Research article

Daily LGARCH model estimation using high frequency data

  • Received: 29 July 2021 Accepted: 27 September 2021 Published: 08 October 2021
  • JEL Codes: C13, C22

  • In this paper, we introduce the intraday high frequency data to estimate the daily linear generalized autoregressive conditional heteroscedasticity (LGARCH) model. Based on the volatility proxies constructed from the intraday high frequency data, the quasi maximum likelihood estimation (QMLE) of the daily LGARCH model and its asymptotic distribution are studied under some regular assumptions. One criterion is also given to choose the optimal volatility proxy according to the asymptotic results. Simulation studies show that the QMLE of the parameters performs well. It is also found that introducing the intraday high frequency data can significantly improve the estimation precision. The proposed method is applied to analyze the SSE 50 Index, which consists of the 50 largest and most liquid A-share stocks listed on Shanghai Stock Exchange. Empirical results show the method is of potential application value.

    Citation: Xiaoling Chen, Xingfa Zhang, Yuan Li, Qiang Xiong. Daily LGARCH model estimation using high frequency data[J]. Data Science in Finance and Economics, 2021, 1(2): 165-179. doi: 10.3934/DSFE.2021009

    Related Papers:

    [1] Morten Risstad, Mathias Holand . On the relevance of realized quarticity for exchange rate volatility forecasts. Data Science in Finance and Economics, 2024, 4(4): 514-530. doi: 10.3934/DSFE.2024021
    [2] Manveer Kaur Mangat, Erhard Reschenhofer, Thomas Stark, Christian Zwatz . High-Frequency Trading with Machine Learning Algorithms and Limit Order Book Data. Data Science in Finance and Economics, 2022, 2(4): 437-463. doi: 10.3934/DSFE.2022022
    [3] Kasra Pourkermani . VaR calculation by binary response models. Data Science in Finance and Economics, 2024, 4(3): 350-361. doi: 10.3934/DSFE.2024015
    [4] Obinna D. Adubisi, Ahmed Abdulkadir, Chidi. E. Adubisi . A new hybrid form of the skew-t distribution: estimation methods comparison via Monte Carlo simulation and GARCH model application. Data Science in Finance and Economics, 2022, 2(2): 54-79. doi: 10.3934/DSFE.2022003
    [5] Jin Zeng, Yijia Zhang, Yun Yin, Peter G Moffatt . The effect of the Covid pandemic on stock market volatility: Separating initial impact from time-to-recovery. Data Science in Finance and Economics, 2024, 4(4): 531-547. doi: 10.3934/DSFE.2024022
    [6] Maria Francesca Carfora, Albina Orlando . Application of the Gordon Loeb model to security investment metrics: a proposal. Data Science in Finance and Economics, 2024, 4(4): 601-614. doi: 10.3934/DSFE.2024025
    [7] Yue Yuin Lim, Sie Long Kek, Kok Lay Teo . Efficient state estimation strategies for stochastic optimal control of financial risk problems. Data Science in Finance and Economics, 2022, 2(4): 356-370. doi: 10.3934/DSFE.2022018
    [8] Dominic Joseph . Estimating credit default probabilities using stochastic optimisation. Data Science in Finance and Economics, 2021, 1(3): 253-271. doi: 10.3934/DSFE.2021014
    [9] Chong Li, Guoqiong Long, Shuai Li . Research on measurement and disequilibrium of manufacturing digital transformation: Based on the text mining data of A-share listed companies. Data Science in Finance and Economics, 2023, 3(1): 30-54. doi: 10.3934/DSFE.2023003
    [10] Nitesha Dwarika . Asset pricing models in South Africa: A comparative of regression analysis and the Bayesian approach. Data Science in Finance and Economics, 2023, 3(1): 55-75. doi: 10.3934/DSFE.2023004
  • In this paper, we introduce the intraday high frequency data to estimate the daily linear generalized autoregressive conditional heteroscedasticity (LGARCH) model. Based on the volatility proxies constructed from the intraday high frequency data, the quasi maximum likelihood estimation (QMLE) of the daily LGARCH model and its asymptotic distribution are studied under some regular assumptions. One criterion is also given to choose the optimal volatility proxy according to the asymptotic results. Simulation studies show that the QMLE of the parameters performs well. It is also found that introducing the intraday high frequency data can significantly improve the estimation precision. The proposed method is applied to analyze the SSE 50 Index, which consists of the 50 largest and most liquid A-share stocks listed on Shanghai Stock Exchange. Empirical results show the method is of potential application value.



