RISK,RETURN,AND EQUILIBRIUM 617 above are repeated.That is,7 years of data are used to form portfolios; the next 5 years are used to compute initial values of the independent variables in (10);and then the risk-return regressions of (10)are fit month by month for the following 4-year period. The nine different portfolio formation periods (all except the first 7 years in length),initial 5-year estimation periods,and testing periods (all but the last 4 years in length)are shown in table 1.The choice of 4-year testing periods is a balance of computation costs against the desire to reform portfolios frequently.The choice of 7-year portfolio formation periods and 5-8-year periods for estimating the independent variables Be.-and 5.-1()in the risk-return regressions reflects a desire to bal- ance the statistical power obtained with a large sample from a stationary process against potential problems caused by any nonconstancy of the B. The choices here are in line with the results of Gonedes (1973).His results also led us to require that to be included in a portfolio a security available in the first month of a testing period must also have data for all 5 years of the preceding estimation period and for at least 4 years of the portfolio formation period.The total number of securities available in the first month of each testing period and the number of securities meeting the data requirement are shown in table 1. C.Some Observations on the Approach Table 2 shows the values of the 20 portfoliosand their standard errors s(B.)for four of the nine 5-year estimation periods.Also shown are:r(Rp,Rm)2,the coefficient of determination between Rpt and Rmt; s(R),the sample standard deviation of R:and s(p),the standard devia- tion of the portfolio residuals from the market model of (8),not to be confused with 5.(),the average for individual securities,which is also shown.The B2.and 3.1()are the independent variables in the risk return regressions of (10)for the first month of the 4-year testing periods following the four estimation periods shown. Under the assumptions that for a given security the disturbances a in (8)are serially independent,independent of Rmt,and identically distrib- uted through time,the standard error of B is ()= 0() Vno(Rm) where is the number of months used to compute BLikewise, (序-1）= o() Vna(R) Thus,the fact that in table 2,s(p)is generally on the order of one-third to one-seventh 5p.1()implies that s(B2.1)is one-third to one-seventh