MPE Mathematical Problems in Engineering 1563-5147 1024-123X Hindawi 10.1155/2019/4267532 4267532 Research Article Bayesian Semiparametric Double Autoregressive Modeling https://orcid.org/0000-0003-1525-7047 Wang Min 1 2 Liu Ji-Chun 1 Cruz Frederico 1 School of Mathematical Sciences Xiamen University Xiamen 361005 China xmu.edu.cn 2 School of Mathematical Sciences Guizhou Normal University Guiyang 550001 China gznu.edu.cn 2019 1572019 2019 19 01 2019 27 03 2019 28 04 2019 1572019 2019 Copyright © 2019 Min Wang and Ji-Chun Liu. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

This paper proposes a Bayesian semiparametric modeling approach for the return distribution in double autoregressive models. Monte Carlo investigation of finite sample properties and an empirical application are presented. The results indicate that the semiparametric model developed in this paper is valuable and competitive.

National Natural Science Foundation of China 11071202 11861025 Science and Technology Foundation of Guizhou Province LKS05
1. Introduction

The original Bayesian theory is a parametric method. The parametric model has been long applied in classical statistical and Bayesian statistical inference studies, and its estimation is based on the unknown parameters of the overall distribution. In a review of Bayesian statistics, Lindley  identified nonparametric methods as a notable area in that it lacks Bayesian progress. Our purpose here is to present work with mixture priors that substantially contributes to both parametric and nonparametric density estimation. With the development of statistical theory, statisticians have found the need to reduce the overall distribution assumptions in many cases. We find that estimates of volatility using nonparametric models can differ dramatically from those using a normal return distribution if there is evidence of a heavy-tailed return distribution. When using a normal distribution or Weibull distribution in general, we often need to assume that the population distribution is unimodal (see ). The support set must be sufficiently large to capture the fat tail. We find that the Dirichlet process mixture (DPM) model is able to adapt to several frequently used distribution models and also accurately estimates the posterior distribution of the volatilities of the returns, without assuming any underlying distribution.

Mixture models are noted for their flexibility and are widely used in the statistical literature. Nonparametric methods are usually combined Bayesian approaches with the development of more mature technologies in density estimation, regression, survival analysis, hierarchical models, and model validation . The Dirichlet process (DP) is one type of nonparametric method often used in density estimation. Ferguson  took Dirichlet as a random probability measure and noted that if using the mixture distribution to approximate an unknown distribution, the posterior inference must be “controlled metaphysically” and the support set must be sufficiently large. Escobar and West  were the first to propose a modeling style involving the DP to introduce uncertainty in the number of components. The wide application of DPM priors is one of the major success stories of modern Bayesian statistics.

A central property of economic time series that is common to many financial time series is that their volatility varies over time. Describing the volatility of an asset is a key issue in financial economics. The most popular class of models for time-varying volatility is represented by GARCH-type models . GARCH models are commonly used to describe, estimate, and predict the dynamics of financial returns. Recent surveys of the existing GARCH model literature can be found in Davidson  and Rombouts et al. . In contrast to Engle’s  ARCH model, a double autoregressive (DAR) model, which is a special case of the ARMA-ARCH models in Weiss  and an example of weak ARMA models in Francq and Zakoïan [11, 12], is also drawing attention from researchers. Under the assumption of the disturbance following a normal distribution, Ling  considered the structure and the maximum likelihood estimation. In addition, stochastic volatility (SV) models have enjoyed great popularity in analyzing financial data in the last couple of decades (see , etc.). Delatola and Griffin  presented a method for Bayesian nonparametric analysis of the return distribution in a stochastic volatility model, which allows noninformative prior distributions for the parameters and assumes that the mixing component variances are constant. In addition, they  extended the work and developed a Bayesian semiparametric model with a leverage effect allowing for stylized facts such as heavy tails of the distribution of returns and correlation between the returns and changes in volatility. Jensen  solved a semiparametric, Bayesian estimator of the long-memory stochastic volatility model using a Markov chain Monte Carlo (MCMC) algorithm. Jensen and Maheu  extended the existing fully parametric Bayesian literature on stochastic volatility to allow for more general return distributions.

