Shadow Prices Rank Competing Restrictions Under Model Uncertainty

When a theory is only partly right, the hard part is deciding how much to trust it. This paper tackles that problem by putting competing restrictions on a model on a price tag: a shadow price that measures how costly each restriction is to the data fit. The authors build a unified framework for weak, noisy, or approximate restrictions, alongside nuisance control covariates, and let the data choose a tolerance level with a Stein-type risk criterion. They also use a debiasing step based on Karush–Kuhn–Tucker conditions, and introduce individual shadow prices to judge the empirical relevance of different restrictions one by one. A plateau rule is proposed to help separate signal from noise. The paper establishes consistency and asymptotic normality for the estimators, and characterizes the individual shadow prices. Simulations and an application to a Solow growth model show how the approach can work in practice when model uncertainty makes simple yes-or-no model selection too brittle.

A growth model can look neat on a whiteboard and messy in data. That gap matters when a theory adds rules you are not sure about. One bad rule can pull the fit off course. One good rule can still help steady the result. This approach gives each rule a price tag. That price is a shadow price. It shows how much the data pay when a restriction stays in place. Think of it like a kitchen rulebook. Some rules help the meal. Some only slow the cook. The method sorts them by real cost, not by pride. That is sharper than a simple yes-or-no vote. It matters when theory is only partly right.

Pricing the rules, not just the fit

The core idea is not to force every restriction into the same box. Some rules are weak. Some are noisy. Some are only partly true. The framework measures the degree of misspecification, or how far a restriction misses the data. It gives each rule its own shadow price. That number tracks the cost of keeping the rule when data resist it. A Stein-type risk rule, a data score that balances fit and noise, picks the tolerance level. Tolerance means how much mismatch the model will allow. A ridge penalty is a shrinkage rule that pulls estimates toward zero. Karush-Kuhn-Tucker conditions, the check points for a constrained best fit, then help undo shrinkage bias. Individual shadow prices, or ISP, compare the rules one by one. A plateau rule helps separate signal from noise. The method is consistent. That means it homes in on the right target as data grow. It is also asymptotically normal. That means the errors settle into a bell curve. Simulations and a Solow growth model show it working in practice.

How the price gets set

The engine here is constrained optimization. That means finding the best fit while keeping some rules in place. A Lagrangian is a way to score fit and rule keeping together. It lets the model trade one against the other. When a rule bites hard, its shadow price rises. When a rule barely changes the fit, its price stays low. The individual shadow prices compare those costs one by one. Karush-Kuhn-Tucker conditions mark the edge of the best constrained fit. The debiasing step uses them to undo shrinkage bias. The plateau rule then looks for a flat stretch in those prices. Flat means the data no longer see extra signal. That gives a cleaner break between weak theory and real evidence.

Why this changes model choice

This matters because model choice is often too blunt. A rule is not always right or wrong. It can be almost right and still useful. It can also be costly enough to drop. Shadow prices turn that gray zone into a number. That helps with theory-driven work, where rules come from economic ideas. It also helps when nuisance covariates, extra controls that are not the main target, clutter the picture. The plateau rule gives a fast way to stop chasing tiny changes. So the method does not just pick a model. It explains why one rule deserves more trust than another.

What the next test should show

That makes a practical promise. An analyst can keep an imperfect theory in play without hiding its cost. The shadow price says what each rule costs. The plateau rule says when more detail stops adding signal. In a Solow growth model, a classic model of how economies grow, that means competing growth ideas can be weighed one by one. They no longer need a crude keep-or-drop vote. The next test should be another setting with many candidate restrictions and nuisance covariates. The hard question is whether the same price pattern still separates weak theory from strong evidence there.