Inference on Causal and Structural Parameters using Many Moment Inequalities

Victor Chernozhukov, Denis Chetverikov, Kengo KatoThe Review of Economic Studies 2013

TL;DR

This paper tackles the challenge of testing economic models with a vast number of moment inequalities, a common scenario in partially identified models. The authors introduce a flexible framework for constructing valid statistical tests, even when the number of inequalities ($p$) far exceeds the sample size ($n$), offering a significant advancement for empirical economics.

What Problem Does This Paper Address?

In econometrics, many models are "partially identified," meaning they don't pin down exact parameter values but rather a set of possible values consistent with the data. Moment inequalities are a powerful tool for characterizing these identified sets. However, in complex economic models (like games with multiple equilibria or dynamic markets), the number of moment inequalities implied can explode, becoming exponentially large in the number of underlying parameters or choices. Existing statistical methods struggle with such a high-dimensional inequality setting, often requiring researchers to arbitrarily select a subset of inequalities, potentially leading to biased inference. This paper addresses the critical need for statistically sound inference procedures that can handle a large, potentially much larger than sample size, number of moment inequalities.

Methods

The core of the paper is developing robust methods for computing critical values for a test statistic that captures violations of these moment inequalities. The authors propose several approaches:

  • Self-Normalized (SN) Method: This analytical approach uses a union bound combined with a moderate deviation inequality for self-normalized sums. It's computationally simple and provides a baseline critical value.
  • Bootstrap Methods (Multiplier and Empirical): These simulation-based methods (Multiplier Bootstrap - MB, and Empirical Bootstrap - EB) are more computationally intensive but are designed to better account for the correlation structure among the moment inequalities, potentially leading to less conservative tests.
  • Two-Step and Three-Step Procedures: These methods enhance the power of the SN and bootstrap approaches by incorporating inequality selection. They first identify and discard "uninformative" inequalities (those clearly not binding the null hypothesis) and, in the three-step version, also weakly informative ones, to focus on the most relevant constraints.

The paper rigorously proves the validity of these methods, showing they control the Type I error rate (rejecting a true null hypothesis) polynomially in $n$, even when $p$ grows exponentially with $n$.

Datasets

Not discussed in the paper. The paper focuses on theoretical development and uses simulated data in its Monte Carlo experiments.

Key Insights and Ideas

  • Scalability is Key: The primary innovation is developing methods that remain valid and powerful even when the number of moment inequalities ($p$) is vastly larger than the sample size ($n$). This is crucial for applying these techniques to realistic, complex economic models.
  • Bootstrap's Advantage: While the analytical SN method is simple, the bootstrap methods (MB and EB) are shown to be less conservative and potentially more powerful, especially when the data exhibits correlation. This is because they implicitly account for the dependence structure among the inequalities.
  • Inequality Selection Matters: The two-step and three-step procedures, which intelligently select relevant inequalities, significantly improve test power by avoiding over-conservatism caused by irrelevant or weakly informative constraints.
  • Uniform Validity: The methods are proven to be valid uniformly over large classes of underlying distributions, making them broadly applicable and robust.

Key Results

  • Polynomial Size Control: The proposed tests achieve Type I error control with the error decreasing polynomially in $n$, even when $p$ is exponentially larger than $n$.
  • Improved Power: The bootstrap and multi-step selection procedures lead to tests with better power properties compared to the one-step SN method, particularly in settings with correlated data or when alternatives are close to the null.
  • Theoretical Guarantees: The paper provides non-asymptotic bounds on rejection probabilities, offering strong theoretical backing for the validity of the proposed inference procedures.
  • Application to Economic Models: The methods are demonstrated to be applicable to challenging economic models, such as the market structure model of Ciliberto and Tamer (2009), which generates a very large number of moment inequalities.

Limitations and Future Work

  • Computational Intensity: While the SN method is computationally simple, bootstrap methods are more demanding, especially the multi-step versions.
  • Tuning Parameter Selection: The paper provides some guidance for selecting tuning parameters (e.g., $\beta$) through simulations, but optimal selection can still be a challenge.
  • Extension to Non-Differentiable Functions: The three-step method's extension to non-differentiable functions is left for future work.
  • Dependent Data: While an extension to dependent data is discussed in an appendix, further exploration of its practical implications and broader applicability is suggested.
  • Specific Model Violations: The paper notes that for certain alternatives very close to the null, no test can be uniformly consistent at a fast rate.

The authors suggest future work could explore more general bootstrap procedures and further investigate the performance of their methods in various empirical settings.

Summary updated

Abstract

Abstract This article considers the problem of testing many moment inequalities where the number of moment inequalities, denoted by $p$, is possibly much larger than the sample size $n$. There is a variety of economic applications where solving this problem allows to carry out inference on causal and structural parameters; a notable example is the market structure model of Ciliberto and Tamer (2009) where $p=2^{m+1}$ with $m$ being the number of firms that could possibly enter the market. We consider the test statistic given by the maximum of $p$ Studentized (or $t$-type) inequality-specific statistics, and analyse various ways to compute critical values for the test statistic. Specifically, we consider critical values based upon (1) the union bound combined with a moderate deviation inequality for self-normalized sums, (2) the multiplier and empirical bootstraps, and (3) two-step and three-step variants of (1) and (2) by incorporating the selection of uninformative inequalities that are far from being binding and a novel selection of weakly informative inequalities that are potentially binding but do not provide first-order information. We prove validity of these methods, showing that under mild conditions, they lead to tests with the error in size decreasing polynomially in $n$ while allowing for $p$ being much larger than $n$; indeed $p$ can be of order $\exp (n^{c})$ for some $c > 0$. Importantly, all these results hold without any restriction on the correlation structure between $p$ Studentized statistics, and also hold uniformly with respect to suitably large classes of underlying distributions. Moreover, in the online supplement, we show validity of a test based on the block multiplier bootstrap in the case of dependent data under some general mixing conditions.
BibTeX
@article{chernozhukov2013inference,
  author = {Victor Chernozhukov and Denis Chetverikov and Kengo Kato},
  title = {{Inference on Causal and Structural Parameters using Many Moment Inequalities}},
  journal = {The Review of Economic Studies},
  year = {2013},
}
Figure