Introduction of Machine Learning for Valuation of Unlisted Companies

Rasmus Dahl Lassesen & Frederik Kjær Hansen

Student thesis: Master thesis


In an investment-and advisory industry, analytical processes are critical, to evaluate whether an investment is to be commenced or rejected. Professional corporate finance advisors and investors spends hours of human resources and capital to identify a potential investment opportunity. Through interviews of industry stakeholders and a literature review, this thesis has found, that these primarily still makes use of traditional valuation methods. Same interviews indicate, that technological development, such as machine learning, have not gained ground in the industry yet. A technology which has already proven its worth in similar industries, such as real estate, accounting, and banking etc.
Behind this lies the motivation of the present thesis, which purpose is to examine whether the introduction and development of machine learning models can optimize the current processes of valuation for the target audience. The criteria of success for this thesis, is not to develop a machine learning model, which can be implemented immediately, more so to evaluate whether this method has justification for existence.
The thesis follows a neo-positivist perspective, where methodology consists of both qualitative- and quantitively gathering of data to answer the issues of the thesis. The development of machine learning models is based on financial data from almost 1.200 publicly listed companies in the selected food-and drinks industry and their corresponding market capitalization. To evaluate the prediction accuracy of the machine learning models, the estimates are tested and benchmarked against a traditional valuation of an unlisted company in the selected industry.
The analysis shows that the versions of machine learning that indicate best predictability of estimates, is linear regression and neural network models. Based on discussion and comparison of benchmark valuation, the neural network is considered the model with greatest potential to uncover the purpose of the thesis. The thesis concludes, that the predictability of the neural network is satisfactory, but the limitations of the model at its current status, makes it insufficient for implementation in the processes of valuation at the target audience. Given that the model is rectified, the thesis believes that the model can be useful in the screening process and thereby optimize the valuation process and free up resources at the investment-and advisory industry.
The thesis ultimately concludes, that machine learning models show justification for existence.

EducationsMSc in Finance and Accounting, (Graduate Programme) Final Thesis
Publication date2020
Number of pages143