Model selection and forecasting for long-range dependent processes

Fractionally integrated autoregressive moving-average (ARFIMA) models have proved useful tools in the analysis of time series with long-range dependence. However, little is known about various practical issues regarding model selection and estimation methods, and the impact of selection and estimation methods on forecasts. By means of a large-scale simulation study, we compare three different estimation procedures and three automatic model-selection criteria on the basis of their impact on forecast accuracy. Our results endorse the use of both the frequency-domain Whittle estimation procedure and the time-domain approximate MLE procedure of Haslett and Raftery in conjunction with the AIC and SIC selection criteria, but indicate that considerable care should be exercised when using ARFIMA models. In general, we find that simple ARMA models provide competitive forecasts. Only a large number of observations and a strongly persistent time series seem to justify the use of ARFIMA models for forecasting purposes.     

Linear and Non-linear (Non-)Forecastability of High-frequency Exchange Rates

This paper forecasts Daily Sterling exchange rate returns using various naive, linear and non-linear univariate time-series models. The accuracy of the forecasts is evaluated using mean squared error and sign prediction criteria. These show only a very modest improvement over forecasts generated by a random walk model. The Pesaran–Timmerman test and a comparison with forecasts generated artificially shows that even the best models have no evidence of market timing ability.©1997 John Wiley & Sons, Ltd.     

Judging the probability that the next point in an observed time-series will be below,or above,a given value

We presented people with trended and untrended time series and asked them to estimate the probability that the next point would be below each of seven different reference values. The true probabilities that the point would be below these values were 0.01, 0.10, 0.25, 0.50, 0.75, 0.90 and 0.99. People overestimated probabilities of less than 0.50 and underestimated those of more than 0.50. Consequently, their subjective probability distributions were flatter than they should have been: people appeared to be under confident in their estimates of where the next point would lie. This bias was greater for the trended series. It was also greater in a second experiment in which people estimated the probability that the next item would be above the reference values. We discuss reasons for these effects and consider their implications for decision making.     

Multi‐step forecasting for long‐memory processes

In this paper we present results of a simulation study to assess and compare the accuracy of forecasting techniques for long‐memory processes in small sample sizes. We analyse differences between adaptive ARMA(1,1) L‐step forecasts, where the parameters are estimated by minimizing the sum of squares of L‐step forecast errors, and forecasts obtained by using long‐memory models. We compare widths of the forecast intervals for both methods, and discuss some computational issues associated with the ARMA(1,1) method. Our results illustrate the importance and usefulness of long‐memory models for multi‐step forecasting. Copyright © 1999 John Wiley & Sons, Ltd.     

Variance estimation for multivariate dynamic linear models

The problem of estimating unknown observational variances in multivariate dynamic linear models is considered. Conjugate procedures are possible for univariate models and also for special very restrictive common components models but they are not generally applicable. However, for clarity of operation and in order to avoid numerical integration, it is desirable to have conjugacy or approximate conjugacy. Such an approximate procedure is proposed based upon a simple analytic approximation. It is exact for the sub-class of conjugate models and improves on a previous procedure based upon the Robust filter.     

Why do EMD‐based methods improve prediction? A multiscale complexity perspective

Empirical mode decomposition (EMD)‐based ensemble methods have become increasingly popular in the research field of forecasting, substantially enhancing prediction accuracy. The key factor in this type of method is the multiscale decomposition that immensely mitigates modeling complexity. Accordingly, this study probes this factor and makes further innovations from a new perspective of multiscale complexity. In particular, this study quantitatively investigates the relationship between the decomposition performance and prediction accuracy, thereby developing (1) a novel multiscale complexity measurement (for evaluating multiscale decomposition), (2) a novel optimized EMD (OEMD) (considering multiscale complexity), and (3) a novel OEMD‐based forecasting methodology (using the proposed OEMD in multiscale analysis). With crude oil and natural gas prices as samples, the empirical study statistically indicates that the forecasting capability of EMD‐based methods is highly reliant on the decomposition performance; accordingly, the proposed OEMD‐based methods considering multiscale complexity significantly outperform the benchmarks based on typical EMDs in prediction accuracy.