Communicating Credit Risk with Large Language Models: Evaluation of Explanations from Standard and Alternative Data-Based Models

Sahab Zandi; Noah Kostesku; Christophe Mues; María Óskarsdóttir; Cristián Bravo

q-fin.RMAug 18, 2026

Communicating Credit Risk with Large Language Models: Evaluation of Explanations from Standard and Alternative Data-Based Models

Authors: Sahab Zandi, Noah Kostesku, Christophe Mues, María Óskarsdóttir, Cristián Bravo

Organizations: Department of Statistical and Actuarial Sciences, Western University, London, Canada · Department of Computer Science, Western University, London, Canada · University of Southampton Business School, University of Southampton, Southampton, United Kingdom · Centre for Operational Research, Management Sciences and Information Systems, University of Southampton, Southampton, United Kingdom · School of Mathematical Sciences, University of Southampton, Southampton, United Kingdom · Department of Computer Science, Reykjavik University, Reykjavik, Iceland

Abstract

Credit decisioning is a high-stakes task in which model outputs must be accurate and explainable to support compliant decisions. Although modern credit risk models such as eXtreme Gradient Boosting (XGBoost) and Graph Neural Networks (GNNs) improve predictive performance, their explanations are often too technical for stakeholders creating communication gaps that can shape approvals, denials, and fairness judgments. We examine whether Large Language Models (LLMs) can serve as explanation layers that translate post-hoc explanation artefacts into stakeholder-appropriate risk narratives. Using Freddie Mac single-family loan-level data, we develop three pipelines: standard tabular (XGBoost + SHAP), and two with alternative data, a pure network-based (GNN + GNNExplainer), and a bimodal one (combining tabular and network data). We generate narratives with three LLM configurations: a small fine-tuned LLM (Gemma 3 4B), a large fine-tuned LLM (DeepSeek R1 70B), and a zero-shot commercial LLM (Gemini 2.5). Explanation quality is evaluated through automated checks across all pipelines and a human study of bimodal explanations comparing credit risk professionals and non-professionals on eight decision-relevant dimensions. We have three main findings. First, the pipeline accounts for higher variance in evidence-grounding scores than the language model, meaning that the binding constraint on explanation quality is the evidence representation, not the model used. Second, the explanation narratives reliably name the influential factors but are less reliable when stating the direction of influence, which may be consequential for adverse-action communication. Finally, professionals apply stricter evidentiary standards than non-professionals. We discuss implications for the governance of risk models, including deployment considerations and the value of domain-aligned LLMs in regulated credit settings.

Figures & tables

LLM	Size	Fine-tuned	Deployment Mode	Source
Gemma 3	Small (4B)	Yes	Open-source	Google DeepMind
DeepSeek R1	Large (70B)	Yes	Open-source	DeepSeek AI
Gemini 2.5	Undisclosed	No	API-based	Google

Table 1: LLMs used for narrative generation.

Panel of Figure 1 — Figure 1: The architecture of the three explanation pipelines.

Pipeline	Prompt Components
Tabular- based	Predicted outcome; top- $k$ SHAP feature attributions with descriptions; instructional step-by-step reasoning; guided subgoal structure; borrower-level counterfactual prompt; output format constraint.
Network- based	Predicted outcome; network connectivity analysis; instructional step-by-step reasoning for structural and relational reasoning; guided subgoal structure; network-level counterfactual prompt; output format constraint.
Bimodal	Predicted outcome; top- $k$ SHAP feature descriptions; GNNExplainer subgraph descriptions; combined instructional reasoning linking borrower and relational characteristics; combined borrower-level and network-level counterfactual prompt; disambiguation hint for conflicting contributions; output format constraint.

Table 2: Prompt design components for each explanation pipeline.

Figure 2: Workflow for fine-tuning Gemma 3 and DeepSeek R1 using LoRA-based instruction tuning on 8-bit quantised base models. SHAP and GNNExplainer outputs are paired with ChatGPT-4o-generated explanations to create the synthetic supervision dataset.

