Equity Valuation - Seeking an Understanding through Alternative Approaches

Abstract

This dissertation involves measuring and testing the empirical performances of equity pricing models. The first paper extends the constant discount factor model with intrinsic bubbles developed in Froot and Obstfeld (1991) to account for autocorrelation in dividend growth rates. We derive an analytical expression for both the present value stock price and an intrinsic bubble component when dividend growth rates evolve as a Gaussian AR(1) process. Hypotheses tests favor an AR(1) process for dividend growth rates and an AR(1)-based model developed here for price-dividends ratios over a benchmark case. Hypotheses tests also reject the absence of a bubble component in stock prices. Incorporating the bubble component into our model provides a significant improvement in fit to observed P/D ratios and stock prices. The second paper assesses the empirical implications of the residual income model developed in Ohlson (1995). A key assumption stipulates that next period t+1 residual income is a linear function of current period residual income and a latent variable referred to as ‘other information’. This ‘other information’ is posited to contain information on next period t+1 residual income and reflected in current stock prices. We propose to estimate this latent ‘other information’ variable using a state space framework. We estimate the valuation model, within the embedded state space framework, using the Kalman filter. Performance yardsticks indicate that our state space estimation approach shows promise in valuing stocks. The third paper attempts to estimate and study the role of ‘other information’ vt, as theorized in the Ohlson (1995), for tracking contemporaneous returns and predicting future returns of the S&P 500. vt is unobserved and is defined as a summary of value-relevant information about future profitability. This suggests a potential to predict subsequent returns. We apply a factor augmented vector autoregression (FAVAR) to estimate vt and evaluate its predictive performance. The FAVAR model enables us to estimate unobserved factors that are broadly captured by big data. We use principal components estimation to extract the unobserved factors from a rich set of data. Our analysis shows that the estimated vt has statistically reliable power to predict future returns.