Stock price forecasting and trading strategy implementation : a fundamental analysis approach using hierarchical bayesian models and univariate time-series models

Abstract

The equity valuation model proposed by Ohlson (1995) seeks to use quarterly data from public companies’ financial statements in order to give adequate confidence intervals for forecasted stock prices. This model does so by applying a time-series regression model, using the company’s book value per share and expected abnormal earnings per share for the following four quarters as inputs, with the company’s stock price serving as this model’s output. By making use of analysts’ forecasted earnings, Ying et al. (2005) apply a Hierarchical Bayesian approach to the Ohlson model, showing that this method improves upon the classical Frequentist approach. In this dissertation, we apply a simplified version of the Hierarchical Bayesian model proposed by Ying et al. (2005) to the constituent companies of the S&P 100 index. Instead of using analysts’ forecasted earnings, we make use of univariate seasonal ARIMA models, whose specifications are decided through the use of Genetic Algorithms, thus building on the works of Lai and Li (2006). The book value per share is also estimated in a similar fashion. A combination of yield curve functions and linear regressions are also used to determine estimates for the risk-free rate, as described in Svensson (1995) and Dominguez and Novales (2002), respectively. In this regard, we propose a self-sufficient equity valuation model that does not make use of analysts’ consensus estimates. Following the implementation of our model to the constituent companies of the S&P 100 index, we explore the viability of creating a trading strategy that manages to consistently beat the market. This is done by backtesting our proposed trading strategy using a 36-month validation dataset, subsequently comparing the results to a simple buy-and-hold approach of the S&P 100 index. Our conclusions present a trading strategy with an average annual return of 23.23%, an alpha of 11.28%, a beta of 1.33, and an R-Squared of 0.79. Given such exceptional results, the validity of the Efficient Market Hypothesis comes into question.