Metric	Acronym	CRP Wording	NCRP Wording
Understandability	UND	The explanation is easy to understand.	The explanation is easy to understand.
Trustworthiness	TRU	The explanation can be trusted because it presents sufficient and reliable evidence to support the loan decision.	I trust this explanation given by the system.
Insightfulness	INS	This explanation reveals insightful risk factors that influence loan decisions.	This explanation provides useful insight into why the decision was made.
Satisfaction	SAT	The level of detail satisfies my expectations for a risk assessment.	I am satisfied with how the explanation addressed my concerns about the decision.
Confidence	CON	This explanation increases my confidence for loan approval/denial.	I am confident in the explanation provided for the system’s decision.
Convincingness	CVN	The justification is convincing and argues for the decision using well-weighted risk evidence.	The explanation is convincing and makes the decision seem reasonable.

Table 3: Perceived explanation-quality criteria across CRPs and NCRPs.

Pipeline	LLM	TFC	TDC	NFC	NDC
Tabular- based	Gemma 3	3.77 $\pm$ 0.12	3.68 $\pm$ 0.15	1.73 $\pm$ 0.09	2.14 $\pm$ 0.17
	DeepSeek R1	4.35 $\pm$ 0.10	3.48 $\pm$ 0.13	1.58 $\pm$ 0.10	1.91 $\pm$ 0.17
	Gemini 2.5	3.91 $\pm$ 0.11	3.94 $\pm$ 0.15	1.90 $\pm$ 0.09	2.42 $\pm$ 0.15
Network- based	Gemma 3	1.02 $\pm$ 0.03	1.10 $\pm$ 0.07	4.83 $\pm$ 0.07	3.20 $\pm$ 0.09
	DeepSeek R1	1.00 $\pm$ 0.00	1.15 $\pm$ 0.10	4.46 $\pm$ 0.11	3.22 $\pm$ 0.09
	Gemini 2.5	1.00 $\pm$ 0.00	1.01 $\pm$ 0.02	4.88 $\pm$ 0.07	2.89 $\pm$ 0.10

Table 4: LLM-as-a-judge scores across pipeline–LLM configurations. Boldface indicates the highest mean per column within each pipeline.

Metric	$F_{\text{pipe}}$	$\eta^{2}_{\text{pipe}}$	$F_{\text{llm}}$	$\eta^{2}_{\text{llm}}$	$F_{\text{int}}$	$\eta^{2}_{\text{int}}$
TFC	328.96	0.391	11.81	0.014	10.04	0.024
TDC	299.02	0.352	26.08	0.031	22.07	0.052
NFC	354.27	0.377	62.95	0.067	21.66	0.046
NDC	39.19	0.069	28.94	0.051	11.63	0.041

Table 5: Two-way ANOVA results for LLM-as-a-judge fidelity metrics. All effects are statistically significant (

p<10^{-6}

Metric	Gemma 3	DeepSeek R1	Gemini 2.5
UND	4.65 $\pm$ 0.14	4.24 $\pm$ 0.23	4.41 $\pm$ 0.19
TRU	4.23 $\pm$ 0.21	3.93 $\pm$ 0.25	3.98 $\pm$ 0.22
INS	4.09 $\pm$ 0.25	3.78 $\pm$ 0.25	3.88 $\pm$ 0.23
SAT	3.97 $\pm$ 0.24	3.75 $\pm$ 0.29	3.57 $\pm$ 0.27
CON	4.10 $\pm$ 0.22	3.70 $\pm$ 0.27	3.87 $\pm$ 0.24
CVN	4.08 $\pm$ 0.22	3.64 $\pm$ 0.29	3.74 $\pm$ 0.25

Table 6: Human (CRP) mean Likert ratings across explanation-quality metrics. Boldface indicates the highest mean per column.

