Cluster analysis and firm patterns of earnings persistence : a new approach : a thesis presented in partial fulfilment of the requirements for the degree of Doctor of Philosophy in Finance at Massey University, Manawatū, New Zealand
Loading...
Date
2019
DOI
Open Access Location
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Massey University
Rights
The Author
Abstract
The development of a method to appropriately address the problem of heterogeneous-group
specific coefficients (HGSC) is of paramount importance for any studies where there are
concerns of HGSC. Accordingly, the goal of this thesis is to investigate a solution to the
prevalent problem of HGSC within the context of the finance discipline. Specifically, this
thesis introduces a novel clustering procedure called regression oriented-weighted K-
means clustering (ROWK). This new method employs the regression mean absolute
residuals (MAR) to inform the cluster analysis identification of optimal feature weights.
The performance of ROWK clustering is examined via both simulated and real data.
Simulation results show significant improvements from the adoption of ROWK relative to
K-means clustering and weighted K-means clustering through three channels. Specifically,
through the examination of three case studies, this thesis finds that ROWK places more (less)
weight on more (less) relevant features; reduces the influence of multicollinearity by
reducing the weights of irrelevant features which are highly correlated with relevant features;
and captures relevance not only by its contribution to cluster recognition but also by
regression estimation. The thesis further examines the performance of ROWK clustering
using real data for earnings persistence models. ROWK outperforms other standard
techniques in the sense of correctly identifying the underlying clusters on earnings
persistence models. The thesis also documents that analysts’ forecasts only partially
incorporate the information from cluster patterns in the short run, while ignoring impacts of
these patterns on long-term future earnings. As a result, conditioning on such information
allows the identification of reliable and economically important patterns in analyst forecast
errors.
Description
Keywords
Finance, Econometric models, Cluster analysis, Regression analysis, K-means, feature weightings, group-specific coefficients, firm patterns, earnings persistence