Research
On-device research index

arXiv research

A locally-built, LLM-digested index of recent arXiv papers in quant finance, geometry/topology, and statistical ML — keyword search served straight from SQLite on this machine.

169,181 papers · 148 categories

Trend · papers per month

14284256 · May 202619922001200920182026
48 results for narrative reports

Developing an AI economist agent using RAG, knowledge graphs, and LLMs for economic scenario analysis.

problem Economic scenario analysis using large language models and knowledge graphs.
method Proposing an RAG-based AI economist framework that utilizes knowledge graphs and LLMs.
result Improves economic coherence and traceability in generated reports.

Narrative disclosures in 10-K filings improve bankruptcy prediction beyond accounting ratios.

problem Traditional bankruptcy prediction models rely on accounting ratios, which may not capture early warning signals.
method Developed a PB Stress Score based on distress-specific language in 10-K narratives, evaluated against accounting and dictionary benchmarks.
result Adding the PB Stress Score increases AUC from 0.8323 to 0.9019 and improves top-decile bankruptcy capture from 44.12% to 64.71%.

Predicting emotional state from narratives using contextual information.

problem Predicting valence from personal narratives using contextual information.
method Investigated multiple machine learning techniques to model narratives, focusing on textual information.
result Models capture inter-individual differences, leading to more accurate predictions of emotional state.

Platform combines RL and language models to study narrative influence on AI decisions.

problem Understanding how narrative elements shape AI decision-making.
method Dual-system architecture with reinforcement learning and language model integration.
result Initial experiments show narrative frameworks can influence AI decision-making.

Paper fine-tunes a language model to predict long-term stock buy signals.

problem Predicting long-term stock price movements with narrative text.
method Fine-tuning a small language model on 10-K reports for buy/sell decisions.
result Buy signals generated from 10-K text are most precise at 6 and 9 months, providing 4.8-9% improvement over random selection.

CrystalCandle creates user-friendly explanations for machine learning models.

problem Low trust in predictive models due to lack of interpretability.
method End-to-end pipeline for model interpretation, including Model Importer, Interpreter, Narrative Generator, and Exporter.
result CrystalCandle leads to higher adoption rates and improved downstream metrics.

Study finds whitepaper narratives do not predict market factor structure.

problem Predicting market behavior from cryptocurrency whitepaper claims.
method Zero-shot NLP classification combined with CP tensor decomposition of market data.
result Weak alignment between whitepaper claims and market statistics and latent factors.

Study uses LLM to extract and compare segment disclosures from financial filings.

problem Challenges in completeness and comparability of segment disclosures in financial reports.
method Developed a large language model framework to extract and preserve segment information from Form 10-K filings.
result The LLM accurately extracts segment-level information and addresses cross-period knowledge questions.

Study finds no significant alignment between whitepaper claims and market structure.

problem Correlation between cryptocurrency whitepaper narratives and market behavior.
method Developed a contamination-aware pipeline for measuring structural correspondence, combining NLP classification and market statistics.
result No significant claims-market alignment detected in the sample.

Paper uses Apprenticeship Learning to model player behavior in interactive narratives.

problem Understanding and simulating player behavior in interactive narratives.
method Receding Horizon IRL (RHIRL) to learn reward functions and policies.
result RHIRL can learn action sequences and generate behavior similar to specific players.

QRAFTI uses multi-agent framework to improve equity factor research.

problem Replicating and developing new equity factors in large financial datasets.
method Integrates a research toolkit with MCP servers for data access and custom coding operations.
result Improves performance and explainability in multi-step empirical tasks.

Study detects SLI in children from spontaneous narrative transcripts.

problem Detecting Specific Language Impairment (SLI) in children.
method Three-stage pipeline: feature extraction, dimensionality reduction, and classification.
result 97.13% accuracy in identifying SLI from transcripts.