Metric	Gemma 3	DeepSeek R1	Gemini 2.5
UND	4.36 $\pm$ 0.11	4.22 $\pm$ 0.15	4.18 $\pm$ 0.17
TRU	4.24 $\pm$ 0.11	4.19 $\pm$ 0.14	4.01 $\pm$ 0.16
INS	4.11 $\pm$ 0.13	4.03 $\pm$ 0.15	4.06 $\pm$ 0.16
SAT	4.08 $\pm$ 0.14	4.12 $\pm$ 0.14	4.01 $\pm$ 0.17
CON	4.24 $\pm$ 0.12	4.17 $\pm$ 0.15	4.11 $\pm$ 0.16
CVN	4.14 $\pm$ 0.13	4.01 $\pm$ 0.15	4.04 $\pm$ 0.16

Table 7: Human (NCRP) mean Likert ratings across explanation-quality metrics. Boldface indicates the highest mean per column.

Panel of Figure 3 — Figure 3: Between-cohort comparisons by LLM.

	CRPs			NCRPs
Metric	Gemma 3	DeepSeek R1	Gemini 2.5	Gemma 3	DeepSeek R1	Gemini 2.5
UND	4.05 (0.87)	3.90 (0.88)	3.94 (0.88)	4.10 (0.76)	3.78 (1.04)	4.07 (0.85)
TRU	4.17 (0.82)	3.92 (1.00)	3.68 (1.05)	4.30 (0.60)	4.01 (0.73)	4.07 (0.78)
INS	4.31 (0.82)	3.77 (1.08)	3.68 (0.99)	4.39 (0.55)	4.14 (0.67)	4.07 (0.88)
SAT	4.40 (0.67)	3.81 (1.03)	3.65 (0.93)	4.38 (0.62)	4.16 (0.71)	4.09 (0.79)
CON	4.35 (0.71)	3.62 (1.16)	3.68 (0.92)	4.36 (0.58)	4.11 (0.72)	4.15 (0.83)

Table 8: Participant-level mean Likert ratings by cohort, LLM, and metric, with standard deviations in parentheses

	CRPs				NCRPs
Metric	Mean diff.	SD diff.	$p_{\mathrm{Holm}}$	$p_{\mathrm{BH}}$	Mean diff.	SD diff.	$p_{\mathrm{Holm}}$	$p_{\mathrm{BH}}$
UND	$-$ 0.050	0.468	1.000	0.681	$-$ 0.121	0.583	1.000	0.733
TRU	$+$ 0.350	0.489	0.250	0.250	$+$ 0.075	0.657	1.000	0.765
INS	$+$ 0.150	0.648	1.000	0.732	$-$ 0.045	0.510	1.000	0.765
SAT	$+$ 0.250	0.635	1.000	0.517	$-$ 0.288	0.455	0.722	0.413
CON	$+$ 0.250	0.687	1.000	0.517	$-$ 0.150	0.555	1.000	0.733

Table 9: Paired differences between DeepSeek R1 and Gemini 2.5 by metric and cohort, with Holm-adjusted and Benjamini–Hochberg FDR-adjusted

p

-values. Positive values indicate higher ratings for DeepSeek R1.

Component	Variance	Share (%)
Participants (leniency/strictness)	0.507	46.3
Explanation cases (LLM $\times$ instance)	0.110	10.0
Residual	0.478	43.7

Table 10: Variance decomposition of overall explanation ratings from a random-intercepts model with participant and explanation-instance effects.

Category	Feature	Description
Sentiment	sentiment_polarity	TextBlob polarity score $\in[-1,1]$ (negative to positive)
	sentiment_subjectivity	TextBlob subjectivity $\in[0,1]$ (objective to subjective)
	vader_compound	VADER composite sentiment score $\in[-1,1]$
	vader_pos	VADER positive sentiment proportion
	vader_neg	VADER negative sentiment proportion
	vader_neu	VADER neutral sentiment proportion

Table 11: Linguistic feature categories used in the textual analysis.

Figure 4: Cohen’s $d$ effect sizes comparing linguistic features in high-rated (top quartile) and low-rated (bottom quartile) explanations across quality metrics. Positive values (red) indicate features more prevalent in high-rated explanations; negative values (blue) indicate the reverse.

Appendix figures & tables11 assets

Supplementary material from the paper’s appendix.

