Because expected future returns are unobservable. all survey respondents extrapolated hi torical returns into the future on the presumption that past experience eavily cond ms tu ure expectations. Where chietly was in their use of ariulumelic versus respondents ed trio average historical equity returns and in their choice of realized returns on T-bills Re onl versus T-bonds to proxy for the return on riskless assets The arithmetic mean return is the simple average of past returns. Assuming the distri bution o returns is stable over time and that periodic returns are independent of one an other, the arithmetic return is the best estimator of expected return. The geometric mean return is the internal rate of return between a single outlay and one or more future receipts It measures the compound rate of return investors earned over past periods. It accurately portrays historical investment experience. Unless returns are the same each time period. the ueometric average will always be less than the arithmetic average and the gap widens as re turns become more volatile Based on Ibbotson Associates' (1995) data from 1926 to 1995, the matrix below illus trates the possible range of equity market risk premiums depending on use of the geomet ric as opposed to the a mean equity return and on use of realized returns on T-bills as opposed to T-bonds.' Even wider variations in market risk premiums can arise when one changes the historical period for averaging. Extending U.S. stock experience back to 1802 Siegel (1992) shows that historical market premiums have changed over time and were typ- ically lower in the pre-1926 period. Carleton and Lakonishok (1985) illustrate considerable variation in historical premiums using difierent time periods and methods of calculation even with data since 1920
ol the texts and trade books in our survey 71 percent support use of the arithmetic mean return T-bills as the best surrogate for the equity market risk premium. For long over term projects. Ehrhardt advocates forecasting the T-bill rate and using a different c of equity for each future time period. Kaplan and Ruback (1995) studied the equity risk pre lied by the valuations in highly leveraged transactions and estimated a mean pre niuni Several studies have documented significant negative autocorrelation in returns-- this violates on of the essen tial tenets of the arithmetic calculation, since if returns are not serially independent, the simple arithmetic mean of a distribution will not be its expected value. The autocorrelation findings are reported by Fama and French (1986). Lo and MacKinlay (1988. and Poterba and Summers minus one For large samples of returns the geometric average can be approximated as the arithmetic average current e half the variance of realized return Ignoring sample size adjustments. the variance of returus in the (1994) ample is 09 yielding an estimate of 10 I12009) 055 5.5% versus the actual 5.8% figure. Kritzman provides an interesting comparison of the two types of averages. from the "Large Company i These figures are drawn trom Table 2-1. Ibbotson (1995 where the R was drawn Bills" series Stocks" series. and R drawn from the "Long-Term Government Bonds and "US. Treasury
miul of 7.97 percent, which is with the arithmetic mean and T-bills. A most c illinority view is that of Copeland. Koller and Mlurrin l gyo. pp. 193 9-1) writing on be halt of the Corporate Financial Practice at McKinsey Company: e believe that the geometric average represents a better estimate of vestors' expected returns over long pe riods of Ehrhardt (1994) recommends use of the geometric mean return if one be lieves stockholders are "buy-and-hold" investors. Hall of the linancial advisers queried use a premium consistent with the arithmetic mean and I bill returns. and many specifically mentioned use of the arithmetic mean. Co porate respondents. on the other hand. evidenced more diversity of opinion and tend to fa vor a lower market premium: 37 percent use a premium of to 6 percent and another 1 I percent use an even lower figure
Comments Regarding Market Risk Premium What do you use as your market risk premium A sampling of responses from our best ompanies shows the choice can be a complicated one Prac Ce Our 4000 basis point market premium is based on the historical relationship of returns on an actualized basis and/or investment bankers estimated cost of equity based on analysts earu projections ngs We use an ibbotson arithmetic average starung in 1960. We have talked to investment banks and consulting tirns with advice from 3 to 7 percent. A 60-year average of about 5.7 percent. This number has been used for a long time in the company and is currently the subject of some debate and is under review. We may consider us ing a time horizon of less than 60 years to estimate this premium "We are currently using 6 percent. In 1993 we polled various investment banks and academic studies on the issue as to the appropriate rate and