TESTS OF H(0) - RHO-1=O FOR AUTOCORRELATION ESTIMATORS R(F1) AND R(F2)

Two new tests of the hypothesis of no lag-1 autocorrelation in a time- series process, i.e., H-0: rho1 = 0, are presented for two reduced-bia s autocorrelation estimators that were introduced by Huitema and McKea n in 1994. The performance of the new tests, z(F1) and t(F2), Was eval uated using Monte Carlo methods. Both new tests were superior to the c onventional Bartlett asymptotic test in the case of small and intermed iate N; t(F2) outperforms z(F1) substantially at small and intermediat e values of N. The true probability of Type I error associated with te st t(F2) is exceedingly close to the nominal value for all values of a lpha (.01, .05, and .10) and N (6-500) investigated. The t(F2) test is also more powerful against positive values of rho1, than are the Bart lett and z(F1) tests.