Heuristics, Judgement, and Public Policy of Bounded Rationality

Heuristics, Judgement, and Public Policy of Bounded Rationality Presented by: Sanjit Dhami & Cass R. Sunstein Date: 25 October 2022 Bayesian Rationality Approach - The dominant framework in economics, neoclassical economics, uses the Bayesian rationality approach (BRA). - The dominant framework in economics, neoclassical economics, uses the Bayesian rationality approach (BRA). - The BRA has been widely adopted in political science (rational actor model); law; management; finance; neuroscience. - BRA in its actual practice has a set of core assumptions: Unlimited computational and cognitive abilities; considered choices that respect axioms of rationality, such as completeness and transitivity; perfect memory; perfect attention; know all math/stats. - BRA also typically ignores emotions and morality in humans (but not part of the rationality requirements). - BRA makes precise and testable predictions. Problem: Several core features of BRA are empirically rejected. BRA: An Insular Economics - From around the second half of the 20th century the BRA gradually purged insights from psychology, sociology, anthropology, and sociobiology. - In addition, the rationality requirements on individual behavior under BRA have become more stringent. - The BRA is motivated by models in theoretical physics that deal with inanimate particles. - Yet, the BRA does not adopt the stringent standards of testing in physics and the other natural sciences. It claims a distinct status for economics, protecting it from refutation. - Mathematical optimization is a central feature of BRA: Humans maximize/minimize well defined objective functions in a mathematical sense. For instance, using engineering methods to find the top of a hill or the depth of an ocean. So What Does the Evidence Show? - Limited attention & willpower; often make choices they regret; limited financial literacy and computational ability; imperfect and motivated memory; motivated recall, overconfidence. - Also emotional, moral, and other-regarding (not a RR). - Mathematical optimization imposes unreasonable cognitive and computational requirements. Consider a simple dynamic programming problem (invented in 1950). - Thirty periods. 50-50 chance income in each period is high or low. Consider a consumption/savings choice for each income level. This requires 230 or about a trillion calculations. - Most problems in economics are more complex than this. But the man on the street is assumed in the BRA to be able to solve these problems. - Mounting empirical refutations of the BRA. - So we need to develop concepts in Bounded Rationality. But what explains the adherence to the BRA for so long? Why the adherence to the BRA? - Homegrown methodological positions within economics. - Friedman's "as if' approach dominates: People may not literally obey empirically rejected assumptions, but the predictions of such models might still be accurate. So people act "as if" they follow the BRA. - However, when tested, the predictions of BRA often fail! Kahneman-Tversky showed this in several papers in the 1970's. The evidence is now too strong, too extensive. - Gilboa et al. (2014): "...economic models are... expected to convey a message much more than to describe a well-defined reality; and the economic theorist is typically not required to clearly specify where his model might be applicable and how." - Rubinstein (2006): "As in the case of fables, models in economic theory are not meant to be testable...a good model can have an enormous influence... not by predicting the future, but rather by influencing culture. Yes, I do think we are simply the tellers of fables, but is that not wonderful?" Our book - We survey the literature on bounded rationality and implications for public policy. We outline the challenges, speculate on which theories might work better, and offer a research agenda for modern bounded rationality. - We define rationality in its various contexts within the BRA, and document extensively the empirical rejections of BRA. - The evidence is far more supportive of models of behavioral economics which is an interdisciplinary approach richly grounded in the social and behavioral sciences. - We do not shy away from challenging topics: Evolutionary game theory; stochastic social dynamics; formal dynamic epidemiological models of the spreads of epidemics; dynamic programming; axiomatic foundations of utility functions; behavioral decision theory; complexity and chaos; and the bias-variance dilemma in machine learning. Towards Bounded Rationality: Two Approaches - Incremental program: Keep fixed other features of the BRA. Relax one feature (e.g., unlimited attention) at a time hoping to get better models. But the end result might be influenced by the order in which assumptions are relaxed. - Big-push program: Give up most of the rejected features of BRA in one go and make empirically founded assumptions. - Most economists have preferred the incremental program; largely 'as if' approach based on substantive rationality. - We make a case for also taking more seriously the big-push program, with a particular emphasis on replacing mathematical optimization with judgement heuristics. - A central feature of the big-push program is procedural rationality (e.g., aspiration adaptation). The Kahneman-Tversky Research Program - Heuristics are simple rules of thumb that are fast and frugal in the use of information. Allow cognitive simplification. - In pathbreaking work in the 1970's K-T showed that people use simple heuristics in making choices (see also Herbert Simon). And that they do not follow the laws of classical statistics. BRA does not hold even in an "as if" sense. - We survey/formalize the relevant heuristics; and evaluate the criticism & empirical evidence for them. - Heuristics examined include representativeness; anchoring; availability; confirmationbias; hindsight-bias; ... We also resolve some important confusions in the literature on heuristics. - We also critically examine the approach to heuristics by Gerd Gigerenzer and show that the two frameworks are not necessarily adversarial. Base Rate Neglect: A K-T finding - Prior beliefs. New Information. Posterior beliefs. How? - Bayes' Law is the main tool used in BRA. But empirical evidence typically rejects it. - Consider the following problem given to students at Harvard Medical School (Casscells et al., 1978). Suppose that a police Breathalyzer test discovers false drunkenness in 5% of the cases when the driver is sober. - However, the Breathalyzers are always able to detect a truly drunk person with complete certainty. In the general population, 1/1000 of drivers engage in driving while drunk. - A driver is checked at random, and the test is positive. What is the probability the driver is drunk? - The modal response was 95%, the mean response was 56%. The correct Bayesian response is approximately 2%. Dealing with True Uncertainty - Donald Rumsfeld (2002) "..as we know, there are known knowns; there are things we know we know. We also know there are known unknowns; that is to say we know there are some things we do not know. But there are also unknown unknowns-the ones we don't know we don't know. And if one looks throughout history...it is the latter category that tends to be the difficult ones." - Economists call the three situations described above as (1) risk, (2) subjective uncertainty, and (3) True uncertainty. - Economics makes no clear predictions under true uncertainty. Mathematical optimization fails. - We propose that heuristics and social norms, because they offer cognitive simplifications, may offer a powerful tool for dealing with true uncertainty. Private and Social Rationality - Modern game theory relies on private rationality. - Prisoner's dilemma game- metaphor for human cooperation. - Simultaneously choose to cooperate C (not snitch) or to defect D (snitch). Payoffs for each pair of choices in matrix. - Private rationality predicts (D, D), but 60% of the choices are (C, C). We often observe more cooperation than can be accounted for by private rationality. - Social rationality may explain the data. Prisoner's might have a social norm or code of conduct so that they do not snitch. They might de-facto behave as if they are maximizing some sort of team payoff. - We outline several concepts in social rationality (team reasoning, correlated equilibria as social norms, Kantian rationality). Conclusions - The BRA in its current form is untenable. This has serious implications for economics, political science, law, finance, and management. - The way forward is developing models in bounded rationality. - We outline what has already been achieved, what the likely challenges are, and what the way forward might be. We do not have all the answers, but this is the nature of science. - Bounded rationality also has critical and hugely important implications for welfare and public policy.