| 965 | 5 | 204 |
| 下载次数 | 被引频次 | 阅读次数 |
本文首次提出信息不对称背景下成交量和波动率关系的日度完备模型。新模型假设当日成交量和已实现测度为次日隐信息流带来新信息,而隐信息流驱动知情交易者交易。基于此模型,本文给出时间序列平稳遍历性成立的充分条件,并且运用高斯拟极大似然法,建立模型参数估计的渐近理论,模拟研究验证了渐近理论的正确性。本文将新模型应用于中国沪深A股市场、美国纽交所和纳斯达克市场的大型公司。基于全样本的模型拟合结果发现:成交量的放大将导致次日条件波动率增大,而条件波动率的增大伴随着成交量的放大,同时,大部分个股中存在信息交易;在存续时间长、市值靠前且交易活跃的中国个股中,相较于基准波动率模型——已实现GARCH模型,新模型大多具有更好的样本内拟合能力;而在市值靠前且交易活跃的美股中,新模型的表现并不优于已实现GARCH模型。在滚动窗口的波动率预测上,较其他被广泛使用的已有模型,新模型在本文研究的中国个股中普遍具有更好的表现;而在本文研究的美股中,已实现GARCH模型表现较好。这反映了新模型更加适用于存续时间长、市值靠前且交易活跃的中国股票市场数据。
Abstract:We propose a realized Mixture of Distribution Hypothesis model for volume-volatility relation under information asymmetry. The model hypothesizes that the trading volume of the day and the realized measurement provides new information for the second-day potential information flow, which drives informed trading. Sufficient conditions for the stationarity and ergodicity of the time series under our modelling framework are provided. A Gaussian quasi-maximum likelihood estimation method for parameter estimation is proposed and its asymptotic theory is established. Simulation results corroborate our theoretical findings. The proposed model and method are applied to both the U.S. and China stock data.Empirical results show that the increase of trading volume will lead to the increase of the second-day conditional volatility, and that our model fits the data of large-cap, long listed and highly liquid stocks from the Chinese market better than the realized GARCH model as the benchmark volatility model. We find that information trading exists in most stocks. Moreover, our model outperforms widely used existing models in volatility forecasting when applied to Chinese stocks, while the realized GARCH model performs better in U.S. stocks. In summary, our model is more appropriate for describing data of large-cap, long listed and highly liquid stocks from the Chinese market than existing models.
[1]孙彦林,陈守东,刘洋.基于股市和汇市成交量信息视角的股价波动预测[J].系统工程理论与实践,2019,39(4):935-945.
[2]于孝建,王秀花.基于混频已实现GARCH模型的波动预测与Va R度量[J].统计研究,2018,35(1):104-116.
[3]Ahern Kenneth R.Do Proxies for Informed Trading Measure Informed Trading?Evidence from Illegal Insider Trades[J].The Review of Asset Pricing Studies,2020,10(3):397-440.
[4]Ahmad A,Francq C.Poisson QMLE of Count Time Series Models[J].Journal of Time Series Analysis,2016,37(3):291-314.
[5]Andersen T G.Return Volatility and Trading Volume:An Information Flow Interpretation of Stochastic Volatility[J].The Journal of Finance,1996,51(1):169-204.
[6]Andersen T G,Bollerslev T.Answering the Skeptics:Yes,Standard Volatility Models Do Provide Accurate Forecasts[J].International Economic Review,1998:885-905.
[7]Andersen T G,Bollerslev T,Diebold F X,et al.Modeling and Forecasting Realized Volatility[J].Econometrica,2003,71(2):579-625.
[8]Bollerslev T.Generalized Autoregressive Conditional Heteroskedasticity[J].Journal of Econometrics,1986,31(3):307-327.
[9]Bollerslev T,Li J,Xue Y.Volume,Volatility,and Public News Announcements[J].The Review of Economic Studies,2018,85(4):2005-2041.
[10]Brooks C.Introductory Econometrics for Finance[M].Cambridge University Press,2019.
[11]Chiang T C,Qiao Z,Wong W K.New Evidence on the Relation Between Return Volatility and Trading Volume[J].Journal of Forecasting,2010,29(5):502-515.
