So let's start with a basic decision problem under uncertainty. Suppose you're choosing between two gambles, gamble x, gamble y. Gamble x allows you to win $1,000 or to get nothing with 50/50 odds. Gamble y gives you a 70% chance of winning $600 or 30% chance of winning $200. Which of these choices is preferred? Now, a naive approach would be to compute expected payoffs and then choose the higher one. But what about randomness? What about risk? We know that a 50/50 gamble of winning $1,000 or getting nothing is not the same as getting $500 for sure. Most people are averse to this kind of risk. How should we quantify the trade off between expected payoffs, uncertainty, variance, or other properties of the payoff distributions? That is the question that we are going to answer.
There are two general approaches to building models for decision
making and uncertainty. Let's call them rational and behavioral. The rational approach, which is what we are going to pursue in this class, is prescriptive. It's a model of choice, which gives us consistent decisions and uncertainty. This internal consistency is desirable. It's basically a way of avoiding costly mistakes. Behavioral approach has a different objective. Rather than figuring out what is the right consistent way of making decisions, the behavioral approach aims to describe how individuals make decisions in all of their imperfections. For example, if you're an investor, you're trying to avoid mistakes in your portfolio choice. You would want to use a rational approach to make sure that your own decision making is consistent. If you're a financial advisor thinking about how to best advise your clients, you may want to understand what behavioral biases are most relevant and help the client correct them. If you're a trader, thinking about what other traders are doing in the market, whoever's taking the other side of your trade, understanding behavioral biases could be at the core of your profitability. So we're going to proceed developing a model of consistent decision making, and we'll call it a rational model. Our model is going to make a few basic assumptions. We are going to think about preferences over outcomes only. Outcomes in our case are cash flows. This is important. It's a restriction which could be relaxed. But under the basic model, we are going to make this assumption. What this means is that we are going to think about the value of the pay off-- $100, $1,000 payoff. We're not going to think about the circumstances under which the payoffs are realized. That is a more general problem. We will also model preferences in a way that is independent of how cash flows are generated. Randomness may come from many sources. It could be algorithmically generated. It could be a natural outcome of economic forces. At the end, our preferences are going to be developed to deal with the distribution of cash flows and not with the mechanism for generating randomness. These are the two basic assumptions that we are going to start with. We're going to maintain them. The expected utility theory, which we are going to present next, is at the core of the rational approach to decision making. It's the leading model for decision making and uncertainty. Under the expected utility theory, investors do not use expected payoff of a gamble to compare gambles to each other. They don't think about just the mean of the return. They don't think about the expected pay off of an investment. They think about the entire distribution. Expected utility theory applies a transformation to the payoff. It's a nonlinear transformation. For example, $1,000 payoff is not going to be perceived as twice as valuable as $500. An investor, depending on his or her preferences, is going to adjust the internal value of each payoff, the utility of each payoff, to reflect their preferences. Once the payoffs are transformed nonlinearly, as you can see in this graph, the probabilities of payoffs are then applied as weights. So under the expected utility theory, an investor evaluates a particular risk investment based on the expected utility, where the utility evaluated as a function of outcomes is averaged using probabilities of different events as weights. For example, if you are faced with a binary gamble and the curve presented on the graph, the blue line is the utility function. The gamble has only two possible payoffs, either 1.2 or 2.8, both equally likely. The utility of the payoff is going to be given by the dot in the middle of the segment, connecting the two dots on the utility curve. This is a graphical representation of the expected utility of this particular gamble. An investor transforms payoffs using the utility function and computes expected value of the utility. That's the starting point. That's the description of uncertainty and how an investor feels about different risky outcomes. Now, the consistent approach to choice is to compare risky investments to each other based on their expected utility. When thinking about investment a versus investment b, whichever yields higher expected utility is preferred. That is the expected utility theory. This approach is mathematicaly parsimonious. It's fairly intuitive, and it helps avoid costly mistakes. It respects laws of probability. It leads to investment decisions with some basic desirable properties, such as sensitivity. For example, under the expected utility theory, if an investor prefers x to y and y to z, the same investor will correctly prefer x to z.