Beyond Prices: Institutions as Signal Architectures

Economics has a familiar way of thinking about markets. A market brings together people who each hold only part of the relevant information, and prices allow them to coordinate without reconstructing the full state of the economy. A buyer does not need to know every production cost, inventory condition, or alternative use of resources. A seller does not need to know every buyer’s preferences. Much of that information is compressed into a price.

That relationship is so familiar that it can obscure a broader point. A market is one type of institution, and a price is one type of signal. Other institutions organize economic and social life through different signals. Firms rely on output measures, budgets, forecasts, KPIs, and performance evaluations. Schools use grades, credentials, and rankings. Regulators use ratings, classifications, capital ratios, and stress tests. Political systems convert complex public preferences into votes, polls, party labels, and mandates.

Digital platforms make the point especially clearly because they often operate through many signals at once. Amazon presents a buyer not only with price, but also ratings, reviews, rankings, delivery information, badges, sponsorship, and recommendations. Sellers see still another set of indicators. The institution is therefore not simply a place where a price is formed. It is an environment in which multiple representations of a much richer commercial reality are produced, ordered, and made consequential.

This suggests a useful way to approach a wide range of economic and social problems:

What becomes visible, who acts on it, through what institutional setting, and what changes because they do?

The question is broader than measurement. It is about how institutions make complex systems actionable.

1. Beyond Prices

No economic actor operates on a complete description of reality.

A manager does not directly observe everything employees know, how much judgment sits inside a team, how resilient the organization will be under stress, or whether current production is reproducing future capability. A regulator cannot observe the entire latent risk structure of the financial system. A voter cannot process the full multidimensional state of a country. A consumer does not reconstruct the production history, quality distribution, and reliability of every product before making a purchase.

Decisions therefore depend on compressed representations.

A bank is not its capital ratio, but the ratio can be useful. A student is not a GPA, but a GPA conveys information. An economy is not GDP, but GDP helps organize discussion and policy. An employee is not annual output, but output often enters compensation, promotion, and staffing decisions.

The important point is not that these representations are incomplete. All useful representations are incomplete. A representation that preserved everything would cease to simplify the problem.

The analytical question is instead whether the information being discarded remains irrelevant for the decision being made.

That is where the institution matters. A number acquires economic significance not simply because it exists, but because someone uses it. A rating matters because lenders, investors, or regulators act on it. A ranking matters because students, employers, donors, or administrators respond. A KPI matters because compensation or managerial attention depends on it. A recommendation matters because it changes visibility and choice.

A signal becomes economically important when it enters a decision rule.

This is why the institution and the signal should be considered together. The same observable can have very different consequences in different institutional settings. A performance statistic used privately for diagnosis is not the same object once it determines compensation. A forecast circulated internally is different from one that becomes official policy guidance. A review written by a consumer becomes something else once a platform aggregates thousands of reviews into a ranking that affects future visibility.

Institutions do not merely process information. They help determine which information becomes actionable.

2. Institutions as Architectures of Visibility

It is useful to think of institutions as partly architectures of visibility.

This does not replace the conventional view of institutions as systems of rules, incentives, rights, authority, and constraints. It adds another dimension. Institutions also determine what parts of a complex underlying reality become sufficiently legible to enter decisions.

Some of this visibility is designed. Firms choose accounting systems and performance measures. Regulators define reporting categories. Platforms design rankings and recommendation systems.

Some of it emerges. Market prices arise from decentralized exchange. Reputation develops through repeated interaction. Professional status can evolve without a central designer.

The distinction between designed and emergent signals matters, but both perform a similar economic function: they reduce complexity enough for people to act.

The result is usually not one signal but a signal architecture.

A university is observed through grades, degrees, admissions selectivity, research reputation, rankings, placement outcomes, recommendations, and alumni networks. A financial institution is observed through market prices, credit spreads, ratings, reported earnings, capital ratios, liquidity measures, analyst forecasts, and supervisory assessments. A political system produces polls, vote shares, fundraising totals, endorsements, media attention, party identification, and issue salience.

These signals may reinforce one another, substitute for one another, or conflict.

A high ranking can attract stronger applicants, which helps sustain the ranking. A credit downgrade can increase borrowing costs and thereby worsen the financial condition that justified the downgrade. A recommendation system can increase demand for a product, and the additional purchases then generate reviews and behavioral data that feed future recommendations.

Once we look at institutions this way, a different set of questions becomes natural. Which signals dominate attention? Which are redundant? Which represent distinct dimensions of the underlying state? Who sees which signals? Which ones carry authority? Which ones trigger action?

