This thesis has presented the insight into volatility forecasting covering from basic theory, simulation studies, practical implementations, to the new proposed models. We have analyzed the data with alternative tools that can make us aware of different factors that have been observed with other common methods. We aim to bring new light, rather than replace settled models. The normal inverse Gaussian (NIG) distribution has been employed to analyze the exploratory data and we have concluded that it is capable of describing the marginal distribution of return during the crisis of 2008 and it can be estimated plainly with either the method of moments or the maximum likelihood estimation. The results show that the NIG distribution has attractive potential to model financial data. It grants an alternative way of modeling financial data in such a way that GARCH-type models, popular models in the financial literature, do not supply._x000D_ The NIG distribution has desirable properties such that it has been used to construct the NIG stochastic volatility (NIG-SV) model. We introduce the hierarchical generalized linear model (HGLM) method for estimation that can be applied to the NIG-SV model. The empirical results show that the HGLM method is as accurate as the maximum likelihood method but it does not require complicated integration. Moreover, the h-likelihood associating the HGLM method provides us the estimates of random effects that are latent information. Consequently, we apply the random effects to estimate and forecast volatility. The NIG-SV forecasting models overcome the standard forecasting models in some occasions._x000D_ Apart from the return-based models, we also investigated the properties of exogenous variables such as open, close, high and low prices as volatility estimators. The range calculated from the difference between daily high and low prices is of interest. Several range-based volatility estimators have been studied and corrected the bias generated from discretization. We test by simulations whether these estimators are relevant in different scenarios when the theoretical conditions do not hold perfectly. It turns out that the Garman-Klass estimator performs impressively and the other estimators also provide proper estimates for volatility. We conclude that the range contains substantial information and it is relevant to incorporate into the new model._x000D_ In the end, all information obtained from previous studies is summarized into the new model. The dynamic normal inverse Gaussian stochastic volatility (DNIG) model has been proposed. It is the extension of the NIG-SV model where the dynamics of the volatility is driven by the range. The DNIG model can be extended to higher orders and it can be estimated simply by the HGLM method. Other researchers have never discovered the relevant information of the autoregressive process of order two in the standard SV model. However, we have shown in this thesis that in most cases, the DNIG of order two models fit better to the series than the DNIG of order one models and the NIG-SV models. It is also remarkable that the coefficient of the range in the DNIG models might be the indication of the crisis. _x000D_ Estimating the DNIG model with the HGLM method yields the random effects estimates and consequently the estimates for volatility. We have shown that the returns standardized by volatilities estimated from DNIG models do not exhibit heavy tails. These results indicate that DNIG models are capable in capturing the relevant information from the returns. DNIG models with HGLM method also allow us to forecast volatility with ease. The DNIG forecasting models have been tested with real data in comparisons with the standard models and they perform nicely. In many cases, the DNIG forecasting models are more accurate than the popular GARCH(1,1) models.
| Date of Award | 8 May 2015 |
|---|
| Original language | Undefined/Unknown |
|---|
| Supervisor | Joan del Castillo Franquet (Director) |
|---|
Volatility Forecasting with Latent Information and Exogenous Variables
Kesamoon, C. (Author). 8 May 2015
Student thesis: Doctoral thesis
Kesamoon, C. (Author), del Castillo Franquet, J. (Director),
8 May 2015Student thesis: Doctoral thesis
Student thesis: Doctoral thesis