    Volatility clustering is a well-known characteristic of financial time series. Accurately describing volatility is helpful for pricing and risk management of financial assets. Many conditional heteroscedasticity models have been proposed to describe the time varying volatility. Among them, the autoregressive conditional heteroscedasticity (ARCH) model proposed by Engle (1982) and the generalized autoregressive conditional heteroscedasticity (GARCH) model proposed by Bollerslev (1986) have been widely studied, especially in the financial industry. For example, Nelson (1991) applied GARCH model to asset pricing, Zou et al. (2015) used GARCH model to estimate the combination of market investment risk value, and De Davide (2019) used GARCH model to analyze and predict S & P 500 index. In many cases, the impact of assets on the market is asymmetric. That is to say, investors react differently to the same amount of good news and bad news. Therefore, in order to reflect this phenomenon, more and more scholars are involved in the study of asymmetric GARCH model, see, e.g., Hentschel (1995), Pan et al. (2008), Gyamerah (2019) and Linton et al. (2020). In addition, considering the cyclical factors of market fluctuations, many scholars began to study periodic GARCH model, such as Zhao et al. (2016).

    As noted by Duffie and Pan (1997), maximum likelihood estimation of the GARCH type model has the potential disadvantage of being overly sensitive to extreme returns. For example, if we consider a market crash, then extreme daily absolute returns may be 10–20 times the normal daily fluctuation, so the quadratic form of GARCH model yields a return effect that is 100–400 times the normal variance, resulting in excessive fluctuation prediction. Therefore, in order to avoid the impact of extreme returns, Xiao and Koenker (2009) proposed a LGARCH model. It has been shown that LGARCH model can produce more robust inferences, compared to the previous GARCH type models. The specific form of LGARCH (1,1) model is as follows:

    yt= htεt, (1)
    ht=ω+α|yt1|+βht1, (2)

    where, ω>0,α,β0, {εt}t=1 is an independent identically distributed sequence with mean 0 and variance 1, namely, {εt}i.i.d(0,1), and ys is independent of {εt:t1} for t>s. Let Ft be the σ field generated by {εt,,ε1,y0,y1}. Given Ft1, the conditional mean of yt is E(yt|Ft1)=0, and the conditional variance of yt is Var(yt|Ft1)=E(h2tε2t|Ft1)=h2tE(ε2t|Ft1)=h2t.

    With the development of electronic information technology, it is easier to obtain intraday high frequency data in the financial market, and such data usually contain lots of useful information and are valuable in improving model estimation. To achieve a more precise parameter estimator of common GARCH model, Visser (2011) proposed a volatility proxy model, embedding intraday high frequency data into the framework of daily GARCH model. The volatility proxy model not only maintains the parameter structure of daily GARCH model, but also introduces the intraday high frequency data. Available results show that the variance of parameter estimator in GARCH (1, 1) model can be reduced 20 times by selecting an appropriate volatility proxy, which greatly improves the estimation accuracy of model parameters. Many scholars have further extended the results of Visser (2011) to other cases. For example, Wang et al. (2018) proposed a compound quantile regression (CQR) method to estimate the GARCH model based on high frequency data, and proved the asymptotic normality of the estimators without strong moment conditions; Wu et al. (2018) studied the quasi maximum exponential likelihood estimation (QMELE) of non-stationary GARCH (1, 1) model under high frequency data, and obtained the limiting properties under weak moment conditions; Fan et al. (2017) studied the VaR estimation based on periodic GARCH model with high frequency data; Deng et al. (2020) studied the parameter test of GARCH model, where the parameter estimators were obtained from intraday high frequency data, and the corrected likelihood ratio test and Wald test statistics were further investigated.

    In the literature, few studies have been done about introducing the intraday high frequency data to the estimation of daily LGARCH model. However, as mentioned above that LGARCH model is more robust than other GARCH type models. Therefore, it makes sense to introduce the intraday high frequency data to estimate the daily LGARCH model, which is a main contribution of this paper. Another contribution is that the proposed estimation method is adopted to all the parameters of the model and it is not necessary to set ω in (2) to be 1 as before, see, for example, Visser (2011) and Wang et al. (2018). The rest of this paper is organized as follows. In Section 2, we introduce the volatility proxy model and estimators. In Section 3, we derive the asymptotic results of the model estimator. Simulations and empirical studies are respectively shown in Section 4 and Section 5. We conclude the article in Section 6.

