Learning While Doing - Optimizing in the field, illustrated using a mutual fund cash balance problem скачать в хорошем качестве

Learning While Doing - Optimizing in the field, illustrated using a mutual fund cash balance problem

There are many operational problems that can be improved without the expense of building a sophisticated model. In this talk, I use the context of optimizing the amount of cash a mutual fund keeps on hand to meet redemption requests, but the approach can be generalized from the simple rules used in this application, up to tuning the parameters of a large optimization model. It requires recognizing that the tunable parameters represent a decision required to optimize a policy for making whatever decision is required. Tuning the parameter(s) requires its own policy. I review the four classes of policies, and pick two that are relevant to this setting. One, a form of cost function approximation (CFA) is quite simple, but has its own tunable parameter, while the other, a form of direct lookahead approximation (DLA) that estimates the value of information from trying one setting (called the knowledge gradient).