State-Dependent Pricing in FinTech Credit: Evidence from P2P Lending

OverviewThis deposit provides the materials required to reproduce the empirical workflow and figures of the study: “State-Dependent Pricing in FinTech Credit: Evidence from P2P Lending.” The study is part of the doctoral dissertation “Risk Management in Digital Finance: Assessment and Pricing in an Emerging Fintech Era” (Chapter 5) by Lennart John Baals. The repository contains (i) the LaTeX source of the manuscript / chapter, (ii) the Python implementation of the macro-factor extraction and regime detection pipeline (robust rolling PCA + Gaussian HMM), (iii) explainability utilities (surrogate/manual decision trees), and (iv) the resulting figures used in the accompanying PDF. Contents of this deposit (file-level summary) Manuscript / thesis chapter source: main_WP5_PhD_Lennart_Baals.tex, references.bib Reproducible analysis notebook: Auto_R_I_XAI_R2_RD_Tree_v010.ipynb Core Python modules (pipeline): Robust PCA utilities: r2pca_utils.py, step0_utils.py Regime detection / HMM: r2_gaussian_hmm.py, r2rd_utils.py Explainability / trees: xaiR2_utils.py, manual_tree_utils.py Minimal helpers: utils.py Paper / chapter PDF: WP5_SNF_P2P_Research_Lennart_Baals.pdf Figures (as used in the manuscript):avg_spread_by_grade_term.png, indicator_boxplots.png, smoothed_state_probs.png, bic_selection.png, cv_selection.png, corr_regime1.png, corr_regime2.png, regime_coef_by_grade.png, regime_coef_by_credit_term.png, regime_labels.png, fig_pc_cumvar.png, fig_rolling_scores.png, manual_tree_depth3.png, surrogate_tree_depth2.png, surrogate_tree_depth3.png Data sources and access conditionsThe workflow combines: Macroeconomic time series (typically retrievable from public sources such as FRED or equivalent macro databases, depending on the exact configuration used in the study), and Loan-level P2P lending data (e.g., LendingClub-originated records, subject to the data provider’s terms and redistribution constraints. A public data set can be found on Kaggle (https://www.kaggle.com/code/danofer/lendingclub-data-defaults/input)). To support responsible reuse, this deposit focuses on code, documentation, and figure outputs. If any raw loan-level dataset cannot be redistributed, users should obtain it directly from the original provider under their own access rights and then map it to the variable names expected by the notebook/scripts (the notebook documents the required columns and preprocessing steps). Any included data descriptions are intended to make this mapping transparent and reproducible. Reproducibility (how to run)A typical reproduction path is: Create a Python environment (tested in a standard scientific Python stack). Open and run Auto_R_I_XAI_R2_RD_Tree_v010.ipynb, which orchestrates preprocessing, factor extraction, regime inference, and plot generation. Alternatively, import the modular pipeline components (robust PCA, HMM, explainability utilities) from the included .py files. To maximize portability, the code is organized as small, readable modules corresponding to each methodological step (factor extraction → regime detection → explainability → plotting). Intended useThis deposit is intended for: replication of the reported figures and regime classification logic, inspection and reuse of the robust PCA + HMM regime detection pipeline, and extension to other loan datasets or macro indicator panels under comparable conditions.