Replication data and code for AI Assistance in Mortgage Lending: Completion, Selection, and Credit Decisions
Year: 2026
Type: dataset
Abstract: Replication data and analysis code for the theoretical and descriptive measurement study ‘AI Assistance in Mortgage Lending: Completion, Selection, and Credit Decisions’ (FinTech manuscript fintech-4651450). Version 1.0.1. This release contains public-HMDA-derived aggregate count inputs, annual reporting-entity mappings, sample selection and exclusion procedures, exact source-vintage hashes, executable estimation and classification code, complete sensitivity schedules, theoretical numerical checks, and validation documentation. It supports principal-table reproduction from supplied aggregates and reconstruction from archived public HMDA Loan/Application Register and Transmittal Sheet files for 2018–2025. Separate historical replay drivers reproduce the supporting appendix analyses. Validation covers the principal tables, all 7,392 classification ranges, 408 historical geographic-regression rows, and 17 additional analysis modules; all 61 archived CSV files reconcile with the frozen archive. The empirical exercise concerns recorded lending outcomes and conditional denominator-classification scenarios. Original public sources and acquisition instructions are documented in the release. The supplied files contain aggregate cells without loan/application identifiers, borrower names or property addresses. Corporate disclosures are cited in the accompanying manuscript. License scope: CC BY 4.0 applies to authored documentation and derived aggregate summaries; the authors’ analysis code is supplied under the MIT License. Source-provider notices and terms apply separately to upstream materials. Contact: Yang Li, School of Business, Singapore University of Social Sciences, [email protected]. Reproduction map for the minor revision (6 November 2026) Exact archive: Version 1.0.1, published 29 October 2026, DOI 10.5281/zenodo.23750651. The manuscript is now titled Mortgage Application Outcomes: Theory of AI Assistance and HMDA Measurement Sensitivity. The original dataset title, version, DOI and all seven archived files are unchanged. This addition documents the existing release. Extract AI_Lending_Replication_v1.0.1.zip. All paths below are relative to that archive. Default: python reproduce.py. Optional modes: --raw-root PATH, --historical, --appendices. Read the archived README.md and TABLE_COVERAGE.md. Exhibit Executable script Aggregate or model input Table 1, Table 2 Analytical definitions / cited sources No empirical input: model/recording definitions Table 3 code/documentation_cost_model.py No empirical input: stated model parameters Table 4 code/rebuild_public_aggregates.py; code/empirical_revision.py inputs/sample_flow.csv; inputs/cohort.csv; inputs/raw_data_inventory.csv; inputs/ts_source_inventory.csv Table 5 code/empirical_revision.py; code/common_mix.py inputs/joint_2024.parquet; inputs/joint_2025.parquet; inputs/cohort.csv; inputs/national_2019.parquet Table 6 code/check_principal_exhibits.py; code/baseline_estimator.py inputs/joint_2024.parquet; inputs/joint_2025.parquet; inputs/cohort.csv; inputs/documentation_resolution_sensitivity.csv Table 7 code/empirical_revision.py inputs/reason_cells.parquet; inputs/cohort.csv Table 8 code/check_principal_exhibits.py inputs/joint_2024.parquet; inputs/joint_2025.parquet; inputs/cohort.csv; inputs/origination_count_decomposition.csv Table 9 code/empirical_revision.py inputs/joint_2024.parquet; inputs/joint_2025.parquet; inputs/location_route_common_pool.csv; inputs/cohort.csv; inputs/national_2019.parquet Figure 1 code/general_costs.py No empirical input: stated model parameters Figure 2 historical/Topic2_research/code/cost_approval_dependence.py No empirical input: stated model parameters Figure 3, Table A19, Table A20, Table A21, Table A23 code/analyze_frontier.py; code/frontier.py; code/validate_frontier.py inputs/reason_cells.parquet; inputs/cohort.csv; inputs/baseline_full_schedule.csv Figure A1 code/documentation_cost_model.py No empirical input: stated model parameters Figure A2 historical/Topic2_research/code/screening_extension.py No empirical input: stated model parameters Figure A3 historical/Topic2_research/code/plot_annual.py historical/Topic2_research/focal_national_funnel.csv Figure A4, Table A17 code/check_principal_exhibits.py; code/baseline_estimator.py inputs/joint_2024.parquet; inputs/joint_2025.parquet; inputs/cohort.csv; inputs/documentation_resolution_sensitivity.csv Table A22 code/validate_frontier.py inputs/denial_pattern_counts.parquet; inputs/cohort.csv Figure reconstruction: the listed scripts reconstruct numerical content. Figures 3 and A4 retain frozen manuscript PNGs; their archived scripts reproduce counts/ranges and endpoints/crossings rather than byte-identical image rendering. Other listed figure modules also create PNGs. Historical replays use isolated copies; the archived README explains their commands. The original data, analysis code and numerical results have not changed. Comparison populations: Better-specific annual ratios and standardized changes do not subtract peers. The institutional relative comparison subtracts the original 30-peer change. Each focal entity in the 33-entity exercise is compared with its other 32 entities. Five- and ten-percentage-point classification spreads are transparent sensitivity benchmarks, not empirically calibrated bounds.
Institution: Singapore University of Social Sciences
Language: English
Subfield: Management Information Systems
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7224670243