    Let θ=(ω,α,β) be the parameter vector for model (1)–(2). In order to introduce intraday high-frequency data, it is necessary to extend the LGARCH model to the volatility proxy model. Denote Yt(u) to be the logarithmic return of an asset at time u on day t, where the time of each trading day is standardized to the interval of [0, 1]. From Visser (2011), we firstly consider the following scale model:

    Yt(u)= htZt(u), (3)
    ht=ω+α|yt1|+βht1. (4)

    here, 0u1, ht is the volatility of day t. Standard process Zt(u) satisfies: when ts, Zt(u) is independent of Zs(u) and has the same distribution as Zs(u). When u=1, (ytYt(1), εtZt(1), EZ2t(1)=1), the model (3)–(4) degenerates into model (1)–(2). It is easy to see that the scale model introduces intraday data information Yt(u), and it retains the parameter structure of the daily LGARCH model. Unfortunately, model (3)–(4) can not be directly estimated due to the inconsistent frequency between ht and Yt(u). In order to estimate θ, we further need to construct the intraday high-frequency data into a daily volatility proxy.

    The volatility proxy is a daily sequence based on the intraday data. That is, for intraday yield process Yt(u), let HtH(Yt(u)) be a volatility proxy for Yt(u), where H() is a given function. Common volatility proxies include realized volatility and intraday price range. Positive homogeneity is an important property of volatility proxy. Namely, for a non-zero constant ρ (ρ>0), the following equality holds:

    H(ρYt(u))=ρH(Yt(u))>0. (5)

    Consider a given function H() satisfying the positive homogeneity and apply H() to equation 3. Then it is obtained that

    HtH(Yt(u))=H(htZt(u))=htH(Zt(u))>0. (6)

    Define

    zH,t=H(Zt(u)),μ=Ez2H,t,εt=zH,tμ. (7)

    Because the standard process Zt() is independently and identically distributed, hence εt is an i.i.d. sequence with Eε2t=1. Then, combined with equations (3)–(7), we have the following volatility proxy model,

    Ht= htzH,t=htμεt, (8)
    ht=ω+α|yt1|+βht1. (9)

    In the above model, a redundant parameter μ appears due to the setting of εt, which makes it impossible to directly apply the QMLE method to estimate θ and μ simultaneously. Next, we give an indirect approach to estimate the parameters. Let ht=htμ, then (8)–(9) can be rewritten as follows:

    Ht= htεt, (10)
    ht=ω+α|yt1|+βht1, (11)

    where

    ω=ωμ,α=αμ,β=β. (12)

    When volatility proxy Ht=|yt|, it is easy to have μ=1,ht=ht. Therefore, model (1)–(2) is a special case of model (10)–(11). According to (10)–(11), we can use the QMLE method to estimate the θ=(ω,α,β). Once an estimator for μ is given, then we can get the estimation of the parameter θ=(ω,α,β) by using (12). The detailed estimation procedures are given in the next section.

    Define θ=(ω,α,β)Θ, where ΘR3 is a parameter space for model (10)–(11). In addition, suppose that θ0=(ω0,α0,β0) is the true value of the parameter θ, which is an interior point of the parameter space Θ.

    Following the convention in the literature, see Visser (2011), we consider the quasi conditional log-likelihood function (apart from a constant term).

    LT(θ)=1nnt=1lt(θ),lt(θ)=logh2t(θ)+H2th2t(θ). (13)

    Then, the QMLE of parameter θ can be defined as follows:

    ˆθ=argminθΘLT(θ). (14)

    According to (12), to further estimate θ, we need to know μ. After ˆθ is estimated, the fitting sequence {ˆht} is obtained from (11). It is already known that the absolute value of return |yt| can also be regarded as a special volatility proxy (Ht=|yt|, εt=|εt|). When Ht=|yt|, the estimated parameter obtained by the likelihood function (13) is actually the estimator of θ in the LGARCH (1, 1) model (1)–(2), which only uses the information of daily data {yt} and no intraday high frequency data is introduced. When Ht=|yt|, the estimated value of θ is denoted as ˜θ=(˜ω,˜α,˜β) and the corresponding fitting series for ht is denoted as {˜ht}. According to ht=htμ or μ=h2t/h2t, we can get an estimator of μ as followed:

    ˆμ=1TTt=1ˆh2t˜h2t. (15)

    Finally, the parameter estimators of LGARCH model with high-frequency information, denoted by ˆθ=(ˆω,ˆα,ˆβ), are given by

    ˆω=ˆωˆμ,ˆα=ˆαˆμ,ˆβ=ˆβ. (16)

    Different from the usual estimator ˜θ=(˜ω,˜α,˜β) corresponding to Ht=|yt|, the estimator ˆθ=(ˆω,ˆα,ˆβ) generally contains intraday high frequency data information and hence is expected to have a better performance.

