736 CHAPTER 19. SETTING THE COST OF INTERNATIONAL CAPITAL
19.4.3 Estimating the Risk premia
Assuming that we have an approximate idea of the beta and gammas, we need
estimates of the expected risk premia per unit of risk. The expected excess return
on the world market portfolio is still rather hard to estimate, even though it is not
quite as bad as a typical currency-risk premium. The sample averages of returns
observed in the past differ substantially across sample periods, and it is also known
that the expected return changes over time.^15 Still, we know that there is a positive
risk premium on the world stock market, and variations over time in the expected
excess return may not be overly important when theNPVevaluation horizon is, say,
one decade rather than a month or two days.
Turning to the expected excess return on the various foreign T-bills, these risk
premia also change over time, as we have seen in Chapter 10—and, unlike for the
world market risk premium, we are not even sure whether the long-run mean actually
differs from zero. Since exchange risk premia are small in the short run and close
to zero in the long run, for practical applications one might have to be content with
an approach that ignores these and use the following simplified version of Equation
[19.36]:
E( ̃rj−r)≈βj,w;sE( ̃rw−r), (19.39)
where the beta is still estimated from a multivariate regression (Equation [19.37])
rather than from a bivariate regression).
You should not be overly discouraged by these approximations. No model is
perfect; and the InternationalCAPMdoes work better than competing models. Still,
the cost of capital is measured imperfectly, andNPVcomputations should always be
undertaken for a whole range of reasonable discount rates, to see to what extent the
accept/reject recommendation is sensitive to the estimate of the cost of capital.
(^15) The return is partially predictable on the basis of (1) the risk spread (the difference between
low-grade bond yields and government bond yields), (2) the term spread (the difference between
short-term and long-term bond yields), and (3) the dividend yield.