
on Econometric Time Series 
By:  Clifford Hurvich (New York University USA); Eric Moulines (ENST, Paris, France); Philippe Soulier (Universite Paris X, France) 
Abstract:  We consider semiparametric estimation of the memory parameter in a model which includes as special cases both the longmemory stochastic volatility (LMSV) and fractionally integrated exponential GARCH (FIEGARCH) models. Under our general model the logarithms of the squared returns can be decomposed into the sum of a longmemory signal and a white noise. We consider periodogrambased estimators using a local Whittle criterion function. We allow the optional inclusion of an additional term to account for possible correlation between the signal and noise processes, as would occur in the FIEGARCH model. We also allow for potential nonstationarity in volatility, by allowing the signal process to have a memory parameter d^* >= 1/2. We show that the local Whittle estimator is considtent for d^* in (0,1). We also show that the local Whittle estimator is asymptotically normal for d^* in (0,3/4) and asymptotically recovers the optimal semiparametric rate of convergence for this problem. In particular, if the spectral density of the short memory component of the signal is sufficiently smooth, a convergence rate of n^{2/5\delta} for d^* in (0,3/4) can be attained, where n is the sample size and \delta > 0 is arbitrarily small. This represents a strong improvement over the performance of existing semiparametric estimators of persistence in volatility. We also prove that the standard Gaussian semiparametric estimator is asymptotically normal if d^*=0. This yields a test for long memory in volatility. 
Keywords:  LMSV, FIEGARCH 
JEL:  C1 C2 C3 C4 C5 C8 
Date:  2004–12–14 
URL:  http://d.repec.org/n?u=RePEc:wpa:wuwpem:0412006&r=ets 
By:  Mengchen Hsieh (New York University); Clifford Hurvich (New York University); Philippe Soulier (Universite Paris X) 
Abstract:  We consider processes with second order long range dependence resulting from heavy tailed durations. We refer to this phenomenon as duration driven long range dependence (DDLRD), as opposed to the more widely studied linear long range dependence based on fractional differencing of an $iid$ process. We consider in detail two specific processes having DDLRD, originally presented in Taqqu and Levy (1986), and Parke (1999). For these processes, we obtain the limiting distribution of suitably standardized discrete Fourier transforms (DFTs) and sample autocovariances. At low frequencies, the standardized DFTs converge to a stable law, as do the standardized autocovariances at fixed lags. Finite collections of standardized autocovariances at a fixed set of lags converge to a degenerate distribution. The standardized DFTs at high frequencies converge to a Gaussian law. Our asymptotic results are strikingly similar for the two DDLRD processes studied. We calibrate our asymptotic results with a simulation study which also investigates the properties of the semiparametric log periodogram regression estimator of the memory parameter. 
Keywords:  Long Memory; Structural Change 
JEL:  C1 C2 C3 C4 C5 C8 
Date:  2004–12–15 
URL:  http://d.repec.org/n?u=RePEc:wpa:wuwpem:0412009&r=ets 
By:  Willa Chen (Texas A&M University); Clifford Hurvich (New York University) 
Abstract:  We consider a common components model for multivariate fractional cointegration, in which the s>=1 components have different memory parameters. The cointegrating rank is allowed to exceed 1. The true cointegrating vectors can be decomposed into orthogonal fractional cointegrating subspaces such that vectors from distinct subspaces yield cointegrating errors with distinct memory parameters, denoted by d_k for k=1,...,s. We estimate each cointegrating subsspace separately using appropriate sets of eigenvectors of an averaged periodogram matrix of tapered, differenced observations. The averaging uses the first m Fourier frequencies, with m fixed. We will show that any vector in the k'th estimated coingetraging subspace is, with high probability, close to the k'th true cointegrating subspace, in the sense that the angle between the estimated cointegrating vector and the true cointegrating subspace converges in probability to zero. The angle is O_p(n^{ \alpha_k}), where n is the sample size and \alpha_k is the shortest distance between the memory parameters corresponding to the given and adjacent subspaces. We show that the cointegrating residuals corresponding to an estimated cointegrating vector can be used to obtain a consistent and asymptotically normal estimate of the memory parameter for the given cointegrating subspace, using a univariate Gaussian semiparametric estimator with a bandwidth that tends to \infty more slowly than n. We also show how these memory parameter estimates can be used to test for fractional cointegration and to consistently identify the cointegrating subspaces. 