    Before stating the asymptotic results for ˆθ, we firstly make the following assumptions.

    A1 The parameter space Θ is compact, and θ0 is an interior point of Θ.

    A2 The i.i.d. random sequence {εt} satisfies E(ε4t)<, and there exists a positive and continuous probability density function almost everywhere.

    A3 The series {Ht,ht} generated from model (10)–(11) are strictly stationary and geometrically ergodic for the considered parameter space Θ.

    Based on assumptions (A1)–(A3), adopting similar arguments to Visser (2011), it is not difficult to prove the asymptotic distribution of ˆθ,

    T(ˆθθ0)LN(0,Σ),T, (17)

    where Σ=Ω1IΩSΩ1I,

    ΩI=E(2lt(θ0)θiθj), (18)
    ΩS=E(lt(θ0)θilt(θ0)θj). (19)

    To obtain the asymptotic variance of ˆθ, we give the partial derivatives of the likelihood function lt(θ) with respect to θ as follows:

    lt(θ)θ=2(1H2th2t(θ))1ht(θ)ht(θ)θ=2(1ε2t)1ht(θ)ht(θ)θ. (20)
    2lt(θ)θiθj=2h2t(θ)(13H2th2t(θ))ht(θ)θiht(θ)θj+2ht(θ)(1H2th2t(θ))2ht(θ)θiθj=2h2t(θ)(13ε2t)ht(θ)θiht(θ)θj+2ht(θ)(1ε2t)2ht(θ)θiθj. (21)

    Further,

    ΩI=E(2lt(θ0)θiθj)=4E(1h2t(θ0)ht(θ0)θiht(θ)θj). (22)
    ΩS=E(lt(θ0)θilt(θ0)θj)=4E(ε2t1)2E(1h2t(θ0)ht(θ0)θiht(θ)θj). (23)

    In terms of (7), Eε2t=1, E(ε2t1)2=Var(ε2t). Therefore,

    Σ=14Var(ε2t)G(θ0)1, (24)

    where,

    G(θ0)i,j=E(1h2t(θ0)ht(θ0)θiht(θ)θj). (25)

    According to (11), it can be obtained that:

    ht(θ)=ω1β+αj=0(β)j|ytj1|. (26)

    Further,

    ht(θ)ω=11β, (27)
    ht(θ)α=j=0(β)j|ytj1|=j=1(β)j|ytj|, (28)
    ht(θ)β=ω(1β)2+αj=1j(β)j1|ytj1|=j=1(β)j1htj(θ). (29)

    Let s2ω, s2α, s2β be the asymptotic variances of ω, α, β respectively. Based on the asymptotic property of ˆθ, we can get the following conclusion provided the parameter μ is given:

    T(ˆωω0)LN(0,s2ωμ),T(ˆαα0)LN(0,s2αμ),T(ˆββ0)LN(0,s2β). (30)

    In practice, it is important to choose a proper volatility proxy Ht for parameter estimation. According to (24), it can be seen that the impact of Ht to the asymptotic variance is based on E (ε4t) which equals E(z4H,t)/[E(z2H,t)]2 from (7). Consequently, it is expected to choose a volatility proxy Ht with small value for E(ε4t). From (6)–(7), similar to Liang et al. (2021), we can obtain

    EH4t(EH2t)2=E(h4t)E(z4H,t)[E(h2t)]2[E(z2H,t)]2=E(h4t)[E(h2t)]2E(z4H,t)[E(z2H,t)]2=cE(z4H,t)[E(z2H,t)]2. (31)

    Here, c=E(h4t)/[E(h2t)]2 is a positive constant and hence E(z4H,t)/[E(z2H,t)]2 is proportional to EH4t/(EH2t)2. Let

    MH=(EH4t)/(EH2t)2. (32)

    Then we have smaller MH smaller E(z4H,t)/[E(z2H,t)]2 smaller Var(ε2t).

    From the above, among several candidates, one can choose the volatility proxy Ht according to its MH value. The optimal Ht should have the smallest MH value.