Appendix

Feature	Description
fico	FICO credit score at origination
dti	Original debt-to-income ratio
ltv	Original loan-to-value ratio
mi_pct	Mortgage insurance percentage
orig_upb	Original unpaid principal balance
loan_term	Original term of the loan in months

Appendix

Table A.1 : Features used for model training and analysis.

Component	Concrete instantiation
1) Predicted outcome	Prediction: Loan 31780 default probability $=0.8712$ . Actual outcome: default.
2) Top- $k$ SHAP feature attributions with descriptions	Evidence (top- $k$ SHAP; top-3 shown): • cnt_borr (SHAP:+0.4362): pushed toward default • fico (SHAP:–0.2609): pushed toward no default • dti (SHAP:+0.0799): pushed toward default (Full top- $k$ list is passed to the LLM; only top-3 shown here.) Feature descriptions: cnt_borr = number of borrowers; fico = credit score; dti = debt-to-income ratio.
3–4) Instructional reasoning + guided subgoals	Instructions (follow in order): 1. Explain the predicted outcome in plain language. 2. Use the SHAP ordering; explain each listed factor: what it is, whether it increases/decreases risk, and why it matters to lenders. 3. Combine factors into a short overall justification. Constraints: borrower-facing language; do not mention SHAP/models/algorithms; reference only evidence-block factors; neutral tone; short sentences.
5) Borrower-level counterfactual prompt	Counterfactual guidance: Provide two realistic improvement steps tied to the risk-increasing factors (positive SHAP values). Use cautious phrasing (e.g., “If you were able to…, this would likely improve…”).
6) Output format constraint	Format: No bullet points in the final answer. End with one sentence summarising the main drivers.

Appendix

Table C.1 : Tabular-based prompt example instantiating Table 3 of the main text.

Component	Concrete instantiation
1) Predicted outcome	Prediction: Loan 2885 no-default probability $=0.9489$ . Actual outcome: no default.
2) Network connectivity analysis	Evidence (GNNExplainer summary; top-3 shown): • Top node features (importance scores): – fico (0.6967) – if_corr (0.6736) – ltv (0.3349) • Top edge connections: – Geographic area connection (Loan 2733 $\rightarrow$ 2885, cluster size $\sim$ 75): impact 0.2380 – Area-provider cluster connection (Loan 2622 $\rightarrow$ 2885, cluster size $\sim$ 9): impact 0.2314 – Same loan provider connection (Loan 2493 $\rightarrow$ 2885, cluster size $\sim$ 145): impact 0.2266 (Full lists are passed to the LLM; only top-3 shown here.)
3–4) Instructional reasoning + guided subgoals	Instructions (follow in order): 1. Explain the predicted outcome in plain language. 2. Describe the strongest connections using the edge list and cluster sizes. 3. Explain (at a high level) why geographic/provider groupings may matter, without adding unverifiable claims. Constraints: focus on relational/context evidence; mention at least two connection types; do not introduce attributes not shown in the evidence block; neutral tone.
5) Network-level counterfactual prompt	Counterfactual guidance: Provide one cautious “what-if” statement on how different relational patterns (e.g., fewer links to a given cluster) could change confidence; avoid implying control over other borrowers.
6) Output format constraint	Format: Include at least one sentence referencing a connection type and a cluster size from the evidence list. End with a one-sentence summary.

Appendix

Table C.2 : Network-based prompt example instantiating Table 3 of the main text.

