Sequential algorithm for structural estimations with equilibrium constraints

arXiv:2606.04356 · econ.EM · Submitted 2026-06-03 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Economics. The day in economics: theory, mechanisms and the evidence beneath them.

Ruth: We're looking at the paper: "Sequential algorithm for structural estimations with equilibrium constraints".

Warren: Here's the gist of the paper. This study investigates sequential algorithms possessing the Zero Jacobian Property (ZJP) for estimating structural models constrained by equilibrium conditions, specifically focusing on Maximum Likelihood Estimation (MLE) and the Generalized Method of Moments (GMM).

Ruth: These algorithms demonstrate fast, near-quadratic local convergence in large samples to the constrained optimization solution.

Warren: Furthermore, if consistent initial parameter estimates are available, these methods yield an asymptotically efficient estimator even after a single iteration.

Ruth: The core contribution is the proposal of the Sequential Linearly Constrained (SLC) algorithm.

Paper summary: Ruth: So, to recap what we're hearing from "Sequential algorithm for structural estimations with equilibrium constraints," the core thesis is that you can use sequential algorithms that possess the Zero Jacobian Property or ZJP to estimate structural models even when those models have equilibrium constraints >

Warren: They show this works for Maximum Likelihood Estimation and Generalized Method of Moments, showing fast local convergence in large samples to the constrained optimization solution >

Iris: The paper introduces a specific algorithm called the Sequential Linearly Constrained or SLC algorithm, which they claim is applicable to a much broader class of structural models than previously known methods >

Ruth: And the key advantage they highlight for this SLC algorithm is that it can be implemented without explicitly computing the Jacobian matrix of those equilibrium constraints, which makes it multiple times faster than existing approaches like Nested Fixed Point or NFXP >

Warren: They tested this performance using two numerical examples: a dynamic discrete game with time-varying unobserved heterogeneity and a dynamic demand model >

Iris: And they confirmed that the SLC algorithm performs about four to eight times faster than NFXP, whether you use analytical derivatives or not >

Ruth: They also discussed the theoretical convergence properties, showing that these sequential algorithms with ZJP have nearly quadratic local convergence rates when the sample size is sufficiently large >

Warren: And they noted a significant strength is that this convergence property holds even when there's no consistent initial estimate of Y available for non-standard structural models >

Iris: They proved this unification across Maximum Likelihood Estimation and Generalized Method of Moments, showing that both share the same property ∇Y Q theta b Yb equals Op N minus one two > <ref:2606.04356#pg3>

Ruth: This means you can use these sequential algorithms for GMM too, which broadens the applicability of this kind of method significantly >

Warren: They also showed that if consistent initial estimates for those parameters are available, the estimator becomes asymptotically efficient even after only one iteration >

Iris: But they did flag a limitation: this convergence rate is dependent on having a sufficiently large sample size for it to be reliable >

Ruth: And another point they mentioned is that while the SLC algorithm minimizes the objective function given constraint linearization, it's more robust against multiple solutions to those equilibrium constraints compared to Lagrangian based algorithms like SQP >

Warren: So, basically, they're proposing a method that's faster computationally and has better theoretical convergence guarantees than what we currently have for these constrained problems >

Conclusion: Ruth: So looking at the full title "Sequential algorithm for structural estimations with equilibrium constraints," what this paper really boils down to is that they’ve developed a new sequential approach that handles those tricky equilibrium conditions in structural models without needing the heavy machinery of explicit Jacobian calculations >

Warren: The authors, Takeshi Fukasawa and others, are proposing the SLC algorithm as a way to get better performance than traditional methods like NFXP because it's computationally faster in practice >

Iris: The implication for us is that if you're doing structural estimation in economics or industrial organization where those constraints are common, you have this potentially much more efficient tool available that doesn't require an initial guess of the economic variables Y to start >

Ruth: And it’s not just about speed. They show it can be asymptotically efficient with just one step if you happen to have those good starting parameters, which is a nice practical feature for researchers >

Warren: It changes how we think about these constrained optimization problems by showing that sequential algorithms with ZJP offer superior statistical efficiency and convergence guarantees compared to non-ZJP methods >

Iris: So, the big picture is that for structural estimation with constraints, this paper positions the SLC algorithm as a highly efficient candidate because it gives you strong theoretical guarantees on how fast you'll converge >