    In this section, we carry out Monte Carlo experiments to assess the finite-sample performance of the proposed estimators. In order to simulate the process of (3)–(4), we refer to Visser's (2011) example to simulate the intraday standard stochastic process Zt(u), which is produced by the following stochastic difference equations:

    dΓt(u)=δ(Γt(u)uΓ)du+σΓdB(2)t(u), (33)
    dZt(u)=exp(Γt(u))dB(1)t(u),u[0,1]. (34)

    here, B(1)t and B(2)t are two uncorrelated Brownian motions, Zt(0)=0, and Γt(0) is randomly generated from the stationary distribution N (μΓ,σ2Γ). The time interval [0, 1] within the day is equally divided into 240 cells to correspond to the frequency of 1 min in a real trading day. The settings for parameters in (33)–(34) are δ=1/2,σΓ=1/4 and μΓ=1/16. To further generate the stochastic process Yt(u), the parameters in equation (4) are set in the following two cases: θ0=(0.1,0.4,0.2) and θ0=(0.1,0.3,0.5).

    We consider realized volatility (RV) as the volatility proxies Ht in (8) under different frequencies, namely, 1-minute (RV1), 5-minute (RV5), 10-minute (RV10), 15-minute (RV15) and 30-minute (RV30). For RV1, the formula is given by:

    Ht=RV1t=(240i=1[Yt(ui)Yt(ui1)]2)1/2, (35)

    where, the value of Yt(u0) is replaced by Yt(0)=0. Other volatility proxies can be computed similarly. For comparison, we also consider the case Ht=|yt| where the estimator is reduced to the usual estimator which only uses the daily data. The sample sizes are T=500,1000 and 1500 and the replication time is set to be 1000. For each Ht, its MH value in (32) is estimated as:

    ^MH=T1Tt=1H4t(T1Tt=1H2t)2. (36)

    The mean of 1000 estimated MH values is taken as the final estimated value for MH and is used to judge whether the volatility proxy is optimal.

    Tables 1 and 2 summarize the empirical biases (Bias), empirical standard deviations (SD), asymptotic standard deviations (AD) and the value of MH (^MH) for ˆθ=(ˆω,ˆα,ˆβ). It can be seen that the Bias for each case is generally small. SD and AD of all parameters are very close and decrease when the sample size becomes large, which is consistent with the asymptotic results. Compared to the case Ht=|yt|, the SD and AD of the parameters estimated by RV are significantly smaller, which implies that introducing the intraday high frequency data can effectively improve the precision of the estimator.

    Table 1.  Bias, SD and AD of QMLEs and ^MH, θ0=(0.1,0.4,0.2).
    θ0=(0.1,0.4,0.2) |yt| RV1t RV5t RV10t RV15t RV30t
    T=500 ˆω Bias 0.0030 0.0002 0.0000 0.0000 0.0002 0.0006
    SD 0.0227 0.0083 0.0088 0.0092 0.0099 0.0117
    AD 0.0210 0.0080 0.0085 0.0099 0.0116 0.0131
    ˆα Bias 0.0043 0.0007 0.0010 0.0012 0.0024 0.0012
    SD 0.0685 0.0255 0.0268 0.0285 0.0300 0.0352
    AD 0.0851 0.0249 0.0264 0.0308 0.0347 0.0378
    ˆβ Bias 0.0136 0.0024 0.0032 0.0031 0.0033 0.0061
    SD 0.1228 0.0407 0.0440 0.0466 0.0514 0.0621
    AD 0.1082 0.0480 0.0506 0.0590 0.0668 0.0739
    ^MH 5.9783 2.2481 2.3255 2.4233 2.5264 2.8541
    T=1000 ˆω Bias 0.0022 0.0002 0.0001 0.0002 0.0001 0.0006
    SD 0.0167 0.0057 0.0061 0.0064 0.0069 0.0081
    AD 0.0200 0.0053 0.0058 0.0062 0.0071 0.0082
    ˆα Bias 0.0012 0.0009 0.0007 0.0005 0.0004 0.0011
    SD 0.0473 0.0179 0.0191 0.0202 0.0214 0.0251
    AD 0.0490 0.0155 0.0168 0.0177 0.0199 0.0236
    ˆβ Bias 0.0105 0.0013 0.0011 0.0010 0.0009 0.0035
    SD 0.0925 0.0274 0.0301 0.0327 0.0357 0.0431
    AD 0.1023 0.0295 0.0324 0.0345 0.0396 0.0450
    ^MH 5.9892 2.1409 2.2125 2.3067 2.4062 2.7358
    T=1500 ˆω Bias 0.0011 0.0000 -0.0000 -0.0000 -0.0000 0.0002
    SD 0.0140 0.0046 0.0048 0.0051 0.0054 0.0066
    AD 0.0133 0.0038 0.0041 0.0045 0.0048 0.0056
    ˆα Bias 0.0005 0.0005 0.0004 0.0002 0.0005 0.0000
    SD 0.0423 0.0151 0.0163 0.0172 0.0185 0.0217
    AD 0.0463 0.0125 0.0135 0.0148 0.0154 0.0186
    ˆβ Bias 0.0060 0.0011 0.0008 0.0008 0.0005 0.0021
    SD 0.0766 0.0231 0.0247 0.0269 0.0292 0.0357
    AD 0.0719 0.0213 0.0231 0.0253 0.0268 0.0318
    ^MH 5.9796 2.0980 2.1750 2.2722 2.3712 2.7094

