基于稳健Cook距离的时间序列异常值诊断

Cook距离公式常用于回归模型的异常值诊断,但由于公式中的样本方差■对异常值敏感,导致公式缺乏稳健性,使得诊断效果不理想。基于以上问题,文章选取绝对离差中位数作为样本标准差的稳健估计量,得到了样本方差■的稳健估计量,进而构造出稳健Cook距离公式;借鉴传统Cook距离的回归模型异常值诊断理论,将稳健Cook距离公式应用于时间序列异常值诊断,拓展了传统Cook距离公式的异常值诊断领域。通过选取模拟样本量分别为50、100、200,污染率分别为0、1%、5%、10%的ARMA(1,1)序列及金融时间序列进行实例分析,结果发现:(1)在无污染时,稳健Cook距离法与常规Cook距离法的诊断正确率均为100%,两者没有出现"误诊"现象;(2)在样本量、污染率同时增大时,常规Cook距离诊断正确率急剧下降,当污染率达到5%及以上时,已基本无诊断力,而稳健Cook距离法依然能保持较高的诊断力。稳健Cook距离法不仅能应用于时间序列异常值诊断,也能应用于回归分析的异常值诊断。     

A Note on Analysing Time Series Data which Accumulate Over Unequal Observation Intervals

The problem of analysing flow series observed over unequal periods of time is discussed, and a suggestion is made for dealing with situations where the lengths of the periods follow a simple pattern     

A simple diagnostic method of outlier detection for stationary Gaussian time series

In this paper we present a "model free' method of outlier detection for Gaussian time series by using the autocorrelation structure of the time series. We also present a graphic diagnostic method in order to distinguish an additive outlier (AO) from an innovation outlier (IO). The test statistic for detecting the outlier has a χ ² distribution with one degree of freedom. We show that this method works well when the time series contain either one type of the outliers or both additive and innovation type outliers, and this method has the advantage that no time series model needs to be estimated from the data. Simulation evidence shows that different types of outliers can be graphically distinguished by using the techniques proposed.     

Graphical comparison of means with a standard in the one-way layout

Some simple test procedures are considered for comparing several group means with a standard value when the data are in a one-way layout. The underlying distributions are assumed to be normal with possibly unequal variances. The tests are based on a union-intersection formulation and can be applied in a form similar to a Shewhart control chart. Both two-sided and one-sided alternatives are considered. The power of the tests can be obtained from tables of a non-central t distribution. Implementation of the tests is illustrated with a numerical example. The tests help identify any group means different from the standard and might lead to a decision about rejecting the null hypothesis before all the group means are observed. The resulting savings in time and resources might be valuable in applications where the number of groups is large and the cost of acquiring data is high. For situations where the normality assumption is untenable, a non-parametric procedure, based on one-sample sign tests is considered.     

Detecting possibly non-consecutive outliers in industrial time series

A method for robust estimation and multiple outlier detection in time series generated by autoregressive integrated moving average processes in industrial environments is developed. The procedure is based on reweighted maximum likelihood estimation using Huber or redescending weights and, therefore, generalizes the well-established robust M -estimation procedures used in the regression framework. When the scalar process is non-stationary, the computations required can be performed equally well using either rhe original undifferenced series or auxiliary differenced series. Whereas the latter alternative may be preferred for scalar series, the former might be extended to cope with vector partially non-stationary time series without differencing the series, thus avoiding non-invertibility and parameter identifiability problems caused by overdifferencing. The overall strategy is applied in two real industrial data sets.     

A note on time series model specification in the presence of outliers

We investigate the usefulness of sample autocorrelations and partial autocorrelations as model specification tools when the observed time series is contaminated by an outlier. The results indicate that the specification power of these statistics could be significantly jeopardized by an additive outlier. On the other hand, an innovational outlier seems to cause no harm to them.     

