Time-varying volatilities, CAPM betas, and factor loadings: A high-frequency data perspective

Betas or Factor Loadings in Multifactor Pricing models are the most fundamental risk measures of equity returns. I propose a nonparametric measure of latent betas---Realized Beta, or Realized Factor Loadings in multifactor pricing model.; This idea originated from the emerging concept of realized volatility that has produced new insights in modeling financial market volatility: by summing squared intraday returns from tick-by-tick prices, it is possible to measure more accurately the ex-post volatility and covariance, and therefore the beta over a fixed time interval. The realized beta has three features: it is observable; it is time-varying, and it will converge to true beta as the sampling frequency becomes infinitesimal under standard assumptions. Therefore, the realized beta approach provides researchers and practitioners with a powerful method in measuring, modeling and forecasting betas.; I apply the realized beta approach to two important subjects in financial economics: (1) News Asymmetry property of realized betas in Capital Asset Pricing Model and (2) Factor Representation and Return Forecasting in the Fama-French three-factor model.; It has been long suspected that the conditional beta of an asset usually goes up after negative news hit the market---the so-called leverage effect. However, a host of studies using multivariate LARCH type models found little supporting evidence from the data. In the chapter titled "News Asymmetry in Volatility and CAPM Betas---a High Frequency Approach", I tackle the beta asymmetry issue by using two approaches: the bivariate EGARCH model proposed in Braun, Nelson and Sunier (1995, Journal of Finance) and the realized beta approach. I find that the EGARCH model only identifies one industry (the construction sector) as exhibiting beta asymmetry at a daily level between 1993-2001; however, models built on realized beta reveal a systematic pattern of news asymmetry: the betas of cyclical industries (including construction) rise after bad news hits the market. Furthermore, compared to bivariate EGARCH dynamic betas, the high-frequency-based realized betas yield smaller in-sample pricing errors (alphas) for most of the 12 industry portfolios.; In "Measuring and Modeling Systematic Risk in Factor Pricing Models using High-Frequency Data", I construct realized factor loadings of the popular Fama-French three-factor model. Monthly realized factor loadings and returns are extracted from 5-minute intraday returns series of 25 size and book-to-market sorted portfolios between 1993--1999. Once again, the high-frequency based realized factor loadings outperform the conventional rolling regression loadings and constant loadings---not only in factor representations, but also in out-of-sample return predictions. In a mean-variance optimization experiment, the dynamic trading strategy based on realized factor loadings forecasts produces consistently higher Sharpe ratios than does the strategy based on conventional loading forecasts. Furthermore, with a quadratic utility function and risk aversion coefficient of 10, a representative investor is willing to pay up to 4.5% per year of returns in order to switch from the trading strategy of rolling regression forecast loadings to the ones using high-frequency-based realized factor loading forecasting.