Achim Ahrens, Christian B. Hansen, Mark E. Schaffer, Thomas Wiemann
{"title":"模型平均和双机器学习","authors":"Achim Ahrens, Christian B. Hansen, Mark E. Schaffer, Thomas Wiemann","doi":"10.1002/jae.3103","DOIUrl":null,"url":null,"abstract":"<p>This paper discusses pairing double/debiased machine learning (DDML) with <i>stacking</i>, a model averaging method for combining multiple candidate learners, to estimate structural parameters. In addition to conventional stacking, we consider two stacking variants available for DDML: <i>Short-stacking</i> exploits the cross-fitting step of DDML to substantially reduce the computational burden, and <i>pooled stacking</i> enforces common stacking weights over cross-fitting folds. Using calibrated simulation studies and two applications estimating gender gaps in citations and wages, we show that DDML with stacking is more robust to partially unknown functional forms than common alternative approaches based on single pre-selected learners. We provide Stata and <span>R</span> software implementing our proposals.</p>","PeriodicalId":48363,"journal":{"name":"Journal of Applied Econometrics","volume":"40 3","pages":"249-269"},"PeriodicalIF":2.3000,"publicationDate":"2025-01-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1002/jae.3103","citationCount":"0","resultStr":"{\"title\":\"Model Averaging and Double Machine Learning\",\"authors\":\"Achim Ahrens, Christian B. Hansen, Mark E. Schaffer, Thomas Wiemann\",\"doi\":\"10.1002/jae.3103\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<p>This paper discusses pairing double/debiased machine learning (DDML) with <i>stacking</i>, a model averaging method for combining multiple candidate learners, to estimate structural parameters. In addition to conventional stacking, we consider two stacking variants available for DDML: <i>Short-stacking</i> exploits the cross-fitting step of DDML to substantially reduce the computational burden, and <i>pooled stacking</i> enforces common stacking weights over cross-fitting folds. Using calibrated simulation studies and two applications estimating gender gaps in citations and wages, we show that DDML with stacking is more robust to partially unknown functional forms than common alternative approaches based on single pre-selected learners. We provide Stata and <span>R</span> software implementing our proposals.</p>\",\"PeriodicalId\":48363,\"journal\":{\"name\":\"Journal of Applied Econometrics\",\"volume\":\"40 3\",\"pages\":\"249-269\"},\"PeriodicalIF\":2.3000,\"publicationDate\":\"2025-01-19\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"https://onlinelibrary.wiley.com/doi/epdf/10.1002/jae.3103\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Journal of Applied Econometrics\",\"FirstCategoryId\":\"96\",\"ListUrlMain\":\"https://onlinelibrary.wiley.com/doi/10.1002/jae.3103\",\"RegionNum\":3,\"RegionCategory\":\"经济学\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q2\",\"JCRName\":\"ECONOMICS\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Journal of Applied Econometrics","FirstCategoryId":"96","ListUrlMain":"https://onlinelibrary.wiley.com/doi/10.1002/jae.3103","RegionNum":3,"RegionCategory":"经济学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"ECONOMICS","Score":null,"Total":0}
This paper discusses pairing double/debiased machine learning (DDML) with stacking, a model averaging method for combining multiple candidate learners, to estimate structural parameters. In addition to conventional stacking, we consider two stacking variants available for DDML: Short-stacking exploits the cross-fitting step of DDML to substantially reduce the computational burden, and pooled stacking enforces common stacking weights over cross-fitting folds. Using calibrated simulation studies and two applications estimating gender gaps in citations and wages, we show that DDML with stacking is more robust to partially unknown functional forms than common alternative approaches based on single pre-selected learners. We provide Stata and R software implementing our proposals.
期刊介绍：
The Journal of Applied Econometrics is an international journal published bi-monthly, plus 1 additional issue (total 7 issues). It aims to publish articles of high quality dealing with the application of existing as well as new econometric techniques to a wide variety of problems in economics and related subjects, covering topics in measurement, estimation, testing, forecasting, and policy analysis. The emphasis is on the careful and rigorous application of econometric techniques and the appropriate interpretation of the results. The economic content of the articles is stressed. A special feature of the Journal is its emphasis on the replicability of results by other researchers. To achieve this aim, authors are expected to make available a complete set of the data used as well as any specialised computer programs employed through a readily accessible medium, preferably in a machine-readable form. The use of microcomputers in applied research and transferability of data is emphasised. The Journal also features occasional sections of short papers re-evaluating previously published papers. The intention of the Journal of Applied Econometrics is to provide an outlet for innovative, quantitative research in economics which cuts across areas of specialisation, involves transferable techniques, and is easily replicable by other researchers. Contributions that introduce statistical methods that are applicable to a variety of economic problems are actively encouraged. The Journal also aims to publish review and survey articles that make recent developments in the field of theoretical and applied econometrics more readily accessible to applied economists in general.
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号