That perspective can matter even when every individual measure is statistically respectable. A system can become distorted not because a particular indicator is false, but because some dimensions of reality become highly visible and others remain institutionally quiet.

Visibility itself can shape allocation.

3. When Signals Become Part of the System

The deeper issue begins when people adapt to the signals through which they are evaluated, coordinated, or governed.

A university ranking initially summarizes characteristics of universities. Once applicants, donors, faculty, and administrators respond to it, the ranking becomes part of the process producing future university characteristics. A credit rating affects financing conditions. A public forecast changes expectations. A political poll changes campaign strategy. A product ranking changes customer traffic.

At that point observation is no longer outside the system.

The signal has entered the causal structure.

This feedback can take a relatively straightforward form. People respond to a signal, and their response changes the underlying state. That mechanism is familiar in economics. Incentives redirect effort. Public information coordinates behavior. Prices alter supply and demand. Social proof affects adoption.

But a more difficult case arises when response changes what the signal itself tells us.

Artificial intelligence offers a useful example.

Organizations have historically inferred something about human capability from visible output. A strong analysis, functioning computer program, legal memorandum, or technical design normally tells us something about the knowledge and judgment of the person or team that produced it.

Generative AI changes that relationship.

Two workers can now produce similarly polished outputs while relying on very different levels of independent expertise. The visible output may improve while carrying less information about the human capability underneath it.

Nothing necessarily went wrong with the output measure. Output still measures output.

What changed is the inference being drawn from it.

This distinction becomes consequential when organizations use strong output to justify further delegation, reduced apprenticeship, smaller junior teams, or different training investments. Those decisions then affect the future stock of human capability. The organization can therefore experience strong current performance while gradually changing the relationship between performance and the capacity required to reproduce it.

The same logic can appear elsewhere.

Infrastructure can continue functioning while deferred maintenance accumulates. A financial institution can report stable current performance while hidden exposures migrate. A low accident count can coexist with increasing exposure to rare catastrophic failures. An educational credential can retain the same formal definition while the production process behind student work changes.

The important question is therefore not only whether an observable is accurate.

It is also:

What does this observable allow us to infer about the underlying state we actually care about?

And, more importantly:

Has that relationship changed because people, technology, or institutions adapted to the signal system?

That is a different question from ordinary measurement error.

4. Ownership, Time, and Horizon

Three features of signal architectures are especially useful for understanding when such changes matter: ownership, update speed, and horizon.

Ownership concerns who controls the representation. Who defines the categories? Who supplies the inputs? Who validates the measure? Who can change the formula? Who determines how prominently the signal is displayed and what consequences follow from it?

A decentralized market price has no single owner in the same sense as a corporate KPI. A public statistical series differs from a private scoring algorithm. A platform ranking differs from a professional credential. A measure constructed partly from self-reported inputs differs from one maintained independently.

Ownership affects incentives, transparency, manipulability, and adaptation.

AI benchmarks illustrate the point. A public benchmark may initially provide useful information about model capability. Once the benchmark becomes important for investment, procurement, competition, and reputation, developers naturally optimize around it. Compare that with an evaluation whose questions are privately held by an independent evaluator until testing occurs. Both attempt to assess capability, but the surrounding institutional arrangements are different.

That comparison can be more informative than simply asking whether the benchmark is “good.”

Update speed concerns how quickly the signal architecture adapts when the world changes. Institutions often inherit measures from an earlier technological or behavioral environment. A metric can remain formally unchanged even after the process generating it has been transformed.

The relevant mismatch is often between how quickly the underlying system adapts and how quickly the institution revises the representation through which it sees the system.

Technology can alter work practices within months while professional training changes over years. Financial activity can migrate faster than regulatory categories. Business models can change faster than accounting conventions. Political attention can move much faster than the identities and values underneath it.

A representation can therefore become stale without becoming obviously wrong.

Horizon concerns what the signal is informative about and for how long.

Current output may be highly informative about current production and weakly informative about long-run capability. Current earnings can describe profitability today while revealing little about whether a firm is maintaining the productive assets required for the future. A low default rate can describe benign current conditions while carrying little information about resilience under stress.

There is rarely a meaningful answer to the question, “Is this a good signal?” without specifying the decision and the horizon.

A better question is:

Good for what, and for how long?

Once these three dimensions are visible, many apparently technical measurement problems become institutional problems.

5. Changing the Representation

The most useful practical move is often simply to change what we are looking at.

Suppose a firm concludes that AI adoption has raised productivity because output per employee has increased sharply. Now observe something else: the ability to diagnose an unfamiliar error without assistance, reconstruct the reasoning behind a solution, or teach the underlying task to another person.

