156 R.Mehra and E.C.Prescott,The equity premium Index over the ninety years considered was 6.98 percent per annum.This leads to an average equity premium of 6.18 percent(standard error 1.76 percent). Given the estimated process on consumption,fig.4 depicts the set of values of the average risk-free rate and equity risk premium which are both consistent with the model and result in average real risk-free rates between zero and four percent.These are values that can be obtained by varying preference parame- ters a between zero and ten and B between zero and one.The observed real return of 0.80 percent and equity premium of 6 percent is clearly inconsistent with the predictions of the model.The largest premium obtainable with the model is 0.35 percent,which is not close to the observed value. 4.1.Robustness of results One set of possible problems are associated with errors in measuring the inflation rate.Such errors do not affect the computed risk premium as they bias both the real risk-free rate and the equity rate by the same amount.A potentially more serious problem is that these errors bias our estimates of the growth rate of consumption and the risk-free real rate.Therefore,only if the tests are insensitive to biases in measuring the inflation rate should the tests be taken seriously.A second measurement problem arises because of tax consider- ations.The theory is implicitly considering effective after-tax returns which vary over income classes.In the earlier part of the period,tax rates were low. In the latter period,the low real rate and sizable equity risk premium hold for after-tax returns for all income classes [see Fisher and Lorie (1978)]. We also examined whether aggregation affects the results for the case that the growth rates were independent between periods,which they approximately were,given that the estimated was near one-half.Varying the underlying time period from one one-hundredths of a year to two years had a negligible effect upon the admissible region.(See the appendix for an exact specification of these experiments.)Consequently,the test appears robust to the use of annual data in estimating the process on consumption. In an attempt to reconcile the large discrepancy between theory and ob- servation,we tested the sensitivity of our results to model misspecification.We found that the conclusions are not at all sensitive to changes in the parameter u,which is the average growth rate of consumption,with decreases to 1.4 percent or increases to 2.2 percent not reducing the discrepancy.The sensitivity to 8,the standard deviation of the consumption growth rate,is larger.The average equity premium was roughly proportional to 8 squared.As the persistence parameter increased (=0.5 corresponds to independence over time),the premium decreased.Reducing(introducing stronger negative serial correlation in the consumption growth rate)had only small effects.We also modified the process on consumption by introducing additional states that permitted us to increase higher moments of the stationary distribution of the