In this paper, we consider the problem of determining the conditions u
nder which a change in risk increases the optimal value of a decision
variable for all risk-averse agents. For a large class of payoff funct
ions, we obtain the least constraining (necessary and sufficient) cond
ition on the change in risk for signing its effect without any additio
nal restriction on the utility function than risk aversion. It entails
all existing sufficient conditions as particular cases. Our results a
re applied to the linear model which describes the standard portfolio
problem. The necessary and sufficient condition for unambiguous compar
ative statics in this class of problems is termed ''greater central ri
skiness'' (CR). It is shown that CR dominance is neither stronger nor
weaker than second-degree stochastic dominance. (C) 1995 Academic Pres
s, Inc.