Why Aggregate Long Memory Time Series?

This article shows that, for large samples, temporally aggregating a true long memory time series (in order to get an improved estimator) may make little or no sense, as the practitioner can get virtually the same estimates as those from the aggregated series by choosing the appropriate bandwidths on the original one, provided some fairly general conditions apply. Besides, the practitioner has a wider choice of bandwidths than she has of aggregating levels. However, these results apply only to two specific and commonly used estimators, and do not apply to the aggregation procedure undertaken to compute the realized volatility. Also, aggregating a time series in order to test true versus spurious long memory (as in Ohanissian et al., 2008) is a relevant issue, particularly regarding stochastic and/or realized volatility, as many nonlinear processes display spurious long memory where the above result does not apply.     

Long memory in intertrade durations,counts and realized volatility of NYSE stocks

We study the persistence of intertrade durations, counts (number of transactions in equally spaced intervals of clock time), squared returns and realized volatility in 10 stocks trading on the New York Stock Exchange. A semiparametric analysis reveals the presence of long memory in all of these series, with potentially the same memory parameter. We introduce a parametric latent-variable long-memory stochastic duration (LMSD) model which is shown to better fit the data than the autoregressive conditional duration model (ACD) in a variety of ways. The empirical evidence we present here is in agreement with theoretical results on the propagation of memory from durations to counts and realized volatility presented in Deo et al. (2009).     

Some simulations and applications of forecasting long-memory time-series models

In this paper, we show some results of forecasting based on the ARFIMA(p,d,q) and ARIMA(p,d,q) models. We show, by simulation, that the technique of forecasting of the ARIMA(p,d,q) model can also be used when d is fractional, i.e., for the ARFIMA(p,d,q) model. We also conduct a simulation study to compare the two estimators of d obtained through regression methods. They are used in the hypothesis test to decide whether or not the series has long memory property and are compared on the basis of their k-step ahead forecast errors. The properties of long-memory models are also investigated using an actual set of data.     

Refined Inference on Long Memory in Realized Volatility

There is an emerging consensus in empirical finance that realized volatility series typically display long range dependence with a memory parameter (d) around 0.4 (Andersen et al., 2001 Andersen , T. G. , Bollerslev , T. , Diebold , F. X. , Labys , P. ( 2001 ). The distribution of realized exchange rate volatility . Journal of the American Statistical Association 96 ( 453 ): 42 – 55 .[Taylor & Francis Online], [Web of Science ®] , [Google Scholar]; Martens et al., 2004 Martnes , M. , Van Dijk , D. , De Pooter , M. ( 2004 ). Modeling and forecasting S&P 500 volatility: Long memory, structural breaks and nonlinearity. Tinbergen Institute Discussion Paper 2004-067/4 . [Google Scholar]). The present article provides some illustrative analysis of how long memory may arise from the accumulative process underlying realized volatility. The article also uses results in Lieberman and Phillips (2004 Lieberman , O. , Phillips , P. C. B. ( 2004 ). Expansions for the distribution of the maximum likelihood estimator of the fractional difference parameter . Econometric Theory 20 ( 3 ): 464 – 484 . [Google Scholar], 2005 Lieberman , O. , Phillips , P. C. B. ( 2005 ). Expansions for approximate maximum likelihood estimators of the fractional difference parameter . The Econometrics Journal 8 : 367 – 379 . [Google Scholar]) to refine statistical inference about d by higher order theory. Standard asymptotic theory has an O(n ^?1/2) error rate for error rejection probabilities, and the theory used here refines the approximation to an error rate of o(n ^?1/2). The new formula is independent of unknown parameters, is simple to calculate and user-friendly. The method is applied to test whether the reported long memory parameter estimates of Andersen et al. (2001 Andersen , T. G. , Bollerslev , T. , Diebold , F. X. , Labys , P. ( 2001 ). The distribution of realized exchange rate volatility . Journal of the American Statistical Association 96 ( 453 ): 42 – 55 .[Taylor & Francis Online], [Web of Science ®] , [Google Scholar]) and Martens et al. (2004 Martnes , M. , Van Dijk , D. , De Pooter , M. ( 2004 ). Modeling and forecasting S&P 500 volatility: Long memory, structural breaks and nonlinearity. Tinbergen Institute Discussion Paper 2004-067/4 . [Google Scholar]) differ significantly from the lower boundary (d = 0.5) of nonstationary long memory, and generally confirms earlier findings.     