LLMs can help explain credit risk models but not autonomously.

problem Leveraging LLMs for post-hoc explainability in credit risk models.
method Comparison of LLM outputs with SHAP and coefficient-based attributions on three LMs.
result LLMs reliably preserve feature-importance rankings but poorly align with autonomous explanations.

Study shows GPT's earnings forecasts are human-like but not always accurate.

problem Information friction in AI-generated financial analysis.
method Examined GPT's earnings forecasts following corporate earnings releases and proposed a diagnostic framework.
result GPT's narrative attention is consistent and human-like but not always associated with higher forecast accuracy.

New framework tackles deep financial reporting bottleneck by improving hallucination and coherence.

problem Statistical smoothing trap in LLMs limits deep financial reporting quality.
method DeepNews Framework integrates information foraging, schema-guided planning, and adversarial prompting.
result DeepNews system achieves 25% acceptance rate in blind test, significantly outperforming SOTA.

Analyzes news graphs to predict financial market dislocations.

problem Predicting financial market dislocations using news content.
method Extracts entities from news articles, aggregates them into graphs, applies network analysis, and uses sentiment analysis.
result Identifies high entropy in news graphs correlates with financial market dislocations.

Paper develops a method for valid inference using language model predictions from verbal autopsy narratives.

problem Valid inference from verbal autopsy narratives for public health decision-making.
method Develops multiPPI++ method for valid inference using NLP techniques for COD prediction.
result Demonstrates the effectiveness of multiPPI++ in handling transportability issues and recovering ground truth estimates.

Model shows how past consumption affects household confidence, leading to varied economic outcomes.

problem Exploring how past consumption impacts current confidence and economic activity in a multi-household model.
method Developed a DSGE model where past consumption influences individual household confidence and consumption propensity.
result The model demonstrates a range of economic outcomes including high output with no crises, high output with increased volatility, and alternation of high and low output states.

LDA identifies latent topics in CFPB consumer complaints over time.

problem Identify latent topics in CFPB consumer complaints for better regulation effectiveness.
method Latent Dirichlet Allocation (LDA) for topic modeling of consumer complaints.
result Time trends of latent topics reveal regulatory effectiveness and consumer protection issues.

LLMs outperform human analysts in predicting earnings direction.

problem Evaluating financial statements without narrative or industry-specific information.
method Trained GPT4 on standardized, anonymous financial statements and instructed to predict earnings direction.
result LLMs predict earnings directionally with accuracy comparable to narrowly trained ML models.

Enhances neural language processing with a hierarchical context-aware model.

problem Limited context in neural language processing systems.
method Hierarchical recurrent neural network with multi-level context representation.
result Improves semantic error detection by 12.75% relative for unsupervised models and 20.37% relative for supervised models.

Study finds monthly SIPs outperform first-day SIPs in Nifty 50 by 0.5-2.5% annually.

problem Underexplored impact of SIP timing in India's equity market.
method 22-year analysis using multi-layered statistical framework (non-parametric tests, effect size metrics, SSD).
result Monthly SIPs (EXP-SIP) outperform first-day SIPs (FTD-SIP) by 0.5-2.5% annually over short-to-medium-term horizons.

Detects radical content on Twitter using textual, psychological, and behavioral signals.

problem Limit the spread of extremist narratives on social media.
method Analyzed extremist material, created contextual text-based model, inferred psychological properties, evaluated on Twitter.
result Radical users exhibit distinguishable textual, psychological, and behavioral properties.

Algorithm reduces audit costs by identifying best service configurations from biased textual evidence.

problem Designing service systems from textual evidence requires accurate selection despite biased automated scoring.
method Developed PP-LUCB algorithm combining LLM scores and selective audits to minimize costs.
result Correctly identified the best model in 40/40 trials with 90% cost reduction.

Framework improves clinical timeline reconstruction from text and tables.

problem Temporal precision and event timing in clinical narratives and EHRs.
method Retrieval-augmented multimodal alignment framework.
result Consistently improves absolute timestamp accuracy and temporal concordance.