One rational individual may be willing to pay less than another to insure a risk "epsilon" when another risk w is present even though he would pay more to insure any isolated risk, and even though E(" epsilon" w) = 0 for all w. Noticing this, Ross (1981) proposed excluding such reversals and gave equivalent analytical conditions. Reconsidering, we explain why some reversals are natural and show that prohibiting them has peculiar and undesirable properties. Although we also simplify the conditions and prove them necessary for partial-risk portfolio results, we conclude that they represent revealing restrictions on comparative statics rather than natural implications of increased aversion to risk. Copyright 1990 by Kluwer Academic Publishers
