Corporate cash policy and double machine learning - International Journal of Finance & Economics special issue article

Aug 3, 2026

Upload by Aria

This study examines firm-level drivers of corporate cash policy using double machine learning (DML), aiming to uncover how high-dimensional determinants shape firms’ cash holdings. The analysis identifies asset tangibility and R&D spending as key drivers of cash increases when assessed separately and jointly. The findings further align with transaction cost mechanisms and the refinancing risk of long-term debt early in the sample period, while the precautionary motive appears more relevant in more recent times. Robustness checks confirm reliability across alternative machine learners, cash proxies, and estimation methods.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages19

Reading Time48 min

AccessFree to download

Introduction

Economic motives for holding cash

Machine learning approach and double/debiased ML

Estimation strategy and robustness

Received: 29 August 2023 Revised: 13 May 2024 Accepted: 3 August 2024 DOI: 10.1002/ijfe.3039 SPECIAL I SSUE A RTICL E Corporate cash policy and double machine learning Hadi Movaghari | Serafeim Tsoukas | Evangelos Vagenas-Nanos Adam Smith Business School, University of Glasgow, Glasgow, UK Correspondence Hadi Movaghari, Adam Smith Business School, University of Glasgow, Glasgow G11 6EY, UK. [Email: hadi.movaghari@glasgow.ac.uk](Email: hadi.movaghari@glasgow.ac.uk) Abstract We are the first to explore the role of firm-level drivers in corporate cash policy applying cutting-edge double machine learning technique. We identify tangibility of assets and R&D spending as two main driving forces behind the cash increase when they are considered both independently and jointly. Furthermore, our findings support the relevance of the transaction cost model and therefinancing risk of long-term debt at the beginning of the sample period. In contrast, precautionary motive emerges as more pertinent in contemporary times. Our results are robust to alternative machine learners, cash proxies and estimation methods. KEYWOR DS cash holdings, double machine learning, machine learning, precautionary motive, simultaneous causal effect, transaction motive 1 | INTRODUCTION The last two decades have seen phenomenal growth in the theoretical and empirical literature seeking to explain why firms hold cash. The four main explanations offered by the seminal studies are related to precautionary motives, misaligned managers' incentives, transaction costs and taxation deferral (for a detailed review, see Graham & Leary, 2018) . Much of the previous work uses accounting ratios and other publicly available information in reduced-form models in order to determine firms'propensity to accumulate cash. More recent studies propose some new factors (or drivers) such as the cost of carry, debt maturity, intangible assets, R&D spending, asset tangibility, tax costs of repatriating earnings, industrial diversification, relationship with customers, and multinationality to explain corporate cash holdings. This issue creates a high-dimensional set of potential determinants of cash holdings. However, the literature has not settled on a universally accepted set of the most salient determinants of cash holdings. Using machine learning techniques across an extensive sample period, the present study aims to exploit complex patterns and high dimensionality in cash behaviour and quantify the relative importance of its determinants. In addition, we examine the evolution of the factors that cause changes to firms'cash holdings over time and across industries. In this study, we rely on the cutting-edge double (or debiased) machine learning (DML) procedure of Chernozhukov et al. (2017, 2018), which connects the theoretical work on nonparametric and semiparametric methods with machine learning. It is a framework for casual inference that provides estimates that are ‘root-nconsistent’. DML is particularly appealing in the context of cash holdings for the following reasons. First, changes in cash holdings are infrequent, vary over the business cycle, involve discontinuous adjustments, and in the presence of investment lumpiness and costly external finance, there is a nonlinear cash policy (Almeida et al., 2004; Tsoukalas et al., 2017) . Therefore, machine learning models that fit complex, nonlinear functional forms can lead to substantial improvement in prediction This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. © 2024 The Author(s). International Journal of Finance & Economics published by John Wiley & Sons Ltd. 3262 MOVAGHARI ET AL. accuracy. Second, identifying the most relevant predictors from an extensive set of candidate variables can prove challenging leading to omitted variable concerns,

FAQs

What technique does the paper use to study corporate cash policy?

It uses double (debiased) machine learning (DML) to estimate causal effects in a high-dimensional setting with flexible modeling of nonlinear relationships.

Which drivers are identified as main contributors to cash increases?

Asset tangibility and R&D spending emerge as the two main driving forces behind the cash increase, both in separate and joint analyses.

How do the results relate to common theories of cash holdings?

The evidence supports transaction cost modeling and refinancing risk of long-term debt at the start of the sample period, while the precautionary motive becomes more pertinent in contemporary times.

Aug 3, 2026

Upload by Aria