Why I feel so strongly about ordinal utility.
Let me put some more color to this tweet:
Arrow’s impossibility theory clearly shows that ordinal utility cannot be aggregated across different agents and goods. Ordinal utility contains preference information in terms of rankings but not degree of satisfaction, hence it cannot be aggregated. This puts DSGE models in a bind because such models use ordinal utility. The entire premise behind DSGE micro-foundations is the aggregation of individual decisions into macro outcomes. Accordingly, DSGE has had to resort to a number of restrictive and implausible assumptions to overcome the aggregation problem such as a single representative agent, a single representative good, static preferences, etc.
Ordinal utility has been further undermined by advances in behavioral economics. In 1979, Daniel Kahneman and Amos Tversky put forth Prospect Theory and the notion that people make decisions based on gain-loss prospects as opposed to the absolute utility of outcomes. This provided the impetus for a new discipline — behavioral economics. Over the last 30 years, behavioral economics has made significant progress in understanding choice and decision-making. Furthermore, the discipline has amassed substantial experimental evidence that rejects the utility-optimization framework embedded in mainstream models (more reading on behavioral economics here and here). Nonetheless, these advancements have seen few applications in macroeconomics primarily due to the difficulty of framing consumption in terms of gain-loss prospects. Going back to the aggregation problem, calculating gains and losses pre-supposes addition and subtraction from a reference point. By definition, ordinal utility cannot accommodate such simple arithmetic.
Third point, DSGE relies on the axioms of expected utility theory in order to derive a pseudo-cardinal conception of utility. This maneuver allows for the construction of aggregate utility curve that can be plugged in DSGE models. The shape of the utility curve is guided by the notions of declining marginal utility and risk-aversion. Accordingly, utility is specified as some function that is always rising albeit at a declining rate. Declining marginal utility makes intuitive sense for an individual good (addictive drugs being an exception); however, it is difficult, if not impossible, to reconcile declining marginal utility with the unlimited nature of human needs. Clearly, human needs (and the corresponding utility upon the satisfaction of such needs) are no less diminished as we become richer and consume more — simply, we desire and want different things. What this means is that the ordinal utility framework lacks a reference point for needs. Furthermore, declining marginal utility is not a requirement for risk-aversion. Instead, risk-aversion can be explained with the gain-loss asymmetry expressed in terms of needs. In other words, the unsatisfied needs due to a dollar of losses are always greater than the satisfied needs due to a dollar of gains.
Hence, the point of my tweet. The DSGE utility framework suffers from an aggregation problem; it lacks a reference point for needs and last but least, it is not consistent with advances in behavioral economics. As such, it is not even an idealized representation of agent behavior. Accordingly, DSGE models cannot be considered micro-founded.