Jensen and Maheu [19, 20] proposed a Bayesian nonparametric modeling approach for the return distribution multivariate GARCH models (MGARCH). Virbickaite˙  considered an asymmetric dynamic conditional correlation (ADCC) model for time-varying correlations with GJR-GARCH models for individual volatilities, denoted ADCC-GJR-GARCH, and provided a substantially more realistic evaluation of the comovements of the assets’ returns than standard symmetric MGARCH models. Ausín et al.  developed a Bayesian semiparametric approach based on modeling the innovations using the class of scale mixtures of Gaussian distributions with a DP prior on the mixing distribution. Jensen and Maheu [19, 20] extended the asymmetric, stochastic, and volatility models by modeling the return volatility distribution nonparametrically. The novelty of their paper was in modeling the distribution with an infinite mixture of normals, where the mixture of unknowns have a DPM prior. This motivated us to establish a DAR with a DPM model (DAR-DPM).

The remainder of this paper is organized as follows. In Section 2, we describe the DP mixed process and introduce the DAR model with a nonparametric DPM prior for the unknown distribution. An MCMC scheme is used to sample from the posteriors, and the parametric portions of the model are presented in Section 3. Section 4 provides some simulation studies regarding the performance of the proposed method. An empirical example is reported in Section 5, and Section 6 concludes the paper.

2. DAR Models with Dirichlet Process Mixture Prior

The usual structure of a DAR model  assumes that a return series, denoted by rt, can be written as follows:(1)rt=i=1pϕirt-i+ht1/2εt,ht=ω+i=1qβirt-i2,where ht are the conditional variances of rt given the past history, t-p,,0,1,, ω,βi>0; εt is a sequence of independent and identically distributed (IID) random variables with zero mean and unit variance; and rs is independent of εt:t1 for s0. Ling  gave the stationary ergodic conditions of the general model and obtained the necessary and sufficient conditions for the strict stationarity and weak stationarity of the DAR model. Let At be a random matrix of p×p:(2)ϕ1+β1ε1tϕp-1+βp-1εp-1,tϕp+βpεptIp-1Op-1×1where Ir is the unit matrix of r×r and Or×s is the zero matrix of r×s. To facilitate the presentation of results, we need to give the concept of the Lyapunov exponent here, which is defined as (3)r=inf1nElnA1AnFrom εit~N(0,1), we know that there is Eln+A1<, where ln+x=maxx,0, subordinate to the Theorem of Subadditive Ergodicity , and (4)γ=limn1nA1An.Ling  gave a theorem in which the necessary and sufficient condition for the existence of a strictly stationary solution of the DAR model is γ<0. When β=1, γ=Elnϕ1+β1εt, the condition is reduced to ϕ12+β1<1. From the study of Lu and Jiang , the sufficient condition for the weakest stationarity of the model is i=1pϕi+Eξti=1pβi<1.

As noted by Ausín et al. , typical models for the innovation distribution include the Gaussian, Student, and Gaussian mixture models. The purpose of this paper is to construct a robust alternative to these distributional assumptions. Generally, the unit variance restriction of εt in (1) makes it difficult to undertake semiparametric Bayesian inference. To avoid this, we rewrite (1) as follows:(5)rt=i=1pϕirt-i+h~t1/2ξt,h~t=1+i=1qβ~irt-i2where h~t=ht/ω is a rescaled volatility and β~i=βi/ω, and ξt=ω1/2εt are sequences of IID random variables with mean 0 and variance ω.