[12]Clark P K.A Subordinated Stochastic Process Model with Finite Variance for Speculative Prices[J].Econometrica:Journal of the Econometric Society,1973:135-155.
[13]Corsi F.A Simple Approximate Long-Memory Model of Realized Volatility[J].Journal of Financial Econometrics,2009,7(2):174-196.
[14]Engle R.New Frontiers for ARCH Models[J].Journal of Applied Econometrics,2002,17(5):425-446.
[15]Engle R F,Gallo G M.A Multiple Indicators Model for Volatility Using Intra-Daily Data[J].Journal of Econometrics,2006,131(1-2):3-27.
[16]Epps T W,Epps M L.The Stochastic Dependence of Security Price Changes and Transaction Volumes:Implications for the Mixture-ofDistributions Hypothesis[J].Econometrica:Journal of the Econometric Society,1976:305-321.
[17]Ferland R,Latour A,Oraichi D.Integer‐Valued GARCH Process[J].Journal of Time Series Analysis,2006,27(6):923-942.
[18]Fokianos K,Rahbek A,Tj?stheim D.Poisson Autoregression[J].Journal of the American Statistical Association,2009,104(488):1430-1439.
[19]Fokianos K,Tj?stheim D.Log-Linear Poisson Autoregression[J].Journal of Multivariate Analysis,2011,102(3):563-578.
[20]Fuertes A M,Izzeldin M,Kalotychou E.On Forecasting Daily Stock Volatility:The Role of Intraday Information and Market Conditions[J].International Journal of Forecasting,2009,25(2):259-281.
[21]Glosten L R,Milgrom P R.Bid,Ask and Transaction Prices in a Specialist Market with Heterogeneously Informed Traders[J].Journal of Financial Economics,1985,14(1):71-100.
[22]Hansen P R,Huang Z.Exponential GARCH Modeling with Realized Measures of Volatility[J].Journal of Business&Economic Statistics,2016,34(2):269-287.
[23]Hansen P R,Huang Z,Shek H H.Realized GARCH:a Joint Model for Returns and Realized Measures of Volatility[J].Journal of Applied Econometrics,2012,27(6):877-906.
[24]Koubaa Y,Slim S.The Relationship Between Trading Activity and Stock Market Volatility:Does the Volume Threshold Matter?[J].Economic Modelling,2019,82:168-184.
[25]Neumann M H.Absolute Regularity and Ergodicity of Poisson Count Processes[J].Bernoulli,2011,17(4):1268-1284.
[26]Pitt M K,Shephard N.Filtering Via Simulation:Auxiliary Particle Filters[J].Journal of the American statistical association,1999,94(446):590-599.
[27]Shephard N,Sheppard K.Realising the Future:Forecasting with High-Frequency-Based Volatility (HEAVY) Models[J].Journal of Applied Econometrics,2010,25(2):197-231.
[28]Straumann D,Mikosch T.Quasi-Maximum-Likelihood Estimation in Conditionally Heteroscedastic Time Series:a Stochastic Recurrence Equations Approach[J].The Annals of Statistics,2006,34(5):2449-2495.
[29]Suominen M.Trading Volume and Information Revelation in Stock Market[J].Journal of Financial and Quantitative Analysis,2001,36(4):545-565.
[30]Tauchen G E,Pitts M.The Price Variability-Volume Relationship on Speculative Markets[J].Econometrica:Journal of the Econometric Society,1983:485-505.