Keywords:  Fractional Cointegration; Long Memory; Tapering; Periodogram 
JEL:  C1 C2 C3 C4 C5 C8 
Date:  2004–12–15 
URL:  http://d.repec.org/n?u=RePEc:wpa:wuwpem:0412007&r=ets 
By:  JS Armstrong (The Wharton School  University of Pennsylvania); Fred Collopy (Case Western Reserve University) 
Abstract:  We consider how judgment and statistical methods should be integrated for timeseries forecasting. Our review of published empirical research identified 47 studies, all but four published since 1985. Five procedures were identified: revising judgment; combining forecasts; revising extrapolations; rulebased forecasting; and econometric forecasting. This literature suggests that integration generally improves accuracy when the experts have domain knowledge and when significant trends are involved. Integration is valuable to the extent that judgments are used as inputs to the statistical methods, that they contain additional relevant information, and that the integration scheme is well structured. The choice of an integration approach can have a substantial impact on the accuracy of the resulting forecasts. Integration harms accuracy when judgment is biased or its use is unstructured. Equalweights combining should be regarded as the benchmark and it is especially appropriate where series have high uncertainty or high instability. When the historical data involve high uncertainty or high instability, we recommend revising judgment, revising extrapolations, or combining. When good domain knowledge is available for the future as well as for the past, we recommend rule based forecasting or econometric methods. 
Keywords:  statistical methods, statistics, time series, forecasting, empirical research 
JEL:  A 
Date:  2004–12–10 
URL:  http://d.repec.org/n?u=RePEc:wpa:wuwpgt:0412024&r=ets 
By:  Yakov Amihud (New York University); Clifford Hurvich (New York University) 
Abstract:  Standard predictive regressions produce biased coefficient estimates in small samples when the regressors are Gaussian firstorder autoregressive with errors that are correlated with the error series of the dependent variable; see Stambaugh (1999) for the singleregressor model. This paper proposes a direct and convenient method to obtain reducedbias estimators for single and multiple regressor models by employing an augmented regression, adding a proxy for the errors in the autoregressive model. We derive bias expressions for both the ordinary least squares and our reducedbias estimated coefficients. For the standard errors of the estimated predictive coefficients we develop a heuristic estimator which performs well in simulations, for both the singlepredictor model and an important specification of the multiple predictor model. The effectiveness of our method is demonstrated by simulations and by empirical estimates of common predictive models in finance. Our empirical results show that some of the predictive variables that were significant under ordinary least squares become insignificant under our estimation procedure. 
Keywords:  Stock Returns; Dividend Yields; Autoregressive Models 
JEL:  C1 C2 C3 C4 C5 C8 
Date:  2004–12–15 
URL:  http://d.repec.org/n?u=RePEc:wpa:wuwpem:0412008&r=ets 
By:  Ole E. BarndorffNielsen; Neil Shephard 
Abstract:  In this brief note we review some of our recent results on the use of high frequency financial data to estimate objects like integrated variance in stochastic volatility models. Interesting issues include multipower variation, jumps and market microstructure effects. 
Date:  2004 
URL:  http://d.repec.org/n?u=RePEc:sbs:wpsefe:2004fe22&r=ets 
By:  Yakov Amihud (New York University); Clifford Hurvich (New York University); Yi Wang (New York University) 
Abstract:  We propose a new hypothesis testing method for multipredictor regressions with finite samples, where the dependent variable is regressed on lagged variables that are autoregressive. It is based on the augmented regressiom method (ARM; Amihud and Hurvich (2004)), which produces reducedbias coefficients and is easy to implement. The method's usefulness is demonstrated by simulations and by an empirical example, where stock returns are predicted by dividend yield and by bond yield spread. For singlepredictor regressions, we show that the ARM outperforms bootstrapping and that the ARM performs better than Lewellen's (2003) method in many situations. 
Keywords:  Augmented Regression Method (ARM); Bootstrapping; Hypothesis Testing 
JEL:  G 
Date:  2004–12–15 
URL:  http://d.repec.org/n?u=RePEc:wpa:wuwpfi:0412022&r=ets 
By:  Meitz, Mika (Dept. of Economic Statistics, Stockholm School of Economics); Saikkonen, Pentti (Dept. of Statistics, University of Helsinki) 
Abstract:  This paper studies a class of Markov models which consist of two components. Typically, one of the components is observable and the other is unobservable or 'hidden'. Conditions under which (a form of) geometric ergodicity of the unobservable component is inherited by the joint process formed of the two components are given. This immediately implies the existence of initial values such that the joint process is strictly stationary and betamixing. In addition to this, conditions for betamixing and existence of moments for the joint process are also provided in the case of (possibly) nonstationary initial values. All these results are applied to a general model which includes as special cases various first order generalized autoregressive conditional heteroskedasticity (GARCH) and autoregressive conditional duration (ACD) models with possibly complicated nonlinear structures. 