Component	Concrete instantiation
1) Predicted outcome	Prediction: Loan 1007 default probability $=0.9132$ . Actual outcome: default.
2) Top- $k$ SHAP feature descriptions	Evidence (tabular SHAP; top-3 shown): • cnt_borr (SHAP= $+0.26$ ): increases risk • dti (SHAP= $+0.21$ ): increases risk • fico (SHAP= $-0.18$ ): reduces risk (Full top- $k$ list is passed to the LLM; only top-3 shown here.)
3) GNNExplainer subgraph descriptions	Evidence (network summary; compact subset shown): • Edge: Same provider connection (Loan 5021 $\rightarrow$ 1007, cluster size $\sim$ 12) • Edge: Area-provider cluster (Loan 2893 $\rightarrow$ 1007, cluster size $\sim$ 8) • Node: Loan 5021 (defaulted) identified as influential neighbour (Full subgraph/edge lists are passed to the LLM; only a compact subset shown here.)
4) Combined instructional reasoning linking both modalities	Instructions (follow in order): 1. Explain the predicted outcome overall. 2. Explain the strongest tabular drivers first (SHAP ordering), then the strongest network drivers (edge types + cluster sizes). 3. Include one explicit linking sentence describing how borrower-level and relational signals combine.
5) Combined borrower- and network-level counterfactual prompt	Counterfactual guidance: Provide two improvement suggestions: (i) one tied to a positive-SHAP tabular factor; (ii) one stated cautiously about relational context (how different clustering patterns could affect confidence).
6) Disambiguation hint for conflicting contributions	Disambiguation rule: If tabular and network evidence point in different directions, acknowledge both and state that the assessment balances the signals, without adding new evidence.

Appendix

Table C.3 : Bimodal prompt example instantiating Table 3 of the main text.

Pipeline	LLM	Perplexity
	Gemma 3	19.38 $\pm$ 0.67
Tabular-based	DeepSeek R1	15.93 $\pm$ 0.36
	Gemini 2.5	33.20 $\pm$ 1.21
	Gemma 3	21.09 $\pm$ 0.59
Network-based	DeepSeek R1	20.06 $\pm$ 0.54
	Gemini 2.5	28.96 $\pm$ 0.72

Appendix

Table D.1: Perplexity scores across pipeline–LLM configurations. Boldface indicates the lowest mean perplexity within each pipeline.

Effect	$F$	$p$	$\eta^{2}$
Pipeline	18.34	$1.6\times 10^{-8}$	0.010
LLM	1302.39	$<10^{-262}$	0.687
Pipeline $\times$ LLM	68.85	$1.1\times 10^{-50}$	0.073

Appendix

Table D.2: Two-way ANOVA results for perplexity with factors Pipeline and LLM.

Pipeline	LLM	UND	TRU	INS	SAT	CON	CVN	COM	USB
	Gemma 3	3.89 $\pm$ 0.07	2.11 $\pm$ 0.06	2.15 $\pm$ 0.07	1.99 $\pm$ 0.02	2.62 $\pm$ 0.10	2.11 $\pm$ 0.06	2.70 $\pm$ 0.09	1.42 $\pm$ 0.10
Tabular-based	DeepSeek R1	3.79 $\pm$ 0.08	2.38 $\pm$ 0.10	2.34 $\pm$ 0.09	2.14 $\pm$ 0.07	2.75 $\pm$ 0.09	2.38 $\pm$ 0.10	2.95 $\pm$ 0.07	1.79 $\pm$ 0.08
	Gemini 2.5	3.99 $\pm$ 0.08	2.80 $\pm$ 0.08	2.78 $\pm$ 0.09	2.58 $\pm$ 0.10	2.96 $\pm$ 0.08	2.79 $\pm$ 0.08	3.29 $\pm$ 0.12	1.86 $\pm$ 0.07
	Gemma 3	3.62 $\pm$ 0.10	1.97 $\pm$ 0.03	1.98 $\pm$ 0.03	1.96 $\pm$ 0.04	2.15 $\pm$ 0.08	1.97 $\pm$ 0.03	2.78 $\pm$ 0.08	1.12 $\pm$ 0.06
Network-based	DeepSeek R1	2.98 $\pm$ 0.03	1.77 $\pm$ 0.08	1.81 $\pm$ 0.08	1.57 $\pm$ 0.10	1.91 $\pm$ 0.06	1.77 $\pm$ 0.08	2.02 $\pm$ 0.04	1.00 $\pm$ 0.00
	Gemini 2.5	2.80 $\pm$ 0.08	1.40 $\pm$ 0.10	1.55 $\pm$ 0.10	1.29 $\pm$ 0.09	1.78 $\pm$ 0.08	1.41 $\pm$ 0.10	1.90 $\pm$ 0.07	1.00 $\pm$ 0.00

Appendix

Table D.3: Simulated CRP ratings across pipeline–LLM configurations. Boldface indicates the highest mean per column.

