship measured. Moreover, Kendall’s tau is invariant, meaning it consistently shows properties, especially in situations where there are transformations on the random variables, such as when the random variables become non-linear. In the context of risk management, Kendall’s tau can help predict risk levels. Assets with strong correlation relationships imply additional risk that needs to be carefully observed and mitigated. In this thesis, three energy commodity returns data sets from the US stock market are used, modeled through the first-order ARMA-GARCH family. The three chosen assets for the research are New York Harbor No. 2 Heating Oil Spot Price (NYHO), Cushing OK WTI Spot Price (WTIO), and Texas Propane Spot Price (PROP). Parameter estimation through maximum likelihood is employed as a tool to obtain the energy yield model parameters. Based on the analysis and simulations, it is found that the 27 pairs of energy yield and mean-variance model consist of three best-fitted copulas: Frank, Student-t, and BB1. The simulation results show that each pair of energy yield provides a positive Kendall’s tau value. VaR predictions also indicate that the aggregation of NYHO and WTIO assets presents the lowest risk for each model. On the other hand, asset modeling through EGARCH and GJR-GARCH return higher backtesting p-values compared to the GARCH model. In terms of model accuracy, generally, the return modeled with first-order EGARCH volatility results in better accuracy compared to other models.