Economic forces and the stock market revisited

https://doi.org/10.1016/j.jempfin.2005.09.001Get rights and content

Abstract

The pricing of the Chen, Roll, and Ross (CRR) macrovariables is re-examined and found to be surprisingly sensitive to reasonable alternative procedures for generating size portfolio returns and estimating their betas. These methods include the full-period post-ranking return approach used in many recent studies. Strong evidence of pricing is obtained only for their industrial production growth factor and, in another contrast, for the VW market index. In particular, the corporate-government bond return spread, an important factor in CRR, is insignificantly negative for the 1958–1983 period, corroborating the cross-sectional regression results.

Introduction

An important body of research in financial economics is concerned with the forces that determine the prices of risky securities, and there are a number of competing theories of asset pricing. These include the original capital asset pricing models (CAPM) of Sharpe (1964), Lintner (1965) and Black (1972), the intertemporal models of Merton (1973), Long (1974), Rubinstein (1976), Breeden (1979), and Cox et al. (1985), and the arbitrage pricing theory (APT) of Ross (1976). In each case a relation between expected return and one or more measures of exposure to systematic risk is derived.

In a CAPM framework, a security's systematic risk is measured by its beta with respect to a diversified stock index, the latter viewed as a proxy for the value-weighted market portfolio of all assets. Black et al. (1972), Blume and Friend (1973), and Fama and MacBeth (1973) are important early examples of such work. Roll (1977) criticizes the early studies, however, emphasizing that they are really tests of the mathematical hypothesis that the stock index is mean-variance efficient, and would reflect on the CAPM only if the true market portfolio were used in the tests.1

Sobered by Roll's conclusions, many researchers have turned to other approaches to risk-return analysis. Motivated by the APT, Roll and Ross (1980) employ factor-analytic methods to estimate multiple measures of systematic risk. Restrictions on the covariance matrix of returns are used to statistically identify the factor sensitivities, as the underlying factors are not observed in this context. Brown and Weinstein (1983), Connor and Korajczyk (1988) and Lehman and Modest (1988) are other studies of this sort.

A third approach to asset pricing empirical work, pursued in this paper, is advanced by Chan et al. (1985) and Chen et al. (1986), henceforth CCH and CRR. These studies look at pricing relative to a set of observable macroeconomic variables, or factors, selected primarily based on economic intuition.2 Although the authors appeal to the APT in motivating their work, the strong intuition underlying their choice of factors is derived, in large part, from the intertemporal models cited above. Thus, one can also view the factors more formally as a “multivariate proxy” for the unobservable equilibrium benchmark (see Shanken, 1987). In the spirit of Roll's critique, this view makes explicit the joint hypothesis concerning the factors that is inherent in such analyses.

This paper makes several contributions to research in the area. First, we examine some issues that arise when inferring whether a particular macroeconomic variable is a priced factor. Both CCH and CRR employ versions of the two-pass cross-sectional regression (CSR) methodology developed by Fama and MacBeth (1973). The reported “t-statistics” are impressive, suggesting that several of the factors are priced; i.e., the betas on these factors help explain the cross-sectional variation in mean asset returns. However, as shown in Shanken (1992a), the usual Fama–MacBeth standard errors on the estimated prices of risk are generally biased downward (asymptotically) due to the well-known errors-in-variables (EIV) problem in the second pass CSRs.3 We examine the extent to which inferences are affected by taking into account this source of measurement error.

Our main finding concerns the lack of robustness of the CRR/CCH results. Twenty size-based portfolios serve as the assets in their studies. Alternative experiments in which securities are grouped on the basis of market beta, standard deviation of return, or price level were conducted as well (see Footnote 8 of CRR), but were not successful. Although we too employ size-based groupings, the results are surprisingly sensitive to the specific way in which the portfolio returns are generated and βs are estimated. In particular, only the market index is significantly priced in our cross-sectional regressions on full-period post-ranking betas estimated from the size portfolio returns.

While we use the same factors as CRR, Ferson and Harvey (1991) explore several variations on the definition of the bond risk premium factor and report that its pricing is “highly sensitive to the definition of the variable and the subperiod”. In other related work, Warga (1989) finds that the CCH pricing estimates can be quite sensitive to employment of a jacknife procedure in which data is deleted 1 month at a time in estimating the betas. He does not assess the impact on inferences about the risk premia, however.

Several methodological innovations in this paper deal with the use of factors that are derived from portfolio returns. These methods are relevant, in particular, whenever some factors in the model are defined as the spread between two portfolio returns, as is now common in the literature, e.g., Fama and French (1993). Additional economic restrictions implied in this case are incorporated in both estimation and testing of the linear risk-return relation. Shanken's (1985a) CSR T2 test of linear factor pricing is adapted to this context and a new test, focusing directly on the factor portfolio constraints, is presented. Our application of these methods to the factors used by CRR/CCH allows us to combine evidence from the corporate and government bond markets with that from the common stock portfolios. We find that the risk premia estimates and the magnitude of the EIV adjustments to the standard errors are often sensitive to the inclusion of bond returns.

The rest of this paper is organized as follows. Section 2 discusses important aspects of the experimental design of this study. Pricing results and tests of multibeta risk-return linearity are presented in Section 3 for the basic 5-factor model, with size portfolios as the assets. Bonds as well as stocks are treated as assets in the 5-factor “restricted” model of Section 4, and Section 5 considers an expanded model that includes a stock market factor. Section 6 summarizes the paper and offers some conclusions.

