Do you believe them? Suppose we restrict both of the above propositions to cases where the probability of a loss is "small" and where the amount at risk is "not too large?" Both of these propositions follow easily from a recent communication by Szpiro (1985) in this Journal. If Szpiro's results are to be believed, we only need to know the price of insurance and the measure of absolute risk aversion at the individual's level of initial wealth to determine what level of coverage is optimal -or at least as an approximation without any "errors of great magnitudes." One of the culprits in obtaining the above results turns out to be the author's mimicking of Pratt's approximation of his risk premium formula using Taylor expansions. Szpiro argues that reasonable levels of the amount at risk, L, and the loss probability, q, do not yield serious errors in calculating his results. For his examples, Szpiro provides the magnitudes of the errors in estimating utility at post-loss wealth levels when we expand utility around the pre-loss wealth level. Even when these errors are small, what we should really care about is how close our predicted optimal insurance level is to the true optimal level. If L is small in relation to the pre-loss wealth level, W, then any predictor will have a small error as a percent of W. If W is $1 million and L is $1 thousand, then a predictor of insuring half the amount at risk has a maximal error in coverage (as a fraction of W-L) of about 0.0005. Although this may be a good predictor of final total wealth, it is doubtful that it is a good predictor of the level of insurance coverage.