Empirical Bayes confidence intervals shrinking both means and variances

Summary.  We construct empirical Bayes intervals for a large number p of means. The existing intervals in the literature assume that variances     are either equal or unequal but known. When the variances are unequal and unknown, the suggestion is typically to replace them by unbiased estimators     . However, when p is large, there would be advantage in 'borrowing strength' from each other. We derive double-shrinkage intervals for means on the basis of our empirical Bayes estimators that shrink both the means and the variances. Analytical and simulation studies and application to a real data set show that, compared with the t -intervals, our intervals have higher coverage probabilities while yielding shorter lengths on average. The double-shrinkage intervals are on average shorter than the intervals from shrinking the means alone and are always no longer than the intervals from shrinking the variances alone. Also, the intervals are explicitly defined and can be computed immediately.     

The effects of outliers on two nonlinearity tests

Two Lagrange multiplier tests for time series nonlinearities in the presence of outliers are examined by simulation experiments. The nonlinearities studied are autoregressive conditional heteroskedasticity (ARCH) and bilinearity; the outlier types are additive, innovative, temporary change and reallocation outliers. The results show that both the sizes and powers of the tests can be severely distorted by even a single outlier. The severity of the distortions depends on the outlier type and magnitude, but also on the underlying process generating 'the series.     

ANALYSIS OF VECTOR AUTOREGRESSIONS IN THE PRESENCE OF SHIFTS IN MEAN

This paper considers the implications of mean shifts in a multivariate setting. It is shown that under the additive outlier type mean shift specification, the intercept in each equation of the vector autoregression (VAR) will be subject to multiple shifts when the break dates of the mean shifts to the univariate series do not coincide. Conversely, under the innovative outlier type mean shift specification, both the univariate and the multivariate time series are subject to multiple shifts when mean shifts to the innovation processes occur at different dates. We consider two procedures, the first removes the shifts series by series before forming the VAR, and the second removes intercept shifts in the VAR directly. The pros and cons of both methods are discussed.     

Making control charts more effective by time series analysis: three illustrative applications

Control charts contribute to the monitoring and improvement of process quality by helping to separate out special cause variation from common cause variation. By common cause variation we mean the usual variation in an in-control process. Special causes can be thought of as disturbances, possibly transitory, impacting a process that is in a state of statistical control. However, there is no clear place in this scheme of special causes and common causes for systematic non-iid variation, such as trend, seasonal, autoregression variation, and intervention effects from efforts to improve the proess. When systematic non-iid variation is present, time series modeling and fitting can fill in this picture. In the time series framework, observations influenced by special causes can be treated as outliers from the currently-entertained time-series model and can be detected by outlier detection methods. We discuss three data sets that illustrate how this can be done in order to make control charts more effective. We show also how a standard control-chart supplement called "pattern analysis" can be useful in time-series work.     

Comparison of Times Series with Unequal Length in the Frequency Domain

In statistical data analysis it is often important to compare, classify, and cluster different time series. For these purposes various methods have been proposed in the literature, but they usually assume time series with the same sample size. In this article, we propose a spectral domain method for handling time series of unequal length. The method make the spectral estimates comparable by producing statistics at the same frequency. The procedure is compared with other methods proposed in the literature by a Monte Carlo simulation study. As an illustrative example, the proposed spectral method is applied to cluster industrial production series of some developed countries.     

An Application of Discrimination When Covariance Matrices are Proportional

In the context of time series, a situation is considered which leads to the discrimination between two normal populations having unequal means and proportional covariance matrices. It is then shown that this has an application to the important problem of earthquake-explosion differentiation. The latter is done by introducing a non-stationary model for seismic records of P-waves from these two kinds of events.     

Combining non-linear regressions that have unequal error variances and some parameters in common

Methods of estimation and inference are presented for the situation where two non-linear regression models with unequal error variances contain some parameters in common. Such a situation arises in structural chemistry, when bond lengths are available for three nearly collinear atoms in crystals and a model is required to quantify the extent and form of the relationship between the longer and the shorter bond. Some atomic triples are symmetric and require a different model and error variance from those required by the asymmetric triples. The profile likelihood for the regression parameters is a weighted sum of the logarithms of the sums-of-squares functions from each model, and the estimates can be obtained by using a simple modification to a standard non-linear least squares program. A likelihood ratio test for assessing whether the parameters in common are equal is described. When these techniques are applied to two data sets consisting of bond lengths for bromine–tellurium–bromine and sulphur–tellurium–sulphur triples, there is no evidence against the equality hypothesis. An extension to the model to allow for a non-constant variance is required for proper analysis of the sulphur–tellurium–sulphur data.     

