DECS-433: Decision Making under Uncertainty
This version: December 10, 2009
We all use models to arrive at decisions. These models range in sophistication from “everyone but me is a fool” to the full rationality assumed in financial markets. Unfortunately, the models we tend to use are not very good. They can, under some conditions, easily lead us astray. This would not be so terrible if these “conditions” were rare. However, they are not. This course is about how to make good models as well as how to distinguish a good model from a bad one. In the study of models and how they relate to decision making we try to find the middle ground between the perspective of the parachutist and that of the truffle hunter. We focus on evaluating uncertainty, understanding the dynamic nature of decision-making, using historical data and limited information effectively and simulating complex systems. Since the topics are quantitative in nature it may be useful to summarize and dispose of the usual objections against them.
• Objection #1: Not everything can be reduced to numbers.
True. But a great deal of importance can.
• Objection #2: Workable quantitative models cannot capture the complexities of real life.
So what? The question is not whether a particular quantitative model accurately represents reality but whether there is an alternative model that is more accurate. How do we know that a decision arrived at by what we call instinct (or intuition, experience, etc.) has really acknowledged all the complexities of reality? Indeed, one of the beauties of quantitative models is their explicitness. Like Cromwell’s portrait they appear warts and all. • Objection #3: The data requirements of quantitative models are prohibitive. This objection had some merit 20 years ago. Given the state of modern computing, it is an excuse for laziness. One of the useful features of a quantitative model is that it tells the decision maker what information s/he should be collecting.
• Objection #4: Quantitative models ignore the context.
Yes. However, there is no law that obliges one to follow the recommendations of a quantitative model that is not appropriate to the context. Second, ignoring the context can be a good thing. Frequently, the context of a decision is irrelevant. A quantitative model often captures the essence of a decision, say, the choice between a gamble and a sure thing. It does not matter whether the context is the stock market, a medical diagnosis or wildcatting. This allows for the transfer of insight from one setting to another. An example of this is the use by hotels of the same models and principles that airlines use for yield management. A more revealing example comes from experiments on auction behavior: The subjects were managers responsible for submitting bids on construction projects. The goal was to see whether these managers made bids in accordance with the optimal bids of the standard (quantitative) auction model. As long as the auctions were couched in the context of the construction industry, the answer was yes. However, when subjects were presented with the same auction scenarios, but the context changed (for example, to oil leases) they did remarkable badly. Why? The bids they made were derived through the use of rules of thumb that were specific to their industry. When the context switched, their rules of thumb became useless. One might argue that this is not important as few people are going to switch industries mid-career. Not so. Suppose that the rules of thumb are tied to characteristics of the industry at a particular time. As time changes, those characteristics change, until eventually those rules become useless. (Note: this is also the reason for having a course not tied to a particular functional area of business).
Perhaps the most surprising thing that you will learn from this course is that simple...
Please join StudyMode to read the full document