     | Show Table
    DownLoad: CSV
    Table 2.  Bias, SD and AD of QMLEs and ^MH value, θ0=(0.1,0.3,0.5).
    θ0=(0.1,0.3,0.5) |yt| RV1t RV5t RV10t RV15t RV30t
    T=500 ˆω Bias 0.0072 0.0011 0.0008 0.0007 0.0006 0.0001
    SD 0.0375 0.0103 0.0110 0.0119 0.0128 0.0156
    AD 0.0436 0.0097 0.0112 0.0112 0.0135 0.0144
    ˆα Bias 0.0008 0.0005 0.0009 0.0010 0.0007 0.0001
    SD 0.0647 0.0217 0.0227 0.0246 0.0259 0.0305
    AD 0.0681 0.0198 0.0220 0.0230 0.0263 0.0292
    ˆβ Bias 0.0202 0.0008 0.0001 0.0006 0.0004 0.0019
    SD 0.1237 0.0334 0.0357 0.0390 0.0424 0.0529
    AD 0.1394 0.0324 0.0372 0.0378 0.0448 0.0487
    ^MH 5.2476 1.9006 1.9645 2.0490 2.1307 2.4028
    T=1000 ˆω Bias 0.0037 0.0000 0.0001 0.0003 0.0005 0.0007
    SD 0.0243 0.0074 0.0080 0.0086 0.0094 0.0115
    AD 0.0260 0.0064 0.0082 0.0091 0.0096 0.0131
    ˆα Bias 0.0013 0.0004 0.0006 0.0006 0.0010 0.0009
    SD 0.0459 0.0157 0.0167 0.0181 0.0190 0.0223
    AD 0.0431 0.0152 0.0169 0.0188 0.0201 0.0275
    ˆβ Bias 0.0114 0.0014 0.0017 0.0023 0.0030 0.0034
    SD 0.0816 0.0239 0.0263 0.0287 0.0312 0.0388
    AD 0.0830 0.0251 0.0278 0.0310 0.0326 0.0441
    ^MH 5.4240 1.8962 1.9615 2.0488 2.1387 2.4244
    T=1500 ˆω Bias 0.0027 0.0000 0.0000 0.0001 0.0001 0.0003
    SD 0.0199 0.0061 0.0065 0.0071 0.0078 0.0095
    AD 0.0146 0.0055 0.0057 0.0061 0.0072 0.0082
    ˆα Bias 0.0014 0.0002 0.0005 0.0004 0.0007 0.0001
    SD 0.0377 0.0126 0.0135 0.0143 0.0155 0.0186
    AD 0.0353 0.0113 0.0122 0.0132 0.0149 0.0174
    ˆβ Bias 0.0077 0.0002 0.0005 0.0007 0.0008 0.0009
    SD 0.0675 0.0195 0.0211 0.0232 0.0256 0.0316
    AD 0.0557 0.0201 0.0210 0.0223 0.0261 0.0302
    ^MH 5.4975 1.9082 1.9754 2.0684 2.1681 2.4631

     | Show Table
    DownLoad: CSV

    By comparing the MH values, it can be found that in the simulation examples considered, the optimal order of different volatility proxies is: RV1t>RV5t>RV10t>RV15t>RV30t>|yt|. Namely the estimator under RV1t shows the best performance. The simulation results show that the volatility proxies model with intraday high frequency data have a good effect on parameter estimation, which is helpful to improve the estimation accuracy of LGARCH model.