Performance of unit root tests in unbalanced panels: experimental evidence

This paper is about the validity of established panel unit root tests applied to panels in which the individual time series are of different lengths, a case often encountered in practice. Most of the tests considered work well under various types of cross-correlation which is true for both, their application in balanced as well as in unbalanced panels. A Monte Carlo study reveals that in unbalanced panels, procedures involving the computation of individual $p$ -values for each cross-section unit (or the combination thereof) are mostly superior to those relying on a pooled Dickey–Fuller regression framework. As the former are able to consider each unit separately, they do not require cutting back the “longer” time series so as to obtain the smallest “balanced” quadrangle which in turn means that no potentially valuable information is lost.     

The Multiple Outlier Problem in Time Series Analysis

In this paper we consider the multiple outlier problem in time series analysis. The underlying undisturbed time series is assumed to be an autoregressive process. The location of the suspicious values is supposed to be known. We introduce conditional least squares estimators for the parameters. The estimates are shown to be strongly consistent. Using similar arguments as in the theory of linear models, we get a test statistic for the general linear hypothesis. Its asymptotic distribution is derived.     

A mixture-based approach to robust analysis of generalised linear models

A method for robustness in linear models is to assume that there is a mixture of standard and outlier observations with a different error variance for each class. For generalised linear models (GLMs) the mixture model approach is more difficult as the error variance for many distributions has a fixed relationship to the mean. This model is extended to GLMs by changing the classes to one where the standard class is a standard GLM and the outlier class which is an overdispersed GLM achieved by including a random effect term in the linear predictor. The advantages of this method are it can be extended to any model with a linear predictor, and outlier observations can be easily identified. Using simulation the model is compared to an M-estimator, and found to have improved bias and coverage. The method is demonstrated on three examples.     

Bias correction for outlier estimation in time series

The problem of outlier estimation in time series is addressed. The least squares estimators of additive and innovation outliers in the framework of linear stationary and non-stationary models are considered and their bias is evaluated. As a result, simple alternative nearly unbiased estimators are proposed both for the additive and the innovation outlier types. A simulation study confirms the theoretical results and suggests that the proposed estimators are effective in reducing the bias also for short series.     

An approximate entropy test for randomness

We develop an entropy-based test for randomness of binary time series of finite length. The test uses the frequencies of contiguous blocks of different lengths. A simple condition ib the block lengths and the length of the time series enables one to estimate the entropy rate for the data, and this information is used to develop a statistic to test the hypothesis of randomness. This static measures the deviation of the estimated entropy of the observed data from the theoretical maximum under the randomness hypothesis. This test offers a real alternative to the conventional runs test. Critical percentage points, based on simulations, are provided for testing the hypothesis of randomness. Power calculations using dependent data show that the proposed test has higher power against the runs test for short series, and it is similar to the runs test for long series. The test is applied to two published data sets that wree investigated by others with respect to their randomness.     

Extremal analysis of short series with outliers: sea-levels and athletics records

The analysis of extreme values is often required from short series which are biasedly sampled or contain outliers. Data for sea-levels at two UK east coast sites and data on athletics records for women's 3000 m track races are shown to exhibit such characteristics. Univariate extreme value methods provide a poor quantification of the extreme values for these data. By using bivariate extreme value methods we analyse jointly these data with related observations, from neighbouring coastal sites and 1500 m races respectively. We show that using bivariate methods provides substantial benefits, both in these applications and more generally with the amount of information gained being determined by the degree of dependence, the lengths and the amount of overlap of the two series, the homogeneity of the marginal characteristics of the variables and the presence and type of the outlier.     

The problem of outlier tests in sample surveys

Outliers can occur as readily in samples from the finite populations (e.g. in sample surveys) as in samples from infinite populations. However, in the vast literature on outliers there is almost no mention of outlier tests for data from sample surveys. We examine the behaviour of some standard outlier test statistics for infinite populations when these are applied to finite populations, examining their properties by extensive simulation studies. Some anomalous results are obtained Nsuggesting a fundamental difficulty in testing outliers for the finite population case.