We relax all assumptions concerning the distribution of ξt in (5), allowing its distribution to be completely unknown and random as if the distribution were an unknown in addition to the parameters and latent volatilities of the DAR model. This strategy suggests using a Gaussian scale mixture model with a density function given by (6)fξξtG=ϕξt0,ΛtdGΛt,where ϕξt0,Λt denotes the density function of the Gaussian distribution with mean zero and variance Λt and G is a random probability measure of a DP; G0 is a measure function in probability space, and α is a strength parameter. The distribution of G is based on the unknown prior of the DP(α,G0). Employing a Lo  type DPM prior, assume that(7)ξtΛt~N0,Λt(8)ΛtG~G(9)Gα,G0~DPα,G0to model the unknown distribution, where G0~Inv-Gamma (v/2,c/2) and α~Gamma(a,b). In (7)-(9), ξt are normally distributed with a mean zero but a random variance Λt, distributed as G. The nonnegative value α is the strength parameter, and G0 is the base distribution. By the definition of DP(α,G0) in Sethuraman , G is almost necessarily equal to a discrete distribution G(dΛ)=j=1πjδσj(dΛ), where δσj(·) is a degenerative distribution on σj2 with the random variance extracted from the prior distribution G0 of DP. To ensure the conjugation, G0 is a gamma distribution with the parameters v/2 and c/2, i.e., the probability distribution of G0~Inv-Gamma (v/2,c/2). The probability of Λt being equal to a particular σj2 is πj, which satisfies(10)π1=π1,πj=πjl<j1-πlunder α>0, πj~Beta(1,α).

The DAR-DPM model above can also be introduced into the potential distribution of variables S=(S1,S2,,ST), while St=j, Λt equals a particular σj2. Under the prior of the DP, the distribution of St is St~j=1πjδj, where the probabilistic weight πj,j=1,2,, defined by (10). Incorporating St into the definition of the DAR-DPM model, we propose rescaling the mode defined in (5), (7), (8), and (9) as follows: (11)rt=i=1pϕirt-i+ξt1+i=1qβirt-i2ξSt,Λt~N0,ΛStSt~j=1πjδjσSt2~G0.Finally, to complete the Bayesian inference of the model, we also need to deduce distributions for the parameters of the DAR-DPM-type models.

3. Bayesian Inference for DAR-DPM-Type Models

This section describes how to perform Bayesian inference for DAR-DPM-type models. Given an observed time series r=r1,r2,,rT and the priors defined in the previous section, we consider the joint posterior distribution of the model parameters. Unfortunately, the joint posterior distribution does not have a closed analytical expression. To this end, the MCMC algorithms of the DPM model in Escobar and West , which produce samples from a Markov chain whose stationary distribution is the joint posterior distribution of the model parameters, are employed.

Let Θ=(ϕ,β,Λt,α), where ϕ=(ϕ1,ϕ2,,ϕp) and β=(β1,β2,,βq) are the full model parameters and Θθ represent a set of parameters that do not contain the parameter θ. We block the unknown parameters such that the conditional posterior distribution of the controllable block is either known or given by a manageable form. (12)πϕ,β,Λt,αrfrϕ,β,Λt,απϕβ,απβαπΛtαπα.Assuming that other parameters and latent variables have been obtained, we can construct a Markov chain on each piece of the posterior distributions for iterative sampling.

3.1. Sample from <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M88"><mml:mfenced open="{" close="}" separators="|"><mml:mrow><mml:msub><mml:mrow><mml:mi>Λ</mml:mi></mml:mrow><mml:mrow><mml:mi>t</mml:mi></mml:mrow></mml:msub></mml:mrow></mml:mfenced><mml:mrow><mml:mo>∣</mml:mo></mml:mrow><mml:mi>r</mml:mi><mml:mo>,</mml:mo><mml:mi>ϕ</mml:mi><mml:mo>,</mml:mo><mml:mi>β</mml:mi><mml:mo>,</mml:mo><mml:mi>α</mml:mi></mml:math></inline-formula>

{ Λ t } r , ϕ , β , α can be sampled using the two-step method based on the Polya method . Set ξ=ξ1,ξ2,,ξT, where ξt=rt-i=1pϕirt-i/1+i=1qβirt-i2. The posterior distribution of Λt can be written as(13)ΛtΛt,tt,ξt,α~αα+T+1gξtGdΛξt+1α+T-1ttfNξtΛtδΛtdΛtwhere g(ξt)fN(ξt0,Λ)G0(Λ)dΛ. Under the conditional probability rule, (14)GdΛξtfNξt0,ΛG0dΛwhose density function is (15)gΛξtΛ-1/2exp-ξt22Λ·Λ-1+v/2exp-c2Λ=Λ-1+v+1/2exp-ξt2+c2Λ.Assuming known ϕ,β,α, the posterior distribution is (16)πσj,Sξ,απσj,Sr,ϕ,β,α,where σj,j=1,2,,k, kT, are the distinct Λt,t=1,2,,T.