Keywords:   
JEL:  C22 
Date:  2004–10–07 
URL:  http://d.repec.org/n?u=RePEc:hhs:hastef:0573&r=ets 
By:  Mototsugu Shintani (Department of Economics, Vanderbilt University) 
Abstract:  This paper extends the diffusion index (DI) forecast approach of Stock and Watson (1998, 2002) to the case of possibly nonlinear dynamic factor models. When the number of series is large, a twostep procedure based on the principal components method is useful since it allows the wide variety of the nonlinearity in the factors. The factors extracted from a large Japanese data suggest some evidence of nonlinear structure. Furthermore, both the linear and nonlinear DI forecasts in Japan outperform traditional time series forecasts, while the linear DI forecast, in most cases, performs as well as the nonlinear DI forecast. 
Keywords:  Diffusion Index, Dynamic Factor Model, Nonlinearity, Prediction 
JEL:  F31 F41 
Date:  2003–10 
URL:  http://d.repec.org/n?u=RePEc:van:wpaper:0322&r=ets 
By:  OLE E. BARNDORFFNIELSEN; PETER REINHARD HANSEN; ASGER LUNDE; NEIL SHEPHARD 
Abstract:  We consider kernelbased estimators of integrated variances in the presence of independent market microstructure effects. We derive the bias and variance properties for all regular kernelbased estimators and derive a lower bound for their asymptotic variance. Further we show that the subsamplebased estimator is closely related to a Bartletttype kernel estimator. The small difference between the two estimators due to end effects, turns out to be key for the consistency of the subsampling estimator. This observation leads us to a modified class of kernelbased estimators, which are also consistent. We study the efficiency of our new kernelbased procedure. We show that optimal modified kernelbased estimator converges to the integrated variance at the optimal rate, m^1/4, where m is the number of intraday returns. 
JEL:  C13 C22 
Date:  2004 
URL:  http://d.repec.org/n?u=RePEc:sbs:wpsefe:2004fe20&r=ets 
By:  Joshua D. Angrist; Guido M. Kuersteiner 
Abstract:  Time series data are widely used to explore causal relationships, typically in a regression framework with lagged dependent variables. Regressionbased causality tests rely on an array of functional form and distributional assumptions for valid causal inference. This paper develops a semiparametric test for causality in models linking a binary treatment or policy variable with unobserved potential outcomes. The procedure is semiparametric in the sense that we model the process determining treatment  the policy propensity score  but leave the model for outcomes unspecified. This general approach is motivated by the notion that we typically have better prior information about the policy determination process than about the macroeconomy. A conceptual innovation is that we adapt the crosssectional potential outcomes framework to a time series setting. This leads to a generalized definition of Sims (1980) causality. We also develop a test for full conditional independence, in contrast with the usual focus on mean independence. Our approach is illustrated using data from the Romer and Romer (1989) study of the relationship between the Federal reserve's monetary policy and output. 
JEL:  C14 C22 E52 
Date:  2004–12 
URL:  http://d.repec.org/n?u=RePEc:nbr:nberwo:10975&r=ets 
By:  González Gómez, Andrés (Dept. of Economic Statistics, Stockholm School of Economics) 
Abstract:  In this paper we introduce the Smooth Permanent Surge [SPS] model. The model is an integrated non lineal moving average process with possibly unit roots in the moving average coefficients. The process nests the Stochastic Permanent Break [STOPBREAK] process by Engle and Smith (1999) and in a limiting case it converges to Threshold Integrated Moving Average [TIMA] models by Gonzalo and Martinez (2003). A test of SPS against STOPBREAK process is presented. Additionally, we introduce a new test for testing SPS process against the random walk. The small sample properties of these tests are <p> investigated by Monte Carlo experiments. An application to the stock markets is presented. 
Keywords:  Linearity test; Monte Carlo testing; Smooth transitions; Moving Averages Models; Permanent Shock; Transitory Shocks. 
JEL:  C12 C15 C22 C51 C52 
Date:  2004–12–07 
URL:  http://d.repec.org/n?u=RePEc:hhs:hastef:0572&r=ets 
By:  Phillip Gould; Anne B. Koehler; Farshid VahidAraghi; Ralph D. Snyder; J. Keith Ord; Rob J. Hyndman 
Abstract:  A new approach to forecasting seasonal data is proposed where seasonal terms can be updated using the most recent relevant information. It was developed to handle features encountered in hourly electricity load data and daily hospital admissions data. The associated state space model is estimated with methods adapted from exponential smoothing, although the Kalman filter may also be used. It nests existing seasonal models and outperforms them over a range of prediction horizons on the data. The approach is likely to be useful in a wide range of applications involving both high and low frequency data. 
Keywords:  Exponential smoothing; HoltWinters; Seasonality; Structural time series model 
JEL:  C22 
Date:  2004–12 
URL:  http://d.repec.org/n?u=RePEc:msh:ebswps:200428&r=ets 
By:  Guillaume Chevillon (Observatoire Français des Conjonctures Économiques) 
Date:  2004 
URL:  http://d.repec.org/n?u=RePEc:fce:doctra:0412&r=ets 