(1)经过推导,可知已实现MDH模型方程组中vt的k阶自协方差■为■。特别地,若k=1,则■,其中,■。当k≥2时,■就无法被表示为模型参数和t kh-自身的方差及协方差的函数形式。
(1)值得注意的是,本文提出的已实现MDH模型中采用了“GARCH类”模型的建模思路,隐信息流仅包含上一期的信息,这有别于经典MDH模型采用的“随机波动率类”模型的建模方法,其中隐信息流包含了当期新息(Innovations),并由包含新息的隐信息流方程(而非GARCH方程)描述。Ferland等(2006),Fokianos等(2009),Fokianos和Tj?stheim(2011),Neumann(2011)对本文使用的这一“GARCH类”泊松自回归模型进行了研究。当然,在“随机波动率类”模型下讨论收益率、成交量、已实现测度的联合建模及其参数估计理论无疑也是一个重要问题(已有孙彦林等(2019)在不考虑已实现测度的情形下讨论了股市汇市两市成交量在股价波动率预测中的作用),将留待后续研究,而在本文中仅关注“GARCH类”建模方法。
(1)因篇幅所限,证明过程以附录展示,见《统计研究》网站所列附件。下同。
(1)R软件stats包中的nlminb函数是一个采用半牛顿法的优化器。Ahmad和Francq(2016)研究泊松拟极大似然估计量的渐近性质的估计方法适用于Neumann(2011)的整数GARCH模型,同时适用于真实参数取参数空间内点和边界点的情况,但其框架中不包含随机冲击项tw。在模拟中,Ahmad和Francq(2016)使用了和本文同样的优化器。此外,由于本文的目标函数形式较为复杂,故将函数的最大估计次数设置为5000,最大迭代次数设置为2000,提升了结果的收敛概率。
(1)本文经以下步骤得到经验初始值。首先,本文使用对数线性已实现GARCH(1,1)模型对实际数据进行回归,得到条件波动率的估计■,并将估计出的■作为对应参数的估计量。用tv的方差除以tv的均值,作为c的估计。其次,本文进行线性回归,其中因变量为■,自变量为■,■和■,得到■的估计。最后,本文做因变量为■,自变量为■的线性回归,得到■的估计。至此本文得到所有参数初始值的估计,该初始值需要做如下处理。首先,如果模拟中该初始值太接近真实值,则算法未必能够收敛。Brooks(2019)也指出,在最大值的附近,似然函数过于平坦,使得算法不易收敛到真实值。为了解决这一问题,本文将该初始值压缩至原来的二分之一。其次,压缩后初始值可能在优化函数中规定的上、下确界之外,对此本文使用表1中对应的先验初始值点进行替换,最终得到经验初始值。为了增加参数估计的稳健性,本文在模拟和实证中同时考虑先验初始值和经验初始值,对比二者的似然函数大小并选取使得似然函数更大的估计。
(2)因篇幅所限,本小节模拟参数设置与结果以附表1~2展示。
(1)借鉴Hansen和Huang(2016)的做法,本节剔除高频观测缺失率高于10%的交易日。比如,中国市场一天有48个5分钟观测,如果该交易日有效观测数小于等于43个,那么删去该交易日数据。中国个股可能会在一些大事件前停牌,比如重大资产重组,停牌当天无交易数据。
(2)因篇幅所限,已实现MDH模型回归结果以附表3~4展示。
(1)在Hansen等(2012)文献表2中,通过对比已实现GARCH模型的“GARCH型”偏似然和GARCH模型的似然来比较两个模型在样本内外的拟合效果,其想法是已实现GARCH模型对数似然可分解为“GARCH型”偏似然加上另一项条件似然,而“GARCH型”偏似然与GARCH模型似然具有相同形式,因而可比。本文使用类似做法,注意到通过简单计算,已实现MDH对数似然函数可以有以下分解:■这里,■为已实现MDH模型的“已实现GARCH型”偏似然,其与已实现GARCH模型的似然函数形式一样,因而可以此作为标准来对比两模型的样本拟合效果。
基本信息:
DOI:10.19343/j.cnki.11-1302/c.2023.03.008
中图分类号:F832.51;F224
引用信息:
[1]彭烨,张志远.信息不对称下成交量与波动率关系建模与统计推断[J].统计研究,2023,40(03):100-113.DOI:10.19343/j.cnki.11-1302/c.2023.03.008.
基金信息:
国家自然科学基金面上项目“金融高频大数据下的风险推断及其与多元标的衍生品定价和金融风险管理的交叉融合研究”(71871132); 国家自然科学基金委重大研究计划重点项目“金融大数据统计推断理论与方法及应用研究”(91546202); 中央高校基本科研业务费(批准号:CXJJ-2019-412)专项资金资助; 部分受到上海市数据科技与决策前沿科学研究基地(Shanghai Research Center for Data Science and Decision Technology)资助
2023-03-25
2023-03-25