tag:blogger.com,1999:blog-4799264811265759956.post2391387642598487783..comments2022-05-19T04:53:44.810-04:00Comments on Quantitative Finance: Foundations and Applications: Keynes, Bayes, and the lawMatheushttp://www.blogger.com/profile/05386153701958504638noreply@blogger.comBlogger2125tag:blogger.com,1999:blog-4799264811265759956.post-65019244945524093272013-09-09T12:39:49.140-04:002013-09-09T12:39:49.140-04:00I never said that the quality of the priors doesn&...I never said that the quality of the priors doesn't matter. Poor priors lead to poor bets and ultimately poor outcomes (i.e you loose your bets more often than you win). All I'm saying is that Bayesianism is entirely agnostic about where you get your priors from, while offering a well defined prescription to update them. It's all there in one elegant theorem. <br /><br />Also notice that not all evidence is born equally. Strong evidence (as measured by the other conditional probabilities that enter Bayes theorem) correct priors very quickly, whereas weak evidence doesn't. <br /><br />So the quality of both the priors and the evidence presented is hugely important. But there are no other hidden assumptions in the argument (e.g symmetry) in either the way to choose priors or to incorporate evidence. Everything is laid out explicitly. <br /><br />Same goes for ergodicity. If you have masses of data about a roulette, you can use the data to initially assign your priors as if it was fair. But as you continue playing and realize it is wearing (how quickly you realize depends how obvious the wearing is - e.g it spins once and comes to a screeching halt), you use Bayes theorem to update them. Matheushttps://www.blogger.com/profile/05386153701958504638noreply@blogger.comtag:blogger.com,1999:blog-4799264811265759956.post-89443514347687796722013-09-09T12:23:47.297-04:002013-09-09T12:23:47.297-04:00You make some bold claims. I couldn't help thi...You make some bold claims. I couldn't help thinking of the case against the Assad regime. It may be that a Bayesian does not care where Kerry's priors come from, but many do. I also do not get your ergodicity argument. Suppose that you have masses of data about a probabilistic process, and then no data for many years. What difference does it make whether you know that the process is probabilistically fixed, or where it is liable to have changed? For example, a roulette wheel that was fair but is now wearing, but you don't know how. Could you do a toy Bayes calculation in the two case?Anonymousnoreply@blogger.com