Predicting integrals of diffusion processes

Consider predicting the integral of a diffusion process Z in a bounded interval A, based on the observations Z(t_1n),…,Z(t_nn), where t_1n,…,t_nn is a dense triangular array of points (the step of discretization tends to zero as n increases) in the bounded interval. The best linear predictor is generally not asymptotically optimal. Instead, we predict using the conditional expectation of the integral of the diffusion process, the optimal predictor in terms of minimizing the mean squared error, given the observed values of the process. We obtain that, conditioning on the observed values, the order of convergence in probability to zero of the mean squared prediction error is O_p(n⁻²). We prove that the standardized conditional prediction error is approximately Gaussian with mean zero and unit variance, even though the underlying diffusion is generally non-Gaussian. Because the optimal predictor is hard to calculate exactly for most diffusions, we present an easily computed approximation that is asymptotically optimal. This approximation is a function of the diffusion coefficient.     

Nonlinearity,Breaks, and Long-Range Dependence in Time-Series Models

We study the simultaneous occurrence of long memory and nonlinear effects, such as parameter changes and threshold effects, in time series models and apply our modeling framework to daily realized measures of integrated variance. We develop asymptotic theory for parameter estimation and propose two model-building procedures. The methodology is applied to stocks of the Dow Jones Industrial Average during the period 2000 to 2009. We find strong evidence of nonlinear effects in financial volatility. An out-of-sample analysis shows that modeling these effects can improve forecast performance. Supplementary materials for this article are available online.     

Likelihood ratio ordering of order statistics

This paper provides an improvement on the work of Bapat and Kochar (1994, Linear Algebra Appl., 199, 281–291) and strengthens the literature on the likelihood ratio ordering of order statistics. For independent (but possibly nonidentically distributed) absolutely continuous random variables X₁,…,X_n, it is shown under some weak conditions that

X_1:n^lrlrX_n:n,

where ^lr stands for the likelihood ratio ordering and X_k:n represents the kth-order statistic.     

Estimation of Long Memory in Integrated Variance

A stylized fact is that realized variance has long memory. We show that, when the instantaneous volatility is a long memory process of order d, the integrated variance is characterized by the same long-range dependence. We prove that the spectral density of realized variance is given by the sum of the spectral density of the integrated variance plus that of a measurement error, due to the sparse sampling and market microstructure noise. Hence, the realized volatility has the same degree of long memory as the integrated variance. The additional term in the spectral density induces a finite-sample bias in the semiparametric estimates of the long memory. A Monte Carlo simulation provides evidence that the corrected local Whittle estimator of Hurvich et al. (2005 Hurvich , C. M. , Moulines , E. , Soulier , P. ( 2005 ). Estimating long memory in volatility . Econometrica 73 ( 4 ): 1283 – 1328 .[Crossref], [Web of Science ®] , [Google Scholar]) is much less biased than the standard local Whittle estimator and the empirical application shows that it is robust to the choice of the sampling frequency used to compute the realized variance. Finally, the empirical results suggest that the volatility series are more likely to be generated by a nonstationary fractional process.     

Realized Volatility and Long Memory: An Overview

The challenge of modeling, estimating, testing, and forecasting financial volatility is both intellectually worthwhile and also central to the successful analysis of financial returns and optimal investment strategies. In each of the three primary areas of volatility modeling, namely, conditional (or generalized autoregressive conditional heteroskedasticity) volatility, stochastic volatility and realized volatility (RV), numerous univariate volatility models of individual financial assets and multivariate volatility models of portfolios of assets have been established. This special issue has eleven innovative articles, eight of which are focused directly on RV and three on long memory, while two are concerned with both RV and long memory.     

Realized Volatility and Long Memory: An Overview

The challenge of modeling, estimating, testing, and forecasting financial volatility is both intellectually worthwhile and also central to the successful analysis of financial returns and optimal investment strategies. In each of the three primary areas of volatility modeling, namely, conditional (or generalized autoregressive conditional heteroskedasticity) volatility, stochastic volatility and realized volatility (RV), numerous univariate volatility models of individual financial assets and multivariate volatility models of portfolios of assets have been established. This special issue has eleven innovative articles, eight of which are focused directly on RV and three on long memory, while two are concerned with both RV and long memory.     

