["If one is willing to interpret Q̃ [the Goldberger, Nagar, Odeh reduced form coefficient covariance estimate] as a covariance matrix of the random parameter π around the constant extlesstex-math extgreater$ extbackslashtilde\ extbackslashpi$ extless/tex-math extgreater, rather than as a covariance matrix of the random estimates extlesstex-math extgreater$ extbackslashtilde\ extbackslashpi$ extless/tex-math extgreater, then using extlesstex-math extgreater$ extbackslashtilde\ extbackslashpi$ extless/tex-math extgreater for extlesstex-math extgreater$ extbackslashtilde\ extbackslashpi$ extless/tex-math extgreater and Q̃ for Q extlesstex-math extgreater$[ extbackslashtilde\ extbackslashpi$ extless/tex-math extgreater and Q are the mean and covariance matrix of the random parameter π] will provide an approximate solution to the evaluation of expectations required in our optimal control problem" [3, p. 641], italics added).]