    In this section, model (3)–(4) is applied to study a real data set. We analyze SSE 50 Index (Shanghai Stock Exchange 50 Index) return series over the period April 28, 2008–July 24, 2013. The data set consists of the closing prices of the SSE 50 Index at one minute intervals, with a total of 1273 trading days and 240 observations per day. Denote the price sequence as {Pt(u),t[0,1273],u[0,1]}. Then the intraday return at u time on day t can be calculated as

    Yt(u)=[logPt(u)logPt1(1)]×100. (37)

    For demonstration, we plot Yt(1) in Figure 1.

    Figure 1.  Time series plot of {Yt(1)}1273t=1.

    To estimate the model, we choose realized volatility as the volatility proxy based on different intraday sampling frequency: 1-minute, 5-minute, 10-minute, 15-minute and 30-minute. Accordingly, the realized volatility of different frequencies are recorded as RV1, RV5, RV10, RV15 and RV30. Similar to the simulation part, for comparison, we also consider the volatility proxy Ht=|yt| which corresponds to the estimator without using high frequency data. Figure 2 shows the time series plots of different volatility proxies.

    Figure 2.  The time series of different volatility proxies.

    The LGARCH (1, 1) model estimation results among different Ht are given in Table 3. The AD in the table means the asymptotic variance of the estimated parameter which is calculated according to the formula (17) and (30).

    Table 3.  Parameter estimation of LGARCH (1, 1) model based on different volatility proxies.
    Ht ˆω ˆα ˆβ AD(ˆω) AD(ˆα) AD(ˆβ) ^MH ^Var(ε2t)
    |yt| 0.0220 0.0501 0.9499 0.0104 0.0144 0.0144 5.5530 3.5947
    RV1t 0.0836 0.1102 0.8685 0.0094 0.0083 0.0104 1.8320 0.3154
    RV5t 0.0239 0.1175 0.8996 0.0061 0.0099 0.0088 2.2553 0.5578
    RV10t 0.0311 0.1137 0.8980 0.0061 0.0091 0.0085 2.2245 0.4724
    RV15t 0.0217 0.1154 0.9024 0.0058 0.0097 0.0085 2.4282 0.5581
    RV30t 0.0326 0.1116 0.8986 0.0082 0.0121 0.0113 2.6533 0.8288

     | Show Table
    DownLoad: CSV

    It can be observed from Table 3 that the ^MH values of realized volatility are generally much smaller than that of |yt|, where RV1t is the smallest and the corresponding ^Var(ε2t) is also the smallest, which is consistent with the previous theoretical results. Compared with the estimation results of RV1t and |yt|, the asymptotic variance of ˆω, ˆα and ˆβ corresponding to RV1t are obviously smaller.

    From Table 3, when Ht=|yt|, the estimated LGARCH (1, 1) model is

    yt=htεt,ht=0.0220+0.0501|yt1|+0.9499ht1. (38)

    When Ht=RV1t, the fitting model is

    yt=htεt,ht=0.0836+0.1102|yt1|+0.8685ht1. (39)

    To further compare the estimation effect between |yt| and RV1t, the 95% confidence intervals of parameter estimators are calculated based on the AD in Table 3. Let ˆθL=(ˆωL,ˆαL,ˆβL) and ˆθU=(ˆωU,ˆαU,ˆβU) be the lower bound and the upper bound respectively. Then we can calculate the upper and lower bounds of ht as follows:

    hLt=ˆωL+ˆαL|yt1|+ˆβLhL,t1,hUt=ˆωU+ˆαU|yt1|+ˆβUhU,t1. (40)

    For comparison, we plot the computed ht, hLt and hUt based on equations (38) and (39) in Figure 3. It can be seen from Figure 3 that the estimated ht from two models are basically close. However, the interval [hLt,hUt] (circle) of model (39) is significantly narrower than that of model (38) (triangle).

    Figure 3.  Time series plots of ht, hLt and hUt: for model (38), ht (real line), hLt and hUt (triangle); for model (39), ht (dashed line), hLt and hUt (circle).

    It is of some interest to compare the performance between RV1t, RV5t, RV10t, RV15t and RV30t. Due to limitations of space, we only show the comparisons between RV1t and RV30t. We draw the 95% confidence intervals of ht calculated by the RV1t and RV30t in Figure 4. We can find that the confidence interval under RV1t is narrower than that under RV30t, which means that the model estimation effect is better with RV1t.

    Figure 4.  Time series plots of ht, hLt and hUt: for model under RV1t(real line, purple, ) and RV30t(dashed line, green, +).

    To sum, from the estimation results of Table 3 and the plots in Figure 3, it is shown that introducing the intraday high frequency data can help to improve the LGARCH model estimation for the considered data. Hence the proposed approach is of certain practical value.