Comparison of non-parametric and semi-parametric tests in detecting long memory

The first two stages in modelling times series are hypothesis testing and estimation. For long memory time series, the second stage was studied in the paper published in [M. Boutahar et al., Estimation methods of the long memory parameter: monte Carlo analysis and application, J. Appl. Statist. 34(3), pp. 261–301.] in which we have presented some estimation methods of the long memory parameter. The present paper is intended for the first stage, and hence completes the former, by exploring some tests for detecting long memory in time series. We consider two kinds of tests: the non-parametric class and the semi-parametric one. We precise the limiting distribution of the non-parametric tests under the null of short memory and we show that they are consistent against the alternative of long memory. We perform also some Monte Carlo simulations to analyse the size distortion and the power of all proposed tests. We conclude that for large sample size, the two classes are equivalent but for small sample size the non-parametric class is better than the semi-parametric one.     

LOG-PERIODOGRAM ESTIMATION OF LONG MEMORY VOLATILITY DEPENDENCIES WITH CONDITIONALLY HEAVY TAILED RETURNS

Many recent papers have used semiparametric methods, especially the log-periodogram regression, to detect and estimate long memory in the volatility of asset returns. In these papers, the volatility is proxied by measures such as squared, log-squared, and absolute returns. While the evidence for the existence of long memory is strong using any of these measures, the actual long memory parameter estimates can be sensitive to which measure is used. In Monte-Carlo simulations, I find that if the data is conditionally leptokurtic, the log-periodogram regression estimator using squared returns has a large downward bias, which is avoided by using other volatility measures. In United States stock return data, I find that squared returns give much lower estimates of the long memory parameter than the alternative volatility measures, which is consistent with the simulation results. I conclude that researchers should avoid using the squared returns in the semiparametric estimation of long memory volatility dependencies.     

LOG-PERIODOGRAM ESTIMATION OF LONG MEMORY VOLATILITY DEPENDENCIES WITH CONDITIONALLY HEAVY TAILED RETURNS

ABSTRACT

Many recent papers have used semiparametric methods, especially the log-periodogram regression, to detect and estimate long memory in the volatility of asset returns. In these papers, the volatility is proxied by measures such as squared, log-squared, and absolute returns. While the evidence for the existence of long memory is strong using any of these measures, the actual long memory parameter estimates can be sensitive to which measure is used. In Monte-Carlo simulations, I find that if the data is conditionally leptokurtic, the log-periodogram regression estimator using squared returns has a large downward bias, which is avoided by using other volatility measures. In United States stock return data, I find that squared returns give much lower estimates of the long memory parameter than the alternative volatility measures, which is consistent with the simulation results. I conclude that researchers should avoid using the squared returns in the semiparametric estimation of long memory volatility dependencies.     

Maximum likelihood estimation of stationary multivariate ARFIMA processes

This article considers the maximum likelihood estimation (MLE) of a class of stationary and invertible vector autoregressive fractionally integrated moving-average (VARFIMA) processes considered in Equation (26) of Luceño [A fast likelihood approximation for vector general linear processes with long series: Application to fractional differencing, Biometrika 83 (1996), pp. 603–614] or Model A of Lobato [Consistency of the averaged cross-periodogram in long memory series, J. Time Ser. Anal. 18 (1997), pp. 137–155] where each component y _{i, t} is a fractionally integrated process of order d _i , i=1, …, r. Under the conditions outlined in Assumption 1 of this article, the conditional likelihood function of this class of VARFIMA models can be efficiently and exactly calculated with a conditional likelihood Durbin–Levinson (CLDL) algorithm proposed herein. This CLDL algorithm is based on the multivariate Durbin–Levinson algorithm of Whittle [On the fitting of multivariate autoregressions and the approximate canonical factorization of a spectral density matrix, Biometrika 50 (1963), pp. 129–134] and the conditional likelihood principle of Box and Jenkins [Time Series Analysis, Forecasting, and Control, 2nd ed., Holden-Day, San Francisco, CA]. Furthermore, the conditions in the aforementioned Assumption 1 are general enough to include the model considered in Andersen et al. [Modeling and forecasting realized volatility, Econometrica 71 (2003), 579–625] for describing the behaviour of realized volatility and the model studied in Haslett and Raftery [Space–time modelling with long-memory dependence: Assessing Ireland's wind power resource, Appl. Statist. 38 (1989), pp. 1–50] for spatial data as its special cases. As the computational cost of implementing the CLDL algorithm is much lower than that of using the algorithms proposed in Sowell [Maximum likelihood estimation of fractionally integrated time series models, Working paper, Carnegie-Mellon University], we are thus able to conduct a Monte Carlo experiment to investigate the finite sample performance of the CLDL algorithm for the 3-dimensional VARFIMA processes with the sample size of 400. The simulation results are very satisfactory and reveal the great potentials of using the CLDL method for empirical applications.     

