Monetary policy and stock valuation: structural VAR identification and size effects

Abstract

This paper examines the relationship between the US monetary policy and stock valuation using a structural VAR framework that allows for the simultaneous interaction between the federal funds rate and stock market developments based on the assumption of long-run monetary neutrality. The results confirm a strong, negative and significant monetary policy tightening effect on real stock prices. Furthermore, we provide evidence consistent with a delayed response of small stocks to monetary policy shocks relative to large stocks.

Keywords:

• Monetary policy
• Stock market
• Size effect
• SVAR

View correction statement:

Erratum

Acknowledgements

We would also like to thank Tom Doan for providing the estimation code and helpful advice.

Notes

¹ The global financial crisis of 2007–2009 has rekindled the academic debate of the early 2000s regarding the appropriate response of monetary policy to financial developments (Bernanke and Gertler 1999, Cecchetti et al. 2000). Recent empirical studies suggest that monetary policy focus during the crisis may have shifted from the price stabilisation objective towards financial stability (Baxa et al. 2013).

² See Sellin (2001) for a thorough survey of the early studies investigating the link between monetary policy and stock prices.

³ Using a combination of high- and low-frequency data, D’Amico and Farka (2011) also provide evidence supporting the strong interdependence between the US stock market and monetary policy. In particular, they find that monetary policy tightening is associated with declining stock market, while the Fed responds to a positive stock market shock by raising the policy rate.

⁴ Similar results are also reported by Bjornland and Jacobsen (2013) for the longer sample period 1983–2010.

⁵ The combination of short- and long-run restrictions also addresses the criticism towards the identification schemes solely based upon long-run restrictions (Faust and Leeper 1997).

⁶ The price puzzle refers to a positive response of goods’ prices following a contractionary monetary policy shock. According to Sims (1992), the price puzzle may be generated if some information that policy makers have about inflationary pressures is not included in the model.

⁷ The results are robust to the exclusion of this dummy.

⁸ In line with Bjornland and Leitemo (2009), the series of annual consumer price inflation as well as the annual change in the leading economic indicator are differenced to stationarity for the sample period considered.

⁹ This implies that the matrix Σ_ε is the identity matrix I. Alternatively, the diagonal elements of Σ_ε may be left unconstrained and, instead, either the diagonal elements in contemporaneous matrix could be set to unity or two approaches can be combined (Kilian 2011).

¹⁰ After taking into account the short-run restrictions, equation (13(13) ) shrinks to .

¹¹ The sample period (as reported) excludes the months reserved for the lags of endogenous variables.

¹² Our main results are largely unchanged when we use quadratic detrending to measure the output gap. The correlation between the two measures of output gap is 0.44 and 0.52 over the sample periods 1994:2–2007:7 and 1994:2–2008:12, respectively. The output gap based on quadratic detrending exhibits larger extreme values that may be implausible (see figure A1 in Online Appendix A). Thus, the HP filter is our preferred approach. See Weidner and Williams (2009) for a discussion of the size of output gap in the US.

¹³ Figure A2 in Online Appendix A plots these time series over the sample period 1994:2–2008:12.

¹⁴ The data library is accessible via this link: http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html#Research.

¹⁵ Plots of monthly returns on the size-sorted portfolios are provided in figure A3 of Online Appendix A.

¹⁶ According to Sims and Zha (1999), the traditional symmetric confidence intervals, reported as one or two standard errors around the point estimate of an impulse response function, may be misleading. Such error bands confound information regarding the location of the coefficient values with the information about the overall model fit. They show that the Bayesian posterior probability bands, simulated using Monte Carlo integration, may be more useful than the confidence intervals based upon the estimates of standard errors. The fractiles correspond to one standard deviation if the standard error bands were used.

¹⁷ The draws are made from the posterior distribution for VAR parameters and residuals under the standard uninformative (flat) prior for a multivariate regression model (Doan 2015). It is important to note that the structural model depends on the new set of parameters for each Monte Carlo draw and model’s parameters must be reset accordingly prior to the generation of new impulse response functions. This is unlike Bjornland and Leitemo (2009) where the original OLS estimates of VAR coefficients are used in in each draw (Doan 2015).

¹⁸ Specifically, in figure 2 of their paper, Brissimis and Magginas (2006) show that following a contractionary monetary policy shock, consumer prices decline continuously, and significantly so, over a horizon of 48 months. At the same time, the policy rate returns to the initial level very quickly and its impact on industrial production becomes insignificant after one year.

¹⁹ This finding appears to be at odds with monetary policy neutrality. We should point out, though, that in our framework, the neutrality restriction refers to the absence of a long-run effect from monetary policy on the price of stocks. Moreover, the notion of monetary policy non-neutrality is challenged by the findings of some recent studies. Hanson and Stein (2015, p. 430) admit that ‘none of our evidence directly refutes the long-run non-neutrality hypothesis that policy is somehow able to move expected real rates far out into future’. Nakamura and Steinsson (forthcoming) examine the impact of monetary policy shocks, identified using a high-frequency data approach, on real interest rates, expected inflation, and expected output growth. Their evidence suggests that real interest rates increase several years out in the future in response to a monetary policy tightening shock. We thank an anonymous referee for raising this point.

²⁰ Results related to the reaction of monetary policy and macroeconomic variables to price developments in different size deciles are available upon request.

²¹ The tendency for the magnitude of the policy impact to increase as we move from smaller to larger stocks is non-monotonic.

²² This argument is consistent with the greater trading activity found for larger stocks (Chordia, Huh and Subrahmanyam 2007).

²³ We also used two lags and obtain consistent results. These findings are available upon request.

²⁴ Two additional dummies are included as exogenous variables. The first (second) is equal to one in September (October) 2008 and zero otherwise. They capture the effects of the collapse of Lehman Brothers.