Is firm growth random? A machine learning perspective

Research output: Contribution to JournalArticleAcademicpeer-review

Abstract

This study contributes to the firm growth debate by applying machine learning. We compare a prominent machine learning technique – random forest analysis (RFA) – to traditional regression in terms of their goodness-of-fit on a dataset of 168,055 firms from Belgium and the Netherlands. For each of these firms, we have one to six years of historical data involving demographic and financial information. The data show high variation in firm growth rates, which is difficult to capture with traditional linear regression (R2 in the range of 0.05–0.06). The RFA fares three to four times better, achieving a much higher goodness-of-fit (R2 of 0.16–0.23). RFA indicates that perhaps firm growth is less random than suggested by traditional regression analysis. Generally, given the modest selection of variables in our dataset, this demonstrates that machine learning can be of value to firm growth research.

Original languageEnglish
Article numbere00107
JournalJournal of Business Venturing Insights
Volume11
DOIs
Publication statusPublished - 1 Jun 2019

Fingerprint

Learning systems
Linear regression
Regression analysis
Firm growth
Machine learning
Goodness of fit

Cite this

@article{4a14884d31a940d987a8064121dc275c,
title = "Is firm growth random? A machine learning perspective",
abstract = "This study contributes to the firm growth debate by applying machine learning. We compare a prominent machine learning technique – random forest analysis (RFA) – to traditional regression in terms of their goodness-of-fit on a dataset of 168,055 firms from Belgium and the Netherlands. For each of these firms, we have one to six years of historical data involving demographic and financial information. The data show high variation in firm growth rates, which is difficult to capture with traditional linear regression (R2 in the range of 0.05–0.06). The RFA fares three to four times better, achieving a much higher goodness-of-fit (R2 of 0.16–0.23). RFA indicates that perhaps firm growth is less random than suggested by traditional regression analysis. Generally, given the modest selection of variables in our dataset, this demonstrates that machine learning can be of value to firm growth research.",
author = "{van Witteloostuijn}, Arjen and Daan Kolkman",
year = "2019",
month = "6",
day = "1",
doi = "10.1016/j.jbvi.2018.e00107",
language = "English",
volume = "11",
journal = "Journal of Business Venturing Insights",
issn = "2352-6734",
publisher = "Elsevier Inc.",

}

Is firm growth random? A machine learning perspective. / van Witteloostuijn, Arjen; Kolkman, Daan.

In: Journal of Business Venturing Insights, Vol. 11, e00107, 01.06.2019.

Research output: Contribution to JournalArticleAcademicpeer-review

TY - JOUR

T1 - Is firm growth random? A machine learning perspective

AU - van Witteloostuijn, Arjen

AU - Kolkman, Daan

PY - 2019/6/1

Y1 - 2019/6/1

N2 - This study contributes to the firm growth debate by applying machine learning. We compare a prominent machine learning technique – random forest analysis (RFA) – to traditional regression in terms of their goodness-of-fit on a dataset of 168,055 firms from Belgium and the Netherlands. For each of these firms, we have one to six years of historical data involving demographic and financial information. The data show high variation in firm growth rates, which is difficult to capture with traditional linear regression (R2 in the range of 0.05–0.06). The RFA fares three to four times better, achieving a much higher goodness-of-fit (R2 of 0.16–0.23). RFA indicates that perhaps firm growth is less random than suggested by traditional regression analysis. Generally, given the modest selection of variables in our dataset, this demonstrates that machine learning can be of value to firm growth research.

AB - This study contributes to the firm growth debate by applying machine learning. We compare a prominent machine learning technique – random forest analysis (RFA) – to traditional regression in terms of their goodness-of-fit on a dataset of 168,055 firms from Belgium and the Netherlands. For each of these firms, we have one to six years of historical data involving demographic and financial information. The data show high variation in firm growth rates, which is difficult to capture with traditional linear regression (R2 in the range of 0.05–0.06). The RFA fares three to four times better, achieving a much higher goodness-of-fit (R2 of 0.16–0.23). RFA indicates that perhaps firm growth is less random than suggested by traditional regression analysis. Generally, given the modest selection of variables in our dataset, this demonstrates that machine learning can be of value to firm growth research.

UR - http://www.scopus.com/inward/record.url?scp=85058182047&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85058182047&partnerID=8YFLogxK

U2 - 10.1016/j.jbvi.2018.e00107

DO - 10.1016/j.jbvi.2018.e00107

M3 - Article

VL - 11

JO - Journal of Business Venturing Insights

JF - Journal of Business Venturing Insights

SN - 2352-6734

M1 - e00107

ER -