Thread by @oliverbeige

Microeconomic equilibrium models (partial and general) use an almost mirrored approach to modeling supply and demand. The supply side, production output of a profit-maximizing firm, is fairly straightforward other than ignoring data as a factor of production. (History matters.)

On the demand side, the consumer or household goes through a similar linear optimization process to optimize budget-constrained (expected) utility based on the choice between consumption bundles. Where the two meet, perfection reigns.

Utility, subjective value, is itself an improvement on the prior concept of objective value. But it's been shrouded in controversy from the very beginning, since the axiomatic assumption was that we can infer preferences, and in turn utility, only indirectly, via observed choice.

These axiomatic assumptions are currently collapsing in typical Kuhnian fashion, and we not only get to observe the collapse in real time, we are even the subjects of the field experiment. As data-driven behemoths move closer to linking our choices with the preferences we reveal.

While in the process we are slowly getting a clearer picture of how we really make our consumption decisions, we also get to observe the attendant economic community in the state of increasing conflict between theoretical paradigm and empirical reality, exactly as predicted.

Thomas Kuhn's Structure of Scientific Revolutions might be the most prophetic book of all time.

Strangely enough, all three processes: supply, demand, and Kuhn's revolution are tightly coupled. They play off the same playbook: as growth cycles of collective tentative multi-step decision processes with the ongoing goal to minimize negative surprise.

In other words, complex organisms, be it firms, the human brain, or a rookery of scientists, conceive an (optimistic) view of their world which they try to adhere to while responding to negative triggers that the real world does not match the conceived world.

These are processes we get to understand ever better on all fronts, and the mathematical models have astounding overlap. But the most compelling evidence is that those entities who have the most at stake from being precise in their forecasts are switching to Machine Learning.

The first observations that humans don't make their choices in isolation, out of a deep-seated basic need, but are also taking in and digesting signals in order to tentatively close in on their most preferred consumption bundle (or technology) are very old.

But they kicked into overdrive around 1995, when the internet took over and more and more observations confirmed that humans make their choices as the result of interactions with their peers. A firm's production factors are labor, capital, land and data. A human's — other humans.

This doesn't only apply to so-called network effects but also to the ubiquitous recommender and review systems hat help customers buy their favorite products online without risking — negative surprises.

Just how confused the economic profession is about this shines through in this recent article by MIT's Dick Schmalensee on the limits of network effects, who is actually ahead of the curve by having an inkling that the increasing returns from "network effects" eventually decrease

...even though he cannot express why (hint: like production, it is a growth cycle based on the right combination of factor inputs: peers), but then manages to get demand side network effects confused with supply side hype cycles.
mitsloan.mit.edu/shared/ods/doc…

Malthusian growth cycles: tentative beginnings, growing strength as all factors are aligned, maturity, and eventual decay, are ubiquitous inside and outside of economics. Production theory uses them explicitly to point out that marginal returns are decreasing in equilibrium.

They are even ubiquitous in production environments, where things are being put together, eventually run smoothly and efficiently at increasing scale until component decay and build-up reduce performance. Machine replacement is one of the oldest economic puzzles.

And just to top it off, they are also the underlying forces of the hype-disappointment cycles which confused Dick. Investors pour money into a black box betting on a return of uncertain likelihood, size, and delay, and like many lottery ticket buyers, they are often disappointed.

In hype cycles, if only the estimated delay between investment and return is too optimistic the cycle will eventually recover. If the estimate on the return turns out too optimistic as we are closing in in a Bayesian updating process, the cycle will flatline — negative surprise.

Over the last twenty years we have developed the mathematical tools to better model an economy, especially tools that don't decouple "economic agents" from cognitive decision making processes. They are all used in production environments and becoming increasingly sophisticated.

On the sophisticated end we have things like Bayesian belief networks and (closely related) Markov random fields. On the folksy end we have Malthusian cycles. Just learn how to look under the hood of a growth cycle. If you can't do that, you shouldn't call yourself an economist.

Economics is increasingly clutching at the straw called causal inference to justify its recalcitrant refusal to leave the linear world under increasing strain of evidence that the real world is nonlinear and Bayesian. But phony inference is phony.

The supply-demand cross is still an extremely parsimonious way to start investigating how exchange of assets, market-based or not, really works. Its usefulness for prediction or inference is extremely narrow, and for policy implications arbitrarily close to zero.

In 25 years we'll look back, equipped with the toolset developed over the last 25 years and a couple of Nobels along the way, and wonder how folks back in the 2010s couldn't see what is so blatantly obvious in front of them.

That's how Kuhn cycles work.