摘要
在大数据和算力迅速增长的背景下,人工智能,尤其是机器学习在经济金融领域研究得到广泛应用.本文探讨了机器学习方法与传统计量经济学方法在研究范式、主要目的和先验假设方面的异同,并提出其在经济金融领域应用的"三板斧":基于高维和大规模数据预测、另类数据处理和特征重要性识别.已有研究借助机器学习方法独特的"三板斧"对变量预测、另类数据构建和因果推断等做出了有益补充和完善,但数据质量、模型可解释性及结果的潜在偏见和非公平性成为制约其深入应用的挑战.未来研究应聚焦于通过迁移学习等框架改善数据可用性和质量问题,提升模型预测性能和可解释性,以及建立透明公正的模型评估标准,尽可能避免偏见和非公平结果.
Abstract
Amid the rapid growth of big data and computing power,artificial intelligence,especially machine learning,has made significant progress in economic and financial research.This paper explores the similarities and differences between machine learning methods and traditional econometric methods,focusing on research paradigms,main objectives,and prior assumptions.It introduces the"three key tools"of machine learning for its application on economic and financial fronts:high-dimensional and large-scale data prediction,alternative data processing,and feature importance identification.By reviewing existing literature,this paper finds that machine learning methods,through these key tools,have greatly enhanced variable prediction,alternative data construction,and causal inference in economics and finance.However,challenges related to data quality,model interpretability,and potential bias and unfairness in results limit their broader application.Future research should focus on improving data availability and quality through frameworks like transfer learning,enhancing model prediction accuracy and interpretability,and establishing transparent and fair model evaluation standards to mitigate bias and unfair outcomes.
NETL
NSTL
万方数据