Realistic Economic Modeling Would Require Massive Information Collection, Taking Away Freedom
May 13, 2022
Figure. The step response of a process element’s empirical model that has dead time, primary and secondary lags, and process gain. (James Beall, ISA )
Conditions always change. Also, it’s costly to measure information and to use information. These facts greatly hinder economic modeling.
The solution that’s optimal is to not model the economy and to instead just limit governments.
Below, I provide a detailed picture of the feasibility of economic modeling by discussing the real world’s complexity, future advanced modeling, the two basic model types that are currently available, and some striking limitations of current modeling. In the big picture, realistic economic modeling is not needed to guide human action, and would needlessly destroy value and freedom.
People Buy Products, Develop Products, Coerce Others
Every person uses scarce information to forecast what actions are likely to work out best for him.
A person controlling a single process variable by manipulating a single final-control element can be modeled well. A ship helmsman’s direction-control actions were observed to be proportional to the process variable’s error, adjusted if any offset was building up, and adjusted if the error was changing rapidly. Such actions were modeled as proportional-integral-derivative control . A century later, this remains the workhorse algorithm that’s used to control most process variables .
A person manipulating many controls to achieve a single objective, though, hasn’t been modeled well.
Individuals’ actions in various economic roles are complex and are affected by many variables:
All these individual actions, on each transaction, determine which products people produce and buy, and in what quantities.
A person maximizes his satisfaction by making use of his brain’s neural networks.
These neural networks filter the person’s perceptions and distribute control instructions to where they’re needed in the person’s body. The filtering is accomplished by predicting what the person will sense, and the control instruction is accomplished by predicting what precise actions will result from specific instructions. In both cases, the neural networks monitor for how well their predictions are matched by reality, and the neural networks make adjustments as needed, which may include updating their models. In short, the brain’s neural networks continuously model and validate .
These neural networks not only are sophisticated mechanisms but also require extensive training using incoming information and trial-and-error action. Since this processing works in humans, this or other processing plausibly could work in artificial intelligence models.
Models of customers choosing among products, even if motivated only to improve targeting in marketing and sales, can easily be anticipated to develop ever-increasing accuracy.
Models of product developers’ creativity and management skills would be complex, but conceivably could be developed along the same path.
Models of government coercers’ creativity and social interactions would be complex and could be developed similarly.
Neural networks trained to spend household budgets, to develop products, or to expand government controls would be highly anthropomorphic both in their actions and in their needs for huge amounts of information inputs and learning—for very-big data.
To date, in contrast, practical models of human behavior have only captured human behavior that’s much simpler, like the helmsman’s control mentioned earlier.
Current Dynamic Process Modeling
Economic change is a dynamic process, which calls for dynamic models.
Dynamic process models are built up of numerous element models. Each element model is of one of two types—an empirical model or a physical model.
Empirical model :
Physical model  (of the physics; also called a first-principles model):
With both types of element models, the commonly-used models don’t capture control valves’ real-world stick-slip behavior, called stiction . A single control valve with stiction can cause plantwide oscillations in many other process variables .
Stickiness can be routinely neglected in models of process plants, since oscillations will be seen and mitigated, or preventive control-valve positioners can be installed .
Stickiness in economics is normal human behavior, is hard to reduce, and is highly destructive. For example, stickiness changed just another avoidable contraction  into the sustained Great Depression .
The first step in dealing with stickiness is to admit we have a problem.
Regardless of which element types a modeler uses for various elements, the modeler must validate that the resulting model approaches reality closely enough to be useful. For models that are large in scale or that include complex elements or dependencies, validation would demand copious information .
For individual buyers, producers, and coercers, the underlying lifetimes of preparation, the balance sheets, and other factors are not measured, reported, and learned.
A modeler could restrict consideration to, say, the aggregated buyers in a nation and the aggregated producers in each sector. At such levels of aggregation, much information is reported.
A modeler conceivably might correlate the rates of product innovation in different sectors by starting from the general observation that the most innovation that makes its way into practical products for sale is the innovation that’s done in the sectors that are the least regulated . The modeler might quantify the many forms that regulation takes and their many impacts, and correlate these negative inputs to the changes in various sectors’ relative outputs. But such modeling hasn’t been done.
The result is that despite its large scale, current economic modeling takes the product mix as frozen into place across all time . There is no modeling of the innovation, the increasing specialization, and the full frictional drags from governments that most-fundamentally determine how the economic value-added changes over time.
So then at present, at most a modeler might, for instance, try to correlate how aggregate transactions might change by sector if a pandemic would spread, or if a government would increase its debt-financed spending.
Even for such limited questions, what modeling has been done has not been adequate. The most-developed correlations have ignored debt  and have ignored most actions of governments.
Of course individuals—whose aggregate behavior, after all, is what is being modeled—obviously do pay particularly-close attention to debt, and to all kinds of actions of governments that affect them.
Why Model in Economics? At What Costs?
One consideration about economic modeling that should be foremost is the modeling’s purpose:
The answer, from Austrian economics, is straightforward: none at all. Models aren’t needed to show individuals the fact that what’s best is individual freedom .
Individual moral behavior is the greatest contributor to making life, liberty, and property secure .
For further augmenting this self-restraint, government actions currently have greatly displaced private actions . Government actions can make life, liberty, and property further secure, if these government actions are properly limited.
All other government actions beyond these narrow boundaries destroy value and limit opportunities.
A second consideration about economic modeling that’s inescapable is the modeling’s cost:
Losing freedom to a government or a government crony has almost always ended up being deadly .
Even short of that, if any resources would be taken from individuals in order to model their actions, the full amount of these resources could no longer be used by the individuals themselves, as buyers or as producers. Individuals would be coerced into doing less for themselves than they otherwise would be able to do.
Individuals’ resulting definitive losses of property and opportunities would not be offset by any hypothetical gains elsewhere.
The best way to optimize people’s actions is to not make measurements that would make people’s work less productive, and to not make models as a way to try to change people, but to just leave people alone.
Ultimately the best model of the real world is the real world itself.
To focus our energies on living in the real world, making the real world better in the ways that we ourselves each have the powers to do—this is ideal.
James Anthony is the author of The Constitution Needs a Good Party and rConstitution Papers, publishes rConstitution.us, and has written in The Federalist, American Thinker, Foundation for Economic Education, American Greatness, and Mises Institute. Mr. Anthony is an experienced chemical engineer with a master’s in mechanical engineering.