The two steps are sampled as follows.

Step 1.

(17) S t Λ t : t t , ξ t , α ~ α α + T - 1 g ξ t δ 0 d S t 1 α + T - 1 j = 1 k t f N ξ t 0 , σ j δ j d S t k(t) is the distinct number of Λt,tt presenting different σj. If St is drawn to be 0, k increases by 1, St is set for the new value k, and the new σk is extracted from (15); otherwise, St is the random sample value j.

Step 2.

Discarding σ in the extraction of Step 1, new σj,j=1,2,,k, are extracted using the sample S and k iteration, and the posterior distribution is as follows: (18)πσjξ,S,kt:St=jfNξt0,σjG0dσjσj-1+v/2exp-c2σj·t:St=jσj-1/2exp-ξt22σj=σj-t:St=j1+v/2+1exp-t:St=jξt2+c2σj~IGt:St=j1+v2,t:St=jξt2+c2.

3.2. Sample <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M118"><mml:mrow><mml:mi>ϕ</mml:mi></mml:mrow></mml:math></inline-formula> and <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M119"><mml:mrow><mml:mi>β</mml:mi></mml:mrow></mml:math></inline-formula>

The sampling of πϕr,β,Λt and πβr,ϕ,Λt is based on the Metropolis-Hastings (MH) method . The use of R language for this method is mentioned in a previous article , which proposed that the distribution is the conditional distribution of the parameters.

Let p(ϕi),i=1,2,,p, denote the prior distribution of ϕi assuming known S=S1,S2,,ST and σ=(σ1,σ2,,σk), ξt/σSt~N(0,1). In addition, write ht1+j=1qβjrt-j2. Then, ξt=(rt-i=1pϕirt-i)/ht; therefore, the posterior distribution of ϕi is(19)πϕir,S,Θϕit=1T12πσSthtexp-rt-j=1pϕirt-j22σSthtpϕiexp-ϕi-μϕi22σϕi2·pϕiwhere σϕi2=i=1T(rt-i2/σStht)-1, μϕi=t=1T(rt-i(rt-jiϕjrt-i2)/σStht)σϕi2. Then, our proposed distribution for ϕi is N(μϕi,σϕi2),i=1,2,,p. By MH sampling, the accepted probability of ϕi, α(ϕi,ϕi)=mingϕifNϕiμϕi,σϕi2/gϕifNϕiμϕi,σϕi2,1, where g(ϕi)πϕir,S,Θϕi. If the suggested value ϕi has been refused, then the prior ϕi remains as the current value drawn from πϕir,S,Θϕi.

Sampling from π(βr,ϕ,{Λt}), set p(βi) as the prior of βi,i=1,2,,q, ϵtrt-j=1pϕirt-j=ξtht1/2, and Ftσ(rs,st). Then, under Ft-1, (20)utϵt2-σStht=ϵt2σStht-1σStht=χ21-1σStht.The conditional mean of ut is EutFt-1=0, and the conditional variance is varutFt-1=2σStht2. Nakatsuma  suggests replacing this ut with N(0,2σSt2ht2). Then, the posterior distribution of βi is (21)πβir,S,Θβit=1T12π·2σSt2ht2exp-ϵt2-1+j=1qβjrt-j2σSt24σSt2ht2·pβiexp-βi-μβi22σβi2pβigh1:Twhere σβi2=t=1Trt-i4/2ht2-1, μβi=σβi2t=1T(rt-i2(ϵt2/σSt-1-jiβjrt-j2)/2ht2). The suggested distribution of βi is N(μβi,σβi2),i=1,,q. Extracting βi, the idea of rejecting or accepting is similar to ϕi.