Section snippets

Experimental design

The two-pass methodology involves (i) estimation of beta(s) for each asset in a first-pass time-series regression of asset returns on the given factor(s), and (ii) estimation of the risk-return parameters, i.e., the zero-beta rate and price(s) of risk, by a CSR of the returns for the given assets on the betas estimated in the first pass. These CSRs are performed each month, and the results are aggregated by averaging the time series of estimates for each of the risk-return parameters.

Often in

The unrestricted 5-factor model

In this section, the pricing of the 5 CRR factors is examined and tests of the validity of the 5-factor multibeta model are presented. We begin by considering the following model for excess returns:10Rpt=αp+β1pMPt+β2pDEIt+β3pUIt+β4pUPRt+β5pUTSt+εptwhere: Rpt = the excess return on size portfolio p for month t; MP = the percentage change in industrial production led by 1 month; DEI = the change in expected

The restricted 5-factor model

Thus far, only size portfolios have been employed as assets in our empirical analysis. The precision of the estimators and the power of the tests may be increased by including other assets. Stambaugh (1982) emphasizes this point in testing the CAPM. Two obvious candidates for inclusion are the long-term government bond and low-grade bond portfolios from which the factors, UTS and UPR, are derived. The analysis below is a natural extension of techniques used by Black et al. (1972), to the

An expanded factor model

The five CRR factors typically account for only about 25% or 30% of the time-series variation in our 20 size portfolio returns. The time-series R2 rises to about 80% when the excess return on the value-weighted CRSP stock index (VW) is included as a sixth factor. Thus, it seems likely that the usual market factor captures potentially important components of systematic risk not reflected in the other macro-factors. CCH/CRR find that the addition of an equity index does not have much of an effect

Summary and conclusions

In this paper, we have examined the relation between expected returns and measures of systematic risk with respect to five macroeconomic factors studied by Chan et al. (1985) and Chen et al. (1986). Like CCH/CRR, we use a version of Fama and MacBeth's (1973) two-pass methodology with securities grouped into portfolios based on annual rankings of the market value of equity (“size”). However, whereas CCH/CRR estimate betas using backward-looking returns, relative to the ranking dates, we employ

References (48)

  • J. Long

    Stock prices, inflation, and the term structure of interest rates

    Journal of Financial Economics

    (1974)
  • R. Roll

    A critique of the asset pricing theory's test: Part I. On past and potential testability of the theory

    Journal of Financial Economics

    (1977)
  • S. Ross

    The arbitrage theory of capital asset pricing

    Journal of Economic Theory

    (1976)
  • J. Shanken

    Multivariate tests of the zero-beta CAPM

    Journal of Financial Economics

    (1985)
  • J. Shanken

    Multivariate proxies and asset pricing relations: living with the roil critique

    Journal of Financial Economics

    (1987)
  • J. Shanken

    Intertemporal asset pricing: an empirical investigation

    Journal of Econometrics

    (1990)
  • R. Stambaugh

    On the exclusion of assets from tests of the two-parameter model: a sensitivity analysis

    Journal of Financial Economics

    (1982)
  • M. Vassalou

    News related to future GDP growth as a risk factor in equity returns

    Journal of Financial Economics

    (2003)
  • F. Black

    Capital market equilibrium with restricted borrowing

    Journal of Business

    (1972)
  • F. Black et al.

    The capital asset pricing model: some empirical tests

  • M. Blume et al.

    A new look at the capital asset pricing model

    Journal of Finance

    (1973)
  • S. Brown et al.

    A new approach to testing asset pricing models: the bilinear paradigm

    Journal of Finance

    (1983)
  • K.C. Chan

    On the contrarian investment strategy

    Journal of Business

    (1988)
  • K.C. Chan et al.

    An unconditional asset-pricing test and the role of firm size as an instrumental variable for risk

    Journal of Finance

    (1988)
  • Cited by (66)

    • The pricing of volatility risk in the US equity market

      2022, International Review of Financial Analysis
      Citation Excerpt :

      Furthermore, the inflation factors are significantly priced in Eqs. (6), (7) and (10) and the robustness factors OG, CG and TED are now priced as well. Interestingly the seasonally adjusted industrial production factor (MP-SA) now has the expected positive market price of risk as in Chen et al. (1986) and Shanken and Weinstein (2006). The premium of the value weighted equity return (VW) is still significant but slightly smaller than before.

    • Complex analytic wavelets in the measurement of macroeconomic risks

      2019, North American Journal of Economics and Finance
    • Regional economic activity and stock returns

      2019, Journal of Financial and Quantitative Analysis
    View all citing articles on Scopus

    Earlier versions of this paper with the title “Macroeconomic Variables and Asset Pricing: Further Results” were presented in seminars at Arizona State University, UC Berkeley, University of British Columbia, University of Chicago, Columbia University, University of Gotemborg, Haifa University, Hebrew University, University of Iowa, London Business School, Norwegian School of Economics, University of Oklahoma, Southern Methodist University, Stockholm School of Economics, Vanderbilt University, the University of Wisconsin, Yale University, the 1986 Western Finance Association meetings, and the 1987 American Finance Association meetings. Thanks to the participants and to S. Brown, N-F. Chen, D. Conway, E. Fama, L. Harris, S.P. Kothari, M. Reinganum, A. Christie, R. Roll, J. Warner and, especially, the editor Wayne Ferson for helpful comments and discussions. Shanken is grateful for financial support under the Batterymarch Fellowship Program and from the Managerial Economics Research Center at the Simon School, University of Rochester, and the Federal Reserve Bank of Atlanta. Weinstein is grateful for support from the Dean's Scholar Program of the U.S.C. Marshall School of Business.

    View full text