Therefore, ∏* attains a unique minimum at S = Pd/Pf. (See Figure 4.) Since U''(∙) < 0,
U'(Щ) reaches a unique maximum at S = Pd/Pf. Thus, there exists a unique S > Pd/Pf
defined by U,(Π*(S)) = E[U,(Π*(S))]. Similarly, there exists a unique S# < Pd∕Pf
defined by U'(Π^(S#)) = E[U'(∏χ(S,))].
First, consider case (a), defined by E[U'(∏:)| S ≥ Pd∕Pf] not exceeding E[U'(Π:)|S ≤
Pd/Pf], where E[∙∣∙] is the conditional expectation operator. By the definition of the
covariance operator, one can write
Cov [U'(∏:), max(S - Pd/Pf, 0)]
= /Pd/P’ {u'(∏'(S)) - e[u'(∏:)]}{ - e[max(S - Pd/Pf,0)]} dG(S)
+ ∣Sm U '(πxs )) - e[u '(∏ : )]}
×{(S - Pd/Pf ) - e[ max(S - Pd/Pf, 0)]} dG(S)
= /Sp, {U'(∏(S)) - E[U'(∏:)]}(S - Pd/Pf) dG(S)
= /Sp, {U'(∏*(S)) - E[U'(∏Z)]}{(S - Pd/Pf) - (S - Pd/Pf)} dG(S)
+1Sp, {U'(∏'(S)) - E[U'(∏Z)] }(S - Pd/Pf) dG(S)
< (S - Pd/Pf) jS/p {U'(∏Z(S)) - E[U'(∏Z)]} dG(S)
= (S - Pd/Pf)1 /S U'(∏Z(S)) dG(S) G(Pd/Pf)
I jPd∕pf `-
- IsFd'F' U'(∏:(S)) dG(S) [1 - G(Pd/Pf )]1
= (S - Pd/Pf ) G(Pd/Pf ) [1 - G(Pd/Pf )]
× 1e[u '(∏ : )∣ S ≥ Pd/Pf ] - e[u '(∏ :)| S ≤ Pd/Pf ] 1,
where the inequality follows from the fact that {U'(∏X(S)) -E[U'(∏:)]} and (S-Pd/Pf ) -
(S - Pd/Pf ) = (S - S) have opposite signs for all S > Pd/Pf. The curly bracketed
term in the last line is non-positive by assumption. Since (S - Pd/Pf ), G(Pd/Pf ) and
[1 - G(Pd/Pf )] are all positive, Cov[U'(Щ), max(S - Pd/Pf, 0)] is negative for case (a).
__ _ __ _______. , . ~. _ , . _ _______. , . ~.
Now, consider case (b), in which E[U'(∏£)| S ≥ Pd/Pf] is greater than E[U'(∏:) ∣S ≤
21