在将来的计量经济学发展趋势中，Machine learning和Big data一定是绕不过去的弯，原因在于他们是AI时代的标配。以下19篇文章可以帮助咱们更好地理解机器学习和大数据在计量经济学中的应用，因此建议收藏。
10.详解聚类, 回归, 分类三种统计方法, 人人都能够懂的方式
14.回归方法深度剖析(OLS, RIDGE, ENET, LASSO, SCAD, MCP, QR）
15.高维回归方法: Ridge, Lasso, Elastic Net用了吗
18.深度学习 vs. 概率图模型 vs. 逻辑学
Machine learning (ML) is most commonly understood as a set of computational techniques applied to big datasets in order to make granular predictions for businesses, from advertising to fraud detection to user recommendations. Yet another, perhaps less appreciated, application comes from academia, where social scientists have slowly but steadily begun leveraging ML techniques to gain new insights from data.
In a recent paper from the National Bureau of Economic Research (NBER), Susan Athey provided a useful assessment of the contributions of ML to economics, summarizing emerging econometric literature combining ML and causal inference, before drawing broader conclusions about the impact of ML on economics as a field.
The crucial difference between econometrics and ML: The former aims to establish causality, while the latter aims to produce accurate and actionable predictions
First, Athey culled through the plethora of buzzwords related to ML to establish a practical definition for economists and policy analysts. She defined ML as “a field that develops algorithms designed to be applied to data sets, with the main areas of focus being prediction, classification and clustering tasks.” For the author, the key was a clear distinction between the goals of ML techniques and the goals of traditional econometric methods of causal inference. In econometrics, Athey wrote, the primary aim is typically to uncover a clean, causal relationship between the outcome variable and another variable of interest. As such, econometricians established a solid empirical framework for answering questions regarding the impact of various policy changes on particular populations. In contrast, ML techniques are not designed to identify causal relationships between variables; rather, their purpose is to make accurate predictions. Athey argued that this constitutes the crucial difference between econometrics and ML: The former aims to establish causality, while the latter aims to produce accurate and actionable predictions.
Fortunately, this does not mean both frameworks cannot work together. In fact, Athey argued that there is much to gain from implementing both frameworks side by side. For instance, as sophisticated ML applications become increasingly skilled at granular prediction, ML practitioners may no longer be able to ignore the question of causality. If they do, they risk losing sight of what drives the predictive success of their models. Because of this, ML experts can benefit from a coordinated application of econometrics to their work.
Likewise, ML techniques can be useful for econometricians. These experts can employ ML techniques to improve, expand and even uncover data to build stronger econometric models. By allowing the design of systematic model-selection processes, ML techniques can help economists avoid inappropriate model selection. Moreover, ML techniques can improve the evaluation of policy interventions by tweaking several standard methodologies, thereby enabling econometricians to identify causal relationships even with small samples. An alliance between ML and econometrics also permits the estimation of more realistic—and therefore more complex—models, due to the computational performance of novel ML techniques. For example, Ruiz et al. (2017) used ML techniques to analyze consumer preferences for bundles selected from more than 5,000 items, an exercise that yielded more than 25,000possibilities. A calculation of this magnitude would computationally have been prohibitively intensive only a few years ago.
The combination of ML techniques and econometric tools for causal inference stands not only to produce a promising new strand of literature, but also to inspire a profound transformation in the field of economics. In the future, the ML-econometrics partnership may provide novel solutions to important econometric problems and enhance the debate around contentious policy research questions.
Article source: Athey, Susan. “The Impact of Machine Learning on Economics.” The Economics of Artificial Intelligence: An Agenda. National Bureau of Economic Research, forthcoming (draft 2018).
SHOPPER: A Probabilistic Model of Consumer Choice with Substitutes and Complements, https://arxiv.org/abs/1711.03560
干货系列：能源环境 | 效率研究 | 空间计量 | 国际经贸 | 计量软件 | 商科研究