News-Based Macroeconomic Asset Pricing via SSRN
Yale University, School of Management, Students
Yale University, Department of Economics, Students
September 25, 2014
News contains valuable conditioning information that captures time-varying investor concern over different aspects of the macroeconomy. We create separate news measures to approximate investor concern about consumption growth, inflation, and unemployment by associating front-page articles in the Wall Street Journal with the respective macroeconomic factors. For each macroeconomic variable, we construct a news conditional macro model and compare its cross-sectional pricing power against that of the corresponding macro 1-factor beta-model. Next, we focus on our consumption-based news measure and simulate a long-short trading strategy based on news beta-loadings of NYSE and AMEX stocks. Strategy returns are significant and robust to adjusting for Fama French 3-factor risk. Finally, we show our news measures capture aspects of investor concern related to changes in macroeconomic forecasts. We discuss the implications of our consumption-based news measure for three important macro asset pricing models: Habit, Rare Disaster, and Long Run Risk. Our empirical findings do not support the Habit or Rare Disaster models. However, we find evidence in favor of the Long Run Risk model.
News-Based Macroeconomic Asset Pricing – Introduction
The asset pricing profession has long sought to connect prices with economic fundamentals. Among the first theoretical models to make this connection are the Capital Asset Pricing Model (CAPM), the Consumption CAPM, and the Arbitrage Pricing Theory. 2 Despite the intuitive link between asset prices and economic conditions, macro-based theoretical models have had limited success in empirical tests. Measuring unobserved state variables related to time-varying investor marginal utility has been a major obstacle in empirical applications.
News-based measures of investor concern offer a largely unexplored channel to capture important aspects of marginal utility related to the time-varying risk premia. Intuitively, the news reflects the important topics and issues of the day. We exploit this property of the news in constructing a measure to approximate investor concern. We contend that shocks which effect changes in investor concern will likely be reflected in news coverage. In this case, our measure of the news can provide a way to incorporate fundamental shocks as conditioning information while remaining agnostic about any underlying model. For example, legislation enacting a strict pollution policy would have an impact on the economy and would draw investor attention but would not be detected in any contemporary macroeconomic series. In this way, news offers a way to measure investor concern that will allow us to incorporate valuable conditioning information reflected in time-varying risk premia.
To this end, we develop a proxy for investor concern using the front page of the Wall Street Journal, a publication which has maintained its role of being US investors’ main source of information about significant market-wide news for over a century. We create three separate news measures to capture time-varying investor concern about the following important macroeconomic variables: consumption growth, federal funds rate, and unemployment rate. Each measure uses news coverage on subjects pertaining to the associated macroeconomic factor to capture investor concern about that macro factor. For example, the consumption growth measure is elevated in the 2001 recession and in the energy price spike in 2006 while the federal funds rate measure is elevated before each Fed Chair transition.
Ultimately, many shocks will impact asset prices through their effect on investor concern. We start from the well-studied one-factor beta-models based on real consumption growth, federal funds rate, and unemployment rate respectively. We augment each of these models with the corresponding news measure and compare the asset pricing performance. If our news measure captures important aspects of time-varying investor concern then the augmented models should provide significant asset pricing improvements over the corresponding one-factor beta-models. As such, the primary purpose of this paper’s empirical tests is to determine whether our news measure actually provides conditioning information with important asset pricing implications.
We construct our augmented models by combining each one-factor beta-model with the corresponding news measure as a state variable to obtain a 3-factor model with a macro
term, a news term, and a cross-term. We test our model using standard two-pass cross-sectional regressions on 51 test portfolios. For all of the tested models we observe significant news related coefficients. More importantly, conditioning on news dramatically improves the pricing power of each of the one-factor beta-models based on the improvements in adjusted R2. This increase is statistically significant for all of the macroeconomic factors when bench-marked against a distribution of adjusted R2 from noise models generated from Monte Carlo simulations. Our results indicate that news pertaining to each macroeconomic variable contains important conditioning information consistent with our hypothesis that news captures variation in investor concern.
See full PDF below.