Optimal trimming proportion in regression analysis for non-normal distributions

Amitava Mitra and Pankush Kalgotra

Journal of Business Analytics, 2022, vol. 5, issue 2, 152-163

Abstract: Regression analysis is a widely used modelling tool in business decision making. However, proper application of this methodology requires that certain assumptions, associated with the model, be satisfied. The assumption we focus on is the normality of the response variable, which is directly related to the assumption of normality of the error component. In a variety of fields in business, such as finance, marketing, information systems, operations, and healthcare, the selected dependent variable does not inherently have a normal distribution. In the regression context, where the model parameters and independent variables are assumed to be constant, the distribution of the random error component thus influences the distribution of the dependent variable. Here, we study the impact of symmetric and asymmetric error distributions on the performance of the estimated model parameters. To create robust estimates, through a process of trimming the response variable, we study the effectiveness of the trimmed estimators with respect to the ordinary least squares estimator (OLS) via a simulation procedure. Accordingly, to minimise the ratio of the mean squared error of the trimmed estimator to that of the OLS, a recommendation is developed for the optimal trimming proportion.

Date: 2022
References: Add references at CitEc
Citations:

Downloads: (external link)
http://hdl.handle.net/10.1080/2573234X.2021.2007803 (text/html)
Access to full text is restricted to subscribers.

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:taf:tjbaxx:v:5:y:2022:i:2:p:152-163

Ordering information: This journal article can be ordered from
http://www.tandfonline.com/pricing/journal/tjba20

DOI: 10.1080/2573234X.2021.2007803

Access Statistics for this article

Journal of Business Analytics is currently edited by Dursan Delen

More articles in Journal of Business Analytics from Taylor & Francis Journals
Bibliographic data for series maintained by Chris Longhurst ().