Collaborative Research: Artificial intelligence and deep learning solution methods for dynamic economic models

人工智能（AI）有许多令人印象深刻的应用，包括自动驾驶汽车、计算机视觉和语音识别。该项目表明，许多具有挑战性的经济模型和应用程序可以通过使用与数据科学家处理其令人印象深刻的应用程序相同的突破性人工智能技术和相同的最先进的软件和硬件组合来成功分析。该项目开发开源人工智能软件，使研究在早期解决方法下难以解决的复杂经济模型成为可能。这一人工智能框架的应用包括货币政策的中央银行模型、财富不平等的增长模型以及社会保障和人口老龄化的模型。开发的人工智能工具将通过提供仔细记录的复制材料、示例和教程在业内传播。这个项目由几个部分组成。首先，该项目展示了如何将经济动力学的三个基本对象——终身奖励、Bellman方程和Euler方程——转换为适合深度学习的目标函数。其次，采用随机梯度下降法在少数随机绘制的网格点上实现目标的最大化，而不是传统的求解方法使用一个大的固定网格。第三，将多个期望算子组合为一个无偏期望算子，构造动态经济模型的期望算子。第四，该项目自动化了人工智能解决方案框架，使其无处不在，可移植到其他应用程序中。第五，该项目解决了一系列与经验相关的应用，这些应用代表了对经济学中现有解决方法的挑战。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

AbstractArtificial intelligence (AI) has many impressive applications, including self-driving cars, computer vision, and speech recognition. This project demonstrates that many challenging economic models and applications can be successfully analyzed by using the same break-ground AI technologies and the same state-of-the-art combinations of software and hardware as those used by data scientists for dealing with their impressive applications. This project develops open-source AI software that makes it possible to examine complex economic models that were intractable under the earlier solution methods. Applications of this AI framework include central-banking models of monetary policy, growth models of wealth inequality, and models of social security and population aging. The developed AI tools will be disseminated in the profession by providing carefully documented replicated materials, examples and tutorials.This project consists of several components. First, the project shows how to convert three fundamental objects of economic dynamics -- lifetime reward, Bellman equation and Euler equation -- into objective functions suitable for deep learning. Second, the project adapts the stochastic gradient descent method to maximizing the objective on few randomly drawn grid points instead of a large fixed grid used by conventional solution methods. Third, the project shows how to construct the expectation operators for dynamic economic models by combining multiple expectation operators into a single unbiased expectation operator. Fourth, the project automates the AI solution framework to make it ubiquitous and portable to other applications. Fifth, the project solves a collection of empirically relevant applications that represent challenges to the existing solution methods in economics.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.