Waseda University

econ.EM

Submitted: 2026-06-03

Updated: 2026-11-16

Code: https://github.com/takeshi-fukasawa/sequential_SLC

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 91/100

The gist: This study investigates sequential algorithms possessing the Zero Jacobian Property (ZJP) for estimating structural models constrained by equilibrium conditions…

Key concepts

Zero Jacobian Property (ZJP)
This property describes a specific mathematical characteristic of sequential algorithms that ensures they possess fast convergence rates, specifically near-quadratic local convergence. This property is crucial for guaranteeing efficient estimation when solving complex constrained optimization problems iteratively.
Sequential Linearly Constrained (SLC) Algorithm
The SLC algorithm is the core contribution of the study. It is designed to solve constrained optimization problems by sequentially solving a linearized version of the original constraint equation, rather than explicitly calculating the Jacobian matrix. This makes it computationally much faster than traditional methods like Nested Fixed Point.
Asymptotic Efficiency
This refers to a theoretical guarantee that an estimator achieves the lowest possible variance among all possible estimators as the sample size becomes very large. The paper shows that the SLC algorithm can reach this high level of efficiency, even after just one iteration, provided initial estimates are consistent.
Jacobian-free Implementation
This technique involves solving linear equations using methods like Krylov-based solvers instead of explicitly calculating the Jacobian matrix of the constraints. This approach significantly reduces computational cost and memory usage, making it practical for large-scale structural estimation problems.

Terminology

Summary

This study investigates sequential algorithms possessing the Zero Jacobian Property (ZJP) for estimating structural models constrained by equilibrium conditions, specifically focusing on Maximum Likelihood Estimation (MLE) and the Generalized Method of Moments (GMM). p. 0

Where the paper says it

Sequential algorithm for structural estimations with equilibrium constraints∗ Takeshi Fukasawa† June 8, 2026 Abstract This study examines sequential algorithms with the Zero Jacobian Property (ZJP) for estimating structural models subject to equilibrium constraints. For the Maximum Likelihood Estimation (MLE) and the Generalized Method of Moments (GMM), the current study shows that these algorithms attains fast (near-quadratic) local convergence in large samples to the solution of the constrained optimization problem.

These algorithms demonstrate fast, near-quadratic local convergence in large samples to the constrained optimization solution. p. 0
Where the paper says it

Sequential algorithm for structural estimations with equilibrium constraints∗ Takeshi Fukasawa† June 8, 2026 Abstract This study examines sequential algorithms with the Zero Jacobian Property (ZJP) for estimating structural models subject to equilibrium constraints. For the Maximum Likelihood Estimation (MLE) and the Generalized Method of Moments (GMM), the current study shows that these algorithms attains fast (near-quadratic) local convergence in large samples to the solution of the constrained optimization problem.

Furthermore, if consistent initial parameter estimates are available, these methods yield an asymptotically efficient estimator even after a single iteration. p. 0
Where the paper says it

If consistent initial estimates of the parameters are available, the algorithms yield an asymptotically efficient estimator even after one iteration. It then proposes a novel algorithm called Sequential Linearly Constrained (SLC) algorithm, which is applicable to a broader class of structural models than existing methods.

The core contribution is the proposal of the Sequential Linearly Constrained (SLC) algorithm. p. 0

Where the paper says it

If consistent initial estimates of the parameters are available, the algorithms yield an asymptotically efficient estimator even after one iteration. It then proposes a novel algorithm called Sequential Linearly Constrained (SLC) algorithm, which is applicable to a broader class of structural models than existing methods.

The SLC algorithm is designed for a broader class of structural models and notably avoids the explicit computation of the Jacobian matrix of the equilibrium constraints, making it multiple times faster than existing methods like Nested Fixed Point (NFXP). p. 0
Where the paper says it

It then proposes a novel algorithm called Sequential Linearly Constrained (SLC) algorithm, which is applicable to a broader class of structural models than existing methods. A key advantage of the SLC algorithm is that it can be implemented without explicitly computing the Jacobian of the equilibrium constraints and can be multiple times faster than the Nested Fixed Point (NFXP) approach.

