Relative Performance Of Estimators Research Articles

Interval estimation of the overall treatment effect in a meta-analysis of a few small studies with zero events.

When a meta-analysis consists of a few small trials that report zero events, accounting for heterogeneity in the (interval) estimation of the overall effect is challenging. Typically, we predefine meta-analytical methods to be employed. In practice, data poses restrictions that lead to deviations from the pre-planned analysis, such as the presence of zero events in at least one study arm. We aim to explore heterogeneity estimators behaviour in estimating the overall effect across different levels of sparsity of events. We performed a simulation study that consists of two evaluations. We considered an overall comparison of estimators unconditional on the number of observed zero cells and an additional one by conditioning on the number of observed zero cells. Estimators that performed modestly robust when (interval) estimating the overall treatment effect across a range of heterogeneity assumptions were the Sidik-Jonkman, Hartung-Makambi and improved Paul-Mandel. The relative performance of estimators did not materially differ between making a predefined or data-driven choice. Our investigations confirmed that heterogeneity in such settings cannot be estimated reliably. Estimators whose performance depends strongly on the presence of heterogeneity should be avoided. The choice of estimator does not need to depend on whether or not zero cells are observed.

The performance of bid-ask spread estimators under less than ideal conditions

Purpose – This paper aims to investigate the performance of estimators of the bid-ask spread in a wide range of circumstances and sampling frequencies. The bid-ask spread is important for many reasons. Because spread data are not always available, many methods have been suggested for estimating the spread. Existing papers focus on the performance of the estimators either under ideal conditions or in real data. The gap between ideal conditions and the properties of real data are usually ignored. The consistency of the estimates across various sampling frequencies is also ignored. Design/methodology/approach – The estimators and the possible errors are analysed theoretically. Then we perform simulation experiments, reporting the bias, standard deviation and root mean square estimation error of each estimator. More specifically, we assess the effects of the following factors on the performance of the estimators: the magnitude of the spread relative to returns volatility, randomly varying of spreads, the autocorrelation of mid-price returns and mid-price changes caused by trade directions and feedback trading. Findings – The best estimates come from using the highest frequency of data available. The relative performance of estimators can vary quite markedly with the sampling frequency. In small samples, the standard deviation can be more important to the estimation error than bias; in large samples, the opposite tends to be true. Originality/value – There is a conspicuous lack of simulation evidence on the comparative performance of different estimators of the spread under the less than ideal conditions that are typical of real-world data. This paper aims to fill this gap.

Clusters of Extreme Observations and Extremal Index Estimate in GARCH Processes

Several methods have been proposed for identifying clusters of extreme values leading to estimators of the extremal index; the latter represents,in the limit, the mean-size of each cluster of thresholds exceedances. The detection of clusters of extremes is relevant for the class of processes commonly used in financial econometrics, such as GARCH processes. The paper illustrates a novel approach to the above identification that exploits additional knowledge of the trajectory of the process around extreme events, and compares it to traditional approaches, using simulation from a GARCH process. We assess the relative performance of estimators in terms of bias, mean square error and distributional properties.

ESTIMATING CENTRAL TENDENCY FROM UNCENSORED TRACE LEVEL MEASUREMENTS1

ABSTRACT: The statistical analysis of data which have trace level measurements has traditionally been a two‐step process in which data are first censored using criteria based on measurement precision, and then analyzed with statistical methods for censored data. The process might be more informative if data were left uncensored. In this paper, information loss attributable to censoring and measurement noise are assessed by comparing the sample mean and median of uncensored measurements with a log regression mean and median based on censored data. Measurements are derived from lognormal parent distributions which have random variability characteristic of trace level measurement. The relative performance of estimators used with error‐free samples and with samples having measurement noise can be explained by differences between the probability distributions of parents and measurements. Measurement introduces bias and dispersion and transforms lognormal parent distributions toward greater symmetry. Estimates using uncensored data are less biased and more accurate than the log regression mean and median when censoring exceeds about 50 percent, and are not much worse at any fraction censored. For data with many (80 percent) results below the limit of detection, bias may be quite severe.