On Assessing Valuation Robots

Abstract

Valuation is the act of determining how much an investment is worth. This analysis is performed on a daily basis by market analysts who try to set the fair price of a company in order to decide whether or not to invest on it. This task is quite challenging since it requires estimating non-observable parameters such as the expected future performance of a company and how much return do the investors expect to get out of it.

This thesis provides a statistical framework for assessing systematic valuation agents. The framework allows for a medium and long term evaluation of quantamental algorithms in a wide range of perspectives such as value-pricing bias, information assimilation, forecast-actual bias, model sensitivity, value components and financial performance.

The main assessment methods here are designed for discounted models, although some are easily adapted to multiple based valuation models. This allows for a rich comparison across the value components of discounted cash flows and multiple based approaches. Additionally, these methods can be applied for assessing human analysts provided that they have enough points on their forecast history.