Yesterday's class Models for Strategic Planning was exceptional; it was the class that Lucky had wanted to hear at some point during the course. Professor Delquié discussed risk, uncertainty, and the value of information in both abstract and concrete ways, inspiring many thoughts about how to look at situations a bit differently.
One example: how often do you miss planes at the airport? If the answer is never, you are probably wasting too much time at airports. This is analagous to a company's credit policy which produces a perfect payment record from its customers. While nonpayments may be minimized, the company's profit certainly is not, and there is money left on the table in the form of lost sales. The lesson is that risk assessment should be done on an explicit and conscious basis. If you do it "on autopilot", you will optimize to an implicit and often incorrect goal.
We reviewed (superficially) the difference between classical statistics and Bayesian statistics in their approaches to solving problems where there are unknown elements. For our purposes, the classical method treats unknown probabilities as unknown, while the Bayesian method allows you to start with a "prior probability" and then modify it in light of additional data.
If a business plan comes across your desk, how do you assess its chance of success? Most people would read the plan, taking into account various factors such as the plan's authors, their prior experience, industry knowledge, etc., and then make a determination about the probability of success. However, in some ways this is a myopic estimation. In reality, the business plan that you are reading can be considered to be a sample from the universe of all business plans. As such, its prior probability is known - it is simply the average success rate of all business plans (which is measurable, and something like 10%). Any additional information that you know can be used to modify this initial probability of success; this is the beauty of Bayesian inference.
Finally, we covered the "expected value of perfect information" (EVPI). Much information has value, but we often don't know how much. Calculation of EVPI shows how much you would benefit by having perfect information, and thus represents a maximum value of how much one should spent on acquiring it. In some ways, this is similar to a call option.
A final digression was made on the subject of opthamology and eye surgery. The professor had a friend who needed to have surgery done on his eyes. He went to one eye surgeon who said that he couldn't predict how accurate the post-surgery eyesight would be. The friend went to another doctor, who told him that his post-surgery eyesight would be 15/20 (ie. would still need some minor correction). The friend asked the doctor how sure he was about the prediction, and the doctor retrieved a scatterplot chart with two axes labelled "Pre-op Predicted Eyesight" vs. "Post-op Actual Eyesight". Impressive! And yet still troubling. Certainly, the data is biased, but the problem is that you don't know how it is biased. You don't know whether all the data is there, or if it is all accurate. Also, you don't have a record of the people who chose not to elect the surgery with that doctor. On balance, is it better to have this information or not? Difficult to say.
3:28:43 PM 
|
|