3.3. Sample <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M161"><mml:mrow><mml:mi>α</mml:mi></mml:mrow></mml:math></inline-formula>

Drawing from the conditional posterior distribution of α can be performed, as explained in Escobar and West , by first sampling an auxiliary variable ζ from a beta distribution, ζ~Beta(α+1,T), and then sampling α from a gamma mixture, πζG(a+S,b-logζ)+(1-πζ)G(a+S-1,b-logζ), where S=maxStt=1T and πζ/(1-πζ)=(a+S-1)/(T(b-logζ)).

3.4. MCMC Algorithm

This part briefly discusses the steps of the MCMC used to fit the DAR-DPM model. Given the priors, we define the following MCMC algorithm:

Initialize ϕ , α , β , k , S , a , b , Λ

Sample ϕ α , β , k , S , a , b , Λ , r

Sample β ϕ , α , k , S , a , b , Λ , r

Sample α ϕ , β , k , S , a , b , Λ , r

Sample k , S ϕ , α , β , a , b , Λ , r

Sample a , b , Λ ϕ , α , β , k , S , r

4. Simulation Study

In this section, the proposed methodology is illustrated using an artificial time series of size T=1000. The simulated data are given by yt=0.2yt-1+ξt1+0.3yt-12, where we set ϕ=0.2 and β=0.3. The following three scenarios are considered: (1) ξt~DP(α,G0); (2) ξt~N(0,1); and (3) ξt~0.9N(0,0.6)+0.1N(0,4.6) (Ausín, 2014). The MCMC algorithm is run for a burn-in of 30000 iterations and subsequently a further 30000 iterations, storing one in 1500 iterations, thus obtaining a total of 1500 realizations from the posterior distribution.

Sensitivity Analysis, Comparison of Different Prior Effects. The set of hyperparameters has been set as a=2,b=4,v=10,c=30. Note that we set the prior α~Gamma(2,4) initially because there is a fair degree of support for values near α=1 (see ), which leads to a rather noninformative prior.

Under the seven different priors in Table 1, we change only the hyperparameters a,b (Priors 2, 3, and 4), simply change the hyperparameters v,c (Prior 5 and 6), or change them all (Prior 7). Table 1 shows that all the estimated values of ϕ and β are closer to the true values of the parameters, with a smaller standard deviation. Therefore, we believe that the parameter estimation of DAR-DPM is not very sensitive to the setting of the hyperparameters. In the following analysis, a significant loss of parameter estimation accuracy is not expected when using the hyperparameters of prior 7.

Sensitivity analysis of the choice of priors.

Prior 1 Prior 2 Prior 3 Prior 4 Prior 5 Prior 6 Prior 7
a = 2, b = 4 a = 6, b = 6 a = 0.1, b = 20 a = 2, b = 4 a = 2, b = 4 a = 2, b = 4 a = 1, b = 5
v = 10, c = 30 v = 10, c = 30 v = 10, c = 30 v = 3, c = 0.003 v = 10, c = 0.01 v = 10, c = 1 v = 3, c = 6
ϕ 0.2008 0.2008 0.2033 0.1999 0.2042 0.1949 0.1919
0.0017 0.0025 0.0015 0.0022 0.0018 0.0023 0.0018
β 0.3164 0.3151 0.3092 0.3089 0.3319 0.3407 0.3146
0.0019 0.0037 0.0026 0.0019 0.0036 0.0036 0.0034
α 0.9625 0.9780 0.9289 0.9187 0.8949 0.9096 0.9492
0.0025 0.0011 0.0048 0.0033 0.0062 0.0045 0.0025
k 5.8512 5.9677 5.8917 3.5473 3.2172 3.7628 5.5979
1.3858 1.4954 2.1101 0.6853 0.4840 0.4602 1.3408

Note. This table presents the posterior mean and standard deviations of the model parameters for prior 1 sets a=2, b=4, v=10, c=30, and k=6. The remaining prior uses the same choices but changes the values of the indicated hyperparameters in the first rows.