The SLC algorithm is defined by a specific mapping that satisfies the ZJP and is equivalent to sequentially solving a constrained optimization problem using a linearization of the original constraint G(Y; theta) = 0. p. 2
Where the paper says it

Building on the unified framework, the present study proposes a novel SLC algorithm as an alternative to previously developed sequential methods. Each iteration is essentially equivalent to solving a surrogate constrained optimization problem in which the constraint is replaced by a linearization of the original constraint G (Y ; θ) = 0.

Theoretical analysis confirms that sequential algorithms with ZJP exhibit nearly quadratic local convergence rates when the sample size is sufficiently large, even without consistent initial estimates of Y. p. 2

Where the paper says it

Concerning the convergence properties of the sequential algorithms with the ZJP, this study demonstrates that their local convergence rate is nearly quadratic, provided the sample size is sufficiently large—even when a consistent initial estimate of Y is unavailable. As it is unclear whether general methods to obtain initial consistent estimate of Y for non-standard structural models are available, the non-dependence on the availability of initial consistent estimates is a notable strength.

Proposition 5 summarizes this convergence rate: gamma k - gamma e = A (gamma e) grad Y Q(theta, e) Y e grad theta gamma'upsilon(theta, e gamma e) + O| gamma k-1 - gamma e | squared.

Numerical experiments on a dynamic discrete game with time-varying unobserved heterogeneity and a dynamic demand model validate the performance of the SLC algorithm, showing it is approximately 4 to 8 times faster than NFXP, irrespective of whether analytical derivatives are employed. p. 0

Where the paper says it

The current study illustrates its performance through two numerical experiments: a dynamic discrete game with time-varying unobserved heterogeneity and a dynamic demand model. Keywords: Sequential estimation; Zero-Jacobian property; SLC algorithm; NFXP; MPEC ∗The current study builds on my prior study titled “Sequential algorithm for estimating structural models without using analytical derivatives.

The study also highlights Jacobian-free implementation methods for SLC and EPL algorithms, which leverage solving linear equations (e. p. 2
Where the paper says it

Each iteration is essentially equivalent to solving a surrogate constrained optimization problem in which the constraint is replaced by a linearization of the original constraint G (Y ; θ) = 0. 3 The current study further discusses that the SLC algorithm can be implemented without explicitly computing the Jacobian of the equilibrium constraint function G (Y ; θ), ∇Y G (Y ; θ), which is very beneficial because computing Jacobians is very cumbersome and sometimes prone to coding mistakes.

g., via Krylov-based methods) instead of explicit Jacobian computation, offering advantages in computational cost and memory usage. p. 5
Where the paper says it

computational cost. 7 Other than the NFXP, MPEC, and sequential algorithms, two-step estimation methods have been proposed for certain standard models.

The paper concludes that while consistent initial estimates enhance efficiency after one iteration, the SLC algorithm can achieve statistical efficiency without them. p. 2

Where the paper says it

The current study also shows that the estimator attains the same (statistical) efficiency as the NFXP estimator even after one iteration when the iteration is started from consistent initial values, in line with the previous methods utilizing ZJP. Building on the unified framework, the present study proposes a novel SLC algorithm as an alternative to previously developed sequential methods.

The study also compares SLC with Lagrangian-based algorithms like SQP, noting that SLC minimizes the original objective function Q(theta, Y) given the constraint linearization, and is more robust against multiple solutions to G(Y; theta) = 0. p. 3
Where the paper says it

5 Both the SQP and the SLC sequentially solve constrained optimization problems where the constraint equation corresponds to the linearization of the equilibrium constraint G(Y ; θ) = 0. The key difference lies in the objective function of the constrained objective function: the SLC uses the original objective function Q (θ; Y ), but the SQP uses a quadratic objective function using the candidate value of the Lagrange multiplier.

Extensions to include inequality constraints on structural parameters theta are also explored. p. 1
Where the paper says it

1 Introduction Structural estimations with equilibrium constraints are prevalent in economics. Examples include dynamic discrete choice models (henceforth DDC; e.

Overall, sequential algorithms with ZJP provide superior statistical efficiency and theoretical convergence guarantees compared to non-ZJP methods, positioning the SLC algorithm as a highly efficient candidate for constrained optimization in statistical applications. p. 5
Where the paper says it

Aguirregabiria and Mira, 2007). 9 Compared to these, sequential algorithms with ZJP offer advantages in terms of statistical efficiency and theoretical guarantees on local convergence speed.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Sequential estimation algorithms can attain fast (near-quadratic) local convergence in large samples to MLE and GMM solutions, even without consistent initial estimates of Y. This allows AI systems to reliably estimate structural parameters from noisy data without requiring perfectly known starting points for nuisance parameters.

