The Modelling and Simulation Society of Australia and New Zealand
Faculty of Business and Law
School of Business / Marketing and Services Research Centre
Australia’s 2000’s decade saw the sharpest rise in mining investments arising from developing Asian emerging economies’ high demand for commodities like coal, iron ore, nickel, oil and gas which drove up prices to a historic level (Connolly & Orsmond, 2011). As of December 2012, 39 % and 9 % of the Australian Securities Exchange’s stocks were of the mining (coal and uranium stocks are included in this category) and energy (e.g. oil, gas and renewable energy stocks) sectors respectively, and investors recently have been considering separate portfolio positions in energy and mining stocks (Jennings, 2010). Facts of these nature set the stage for the task of selecting an optimal portfolio of stock securities where the fundamental questions faced by every investor, individual or institutional, are: a) what is the optimal point in time to go long in the investment position?, b) what are the optimal amounts to invest in every asset of a portfolio? and, c) when is the optimal time to short the portfolio investment position? The focus of the present study is on b) within a one period ahead forecast scenario. Understanding the price and volatility movements of stock securities taking as a basis of study their own dynamics and co-dynamics is a complex task that may be better addressed through a multilateral modelling approach. This paper, in this regard, departs from a single model application by fitting multiple risk measures to the optimization of four portfolios each consisting of 20 ASX’s stocks from the gold, iron ore-nickel, uranium-coal and oil-gas sectors. The five risk measures compared are: the variance, mean absolute deviation (MAD), minimizing regret (Minimax), conditional value at risk (CVaR), and conditional drawdown at risk (CDaR), where the last two are threshold based measures. The risk measure parameters are input into meanvariance quadratic (QP) and differential evolution (DE) portfolio problem specifications. Accurate estimations of the underlying interaction of stocks return series is a crucial element in portfolio allocation and portfolio risk management and frequentist traditional measures of dependence are rather inadequate. Here, with the objective of achieving more accuracy in the estimation of the dependence matrix, a Gaussian pair c-vine copula (PC), the regular graphical lasso (RL) and adaptive graphical lasso (AL) are fitted. Possible advantages from using these recently proposed and sophisticated techniques under model specifications where the covariance matrix is the measure of risk are indicated. The main objectives of the present study are to calculate the optimal weights to be invested in every stock of the portfolios making use of linear and nonlinear model specifications and the risk measures suggested, analyse the weight allocation differences and seek portfolio optimization advantages from using pair vine copulas and the graphical lasso in the estimation of dependence. The present multimodal approach is, therefore, expected to be more robust and as a consequence, provide more complete information that could serve for improved decision making on portfolio selection, allocation and rebalancing. Research questions are answered based on the analysis of gold portfolio outcome values, only. Findings indicate that CDaR is an important risk measure to be considered, along with other measures of risk when optimizing portfolios of stocks and no single measure of risk is suggested alone. The Gaussian pair cvine copula through the use of one different parameter in the modelling of every pair of variables’ joint distribution appears to be more sensitive in capturing data’s distribution characteristics. The adaptive graphical lasso also appears to be more perceptive when grasping the signal of the underlying interaction of the stocks. Therefore, valuable information could be drawn and inferred from applying multiple risk measures and sophisticated statistical techniques for their estimation. The weight allocation from threshold risk measures such as CVar and DaR and Minimax clearly differs from the rest. The models identified stocks with high return relative to risk and vice versa. The originality of the present study lies on the sectors of application and its multi-model nature.