Relative equity valuation: an algorithmic multiple regression approach in Python = Relatív vállalatértékelés: egy többváltozós regressziós megközelítés Python algoritmusokkal

Absztrakt (kivonat)

The main motivation behind the paper is to present multiple linear regression models as an alternative for market multiples, which appear to be the only way of doing relative equity valuation. To be more precise, the goal is to find a method for valuing the equity of private companies based on their accounting metrics by relying on comparison with the valuation of public companies – bypassing a lengthy bottom-up calculation. Despite being seemingly more complicated than traditional techniques, the outlined regressions are far easier to use once understood, as they do not require manual peer selection and are more or less applicable in Excel. According to Damodaran (2006), one of the most famous equity appraisal theorists, “Most equity research reports and many acquisition valuations are based upon a multiple such as a price to sales ratio or the value to EBITDA multiple and a group of comparable firms.” Even when a discounted cash flow model is used to calculate the value of equity, multiples are often used as a sanity check for the results. Despite their popularity, however, the literature on multiples is already very sparse in general and there is barely anything available on other relative valuation methods. Damodaran himself advocates the use of multiple linear regression as an alternative approach (Damodaran, 2006), but his model lacks technical statistical considerations, and is more of a side note than a true demonstration. This deficiency is especially striking as nowadays big data, predictive analytics and machine learning seem to be everywhere. For equity valuation this means hundreds of unique features for every public company, of which there are tens of thousands globally. All of that easily accessed through data vendors like Bloomberg or Capital IQ, to one of which most business schools, corporate finance advisors and investment firms are subscribed to. Yet most of that incredible wealth of information is wasted when relative valuations consists of simple multiples. Now neither the ease of using multiples, nor the business educational background of potential users is lost on the author, therefore the paper will try to remain as non-technical as possible and try to extensively use intuitive reasoning that is easy to understand even with limited statistical knowledge. It is also worth mentioning, that even though the paper relies entirely on Python algorithms, the results are easy to implement is any statistical software or programming language and are also more or less applicable in Excel, which is the golden standard in corporate 2 finance. Python itself is completely open source –and therefore free– which can offset the cost of time invested in learning its use. Since multiples are the “industry” golden standard, any suggested alternative is expected to provide some proof of its usefulness. In this case, that proof will be a demonstration of superior estimation accuracy and generality. The first section of the paper will be dedicated to introducing the multiples approach, its assumptions and the empirical evaluation of its performance. Afterwards the second section will present statistical concepts that can help build a better model, demonstrate multiple example models and measure their accuracy using the multiples as benchmarks. It should be emphasized that the paper does not claim to have found the best possible model or even anything remotely close. In fact, the final section will identify multiple conspicuous problems that need to be solved and should be subject to further investigation. Even so, these models are easy to use and can very well outperform even the best multiple estimations. With the apparent lack of similar research the findings can also be useful from a theoretical point of view.