Risky Business

Abstract

Start with orthodox Bayesianism, whose only (synchronic) constraint on rational credences is that they be coherent. That's remarkably permissive—it licenses unreasonable credences. Bayesians are fond of replying with convergence theorems, but for various reasons, I find them inadequate. So I approve of extra constraints on rational credences. Moreover, orthodox Bayesian decision theory is remarkably permissive about how agents value things. Again, it imposes coherence constraints, but it licenses unreasonable preferences. Now turn to risk attitudes. Which risk attitudes are permissible? Here, orthodox Bayesian decision theory is singularly narrow-minded. It demands risk neutrality: maximising expected utility, rather than some risk-seeking or risk-averse alternative to it. Offhand, this combination of extreme permissiveness about credences and preferences with extreme unpermissiveness about risk attitudes is somewhat surprising. So we need to look at the arguments for risk neutrality—in particular from representation theorems, and from long-run convergence theorems. I find these arguments wanting. Still, there might be extra constraints on rational risk attitudes: analogues of the the extra constraints on rational credences. But I will argue that to the extent these are plausible, they do not privilege risk-neutrality. Indeed, considerations of morality and moral risk undercut such privileging. So we have a reason to be permissive about risk attitudes—even more so than about credences. Hence, we should welcome Lara Buchak's risk-weighted expected utility theory. But in some respects, her theory is not permissive enough. Moreover, even she thinks that some risk attitudes are unreasonable—in other respects her theory is too permissive. Nevertheless, orthodoxy's rigidity about them should be questioned, and we should be grateful to her for doing so. Show more