Pipeline	LLM	UND	TRU	INS	SAT	CON	CVN	COM	USB
	Gemma 3	4.43 $\pm$ 0.11	3.44 $\pm$ 0.10	3.89 $\pm$ 0.06	3.45 $\pm$ 0.10	3.43 $\pm$ 0.10	4.13 $\pm$ 0.10	4.43 $\pm$ 0.11	3.38 $\pm$ 0.10
Tabular-based	DeepSeek R1	4.00 $\pm$ 0.08	3.20 $\pm$ 0.10	3.85 $\pm$ 0.08	3.36 $\pm$ 0.10	3.19 $\pm$ 0.10	3.91 $\pm$ 0.08	4.00 $\pm$ 0.08	3.10 $\pm$ 0.09
	Gemini 2.5	4.39 $\pm$ 0.10	3.62 $\pm$ 0.10	4.19 $\pm$ 0.08	3.81 $\pm$ 0.08	3.62 $\pm$ 0.10	4.29 $\pm$ 0.11	4.39 $\pm$ 0.10	3.54 $\pm$ 0.10
	Gemma 3	3.97 $\pm$ 0.07	2.97 $\pm$ 0.07	3.68 $\pm$ 0.11	3.02 $\pm$ 0.06	2.97 $\pm$ 0.07	3.86 $\pm$ 0.07	3.97 $\pm$ 0.07	2.97 $\pm$ 0.07
Network-based	DeepSeek R1	3.63 $\pm$ 0.10	2.64 $\pm$ 0.10	2.91 $\pm$ 0.06	2.65 $\pm$ 0.10	2.64 $\pm$ 0.10	3.02 $\pm$ 0.06	3.63 $\pm$ 0.10	2.41 $\pm$ 0.10
	Gemini 2.5	3.09 $\pm$ 0.08	2.04 $\pm$ 0.06	2.45 $\pm$ 0.10	2.04 $\pm$ 0.06	2.04 $\pm$ 0.06	2.93 $\pm$ 0.06	3.09 $\pm$ 0.08	2.02 $\pm$ 0.06

Appendix

Table D.4: Simulated NCRP ratings across pipeline–LLM configurations. Boldface indicates the highest mean per column.

Metric	$F_{\text{pipe}}$	$p_{\text{pipe}}$	$\eta^{2}_{\text{pipe}}$	$F_{\text{llm}}$	$p_{\text{llm}}$	$\eta^{2}_{\text{llm}}$	$F_{\text{int}}$	$p_{\text{int}}$	$\eta^{2}_{\text{int}}$
UND	471.115	$8.97\times 10^{-140}$	0.517	51.290	$8.55\times 10^{-22}$	0.105	54.234	$6.95\times 10^{-41}$	0.198
TRU	288.654	$4.30\times 10^{-97}$	0.396	0.575	0.5628	0.001	52.728	$7.66\times 10^{-40}$	0.194
INS	260.980	$1.07\times 10^{-89}$	0.373	3.884	0.0209	0.009	38.091	$2.03\times 10^{-29}$	0.148
SAT	243.487	$7.18\times 10^{-85}$	0.357	0.930	0.3951	0.002	78.371	$7.06\times 10^{-57}$	0.263
CON	445.581	$2.41\times 10^{-134}$	0.503	0.512	0.5996	0.001	22.268	$1.56\times 10^{-17}$	0.092
CVN	279.329	$1.24\times 10^{-94}$	0.389	0.574	0.5632	0.001	50.728	$1.89\times 10^{-38}$	0.188

Appendix

Table D.5: Two-way ANOVA results for simulated CRP ratings.

