Read on Twitter
New paper out today with @MikePricePhD on how biological entities, who have been shaped by natural selection, should make decisions when the payoffs to those decisions are variable and have enduring consequences. ( https://doi.org/10.1017/ehs.2020.28) 1/
This paper has a strong bearing on recent debates in behavioral economics about the nature of human rationality. 2/
What is economic rationality? The standard approach in economics is that a rational actor maximizes a quantity known as expected utility. This is the axiomatic model known as Expected Utility Theory (EUT) formalized by von Neumann, Morgenstern, Savage, and others. 3/
OK, first, what is utility? Utility is a function that assigns a value to a particular outcome. Utility increases with the size of the reward, typically with diminishing marginal returns. 4/
Now, what is expected utility? When faced with an (economic) gamble, you should value the gamble according to the average (i.e., expectation) of the utilities of its associated outcomes. 5/
Think for a minute about what a mathematical expectation is. It's just the summation of a linear combination of weights (probabilities) and outcomes (utilities). There's nothing magical about it. 6/
The economist Mark Machina has noted there are many possible value functions that combine weights and outcomes in various (nonlinear) ways. The question is whether there is a theory that justifies any particular form. 7/
Lots of research has shown that people often violate the axioms of expected utility theory. This empirical base provided the foundation for Tversky and Kahneman’s Prospect Theory, one of the most important theories ever in the behavioral sciences. ( https://doi.org/10.1007/BF00122574) 8/
Does the fact that people regularly violate the axioms of EUT mean that people are, at their core, irrational or should we rethink what we mean by "rationality"? 9/
We note that the human brain was shaped by natural selection. How would a decision-making organ respond to the persistent selective forces of diverse human environments and how does this help us understand the many observed deviations from the canonical predictions of EUT? 10/
Whenever you talk about natural selection, you need to talk about fitness. 11/
What is fitness? This turns out to be a deep philosophical question, but from a practical standpoint we can think of it as proportional representation in future generations. 12/
Another useful way to think about fitness is that it’s a (marginal) rate of increase on an instantaneous or generational time scale. 13/
This highlights a fundamental point about fitness, which is that it is multiplicative, whereas the economic theory of utility is additive. 14/
Think about it: in order to reproduce at age 20, you need to survive independently each of the preceding 19 years. All those survival probabilities multiply. One zero in there and you're done before you have the opportunity to actually reproduce. 15/
The same is true for lineages. If a lineage hits a zero in a given generation, it's over. No amount of averaging will fix this. 16/
This means that an expectation sum is not an appropriate measure of fitness. 17/
This point was first noted in the 1960s by population biologists like Garth Murphy, Dan Cohen & Dick Lewontin. ( http://www.jstor.org/stable/59357 ) 18/
It was formalized in John Gillespie's geometric-mean principle for fitness, which emphasized the trade-off between the mean and variance in multiplicative fitness. ( http://www.genetics.org/content/76/3/601.abstract) 19/
Our focus on the multiplicative nature of fitness aligns us with the ergodic economics approach of @ole_b_peters and colleagues. The unique contribution of our model is the idea of hierarchical evolutionary preferences. This is fundamental. 20/
Natural selection doesn’t maximize utility. However, people’s decisions are guided by some sort of “utility,” for example, proximate psychological mechanisms. 21/
This is an idea in evolutionary biology dating back to Ernst Mayr, proximate vs. ultimate causation, and more fully articulated in Nikko Tinbergen’s masterful four problems of biology. ( https://doi.org/10.1111/j.1439-0310.1963.tb01161.x) 22/
Because the objective function being maximized is fitness, not utility, preferences shaped by selection will appear distorted from the perspective of the utility. 23/
As it happens, this is exactly the observation at the heart of so much behavioral economics: measured preferences frequently are distorted from the predictions of EUT. 24/
We also note the deep connections of our outcomes with the theory of Rank-Dependent Expected Utility Theory, first formulated by @JohnQuiggin. This is an axiomatic alternative to EUT that we find remarkably compatible with an evolutionarily-informed decision theory. 25/
Diecidue and Wakker: “The intuition of rank-dependence entails that the attention given to an outcome depends not only on the probability of the outcome but also on the favorability of the outcome in comparison to the other possible outcomes.” ( https://doi.org/10.1023/a:1011877808366) 26/
The hierarchical structure of decisions opens the possibility for utility and fitness to fall out of alignment. Following the insight of Ken Binmore, we treated this decision structure like a principal-agent problem. 27/
Natural selection (the principal) can’t directly shape decisions but must do so through utilities (the agent). A mechanism must exist to ensure that the potentially-divergent interests of principal and agent are aligned. 28/
We used the great @alanogers observations about the marginal substitution of economic goods equalling the marginal change in fitness to close the system and align fitness and utility. ( http://www.jstor.org/stable/2118062 ) 29/
Finally, we link economic decisions to fitness using (my colleague and mentor) Shripad Tulajpurkar’s formalism for age-structured demography in variable environments. The consequences of economic decisions are entailed in the elements of a stochastic matrix model. 30/
Our results show that fitness-maximizers will use nonlinear probability weighting in making decisions when the payoffs to those decisions are variable and have enduring consequences. 31/
We find that under a wide range of conditions, selection favors pessimistic probability weighting rather than the classic inverse-S shape of CPT. We can create evolutionary scenarios that produce the inverse-S shape, but they are in the minority. 32/
Why is an interesting (and technical!) question that we’ll leave for later. I will point to the great work of Nat Wilcox, who suggested that our pessimistic result ultimately makes more sense, given we’re not eliciting certainty equivalents. ( https://doi.org/10.1007/s40881-017-0042-1) 33/
Pessimism of probability weighting seems consistent with the bet-hedging/life-cycle diversification ideas that came out of the models of Gillespie, Orzack & Tulja ( https://doi.org/10.1086/284959 ), and others. 34/
It’s also related to the work of @tavitonst, which provided an early inspiration for us, showing that herder households’ decisions were most consistent with long-term household survival rather than maximization of herd offtake. ( https://doi.org/10.1007/BF00172931) 35/
Wow, that’s a lot. Probably a good time to stop. If you’re interested, read the paper. It’s open access! https://doi.org/10.1017/ehs.2020.28 end/
