Valoración de startups con Aprendizaje Automático

Abstract

In this project the author will develop a large-scale automatic valuation method for companies (with a strong focus on startups). The main challenge is the lack of nancial information which is usually used when valuating companies with traditional methods. This problem will be overcome using a Machine Learning approach. The models will be fitted using collected data from sources from the Internet using web scrappers & crawlers. After a holistic analysis on the theoretical valuation of companies, it will be concluded that the valuation formula will lay its basis on the Discounted Free Cash Flows model. Also, a research of state-of-the-art Machine Learning techniques will take place. Then, an estimation model for each one of the formula's inputs will be built accordingly with the company's information that is known beforehand: in order to determine the company's economic activity from its description an expert rule-based system will be implemented. The company's revenues will be estimated using a meta-estimator that fits 700 randomized decision trees with data related to the number of employees, economic activity, social traction and other features. A similar estimator will also be used, with the help of an expert system, to transform those revenues into free cash ows. On the other hand, the risk of failure, growth and discount rate will be determined with the development of a system that crawls people's interest trends and matches the already known metrics of companies with the target ones according to its economic activity similarity.