If the conclusion changes, we have learned something. The original output measure was answering one question while perhaps being used implicitly to answer another.

Suppose a platform interprets engagement as evidence of public importance. Compare engagement with population prevalence.

Suppose safety is evaluated through incident counts. Compare the count with incidents relative to exposure.

Suppose political polarization is summarized by a single left-right coordinate. Examine the distribution of beliefs across individual issues.

Suppose two financial institutions report similar capital ratios. Examine whether those ratios conceal radically different asset composition, funding structure, or common exposures.

This is a simple form of representation rotation.

The goal is not to find the one true signal. Complex systems rarely have one. The purpose is to determine whether the phenomenon looks different when the representation changes.

Sometimes it will not. That is useful information. A market price may already be the relevant economic object. A diffusion pattern may be well explained by social learning. A competitive outcome may be driven primarily by participant composition.

In other cases the rotation changes the problem itself. What appeared to be productivity becomes a distinction between current output and capability reproduction. What appeared to be polarization becomes a distinction between underlying beliefs and the public axis on which those beliefs are projected. What appeared to be financial stability becomes a question about exposures hidden inside the same regulatory category.

The representation is then not merely a way of reporting the phenomenon.

It is part of the phenomenon.

A Broader Institutional Perspective

There is a long intellectual history behind this way of thinking.

Hayek emphasized the role of prices in communicating dispersed knowledge. Hurwicz treated institutions as information systems in which messages and rules jointly determine outcomes. Marschak and Radner studied organizations as information-processing systems. Spence and Stiglitz developed signaling and screening. Holmström, Milgrom, and Baker examined incentives built around imperfect performance measures.

Later work moved more directly into feedback. Espeland and Sauder showed how rankings reshape the organizations they rank. Donald MacKenzie described financial models that can become engines of markets rather than merely cameras recording them. Alex Frankel and Navin Kartik have studied how reliance on manipulable information can change what that information reveals. Lucas provided the broader warning that relationships observed under one policy environment may not survive when behavior adjusts to another.

These traditions answer different questions. They should not be collapsed into one theory.

Taken together, however, they point toward a useful habit of economic reasoning.

When confronting a complex problem, ask not only what incentives actors face or what information exists. Ask what part of reality becomes visible enough to enter decisions. Ask who controls that representation, who responds to it, how quickly it updates, and whether the same signal continues to mean the same thing over the horizon that matters.

Sometimes those questions will add little.

Sometimes they will reveal that the most important change is not in the observable itself, but in the relationship between what we can see and the system underneath it.

That is where the Institution–Signal Lens earns its value.

---

References

Aoki, Masahiko. 2011. “Institutions as Cognitive Media between Strategic Interactions and Individual Beliefs.” Journal of Economic Behavior & Organization 79 (1–2): 20–34.

Baker, George P. 1992. “Incentive Contracts and Performance Measurement.” Journal of Political Economy 100 (3): 598–614.

Espeland, Wendy Nelson, and Michael Sauder. 2007. “Rankings and Reactivity: How Public Measures Recreate Social Worlds.” American Journal of Sociology 113 (1): 1–40.

Frankel, Alex, and Navin Kartik. 2019. “Muddled Information.” Journal of Political Economy 127 (4): 1739–1776.

Hayek, F. A. 1945. “The Use of Knowledge in Society.” American Economic Review 35 (4): 519–530.

Holmström, Bengt, and Paul Milgrom. 1991. “Multitask Principal–Agent Analyses: Incentive Contracts, Asset Ownership, and Job Design.” Journal of Law, Economics, & Organization 7: 24–52.

Hurwicz, Leonid. 1960. “Optimality and Informational Efficiency in Resource Allocation Processes.” In Mathematical Methods in the Social Sciences, edited by Kenneth J. Arrow, Samuel Karlin, and Patrick Suppes. Stanford University Press.

Hurwicz, Leonid. 1972. “On Informationally Decentralized Systems.” In Decision and Organization, edited by C. B. McGuire and Roy Radner. North-Holland.

Lucas, Robert E., Jr. 1976. “Econometric Policy Evaluation: A Critique.” Carnegie-Rochester Conference Series on Public Policy 1: 19–46.

MacKenzie, Donald. 2006. An Engine, Not a Camera: How Financial Models Shape Markets. MIT Press.

Marschak, Jacob, and Roy Radner. 1972. Economic Theory of Teams. Yale University Press.

Spence, Michael. 2002. “Signaling in Retrospect and the Informational Structure of Markets.” American Economic Review 92 (3): 434–459.