The Causal Impact of Credit Lines on Spending Distributions

Authors: Yijun Li, Cheuk Hang Leung, Xiangqian Sun, Chaoqun Wang, Yiyan Huang, Xing Yan, Qi Wu, Dongdong Wang, Zhixiang Huang

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental To assess the effectiveness of our methods, we conduct a simulation study. The results reveal that all three estimators are effective, especially for the DML estimator. We finally apply our approach to investigate the causal impact of credit lines on spending distributions based on a real-world dataset collected from a large e-commerce platform.
Researcher Affiliation Collaboration 1 School of Data Science, City University of Hong Kong 2 Department of Financial and Actuarial Mathematics, Xi an Jiaotong Liverpool University 3 Institute of Statistics and Big Data, Renmin University of China 4 JD Digits
Pseudocode Yes Algorithm 1: Computations of ˆ di;w
Open Source Code Yes Our code is available at https://github.com/lyjsilence/The Causal Impact-of-Credit-Lines-on-Spending-Distributions.
Open Datasets No We finally apply our approach to investigate the causal impact of credit lines on spending distributions based on a real-world dataset collected from a large e-commerce platform. ... We collect data from 4,043 platform users. ... The data comprises various variables... Appendix F displays a detailed statistical description.
Dataset Splits Yes We split the N units into K disjoint groups. Let the kth group be Dk of size Nk and form D k. ... 5-fold cross-fitting, i.e., 4, 000 instances are used to train, and 1, 000 instances are used to obtain the three estimators (i.e., DR, IPW, and DML estimator).
Hardware Specification No No specific hardware details (e.g., GPU/CPU models, memory amounts) are mentioned for the experimental setup.
Software Dependencies No The paper mentions 'random forest' and 'MLP' but does not provide specific version numbers for any software, libraries, or frameworks.
Experiment Setup Yes The classification and functional regression models are trained separately. 5, 000 generated instances are trained using 5-fold cross-fitting, i.e., 4, 000 instances are used to train, and 1, 000 instances are used to obtain the three estimators (i.e., DR, IPW, and DML estimator). At last, we average the obtained estimators from the 5 folds as the final results. In the classification task, we use the same classifier (i.e., random forest) to compute IPW for all the estimators. The training details are given in Appendix E.