Modeling Conditional Volatility And Persistence In Variance Under GARCH Models In The Presence Of Structural Shifts

Forecasting and predicting the sterilized factors in finance is now becoming a huge and most prominent field. This area is getting enriched day by day due to the involvement of modern computing software and advancement in complex mathematical and probabilistic or statistical theories. The traders and investors, policy makers and fund managers are always interested to search some reliable hedging and trading strategies. The advancement in volatility forecast has broad influence on managing risk as well. In this scenario GARCH models gained attention of financial researchers due to successful modeling and estimating coverage of all volatility features such as volatility clustering, leverage effect, persistence of shock, mean reverting and risk premium etc. Sudden structural shifts or sudden fluctuations due to national and international global financial and economic events considerably affected the volatility of stock returns. To estimate stochastic volatility we considered symmetric and asymmetric Generalized Auto Regressive Conditionally Heteroscedastic approach for daily stock returns of Asian and European emerging markets, to evaluate and estimate the performance of these models, we have discussed contemporary researchable features such as volatility clustering, leverage effect, fat tail distribution, risk premium and persistence of variance. We applied several specifications of Generalized Autoregressive Conditionally Heteroscedastic models for some initial understanding and to get empirical results. We conclude that asymmetric GARCH successfully compared the silent features as compared to symmetric GARCH, such that exponential GARCH or EGARCH and Threshold GARCH or TGARCH has the ability to capture the Leverage Effect and Volatility Clustering in return data, also in the presence of Student-t distribution and Generalized Error distribution these models can successfully captured the fat tails rather than Gaussian distribution for innovation.In second phase to evaluate the performance of these models in the crisis period, we split the returns data of each market in three period’s i-e pre-crises period, during crises period and post-crises period of2007-2008crash and attempt to estimate and evaluate the stylized effects of conditional volatility features such as volatility clustering, risk Premium, persistency in variance and Leverage Effect. For empirical investigation we applied GARCH in Mean or GARCH-M model and Exponential GARCH or EGARCH model and critically overlook the performance of these models through comparing estimated results. We concluded that the changes in risk premium, persistency in volatility and Leverage effect before, during and post crisis of2007-2008crash are not uniform at all. In few markets we are able to present some empirical evidences of significant association between conditional variance and return but in most cases the association is insignificant indicating that there are some other factors affecting the individual markets.In third phase of this thesis we examined the effect of structural break points in conditional volatility on variance persistency of asymmetric GARCH models. We used Bai and Perron methodology to detect structural break points in conditional variance of daily stock returns of7emerging markets (4-European and3-Asian) from1997to2014. We implied Exponential GARCH or EGARCH and Threshold GARCH or T-GARCH models with and without sudden structural breaks and tried to evaluate persistency in variance and leverage effect while estimating conditional volatility. We concluded that persistence in variance reduces while considering regime shifts in conditional volatility of these models. The half-lives of shock to volatility significantly decline when we consider these sudden break points. Moreover by comparing these two models we concluded that T-GARCH model reduces persistency more gladly than EGARCH model when we account these sudden changes. Finally In this thesis we evaluate the forecast ability of asymmetric GARCH models on the basis of considering structural changes in conditional variance. We applied Bai and Perron methodology to detect the shift points in conditional variance and tried to estimate and forecast the conditional volatility with and without accounting dummy variables for these break points. We applied symmetric and asymmetric GARCH models on the returns data of5-emerging Asian markets, and tried to evaluate the forecast capability of these models in the presence of structural breaks. We concluded that application of these models without consideration of structural breaks can overestimate the persistence in variance.