    With the motivation that to obtain more precise estimators for the well-known daily LGARCH model, this paper studies how to introduce intraday high frequency data into the model estimation. Based on the existing volatility proxy idea, the estimation methods of all parameters of the model and the corresponding asymptotic properties are given.

    The selection criteria of the optimal volatility proxy is also proposed. The simulation results show that the parameter estimators perform well under finite samples. The empirical study based on SSE 50 index shows that using the high frequency data can significantly improve the estimation accuracy of daily frequency LGARCH model, compared to the usual method which only uses the daily data.

    The work is partially supported by National Natural Science Foundation of China 11731015, 11701116, Natural Science Foundation of Guangdong Province 192 2018A030310068, Innovative Team Project of Ordinary Universities in Guangdong Province 2020WCXTD018, Science and Technology Program of GuangZhou 202102020368 and Guangzhou University Research Funding YG2020029, YH202108.

    The authors declare no conflict of interest.



    [1] Bollerslev T (1986) Generalized autoregressive conditional heteroskedasticity. J Econ 31: 307–328. doi: 10.1016/0304-4076(86)90063-1
    [2] De Davide D (2019) Forecasting volatility using combination across estimation windows: An application to S & P500 stock market index. Math Biosci Eng 16: 7195–7216.
    [3] Deng CL, Zhang XF, Li Y, et al. (2020) Garch Model Test Using High-Frequency Data. Mathematics 8: 1922–1939. doi: 10.3390/math8111922
    [4] Engle RF (1982) Autoregressive conditional heteroscdeasticity with estimates of the variance of united kingdom inflation. Econometrica 50: 987–1007. doi: 10.2307/1912773
    [5] Fan P, Lan Y, Chen M, et al. (2017) The estimating method of VaR based on PGARCH model with high-frequency data. Syst Eng Theory Pract 37: 2052–2059.
    [6] Hentschel L (1995) All in the family Nesting symmetric and asymmetric GARCH models. J Financ Econ 39: 71–104. doi: 10.1016/0304-405X(94)00821-H
    [7] Linton O, Wu J (2020) A coupled component DCS-EGARCH model for intraday and overnight volatility. J Econ 217: 176–201. doi: 10.1016/j.jeconom.2019.12.015
    [8] Liang X, Zhang XF, Li Y, et al. (2021) Daily nonparametric ARCH(1) model estimation using intraday high frequency data. AIMS Math 6: 3455–3464. doi: 10.3934/math.2021206
    [9] Nelson DB (1991) Conditional Heteroskedasticity in Asset Returns: A New Approach. Econometrica 59: 347–370. doi: 10.2307/2938260
    [10] Duffie D, Pan J (1997) An Overview of Value at Risk. J Deriv 4: 7–49. doi: 10.3905/jod.1997.407971
    [11] Pan J, Wang H, Tong H (2008) Estimation and tests for power-transformed and threshold GARCH models. J Econ 142: 352–378. doi: 10.1016/j.jeconom.2007.06.004
    [12] Gyamerah SA (2019) Modelling the volatility of Bitcoin returns using GARCH models. Quant Financ Econ 3: 739–753. doi: 10.3934/QFE.2019.4.739
    [13] Visser MP (2011) GARCH parameter estimation using high-frequency data. J Financ Econom 9: 162–197. doi: 10.1093/jjfinec/nbq017
    [14] Wang M, Chen Z, Wang CD (2018) Composite quantile regression for GARCH models using high-frequency data. Econ Stat 7: 115–133.
    [15] Wu S, Feng M, Zhang H, et al. (2018) Quasi-maximum exponential likelihood estimation for non-stationary GARCH (1, 1) models with high-frequency data. Science in China 48: 443–456.
    [16] Xiao Z, Koenker R (2009) Conditional quantile estimation for generalized autoregressive conditional heteroscedasticity models. J Am Stat Assoc 104: 1696–1712. doi: 10.1198/jasa.2009.tm09170
    [17] Zou Y, Yu L, He K (2015). Estimating Portfolio Value at Risk in the Electricity Markets Using an Entropy Optimized BEMD Approach. Entropy 17: 4519–4532. doi: 10.3390/e17074519
    [18] Zhao B, Chen Z, Tao GP, et al. (2016) Composite quantile regression estimation for P-GARCH processes. Sci China Math 59: 977–998. doi: 10.1007/s11425-015-5115-0
  • Reader Comments
  • © 2021 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(2244) PDF downloads(67) Cited by(0)

Other Articles By Authors

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog