Estimation for seasonal fractional ARIMA with stable innovations via the empirical characteristic function method

Seasonal fractional ARIMA (ARFISMA) model with infinite variance innovations is used in the analysis of seasonal long-memory time series with large fluctuations (heavy-tailed distributions). Two methods, which are the empirical characteristic function (ECF) procedure developed by Knight and Yu [The empirical characteristic function in time series estimation. Econometric Theory. 2002;18:691–721] and the Two-Step method (TSM) are proposed to estimate the parameters of stable ARFISMA model. The ECF method estimates simultaneously all the parameters, while the TSM considers in the first step the Markov Chains Monte Carlo–Whittle approach introduced by Ndongo et al. [Estimation of long-memory parameters for seasonal fractional ARIMA with stable innovations. Stat Methodol. 2010;7:141–151], combined with the maximum likelihood estimation method developed by Alvarez and Olivares [Méthodes d'estimation pour des lois stables avec des applications en finance. Journal de la Société Française de Statistique. 2005;1(4):23–54] in the second step. Monte Carlo simulations are also used to evaluate the finite sample performance of these estimation techniques.     

Monte Carlo Maximum Likelihood Estimation for Generalized Long-Memory Time Series Models

An exact maximum likelihood method is developed for the estimation of parameters in a non-Gaussian nonlinear density function that depends on a latent Gaussian dynamic process with long-memory properties. Our method relies on the method of importance sampling and on a linear Gaussian approximating model from which the latent process can be simulated. Given the presence of a latent long-memory process, we require a modification of the importance sampling technique. In particular, the long-memory process needs to be approximated by a finite dynamic linear process. Two possible approximations are discussed and are compared with each other. We show that an autoregression obtained from minimizing mean squared prediction errors leads to an effective and feasible method. In our empirical study, we analyze ten daily log-return series from the S&P 500 stock index by univariate and multivariate long-memory stochastic volatility models. We compare the in-sample and out-of-sample performance of a number of models within the class of long-memory stochastic volatility models.     

On preliminary test and shrinkage estimation in linear models with long-memory errors

This paper discusses the problem of estimating a subset of parameters when the complementary subset is possibly redundant, in a linear regression model when the errors are generated from a long-memory process. Such a model arises due to the overmodelling of a situation involving long-memory data. Along with the classical least-squares estimator and restricted least-squares estimator, preliminary test least-squares estimator and shrinkage least-squares estimator are investigated in an asymptotic set-up and their relative performances are studied under contiguous alternatives. The contiguous alternatives under such dependence are fundamentally different from those under the independent errors case.     

Piecewise FARIMA models for long-memory time series

We consider the problem of modelling a long-memory time series using piecewise fractional autoregressive integrated moving average processes. The number as well as the locations of structural break points (BPs) and the parameters of each regime are assumed to be unknown. A four-step procedure is proposed to find out the BPs and to estimate the parameters of each regime. Its effectiveness is shown by Monte Carlo simulations and an application to real traffic data modelling is considered.     

Tests for structural breaks in memory parameters of long-memory heterogeneous autoregressive models

Tests for structural breaks in the coefficients of the long-memory heterogeneous autoregressive (HAR) models are developed. The tests are based on the partial sum process of the normalized efficient score vector. The tests have the nice property of identifying the parameters of the daily, weekly, and monthly regressors in which breaks occur. Limiting null distributions of the proposed tests are proven to be derived from standard Brownian bridges. A finite sample Monte-Carlo experiment shows reasonable size and power properties of the proposed tests. The proposed method is illustrated by a real data analysis.     

Frequency domain bootstrap for ratio statistics under long-range dependence

A frequency domain bootstrap (FDB) is a common technique to apply Efron’s independent and identically distributed resampling technique (Efron, 1979) to periodogram ordinates – especially normalized periodogram ordinates – by using spectral density estimates. The FDB method is applicable to several classes of statistics, such as estimators of the normalized spectral mean, the autocorrelation (but not autocovariance), the normalized spectral density function, and Whittle parameters. While this FDB method has been extensively studied with respect to short-range dependent time processes, there is a dearth of research on its use with long-range dependent time processes. Therefore, we propose an FDB methodology for ratio statistics under long-range dependence, using semi- and nonparametric spectral density estimates as a normalizing factor. It is shown that the FDB approximation allows for valid distribution estimation for a broad class of stationary, long-range (or short-range) dependent linear processes, without any stringent assumptions on the distribution of the underlying process. The results of a large simulation study show that the FDB approximation using a semi- or nonparametric spectral density estimator is often robust for various values of a long-memory parameter reflecting magnitude of dependence. We apply the proposed procedure to two data examples.     

Semiparametric estimation for seasonal long-memory time series using generalized exponential models

We consider a generalized exponential (GEXP) model in the frequency domain for modeling seasonal long-memory time series. This model generalizes the fractional exponential (FEXP) model [Beran, J., 1993. Fitting long-memory models by generalized linear regression. Biometrika 80, 817–822] to allow the singularity in the spectral density occurring at an arbitrary frequency for modeling persistent seasonality and business cycles. Moreover, the short-memory structure of this model is characterized by the Bloomfield [1973. An exponential model for the spectrum of a scalar time series. Biometrika 60, 217–226] model, which has a fairly flexible semiparametric form. The proposed model includes fractionally integrated processes, Bloomfield models, FEXP models as well as GARMA models [Gray, H.L., Zhang, N.-F., Woodward, W.A., 1989. On generalized fractional processes. J. Time Ser. Anal. 10, 233–257] as special cases. We develop a simple regression method for estimating the seasonal long-memory parameter. The asymptotic bias and variance of the corresponding long-memory estimator are derived. Our methodology is applied to a sunspot data set and an Internet traffic data set for illustration.     

Residual empirical processes for nearly unstable long-memory time series

This paper studies the goodness-of-fit test of the residual empirical process of a nearly unstable long-memory time series. Chan and Ling (2008) showed that the usual limit distribution of the Kolmogorov–Smirnov test statistics does not hold for an unstable autoregressive model. A key question of interest is what happens when this model has a near unit root, that is, when it is nearly unstable. In this paper, it is established that the statistics proposed by Chan and Ling can be generalized to encompass nearly unstable long-memory models. In particular, the limit distribution is expressed as a functional of an Ornstein–Uhlenbeck process that is driven by a fractional Brownian motion. Simulation studies demonstrate that the limit distribution of the statistic possesses desirable finite sample properties and power.     

Sieve bootstrapping the memory parameter in long-range dependent stationary functional time series

We consider a sieve bootstrap procedure to quantify the estimation uncertainty of long-memory parameters in stationary functional time series. We use a semiparametric local Whittle estimator to estimate the long-memory parameter. In the local Whittle estimator, discrete Fourier transform and periodogram are constructed from the first set of principal component scores via a functional principal component analysis. The sieve bootstrap procedure uses a general vector autoregressive representation of the estimated principal component scores. It generates bootstrap replicates that adequately mimic the dependence structure of the underlying stationary process. We first compute the estimated first set of principal component scores for each bootstrap replicate and then apply the semiparametric local Whittle estimator to estimate the memory parameter. By taking quantiles of the estimated memory parameters from these bootstrap replicates, we can nonparametrically construct confidence intervals of the long-memory parameter. As measured by coverage probability differences between the empirical and nominal coverage probabilities at three levels of significance, we demonstrate the advantage of using the sieve bootstrap compared to the asymptotic confidence intervals based on normality.

     

A multivariate long-memory model with structural breaks

This paper introduces a multivariate long-memory model with structural breaks. In the proposed framework, the time series exhibits possibly fractional orders of integration which are allowed to be different in each subsample. The break date is endogenously determined using a procedure that minimizes the residual sum of squares (RSS). Monte Carlo experiments show that this method for detecting breaks performs well in large samples. As an illustration, we estimate a trivariate VAR including prices, employment and GDP in both the US and Mexico. For the subsample preceding the break, our findings are similar to those of earlier studies based on a standard VAR approach in both the countries, such that the variables exhibit integer degrees of integration. On the contrary, the series is found to be fractionally integrated after the break, with the fractional differencing parameters being higher than one in the case of Mexico.     

Large Deviation Results for Statistics of Short- and Long-memory Gaussian Processes

This paper discusses the large deviation principle of several important statistics for short- and long-memory Gaussian processes. First, large deviation theorems for the log-likelihood ratio and quadratic forms for a short-memory Gaussian process with mean function are proved. Their asymptotics are described by the large deviation rate functions. Since they are complicated, they are numerically evaluated and illustrated using the Maple V system (Char et al ., 1991a,b). Second, the large deviation theorem of the log-likelihood ratio statistic for a long-memory Gaussian process with constant mean is proved. The asymptotics of the long-memory case differ greatly from those of the short-memory case. The maximum likelihood estimator of a spectral parameter for a short-memory Gaussian stationary process is asymptotically efficient in the sense of Bahadur.     

Random sampling of long-memory stationary processes

This paper investigates the second order properties of a stationary process after random sampling. While a short memory process gives always rise to a short memory one, we prove that long-memory can disappear when the sampling law has heavy enough tails. We prove that under rather general conditions the existence of the spectral density is preserved by random sampling. We also investigate the effects of deterministic sampling on seasonal long-memory.     

On non parametric statistical inference for densities under long-range dependence

Statistical inference for kernel estimators of the marginal density is considered for stationary processes with long-range dependence. The asymptotic behavior is known to differ sharply between small and large bandwidths. The statistical implications of this dichotomy have not been fully explored in the literature. The optimal rate and a functional limit theorem are obtained for large bandwidths, if the long-memory parameter exceeds a certain threshold. The threshold can be lowered arbitrarily close to the lower bound of the long-memory range. This result is extended to processes with infinite variance, and the construction of simultaneous finite-sample confidence bands is considered.     

Asymptotic expansion for nonparametric M-estimator in a nonlinear regression model with long-memory errors

We consider asymptotic expansion of the nonparametric M-estimator in a fixed-design nonlinear regression model when the errors are generated by long-memory linear processes. Under mild conditions, we show that the nonparametric M-estimator is first-order equivalent to the Nadaraya-Watson (NW) estimator, which implies that the nonparametric M-estimator has the same asymptotic distribution as that of the NW estimator. Furthermore, we study the second-order asymptotic expansion of the nonparametric M-estimator and show that the difference between the nonparametric M-estimator and the NW estimator has a limiting distribution after suitable standardization. The nature of the limiting distribution depends on the range of long-memory parameter α. We also compare the finite sample behavior of the two estimators through a numerical example when the errors are long-memory.     

ON ESTIMATION OF LONG-MEMORY TIME SERIES MODELS

This paper discusses estimation associated with the long-memory time series models proposed by Granger & Joyeux (1980) and Hosking (1981). We consider the maximum likelihood estimator and the least squares estimator. Certain regularity conditions introduced by several authors to develop the asymptotic theory of these estimators do not hold in this model. However we can show that these estimators are strongly consistent, and we derive the limiting distribution and the rate of convergence.     

A wavelet Whittle estimator of generalized long-memory stochastic volatility

We consider a k-GARMA generalization of the long-memory stochastic volatility model, discuss the properties of the model and propose a wavelet-based Whittle estimator for its parameters. Its consistency is shown. Monte Carlo experiments show that the small sample properties are essentially indistinguishable from those of the Whittle estimator, but are favorable with respect to a wavelet-based approximate maximum likelihood estimator. An application is given for the Microsoft Corporation stock, modeling the intraday seasonal patterns of its realized volatility.     

Long Memory in Foreign-Exchange Rates

Using the Geweke–Porter-Hudak test, we find evidence of long memory in exchange-rate data. This implies that the empirical evidence of unit roots in exchange rates may not be robust to long-memory alternatives. Fractionally integrated autoregressive moving average (ARFIMA) models are estimated by both the time-domain exact maximum likelihood (ML) method and the frequency-domain approximate ML method. Impulse-response functions and forecasts based on these estimated ARFIMA models are evaluated to gain insight into the long-memory characteristics of exchange rates. Some tentative explanations of the long memory found in the exchange rates are discussed.     

A bivariate integer-valued long-memory model for high-frequency financial count data

We propose a bivariate integer-valued fractional integrated (BINFIMA) model to account for the long-memory property and apply the model to high-frequency stock transaction data. The BINFIMA model allows for both positive and negative correlations between the counts. The unconditional and conditional first- and second-order moments are given. The model is capable of capturing the covariance between and within intra-day time series of high-frequency transaction data due to macroeconomic news and news related to a specific stock. Empirically, it is found that Ericsson B has mean recursive process while AstraZeneca has long-memory property.     

Two steps generalized maximum entropy estimation procedure for fitting linear regression when both covariates are subject to error

This paper presents a procedure utilizing the generalized maximum entropy (GME) estimation method in two steps to quantify the uncertainty of the simple linear structural measurement error model parameters exactly. The first step estimates the unknowns from the horizontal line, and then the estimates were used in a second step to estimate the unknowns from the vertical line. The proposed estimation procedure has the ability to minimize the number of unknown parameters in formulating the GME system within each step, and hence reduce variability of the estimates. Analytical and illustrative Monte Carlo simulation comparison experiments with the maximum likelihood estimators and a one-step GME estimation procedure were presented. Simulation experiments demonstrated that the two steps estimation procedure produced parameter estimates that are more accurate and more efficient than the classical estimation methods. An application of the proposed method is illustrated using a data set gathered from the Centre for Integrated Government Services in Delma Island – UAE to predict the association between perceived quality and the customer satisfaction.