{"paper_id":"2601.00478","title":"Multimodal Insights into Credit Risk Modelling: Integrating Climate and Text Data for Default Prediction","authors":["Zongxiao Wu","Ran Liu","Jiang Dai","Dan Luo"],"paper_date":"2026-01-01","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2601.00478v1","page_url":"https://thequant.space/flowcharts/multimodal-insights-into-credit-risk-modelling-integrating/","scores":{"empirical_rigor":7.5,"math_complexity":4.5,"quadrant":"Street Traders"},"proposes":{"status":"available","statement":"The paper proposes a deep learning‑based multimodal framework that uses modality‑specific encoders and intermediate representation‑level fusion to combine structured credit variables, climate panel data, and textual loan assessments for credit default classification.","quote":"we propose a multimodal learning framework that employs modality-specific encoders and adopts a representation-level intermediate fusion mechanism to combine heterogeneous representations for credit default classification (Tavakoli et al., 2025).","summary":"Credit risk assessment increasingly relies on diverse sources of information beyond traditional structured financial data, particularly for micro and small enterprises (mSEs) with limited financial histories. This study proposes a multimodal framework that integrates structured credit variables…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"Default is defined as 30 days past payment due.","type":"data","stated":"explicit","quote":"Default is defined as 30 days past payment due."},{"assumption":"The dataset is severely imbalanced with a non‑defaulter to defaulter ratio of 64:1.","type":"data","stated":"explicit","quote":"We note that the dataset is characterised by a severely imbalanced ratio of 64:1 (non-defaulters: defaulters)."},{"assumption":"Missing continuous features are filled with the mean and missing categorical features are replaced with new categories.","type":"data","stated":"explicit","quote":"In cases where continuous features are missing, they are filled with the means of the features, while missing values for categorical features are replaced with new categories."},{"assumption":"Weight of evidence (WoE) encoding assumes a log‑odds relationship between feature bins and default risk.","type":"model","stated":"explicit","quote":"The WoE value in each category/bin for a given feature is calculated as the logarithm of the proportion of non-defaulters to the proportion of defaulters in that particular"},{"assumption":"Climate risk factors are relevant to borrowers’ repayment capacity in the agricultural setting.","type":"other","stated":"implied","quote":"Given that our loan data is based on the agricultural setting, we can likely attribute this superior performance to the high relevance of climate risk factors such as droughts, floods, and temperature extremes to borrowers’ repayment capabilities."},{"assumption":"Textual loan officer assessments contain valuable qualitative information about borrower behavior.","type":"other","stated":"implied","quote":"These results suggest that loan officers’ written assessments may contain valuable insights that reflect borrowers’ repayment intentions and operational capabilities."}]},"datasets":{"status":"available","items":[{"name":"Agricultural loan dataset","provider":"Chinese bank (national basis)","kind":"alternative","access":"proprietary","period":null,"universe":"agricultural micro and small enterprises (mSEs)","quote":"The agricultural loan dataset used in this study was provided by a Chinese bank operating on a national basis"}],"known":[],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"unclear"},"transaction_costs":{"status":"not_applicable","modeled":"n/a","headline_net_of_costs":null,"detail":null,"quote":null,"source":"front_matter","rule":"R3","candidates":{"front_matter":"n/a","evidence_card":"n/a","evidence_card_quoted":false},"meaning":"the paper reports no trading or execution result"},"results":{"status":"available","evaluation_type":"forecast-accuracy","findings":[{"finding":"Models using only climate data outperform structured‑only models, with the GRU achieving an average AUC of 0.658, KS of 0.332 and H‑measure of 0.189.","kind":"performance","quote":"Interestingly, while transformer models generally exhibit superior performance in unimodal analyses, they are slightly outperformed by the GRU model, which achieves average AUC, KS, and H-measure values of 0.658, 0.332, and 0.189, respectively."},{"finding":"Text‑only transformer models also beat structured‑only models, attaining an average AUC of 0.661, KS of 0.371 and H‑measure of 0.224.","kind":"performance","quote":"Among the three models, the transformer model achieves the best performance, with average AUC, KS, and H-measure values of 0.661, 0.371, and 0.224, respectively."},{"finding":"Structured‑only models obtain lower predictive performance (average AUC 0.609, KS 0.301, H‑measure 0.157).","kind":"performance","quote":"The average AUC, KS, and H-measure for the model relying solely on structured data (S) are 0.609, 0.301, and 0.157, respectively."},{"finding":"Integrating climate and text modalities yields significant improvements over any single modality, with transformer‑based multimodal models achieving the highest prediction accuracy.","kind":"predictive-accuracy","quote":"Integrating this information into prediction models may help to capture subtle borrower characteristics that structured data may ignore."},{"finding":"SHAP analysis identifies water‑logging by rain as the most influential climate factor for default prediction.","kind":"statistical","quote":"Using SHAP-based explainability methods, we find that physical climate risks play an important role in default prediction, with water-logging by rain emerging as the most influential factor."}],"reported_performance":null,"out_of_sample":{"method":"holdout","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":42,"citations":{"cited_by_in_corpus":0,"by_independent_authors":0,"how_cited":{"used_as_baseline":0,"uses_method_or_data":0,"extends":0,"discusses_findings":0,"mentions":0,"unclassified":0},"independent_papers":[],"scope":"Citations between papers in this corpus only; citations from outside it are not counted."}},"implementation":{"status":"available","public":false,"code_url":null,"note":"No repository presented by the authors as their own was found in the paper. Absence here is not proof that no code exists."},"replication":{"status":"none_found","card":null,"note":"No replication of this paper has been run by thequant.space yet."},"extraction":{"text_source":"fulltext","model":"openai/gpt-oss-120b","version":1,"extracted_at":"2026-11-14T17:17:25Z"},"provenance":{"claims":"Everything here is what the authors state in the paper. Nothing was re-run or reproduced by the platform.","quotes":"Assumptions, datasets, cost detail and findings were extracted by a language model. Every quote is verbatim paper text. Where an item has `support`, \"full\" means every figure in it appears in the quoted sentences and \"partial\" means a figure is stated elsewhere in the paper; items with a figure found nowhere in the paper were discarded.","datasets_known":"datasets.known lists well-known vendors and public datasets by a stable id, found by a dictionary over the full text where the sentence says the data is used; it is independent of the model and covers only names in the dictionary.","flags":"transaction_costs.modeled, results.evaluation_type and results.out_of_sample.method are platform classifications (structured-decision models); see https://thequant.space/score-guide/#penalty-flags","comparisons":"supports / contradicts / qualifies are a language model's reading of two papers' quoted findings, restricted to closely related papers that share no author. They are pointers to read, not verified replications, and coverage is partial.","implementation":"code_url is a repository the paper itself presents as its own (rule-scored, ambiguous cases judged by a model), checked for liveness when it was recorded."},"docs":"https://thequant.space/agents/"}