Abstract
A common problem in econometrics, statistics, and machine learning is to estimate and make inference on functions that satisfy shape restrictions. For example, distribution functions are nondecreasing and range between zero and one, height growth charts are nondecreasing in age, and production functions are nondecreasing and quasi-concave in input quantities. We propose a method to enforce these restrictions ex post on point and interval estimates of the target function by applying functional operators. If an operator satisfies certain properties that we make precise, the shape-enforced point estimates are closer to the target function than the original point estimates and the shape-enforced interval estimates have greater coverage and shorter length than the original interval estimates. We show that these properties hold for six different operators that cover commonly used shape restrictions in practice: range, convexity, monotonicity, monotone convexity, quasi-convexity, and monotone quasi-convexity. We illustrate the results with two empirical applications to the estimation of a height growth chart for infants in India and a production function for chemical firms in China.
Introduction
In many scientific and economic domains, researchers possess strong prior knowledge regarding the shape of the functions they seek to estimate. For instance, a child’s height should never decrease with age, and a production function should typically exhibit diminishing marginal returns. Despite these logical constraints, modern statistical and machine learning models—ranging from simple regressions to complex neural networks—are often "shape-agnostic."
When trained on noisy or limited data, these models frequently produce "wiggly" estimates that violate fundamental physical or economic principles. While methods exist to incorporate constraints during the training phase, they often introduce significant computational overhead or degrade the convergence rates of the underlying estimators. This paper introduces a generic "ex post" framework that applies functional operators to the outputs of any estimator, ensuring they adhere to desired shape restrictions while simultaneously improving statistical performance.
Methodology: The Power of Functional Operators
The proposed method relies on applying a functional operator $T$ to an initial "raw" estimate $\hat{f}$. For an operator to be effective in this context, the authors identify four critical properties:
- Reshaping: The transformed function $T\hat{f}$ must satisfy the required shape constraint.
- Invariance: If the input $\hat{f}$ already satisfies the constraint, the operator should leave it unchanged ($T\hat{f} = \hat{f}$).
- Order Preservation: If function $f$ is pointwise greater than function $g$, then $Tf$ should remain pointwise greater than $Tg$.
- Distance Reduction (Contraction): The operator must bring the estimate closer to the true function $f_0$ in terms of a chosen norm (e.g., the supremum norm).
Key Operators
The research details six operators that satisfy these properties, covering the most common shape restrictions in practice:
- Range (R): Enforces minimum and maximum bounds via simple thresholding.
- Monotonicity (M): Utilizes "rearrangement" (sorting) to ensure the function is non-decreasing or non-increasing.
- Convexity (C): Employs the Double Legendre-Fenchel transform to identify the "greatest convex minorant."
- Quasi-convexity (Q): A novel operator developed by the authors to handle this economically significant but mathematically challenging shape.
- Monotone Convexity and Monotone Quasi-convexity: Compositions of the above to handle complex, multi-faceted constraints.
Empirical Applications
The framework was tested on two distinct real-world datasets to demonstrate its practical utility:
- Childhood Malnutrition in India: Using data from 37,623 children, the authors estimated height-for-age growth charts. Unconstrained models often showed biologically impossible "dips" in height at certain ages; the shape-enforcing operators corrected these artifacts into smooth, non-decreasing curves.
- Chinese Chemical Production: The method was applied to data from 1,638 firms to estimate production functions mapping labor and capital to output. Enforcing monotonicity and concavity ensured the model aligned with economic theory regarding marginal productivity.
Key Results and Statistical Benefits
The application of these operators provides what the authors demonstrate to be a statistical improvement over raw estimates. Because the operators are distance-reducing, the resulting point estimates are mathematically guaranteed to be at least as accurate as the original estimates. In simulations with a sample size of $n=500$, enforcing both monotonicity and concavity reduced estimation error (sup-norm) by approximately 37%.
Furthermore, the method significantly improves inference. By applying the operators to the upper and lower bounds of a confidence interval, researchers obtain "shape-enforced" bands. These bands are not only narrower (often by more than 20%) but also exhibit higher "coverage" (the probability that the true function lies within the interval).
Conclusion
The ex post shape-enforcement framework offers a powerful, model-agnostic "wrapper" for modern estimation techniques. It allows researchers to leverage the high-speed convergence of advanced machine learning models while ensuring the final results are physically and economically sensible. This approach effectively removes the trade-off between model flexibility and logical consistency, providing a robust tool for empirical researchers across disciplines.
Summary updated