Feedback trading and the behavioural ICAPM: multivariate evidence across international equity and bond markets

Abstract

In this article we develop a ‘behavioural’ Intertemporal Capital Asset Pricing Model (ICAPM) in which the behavioural impetus comes from the feedback trading implications for the autocorrelation of returns. We apply the model in a setting of paired equity and bond investments, employing a bivariate diagonal Berndt–Engle–Kraft–Kroner (BEKK) framework. Our empirics rely on daily equity and bond index returns across six major economies, over the period 1 January 1990 to 30 June 2005. We find evidence supporting the theory that the observed dynamics of serial correlation can be a function of both volatility and conditional covariance (between equity and bonds). Moreover, our behavioural ICAPM shows empirical promise as a useful model of asset pricing in markets that display the feedback trading phenomenon.

Keywords:

• feedback trading
• autocorrelated returns
• behavioural ICAPM
• GARCH-M
• equity and bond markets
• international evidence

JEL Classification::

• G12

Notes

¹ The feedback trading framework has been used in several finance research contexts. For example, Laopodis (2005, 2008) investigates feedback trading and autocorrelation interactions in the foreign exchange market. McKenzie and Kim (2007) and Dean and Faff (2008) combine feedback trading with Markov switching regimes. Antoniou et al. (2005), Chau et al. (2008) and Kurov (2008) consider feedback trading in the context of futures markets.

² Notably, Rubio (1989), Shanken (1990), Song (1994), Elyasiani and Mansur (1998), Scruggs (1998) and others have, with varying degrees of success, empirically investigated such a version of the ICAPM.

³ Merton's (1973) ICAPM has been used to motivate many contemporary asset pricing studies – see, for example, Bali (2008).

⁴ Guo and Whitelaw (2006) employ a similar modelling setup.

⁵ In a US context, Christie (1982), Shanken (1990), and Glosten et al. (1993), who find that both the level and volatility of short-term interest rates are associated with significant shifts in the bond and equity markets.

⁶ GARCH type modelling remains very popular in contemporary finance research – of relevance to the current study, see for example, Chiang et al. (2007), Tsouma (2007), Chen et al. (2008), Li (2009) and Kumar and Dhankar (2010).

⁷ Our choice of the symmetric GARCH (1,1) is in accordance with Koutmos (1997). We experimented with asymmetric GARCH specifications, but encountered nontrivial convergence problems in our bivariate setting for some of our data series. As revealed later, the diagnostics for our reported model estimates indicate that the parsimonious symmetric framework is justified.

⁸ We acknowledge that, ideally, the term to maturity of the risk free rates should be the same across all countries but data availability prevented us from achieving this goal. However, as we span the quite narrow range of 1 month to 3 months, the assumption of a flat term structure is not unreasonable and so our research design choice is justified.

⁹ These findings for equity markets are remarkably similar to Koutmos (1997) – he reports negative coefficients of very similar magnitude to ours, for his earlier/shorter sample period (1986–1991). The only contradictory case is Germany – Koutmos (1997) finds German equities to also have a negative and significant estimated γ₂ coefficient.

¹⁰ The conditional variance estimates are suppressed to conserve space.

¹¹ Application of these tests to the bond markets is problematic and potentially misleading. One way to see this is to note that the results in Table 5 show that the bond market in-mean variance term is only significant in one case (Germany) – but even then the estimated coefficient takes a negative sign, which is difficult to rationalize from an asset pricing point of view. In contrast, the equity in-mean variance term is positive and significant in all six cases (see Table 5) – thereby, making sense of the H1₀ CAPM restriction.