As indicated in my last post, between December 2009 and October 2010, evolutionary biologist David Sloan Wilson wrote a series of posts titled Economics and Evolution as Different Paradigms. Wilson’s basic line of reasoning is that evolutionary biology should play a larger role in economics, and I naturally agree with that position.
For the foundation of his argument, the second post of Wilson’s series contains a reasonably typical attack on modern economics (I’ll accept this monolithic caricature for the moment). I wouldn’t call it the most convincing of attacks, particularly when it starts by providing an example that modern economics deals with quite well:
Consider the following proposition: I’ll give you 1 million dollars for sure or a 50:50 chance at 2.1 million dollars. What’s your choice? If you’re like me, you’ll choose the certain 1 million. Yet, that is a violation of core economic theory that became known as the “Allais Paradox“.
Unfortunately, that is not an example of the Allais Paradox, but an example of risk aversion (and a number of the comments on Wilson’s post pointed this out). The Allais Paradox is, however, nicely set out in the Wikipedia article to which Wilson links. This mistake tends to make much of Wilson’s critique that economics cannot deal with variance of utility fall flat, as for the example given, a simple utility model with risk aversion does this well.
Moving past that hiccup, Wilson suggests that we can find the reason that the Allais Paradox has not been incorporated into modern economic theory in Milton Friedman’s essay The Methodology of Positive Economics (the first essay in his book Essays in Positive Economics). Friedman suggested, among other things, that it is not problematic if the assumptions underpinning the model are unrealistic if the predictive power of the model is good. We should test models by their predictive power. Wilson states that this approach is a recipe for confirmation bias (although, what isn’t?) and that the lax criteria for assessing models allows economists to ignore the Allais Paradox.
Whenever someone uses this essay by Friedman to attack economics, I feel that the reader generally misunderstands what Friedman means when he suggested that the assumptions do not have to be descriptively realistic. Friedman notes that assumptions, by their very nature, never are descriptively realistic, but that they need to be sufficiently good for the purpose at hand. This is a rather mild claim, as any model (and its assumptions) must be more simple than reality or there will be no value to the model.
Regardless, Wilson suggested that the Allais Paradox presented a risk to economic theory. He states:
It violated the principle of maximization of returns, it could not easily be incorporated into the body of formal analytical theory, and it was a move toward realism that would definitely have consequences for economic predictions.
I only partially agree with that assessment. If a paradox like the Allais Paradox is particularly important for a set of decisions that an economic model is seeking to predict, then Friedman’s test of predictability is useful in determining whether it should be incorporated in the model. If it is not important, then it won’t improve the model’s predictive power.
That reflects the general issue I have with the Allais Paradox, and the many other of the biases that behavioural economics has unearthed. Are there tangible examples of how they could be incorporated into an economic model and out-predict a neoclassical model that ignores these biases (they may exist – I am asking this question more of ignorance than confidence that they don’t)? Part of the reason for this is that much of behavioural economics is a catalogue of biases and it lacks an evolutionary framework (and as I noted in my last post, the Evolution Institute seeks to address this).
This is not to say that we should ignore the raft of biases. But instead of pointing out each individual bias and demanding to know why it isn’t incorporated in the model, we should ask whether there is a more systematic problem that the bias is symptomatic of. Instead of trying to add bias by bias, we could be asking more fundamental questions, such as what the agent(s) actual objective is.
I have more sympathy with Wilson’s argument that testing the predictive power of a model can be difficult where the evidence is in a complex system and difficult to collect and document. In this case, Friedman’s test of predictive power may not be a particularly strong filter. Despite this, it is an underused filter for many economic theories. As I have discussed before, economics has no shortage of theories that are still kicking around despite a lack of any empirical support. Use of that filter might get rid some of highly complex, mathematically beautiful but predictively useless models that should have been discarded long ago.