  2. The proposed Sequential Linearly Constrained (SLC) algorithm can be implemented without explicitly computing the Jacobian of equilibrium constraints, which is crucial for high-dimensional problems where storing and computing large matrices is computationally prohibitive. This enables the AI system to handle complex structural models with significantly reduced memory overhead compared to methods relying on explicit Jacobians.

  3. The SLC algorithm can be multiple times faster than the Nested Fixed Point (NFXP) approach, leading to substantial computational savings in estimation time. This speedup allows for faster iteration cycles in AI training or parameter tuning processes involving constrained optimization problems.

  4. The SLC algorithm is applicable to a broader class of structural models because it does not rely on specific structures like the existence of a fixed-point mapping with the Zero Jacobian Property (ZJP). This makes the estimation framework more versatile for various economic and statistical modeling tasks where standard assumptions might not hold.

  5. The SLC algorithm can be implemented without explicitly computing Jacobians, relying instead on solving a linear equation (∇Y G(Yk; θk)) x = b for x in R nY using Krylov-based algorithms. This Jacobian-free approach is beneficial because it avoids the computational cost and potential coding mistakes associated with manually coded Jacobians.

  6. The SLC algorithm can be used to solve constrained optimization problems where the constraint is replaced by a linearization of the original constraint G (Y; θ) = 0, making it similar to Sequential Quadratic Programming (SQP) but using the original objective function Q(θ, Y). This provides a robust and efficient way to handle equilibrium constraints in sequential updates.

  7. The SLC algorithm exhibits good performance even without initial consistent estimates of theta and Y, although incorporating such estimates yields statistically efficient parameter estimates after one iteration. This robustness means the AI system can start its estimation process with less stringent requirements on prior knowledge of the true parameters.

  8. The convergence properties show that for large samples, the local convergence speed is nearly quadratic under conditions where ∇Y Q(θ, Y) is approximately zero, even if it may not be positive definite in finite samples. This guarantees fast parameter estimation once sufficient data is collected.

  9. Incorporating fixed-point iteration acceleration methods like the spectral algorithm can substantially accelerate and stabilize the convergence of sequential algorithms with ZJP, even when they are not contractive near the solution. This allows AI systems to overcome slow convergence issues in practical scenarios where the pseudo-objective function Qe(θ; γ) is flat near the solution.

  10. The SLC algorithm can be combined with a line search procedure to guarantee global convergence under moderate conditions, ensuring that the estimation process does not get stuck in local optima. This feature provides a more reliable path to finding the true structural parameter values compared to unaccelerated sequential methods.

  11. The AI system can utilize Neyman orthogonality-based score functions, which are constructed based on the objective function Q(θ, Y) and its derivatives. These score functions help obtain debiased estimates of parameters when consistent initial estimates of nuisance parameters are available, improving the statistical efficiency of the final results.

  12. The system can leverage numerical derivative-based Jacobian-free methods to compute Jacobian-vector products (JVPs) efficiently, even in high-dimensional settings. This is important because computing the full Jacobian matrix is memory intensive, and JVPs can be approximated using central finite differences or automatic differentiation techniques with acceptable performance trade-offs.

Abstract

This study examines sequential algorithms with the Zero Jacobian Property (ZJP) for estimating structural models subject to equilibrium constraints. For Maximum Likelihood Estimation (MLE) and the Generalized Method of Moments (GMM), the current study shows that these algorithms attain fast (near-quadratic) local convergence in large samples to the solution of the constrained optimization problem. If consistent initial estimates of the parameters satisfying a suitable rate condition are available, the algorithms yield an asymptotically efficient estimator even after one iteration. It then proposes a novel algorithm called the Sequential Linearly Constrained (SLC) algorithm, which is applicable to a broader class of structural models than existing methods. A key advantage of the SLC algorithm is that it can be implemented without explicitly computing the Jacobian of the equilibrium constraints and can be several times faster than the Nested Fixed Point (NFXP) approach. The current study illustrates its performance through two numerical experiments: a dynamic discrete game with time-varying unobserved heterogeneity and a dynamic demand model.

Sources

Related papers