Normal and Finite Normal State Compared with Infinite Normal State. Table 2 shows that when estimating ϕ, the standard normal and finite mixture models behave well. However, both models are unable to estimate β. Clearly, the DPM method exhibits its comparative advantage in all cases and can accurately estimate the two parameters ϕ and β simultaneously, regardless of the data being from the standard Gaussian, finite mixture, or DPM mixed models.

Full sample estimates.

Model Parameter True values DPM Standard Normal Finite mixture
Post.Mean Post.Mean Post.Mean
Post.std.dev Post.std.dev Post.std.dev
DPM ϕ 0.2 0.1909 0.2046 0.1973
0.0364 0.0441 0.0492
β 0.3 0.3095 0.3557 0.3209
0.0367 0.3765 0.0621
α 1 0.9413
0.0472
k 5.4095
1.1744
Standard Normal ϕ 0.2 0.1930 0.1957 0.2048
0.0367 0.0357 0.0344
β 0.3 0.2951 0.2989 0.3071
0.0532 0.0386 0.0704
α 0.9124
0.0491
k 1.5485
0.1827
Finite mixture ϕ 0.2 0.2088 0.1885 0.1956
0.0377 0.0694 0.0647
β 0.3 0.2957 0.6394 0.3283
0.0679 0.3066 0.1954
α 0.9219
0.0370
k 2.4693
0.0774

Note. This table displays the posterior means and standard deviations of the model parameters obtained from the DPM model, Gaussian distribution and mixture of two zero-mean Gaussians.

We have drawn postdensity plots for each parameter based on simulation data for clarity; see Figure 1.

The postdensity plots of each parameter.

5. Empirical Application

This section applies the above method to actual financial data to fit the model. The utilized data are a series of US weekly observations of 3-month Treasury bill data: 1989/03-2019/03 (1566 observations, Data Source: Federal Reserve Economic Data). To make the data smooth, we perform some processing of the original data. Let xt denote the 90-day T-bill weekly returns, and let yt denote the rate of return of the order difference, i.e., yt=xt-xt-1; see Figure 2.

Plots of the observations (a) and the order difference of returns (b) of the weekly observations of American 3-month treasury bills. As shown in (b), the transformed data are smooth.

To evaluate the performance of the different models, first, we give some descriptive statistics of the premodeling data and their autocorrelation function and partial autocorrelation function; see Table 3 and Figure 3.

Descriptive statistics.

Min. 1st Qu. Median Mean 3rd Qu. Max. Variance Sd
y -0.590000 -0.03000 0.000000 0.004051 0.020000 0.980000 0.007110 0.084323

Note. This table shows the summary of the order difference.

(a) is the ACF of y; the first-order lag autocorrelation coefficient is very significant. The PACF diagram (b) shows that we cannot build a simple autoregressive model.

We now illustrate the four models using the above data.

First, assume that ξt~DP(α,G); we may obtain the following fitted DAR(1,1)-DPM model: yt=0.2507yt-1+ξt1+0.7043yt-12, where the standard deviation of each coefficient is 0.0361 and 0.2460 and AIC=-2.1361. Then, assume that ξt~N(0,σ2); the fitted model is yt=0.1776yt-1+ξt1+0.2162yt-12, the standard deviations of the coefficients are 0.2162 and 0.1210, and AIC=-1.7912. Again, suppose that ξt~δN(0,σ12)+(1-δ)N(0,σ22) to arrive at yt=0.2517yt-1+ξt1+0.6161yt-12, where the standard deviations of the coefficients are 0.0385 and 0.2505 and AIC=-2.0394.

The predictive distribution of the series showcases the exibility of the DPM. We computed and plotted the predictive densities for the above models; see Figure 4.

The predictive densities for both the parametric and nonparametric approaches.

According to the three models obtained above, we can see that the parameter values with the DPM model have a smaller variance and minimum value of AIC. In the DPM model, we calculate that α is 0.8628 and that the mean of k is 2.7956. Obviously, under these figures, the disturbance tends to form several groups; therefore, using the DPM model to match these figures is satisfactory. To obtain the above results, we perform 20000 iterations; see Figure 5.