Metric	$F_{\text{pipe}}$	$p_{\text{pipe}}$	$\eta^{2}_{\text{pipe}}$	$F_{\text{llm}}$	$p_{\text{llm}}$	$\eta^{2}_{\text{llm}}$	$F_{\text{int}}$	$p_{\text{int}}$	$\eta^{2}_{\text{int}}$
UND	229.281	$7.38\times 10^{-81}$	0.343	27.297	$3.15\times 10^{-12}$	0.058	53.932	$1.12\times 10^{-40}$	0.197
TRU	302.394	$1.16\times 10^{-100}$	0.408	17.788	$2.67\times 10^{-8}$	0.039	57.377	$4.83\times 10^{-43}$	0.207
INS	685.194	$4.47\times 10^{-180}$	0.609	54.104	$6.93\times 10^{-23}$	0.110	123.595	$1.06\times 10^{-83}$	0.260
SAT	503.172	$2.23\times 10^{-146}$	0.534	16.600	$8.39\times 10^{-8}$	0.036	72.315	$5.50\times 10^{-53}$	0.248
CON	297.011	$2.85\times 10^{-99}$	0.403	16.985	$5.79\times 10^{-8}$	0.037	58.700	$6.06\times 10^{-44}$	0.211
CVN	461.246	$1.08\times 10^{-137}$	0.512	68.807	$1.73\times 10^{-28}$	0.135	75.891	$2.71\times 10^{-55}$	0.257

Appendix

Table D.6: Two-way ANOVA results for simulated NCRP ratings.

Panel of Figure 26 — Figure F.1 : Demographic and background characteristics of survey participants.

Explore similar work

Could Large Language Models work as Post-hoc Explainability Tools in Credit Risk Models?
Feb 21, 2026Wenxi Geng, Dingyuan Liu, Liya Li +1Explainable Artificial Intelligence Explainability Evaluation
Citigroup, Irving, TX, United States · University of North Carolina at Chapel Hill, Chapel Hill, NC, United States · Georgia Institute of Technology, Atlanta, GA, United States +1
Accurate Ensembles, Fragile Narratives: Multi-Scale Stacking and a Fidelity Audit of LLM-Generated Explanations for Credit Risk
Aug 8, 2026Gregorius Reynaldi Pratama, Kuo-Kun TsengFeature Attribution Faithfulness of Language Model Explanations
School of Computer Science and Technology Harbin Institute of Technology, Shenzhen Shenzhen, China
Explanation Fairness in Large Language Models: An Empirical Analysis of Disparities in How LLMs Justify Decisions Across Demographic Groups
May 9, 2026Gautam VeldandaLLM Auditing Algorithmic Fairness
Independent Researcher

Feb 21, 2026J/K move · Enter open · S save

Could Large Language Models work as Post-hoc Explainability Tools in Credit Risk Models?

Wenxi Geng, Dingyuan Liu, Liya Li, Yiqing Wang

Citigroup, Irving, TX, United States · University of North Carolina at Chapel Hill, Chapel Hill, NC, United States · Georgia Institute of Technology, Atlanta, GA, United States+1

Explainable Artificial Intelligence Explainability Evaluation Faithfulness of Language Model Explanations LLM Interpretability

Open paper AskPDF arXiv

Large language models (LLMs) have shown promise in translating model-based explanations into human-readable narratives. This study evaluates whether LLMs can serve as post-hoc explainability interfaces for credit risk models, focusing on their ability to preserve feature-importance rankings and generate autonomous explanations. Using a LendingClub dataset, we compare LLM outputs with SHAP and coefficient-based attributions on three major LLMs, including GPT-4-turbo, Claude-Sonnet-4.5, and Gemini-2.5-Flash. Results indicate that LLMs reliably reproduce reference rankings under controlled prompts but show limited alignment when generating explanations autonomously. These findings suggest that LLMs are best deployed as narrative interfaces rather than substitutes for formal attribution methods in credit risk governance.

Abstract

Explore similar work

Could Large Language Models work as Post-hoc Explainability Tools in Credit Risk Models?

Accurate Ensembles, Fragile Narratives: Multi-Scale Stacking and a Fidelity Audit of LLM-Generated Explanations for Credit Risk

Explanation Fairness in Large Language Models: An Empirical Analysis of Disparities in How LLMs Justify Decisions Across Demographic Groups