Outliers,level shifts,and variance changes in time series

Outliers, level shifts, and variance changes are commonplace in applied time series analysis. However, their existence is often ignored and their impact is overlooked, for the lack of simple and useful methods to detect and handle those extraordinary events. The problem of detecting outliers, level shifts, and variance changes in a univariate time series is considered. The methods employed are extremely simple yet useful. Only the least squares techniques and residual variance ratios are used. The effectiveness of these simple methods is demonstrated by analysing three real data sets.     

Forecasting time series with outliers

Time-series data are often contaminated with outliers due to the influence of unusual and non-repetitive events. Forecast accuracy in such situations is reduced due to (1) a carry-over effect of the outlier on the point forecast and (2) a bias in the estimates of model parameters. Hillmer (1984) and Ledolter (1989) studied the effect of additive outliers on forecasts. It was found that forecast intervals are quite sensitive to additive outliers, but that point forecasts are largely unaffected unless the outlier occurs near the forecast origin. In such a situation the carry-over effect of the outlier can be quite substantial. In this study, we investigate the issues of forecasting when outliers occur near or at the forecast origin. We propose a strategy which first estimates the model parameters and outlier effects using the procedure of Chen and Liu (1993) to reduce the bias in the parameter estimates, and then uses a lower critical value to detect outliers near the forecast origin in the forecasting stage. One aspect of this study is on the carry-over effects of outliers on forecasts. Four types of outliers are considered: innovational outlier, additive outlier, temporary change, and level shift. The effects due to a misidentification of an outlier type are examined. The performance of the outlier detection procedure is studied for cases where outliers are near the end of the series. In such cases, we demonstrate that statistical procedures may not be able to effectively determine the outlier types due to insufficient information. Some strategies are recommended to reduce potential difficulties caused by incorrectly detected outlier types. These findings may serve as a justification for forecasting in conjunction with judgment. Two real examples are employed to illustrate the issues discussed.     

Forecasting time series with long memory and level shifts

It is well known that some economic time series can be described by models which allow for either long memory or for occasional level shifts. In this paper we propose to examine the relative merits of these models by introducing a new model, which jointly captures the two features. We discuss representation and estimation. Using simulations, we demonstrate its forecasting ability, relative to the one‐feature models, both in terms of point forecasts and interval forecasts. We illustrate the model for daily S&P500 volatility. Copyright © 2005 John Wiley & Sons, Ltd.     

Monitoring for outliers and level shifts in kalman filter implementations of exponential smoothing

This paper presents a new application of a Kalman filter implementation of exponential smoothing with monitoring for outliers and level shifts. The assumption is that each observation comes from one of three models: steady, outlier, or level shift. This concept was introduced as a multiprocess model by Harrison and Stevens (1976). However, their handling of the models is different. In this paper four different model-selection criteria are introduced and compared by applying them to data. The new features of the application include the four model-selection criteria and the estimation of the required parameters by maximum likelihood.     

A simultaneous test of unit root and level change

Testing the existence of unit root and/or level change is necessary in order to understand the underlying processes of time series. In many studies carried out so far, the focus was only on a single aspect of unit root and level change, therefore limiting a full assessment of the given problems. Our study aims to find a solution to the given problems by testing the two hypotheses simultaneously. We derive the likelihood ratio test statistic based on the state space model, and their distributions are created by the simulation method. The performance of the proposed method is validated by simulated time series and also applied to two Korean macroeconomic time series to confirm its practical application. This analysis can provide a solution to determine the underlying structure of arguable time series. Copyright © 2009 John Wiley & Sons, Ltd.     

Modelling the advertising-sales relationship through use of multiple time series techniques

When time series data are available for both advertising and sales, it may be worth while to model the two series jointly. Such an analysis may contribute to our understanding of the dynamic relationships among the series and may improve the accuracy of forecasts. Multiple time series techniques are applied to the well-known Lydia Pinkham data to illustrate their use in modelling the advertising-sales relationship. In analysing the Lydia Pinkham data the need for a joint model is established and a bivariate model is identified, estimated and checked. Its forecasting properties are discussed and compared to other time series approaches.     

Monitoring and adjusting forecasts in the presence of additive outliers

The effect of an additive outlier upon the accuracy of forecasts derived from extrapolative methods is investigated. It is demonstrated that an outlier affects not only the accuracy of the forecasts at the time of occurrence but also subsequent forecasts. Methods to adjust for additive outliers are discussed. The results of the paper are illustrated with two examples.     

Identifying the time‐effect factors of multiple time series

The Peña–Box model is considered for finding the time‐effect factors of a multiple time series. This paper first establishes the connection between the Peña–Box model and the vector ARMA model. According to the Peña–Box model, some series can be ignored while modelling the vector ARMA model. A consistent estimator is then proposed to identify the model for nonlinear and nonstationary time series. Finally, the finite‐sample behaviour of the estimator is illustrated via simulations. Copyright © 2005 John Wiley & Sons, Ltd.     

Forecasting demand in international markets: The case of correlated time series

In this paper a data analysis tool for analyzing highly correlated time series data is suggested. The main objective is to unify multiple time series into a single series and then apply a univariate method for the purpose of prediction. This method is essentially efficient for analyzing multiple time series with sparse data. Several time series data of relative demand for black and white television receivers in various countries are analyzed and quite accurate predictions are obtained.     

The use of canonical correlation analysis to identify the order of multivariate ARMA models: simulation and application

This paper is concerned with how canonical correlation can be used to identify the structure of a linear multivariate time series model. We describe briefly methods that use the canonical correlation technique and present simulation results in order to compare and evaluate the performance of these methods. The methods are also applied to a well‐known multivariate time series. Copyright © 2000 John Wiley & Sons, Ltd.     