MCMC results for DAR-DPM: (a) the estimation results of ϕ and (b) the estimation of β; (c) and (d) are their respective density maps. It can be observed that β is left-biased, so we take the median as the estimated value because the median is less affected by skewness.

Finally, we also compare our model with that of Jensen and Maheu [19, 20] using the same datasets. The fitted GARCH(1,1)-DPM model is (22)yt=ht1/2ξtht=0.2613+0.0619yt-12+0.0462ht-1where ξt~DP(α,G). The standard deviations of the coefficients are 0.6421, 0.1288, and 0.1073, and AIC=-2.0979. This model was also applied in Ausín, Galeano, & Ghosh .

Based on this example, the results demonstrate that the DAR-DPM model developed in this paper is valuable and competitive. The support for u is always confined to values below 0.7, as is evident in the posterior density of u plotted in Figure 6 for the equity data, where u=α/(α+1), BF(u=u0)=π(u=u0r,DPM)/π(u=u0)[19, 20]. This indicates that the equity data strongly support a nonparametric specification with only a few.

Posterior density of uα/(1+α).

6. Conclusion

In this article, a semiparametric Bayesian approach has been developed. The innovation distribution has been modeled using a scale mixture of a Gaussian model with a DP prior for the mixed distribution. An MCMC algorithm based on a combination of retrospective and slice sampling has been constructed to obtain samples from posterior distributions of the model parameters. The results that we achieved in each of our experiments in both a simulation study and a real data application are quite encouraging.

Data Availability

The text data used to support the findings of this study are included within the supplementary information file. Data Sources: Federal Reserve Economic Data can be obtained by visiting the following websites: https://www.federalreserve.gov/. Additionally, the Treasury bill data of this study is available at “https://fred.stlouisfed.org/categories/116”.

Conflicts of Interest

The authors declare that they have no conflicts of interest.

Acknowledgments

This research was partially supported by NSFC (11071202&11861025) & Science and Technology Foundation of Guizhou Province (LKS05), China.

