The selection of control variables in capital structure research with machine learning

• Published date: 01 October 2023
• Author: Rumeysa Bilgin
• Date: 01 October 2023
• DOI: http://doi.org/10.1002/jcaf.22647

Received: 22 January 2023 Revised: 1 May 2023 Accepted: 4 June 2023

DOI: 10.1002/jcaf.22647

RESEARCH ARTICLE

The selection of control variables in capital structure

research with machine learning

Control variables in capital structure

Rumeysa Bilgin

Department of Business Administration

Management, Entrepreneurship and

Leadership Research and Application

Center, Istanbul Sabahattin Zaim

University, Istanbul, Turkey

Correspondence

Rumeysa Bilgin, Department of Business

Administration Management,

Entrepreneurship and Leadership

Research and Application Center,

Istanbul Sabahattin Zaim University,

Istanbul, Turkey.

Email: rumeysa.bilgin@izu.edu.tr

Abstract

The previous literature on capital structure has produced plenty of potential

determinants of leverage over the last decades. However, their research mod-

els usually cover only a restricted number of explanatory variables, and many

suffer from omitted variable bias. This study contributesto the literature by advo-

cating a sound approach to selecting the control variables for empirical capital

structure studies. Weapplied linear LASSO inference approaches to evaluate the

marginal contributions of three proposed determinants; cash holdings, non-debt

tax shield, and current ratio. While some studies did not use these variables in

their models, others obtained contradictory results. Our findings have revealed

that cash holdings, current ratio, and non-debt tax shield are crucial factors that

substantially affect the leverage decisions of firms and should be controlled in

empirical capital structure studies.

KEYWORDS

determinants of capital structure, LASSO inference, leverage ratio

1 INTRODUCTION

Factors affecting firms’ capital structure decisions have

attracted the attention of corporate finance researchers

over the last half-century. Early findings reveal that

the leverage ratio decreases with firm profitability and

growth potential and increases with firm size and tangi-

bility (Rajan & Zingales, 1995). More recently, researchers

focused on testing new variables’ role in capital structure

decisions. These efforts have revealed other potential firm,

industry- and country-specific determinants of leverage,

such as firm age, industry munificence, inflation, financial

orientation, and bank concentration (Antoniou et al., 2008;

Baum et al., 2017;Bilgin&Dinc,2019; De Jong et al., 2008;

Frank & Goyal, 2009; Gonzalez & Gonzalez, 2008;Kayo

&Kimura,2011; Kieschnick & Moussawi, 2018; Lim et al.,

2020). However, the empirical studies are far from provid-

ing comparable results since there is no consensus on the

correct model specification, the estimation of the leverage

ratio, and the selection of control variables. Although the

previous literature discusses the first two issues to some

extent, it shows little interest in selecting control variables.

This study contributes to the literature by applying a sound

approach to this selection process.

When the number of explanatory variables is close to the

number of observations, conventional econometric mod-

els, such as OLS, overfit the noise (Nagel, 2021). As a result,

capital structure models usually cover only a restricted

number of explanatory variables and have limited predic-

tive power. Besides, the control variables of these models

are generally selected arbitrarily depending on data avail-

ability and p-hacking. Model misspecification problems

(i.e., omitting relevant variables or including irrelevant

ones) may cause biased and inefficient estimates and

244 © 2023 Wiley Periodicals LLC. J Corp Account Finance. 2023;34:244–255.wileyonlinelibrary.com/journal/jcaf