Asymmetric and cross-sectional effects of inflation on stock returns under varying monetary conditions

Abstract

This study examines the relationship between returns on portfolios, comprised of stocks of various size and book values, and changes in inflation. The relationship is evaluated in the context of positive and negative changes in expected and unexpected inflation, and expansionary and contractionary monetary policy conditions. Results from the panel estimation procedure show that inflation has a strong asymmetric impact on stock returns, and this may explain why simply summing up inflation shocks, as in previous studies, could lead to misleading conclusions. This article concludes that the nature of this asymmetric relationship is complex and contingent on several factors including the state of the monetary policy, whether one is examining expected or unexpected inflation shocks, and the size and book values of the stocks. In general, a positive shock to expected and unexpected inflation has a favourable impact on stock returns during monetary expansion, but not during monetary tightening.

Keywords:

• inflation
• stock returns
• asymmetry
• monetary policy

JEL Classification::

• G10
• E50

Notes

¹ A partial list of studies in this context include Lintner (1975), Nelson (1976), Fama and Schwert (1977), Cohn and Lessard (1981), Fama (1981, 1982), Geske and Roll (1983), Gultekin (1983), Boudoukh and Richardson (1993), Lee (1992) and Ely and Robinson (1997), among others.

² The negative relationship between inflation and real economic activity is consonant with the Keynesian view which posits that higher inflation has an adverse effect on new investments, thus reducing aggregate demand/supply and overall output.

³ In addition to the proxy and policy anticipation hypotheses, several other explanations of the negative relationship between inflation and stock returns have been proffered such as the nominal contracting hypothesis (Bernard, 1986; Pearce and Roley, 1988), taxes (Feldstein, 1980), wealth effects (Stulz, 1986) and money illusion (Modigliani and Cohn, 1979).

⁴ Several studies have documented the importance of size and BE/ME factors in explaining the cross-sectional properties of equity returns. Examples of such studies include Stattman (1980), Banz (1981), Reinganum (1981), Keim (1983), Rosenberg et al. (1985), Dimson and Marsh (1999), Fama and French (1993, 1995, 1996), Barber and Lyon (1997), Lewellen (1999) and Perez-Quiros and Timmerman (2000).

⁵ For a proof, see Johnston (1972, pp. 157–159).

⁶ As an alternative measure of inflation, the Producer Price Index (PPI) variable was considered. The use of PPI did not qualitatively alter the results reported in this article.

⁷ For instance, studies such as by Fama (1981), Fama and Schwert (1977) and Hartzell et al. (1987) among others have used Treasury bill rate as a proxy for expected change in inflation. The measure of unexpected change in inflation, under this framework, is the difference between actual inflation and the bill rate as calculated on an ex post basis. Unfortunately, this approach does not accommodate for time-varying real rates. Still others, such as Yobaccio et al. (1995), have used autoregressive time-series techniques to separate the expected and unexpected components of inflation.

⁸ Studies investigating the properties of survey forecast data have found them to be unbiased and superior to estimates that are produced by autoregressive models by virtue of their lower mean squared errors (McQueen and Roley, 1993; Almeida et al., 1998). Not surprisingly, survey data on macroeconomic forecasts have been used in several recent studies in a variety of contexts, such as the examination of currency markets (Andersen et al., 2003), yield curve modelling (Balduzzi et al., 2001) and Treasury futures prices (Simpson and Ramchander, 2004), among others.

⁹ To simplify the presentation, we use the symbol pi (π) to represent the change in inflation.

¹⁰ See http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html.

¹¹ This measure which is designed to identify a shift in the broad monetary policy stance of the Fed has been adopted by several studies including Booth and Booth (1997), Fujimoto (2003) and Bauer and Vega (2006).

¹² The Fisher hypothesis relates the level of returns to the level of inflation. Given that our inflation-related variables are measured in changes, it would be appropriate to use changes in stock returns on the left-hand side of the regressions, for consistency.

¹³ We also ran all of the models assuming symmetrical responses; the result did not vary much form those presented in Table 2, model 3. Thus, we have omitted these results from Tables 4 and 5 in order to save space.