A looser cointegration concept using fractional integration parameters and quantification of market responsiveness

Many economic variables are fractionally integrated of order d, FI(d) with unequal d's. For modeling their long-run equilibria, we explain why the usual cointegration fails to exist and the unit root type tests have low power. Hence, we propose a looser concept called “tie integration”. A new numerical minimization problem reveals the value of d in the absence of tie integration, denoted by d_null. We use the d from residuals of a regression, as well as, d_null to devise a new index called strength of tie (SOT). An application quantifies market responsiveness.     

On the non-existence of generalised Hadamard matrices

The determinant of a generalized Hadamard matrix over its group ring factored out by the relation Σ_gεG G = 0 is shown to have certain number theoretic properties. These are exploited to prove the non-existence of many generalised Hadamard matrices for groups whose orders are divisible by 3, 5 or 7. For example the GH(15, C₁₅), GH(15, C₃) and GH(15, C₅) do not exist. Also for certain n and G we find the set of determinants of the GH(n, G) matrices.     

Search with small sets in presence of a liar

One unknown element of an n-element set is sought by asking if it is contained in given subsets. It is supposed that the question sets are of size at most k and all the questions are decided in advance, the choice of the next question cannot depend on previous answers. At most l of the answers can be incorrect. The minimum number of such questions is determined when the order of magnitude of k is n with <1. The problem can be formulated as determination of the maximum sized l-error-correcting code (of length n) in which the number of ones in a given position is at most k.     

Adaptive wavelet empirical Bayes estimation of a location or a scale parameter

Assume that in independent two-dimensional random vectors (X₁,θ₁),…,(X_n,θ_n), each θ_i is distributed according to some unknown prior density function g. Also, given θ_i=θ, X_i has the conditional density function q(x−θ), x,θ(−∞,∞) (a location parameter case), or θ⁻¹q(x/θ), x,θ(0,∞) (a scale parameter case). In each pair the first component is observable, but the second is not. After the (n+1)th pair (X_n+1,θ_n+1) is obtained, the objective is to construct an empirical Bayes (EB) estimator of θ. In this paper we derive the EB estimators of θ based on a wavelet approximation with Meyer-type wavelets. We show that these estimators provide adaptation not only in the case when g belongs to the Sobolev space H with an unknown , but also when g is supersmooth.     

基于混频已实现GARCH模型的波动预测与VaR度量

本文将Hansen等（2012）的Realized GARCH模型扩展为包含日内收益率、日收益率以及已实现波动率的混频已实现GARCH模型（M-Realized GARCH模型）。该模型将日内交易分为前后两段,引入了混频均值方程,并对混频均值方程的残差分别建立条件波动率方程和已实现日波动率方程。本文采用2013-2016年沪深300指数混频数据,分别在扰动项服从正态分布、t分布和广义误差分布的假设下,采用损失函数、SPA检验、kupiec检验和动态分位数检验法,对GARCH、Realized GARCH和M-Realized GARCH模型的波动率预测和VaR度量效果对比研究,得出M-Realized GARCH模型能提高预测精度,且VaR实际失败率与理论失败率一致,失败发生之间不相关。最后,本文利用Block bootstrap方法抽样得到混频数据,模拟证明了M-Realized GARCH模型比Realized GARCH模型具有更高的预测精度。