Ex-post and ex-ante prediction of unobserved economic time series: a case study

In several countries, some macro-economic variables are not observed frequently (e.g. quarterly) and economic authorities need estimates of these high-frequency figures to make econometric analyses or to follow closely the country's economic growth. Two problems are involved in this context. The first is to make these estimates after observing low-frequency values and some related indicators, and the second is to obtain predictions using just the observed indicators, i.e. before observing a new low-frequency figure. This paper gives a new optimal solution to the first problem, and solves the second using a recursive optimal approach. In the second situation, additionally, statistical tests are developed for detecting structural changes at current periods in the macro-economic variable involved. © 1998 John Wiley & Sons, Ltd.     

Multiple prediction intervals for time series: Comparison of simultaneous and marginal intervals

Simultaneous prediction intervals for forecasts from time series models that contain L (L ≤ 1) unknown future observations with a specified probability are derived. Our simultaneous intervals are based on two types of probability inequalities, i.e. the Bonferroni- and product-types. These differ from the marginal intervals in that they take into account the correlation structure between the forecast errors. For the forecasting methods commonly used with seasonal time series data, we show how to construct forecast error correlations and evaluate, using an example, the simultaneous and marginal prediction intervals. For all the methods, the simultaneous intervals are accurate with the accuracy increasing with the use of higher-order probability inequalities, whereas the marginal intervals are far too short in every case. Also, when L is greater than the seasonal period, the simultaneous intervals based on improved probability inequalities will be most accurate.     

The role of time series analysis in the evaluation of econometric models

This paper compares the properties of a structural model—the London Business School model of the U.K. economy—with a time series model. Information provided by this type of comparison is a useful diagnostic tool for detecting types of model misspecification. This is a more meaningful way of proceeding rather than attempting to establish the superiority of one type of model over another. In lieu of a better structural model, the effects of inappropriate dynamic specification can be reduced by combining the forecasts of both the structural and time series models. For many variables considered here these provide more accurate forecasts than each of the model types alone.     

Causal implications: The linkage between time series and econometric modelling

Construction of causally and structurally adequate simultaneous equations models can be accomplished by determining causal relations between potential variables and balancing these statistically derived inferences with economic theory to relate behavioural or technological forces among the variables. An appropriate lag structure for each of the equations can be determined by a two step multiple transfer function approach involving reduced form equations. Testing the specification of already existing simultaneous equations models is done by constructing multiple transfer function models of the reduced form equations of the simultaneous equations models which permit incorporation of lead cross correlations.     

Recursive estimation and forecasting of non-stationary time series

The paper presents a unified, fully recursive approach to the modelling and forecasting of non-stationary time-series. The basic time-series model, which is based on the well-known ‘component’ or ‘structuraL’ form, is formulated in state-space terms. A novel spectral decomposition procedure, based on the exploitation of recursive smoothing algorithms, is then utilized to simplify the procedures of model identification and estimation. Finally, the fully recursive formulation allows for conventional or self-adaptive implementation of state-space forecasting and seasonal adjustment. Although the paper is restricted to the consideration of univariate time series, the basic approach can be extended to handle explanatory variables or full multivariable (vector) series.     

Minimum variance prediction of bilinear time series: Direct and adaptive versions

Bilinear models of time series are considered. Minimum variance predictor for bilinear time series, homogeneous in the input and output, is proposed. Results of minimum variance prediction of bilinear time series are included. They are compared to the results of linear prediction of bilinear time series. A minimum variance prediction algorithm for bilinear time series of the general form is developed and an adaptive version of minimum variance algorithm is derived.     

Detection of regime switches between stationary and nonstationary processes and economic forecasting

It often occurs that no model may be exactly right, and that different portions of the data may favour different models. The purpose of this paper is to propose a new procedure for the detection of regime switches between stationary and nonstationary processes in economic time series and to show its usefulness in economic forecasting. In the proposed procedure, time series observations are divided into several segments, and a stationary or nonstationary autoregressive model is fitted to each segment. The goodness of fit of the global model composed of these local models is evaluated using the corresponding information criterion, and the division which minimizes the information criterion defines the best model. Simulation and forecasting results show the efficacy and limitations of the proposed procedure. Copyright © 2005 John Wiley & Sons, Ltd.     

Linear combination of restrictions and forecasts in time series analysis

An important tool in time series analysis is that of combining information in an optimal way. Here we establish a basic combining rule of linear predictors and show that such problems as forecast updating, missing value estimation, restricted forecasting with binding constraints, analysis of outliers and temporal disaggregation can be viewed as problems of optimal linear combination of restrictions and forecasts. A compatibility test statistic is also provided as a companion tool to check that the linear restrictions are compatible with the forecasts generated from the historical data. Copyright © 2000 John Wiley & Sons, Ltd.     

On detecting and estimating a major level or slope change in general exponential smoothing

Exponential smoothing methods do not adapt well to a major level or slope change. In this paper, Bayesian statistical theory is applied to the dynamic linear model, altered by inclusion of dummy variables, and statistics are derived to detect such changes and to estimate both the change-point and the size. The paper also gives test statistics for such problems related to exponential smoothing. The statistics are simple functions of exponentially weighted moving averages of the forecast errors, using the same discount factor used in the exponential smoothing. Gardner has derived an approximate test statistic to detect a mean change in the constant mean model. When the present results are applied to this model they give the exact statistic.     

Non-linear forecasting of financial time series: An overview and some new models

We provide an overview of the papers contained in this Special Issue of the Journal of Forecasting and also discuss some new models for analysing financial time series that have recently been proposed. These are illustrated by empirical examples using 60 years of daily data on the London Stock Exchange's FT30 index.