Lindley D. V. Bayesian Statistics. A Review 1972 Pennsylvania, Pa, USA SIAM Feizjavadian S. H. Hashemi R. Analysis of dependent competing risks in the presence of progressive hybrid censoring using Marshall-Olkin bivariate Weibull distribution Computational Statistics and Data Analysis 2015 82 19 34 10.1016/j.csda.2014.08.002 MR3282163 Griffn J. E. An adaptive truncation method for inference in Bayesian nonparametric models Statistics and Computing 2016 26 1 1 19 Ferguson T. S. A Bayesian analysis of some nonparametric problems The Annals of Statistics 1973 1 209 230 MR0350949 10.1214/aos/1176342360 Zbl0255.62037 Escobar M. D. West M. Bayesian density estimation and inference using mixtures Journal of the American Statistical Association 1995 90 430 577 588 MR1340510 10.1080/01621459.1995.10476550 Zbl0826.62021 Bollerslev T. Generalized autoregressive conditional heteroskedasticity Journal of Econometrics 1986 31 3 307 327 10.1016/0304-4076(86)90063-1 MR853051 Davidson J. Moment and memory properties of linear conditional heteroscedasticity models, and a new model Journal of Business and Economic Statistics 2004 22 1 16 29 10.1198/073500103288619359 2-s2.0-0742324055 Rombouts J. Stentoft L. Violante F. The value of multivariate model sophistication: an application to pricing dow jones industrial average options International Journal of Forecasting 2014 30 1 78 98 10.1016/j.ijforecast.2013.07.006 Engle R. F. Autoregressive conditional heteroscedasticity with estimates of the variance of U.K. inflation Econometrica 1982 50 4 987 1007 10.2307/1912773 MR666121 Weiss A. A. Asymptotic theory for ARCH models: estimation and testing Econometric Theory 1986 2 1 107 131 10.1017/S0266466600011397 Francq C. Zakoian J. Estimating linear representations of nonlinear processes Journal of Statistical Planning and Inference 1998 68 1 145 165 10.1016/S0378-3758(97)00139-0 Zbl0942.62100 Francq C. Zakoian J. Covariance matrix estimation for estimators of mixing weak ARMA models Journal of Statistical Planning and Inference 2000 83 2 369 394 10.1016/S0378-3758(99)00109-3 Zbl0976.62086 Ling S. A double AR(p) model: structure and estimation Statistica Sinica 2007 17 1 161 175 MR2352507 Jacquier E. Polson N. G. Rossi P. E. Bayesian analysis of stochastic volatility models with fat-tails and correlated errors Journal of Econometrics 2004 122 1 185 212 10.1016/j.jeconom.2003.09.001 MR2083256 Zbl06524523 2-s2.0-3042777110 Delatola E.-I. Griffin J. E. Bayesian nonparametric modelling of the return distribution with stochastic volatility Bayesian Analysis 2011 6 4 901 926 Delatola E.-I. Griffin J. E. A Bayesian semiparametric model for volatility with a leverage effect Computational Statistics and Data Analysis 2013 60 97 110 10.1016/j.csda.2012.10.023 MR3007022 Jensen M. J. Semiparametric bayesian inference of long-memory stochastic volatility models Journal of Time Series Analysis 2004 25 6 895 922 10.1111/j.1467-9892.2004.00384.x Zbl1062.62232 Jensen M. J. Maheu J. M. Bayesian semiparametric stochastic volatility modeling Journal of Econometrics 2010 157 2 306 316 10.1016/j.jeconom.2010.01.014 Zbl06609152 Jensen M. J. Maheu J. M. Bayesian semiparametric multivariate GARCH modeling Journal of Econometrics 2013 176 1 3 17 10.1016/j.jeconom.2013.03.009 Zbl1284.62559 Jensen M. J. Maheu J. M. Estimating a semiparametric asymmetric stochastic volatility model with a Dirichlet process mixture Journal of Econometrics 2013 178 523 538 10.1016/j.jeconom.2013.08.018 Zbl1293.91141 Virbickaite A. Ausin M. C. Galeano P. A Bayesian non-parametric approach to asymmetric dynamic conditional correlation model with application to portfolio selection Computational Statistics & Data Analysis 2016 100 814 829 10.1016/j.csda.2014.12.005 Zbl06918367 Ausin M. C. Galeano P. Ghosh P. A semiparametric Bayesian approach to the analysis of financial time series with applications to value at risk estimation European Journal of Operational Research 2014 232 2 350 358 10.1016/j.ejor.2013.07.008 Kingman J. F. Subadditive ergodic theory Annals of Probability 1973 1 6 883 899 10.1214/aop/1176996798 Zbl0311.60018 Lu Z. Jiang Z. L1 geometric ergodicity of a multivariate nonlinear AR model with an ARCH term Probability Letters 2001 51 2 121 130 10.1016/S0167-7152(00)00138-3 Lo A. Y. On a class of Bayesian nonparametric estimates. I. density estimates The Annals of Statistics 1984 12 1 351 357 10.1214/aos/1176346412 MR733519 Sethuraman J. A constructive definition of Dirichlet priors Statistica Sinica 1994 4 2 639 650 MR1309433 Zbl0823.62007 Maceachern S. N. Muller P. Estimating mixture of dirichlet process models Journal of Computational and Graphical Statistics 1998 7 2 223 238 10.1080/10618600.1998.10474772 Nakatsuma T. Bayesian analysis of ARMA-GARCH models: a Markov chain sampling approach Journal of Econometrics 2000 95 1 57 69 10.1016/s0304-4076(99)00029-9 MR1746617 Zbl0970.62014 2-s2.0-0347318519 Ming W. Yunshun W. Estimating GARCH modeling using metropolis-hastings method in R Open Journal of Statistics 2018 08 06 931 938 10.4236/ojs.2018.86062 Nakatsuma T. A Markov-chain sampling algorithm for GARCH models Studies in Nonlinear Dynamics and Econometrics 1998 3 107 117