Advances and Applications in Statistics
Volume 63, Issue 2, Pages 119 - 140
(August 2020) http://dx.doi.org/10.17654/AS063020119 |
|
MODELING STOCK MARKET DAILY RETURNS VOLATILITY USING SYMMETRIC AND ASYMMETRIC GARCH MODELS WITH THREE DIFFERENT DISTRIBUTIONS
Rama Krishna Yelamanchili
|
Abstract: In this paper our aim is to ascertain best fit model specification with appropriate distribution process that accurately captures conditional variance of high frequency financial data. We compile a daily stock market returns series with 7,020 observations spread over 29 years from 1991 to 2019. We test four conditional variance models under three different distribution processes. Results indicate student’s t‑distribution as appropriate distribution process for symmetric and asymmetric volatility models considered in this paper. GARCH (1,1) with student’s t-distribution captures all serial correlations and conditional variance up to 20 lags. Whereas, APARCH-ARMA (1,1) model with student’s t-distribution captures all serial correlations, conditional variance, and leverage effect. This model outperforms all other symmetric and asymmetric models. There is remaining conditional variance in EGARCH (1,1) and EGARCH-ARMA (1,1) models at 10 and 20 lags. We recommend use of APARCH-ARMA model with t-distribution for large volume of high frequency data that contains volatility clustering, persistence, and leverage effect. GARCH (1,1) model with t-distribution is recommended to measure short-run and long-run persistence of volatility shocks. |
Keywords and phrases: ARMA process, conditional variance, distribution process, GARCH models, information criteria.
|
|
Number of Downloads: 223 | Number of Views: 813 |
|