Risk analysis in the evaluation of the international investment opportunities. Advances in modelling and forecasting volatility for risk assessment purposes

The thesis proposes to assess the risk topic in the context of foreign investment decisions. In identifying two main risk-related concepts, I have split risks in two categories using a unique criterion: the ratio between the endogenous and exogenous content of the problem. According to it, I have built a pool of risks that the company may have entirely or partially under control (forming the endogenous part of the problem), and a pool with exogenous risks that the company cannot control at all, but can assess and build strategies for their management (forming the exogenous part of the problem). In each category I have identified one source of risk, representing the most important of all risks belonging to the same pool. For the endogenous risks part, credit risk (in its extensive version counterparty risk) was selected. Related to this, there have been additionally discussed the topics of systemic risk and of the risk associated to the impact of the activity of the international rating agencies on the firm financing problem when a company proceeded to debt issuance. The other half of the problem involves the risk of the sector the company activates in. I have found that the risk assessment in this category became an econometric problem of volatility forecasting for a portfolio of a number of selected returns. The discussion complicates given the following factors: 1. The scientific world has not reached yet to a consensus on the superiority of a certain model or group of models that measures volatility. As such, forecasted volatility estimates may depend on the model or methodologies to be used, type of data frequency (high or low), selection of the error statistics etc. As such, decision making as regards the opportunity of the investment becomes highly dependent on econometric choices to be made. 2. Multivariate models are computationally intensive due to the parameter estimation problem. If a large number of stocks are included in the portfolio, the number of estimations to be done would be so high that the problem would be extremely difficult to be technically undertaken. 3. Due to high correlation of stocks, the estimation problem becomes particularly imprecise and computationally difficult. As a solution to such problems, I have justified the superiority of one autoregressive heteroskedastic model (PC-GARCH) considering not only estimation performance but also cost saving component. For this purpose, I have run an empirical exercise with a portfolio formed of seven stocks belonging to the US IT sector (Adobe, Apple, Autodesk, Cisco, Dell, Microsoft and 3M) in order to evidentiate advantages of this model. They may be summarized as it follows: PC-GARCH • Minimizes computational efforts (by transforming multivariate GARCH models into univariate ones), by reducing significantly the computational time and getting rid of any problem that may arise from complex data manipulations; • Ensures a tight control of the amount of “noise” due to reducing the number of variables to fewer principal components. This may prove benefic since it may result in more stable correlation estimates; • Produces volatilities and correlations for all variables in the system, including those for which direct GARCH estimation is computationally difficult. As such, I’ve concluded that when using large portfolios formed of hundreds or thousands of stocks, for the scope of volatility (and therefore risk) forecasting, PCGARCH is the most appropriate model to be used.