Comparing of Bayesian Model Selection Based on MCMC Method and Finance Time Series(GARCH Model)
Subject Areas : Financial Knowledge of Securities Analysisمحمدرضا صالحی راد 1 , نفیسه حبیب یفرد 2
1 - مسئول مکاتبات
2 - ندارد
Keywords: Bayesian model selection, Garch model, Model likelihood, Markov
, 
, Chain Monte Carlo (MC,
Abstract :
By using the time series models, we can analysis financial data(in last and futuretime). In financial discussions, because of heteroskedastic observations, we can notuse the classical time series models.We focus on popular practical models for financial time series, GARCH- typemodels, that were introduced for the first time by Bollerslev(1986). These modelsrepresent a very wide class of heteroskedastic econometric models. Time seriesmodels(GARCH models too), like regression models, have random errors. Theseerrors have specific distributions.Since that, the GARCH models variability is not clear, thus, we use the Bayesianmodel selection methods to estimate the parameters of the model. In this method, byusing the prior distributions on the parameters, we find the posterior distributionwhich has integral. Then, we can inference about the parameters.To explore the role of the posterior distribution, the most powerful technique is touse Markov Chain Monte Carlo (MCMC) computing methods such as the Gibbssampler and the Metropolis Hasting (MH) algorithm. These algorithms enable toestimate the posterior distribution, but, they don’t readily lend themselves to estimateaspects of the model probabilities. The most widely used one is the group of directmethods, such as the harmonic mean estimator, importance sampling and bridgesampling. Chib(1995 and 2001) proposed an indirect method for estimating modellikelihoods from Gibbs sampling output. This idea has recently been extended to theoutput of the MH algorithm.We use a reversible jump MCMC strategy for generating samples from the jointposterior distribution based on the standard MH approach.
