Paper: arXiv 2502.17011
Authors: Jaskaran Singh Walia, Aarush Sinha, Naman Saraswat, Srinitish Srinivasan, Srihari Unnikrishnan
Abstract
Financial bond yield forecasting is challenging due to data scarcity, nonlinear macroeconomic dependencies, and evolving market conditions. In this paper, we propose a novel framework that leverages Causal Generative Adversarial Networks (CausalGANs) and Soft Actor-Critic (SAC) reinforcement learning (RL) to generate high-fidelity synthetic bond yield data for four major bond categories (AAA, BAA, US10Y, Junk). By incorporating 12 key macroeconomic variables, we ensure statistical fidelity by preserving essential market properties. To transform this market dependent synthetic data into actionable insights, we employ a finetuned Large Language Model (LLM) Qwen2.5-7B that generates trading signals (BUY/HOLD/SELL), risk assessments, and volatility projections. We use automated, human and LLM evaluations, all of which demonstrate that our framework improves forecasting performance over existing methods, with statistical validation via predictive accuracy, MAE evaluation(0.103%), profit/loss evaluation (60% profit rate), LLM evaluation (3.37/5) and expert assessments scoring 4.67 out of 5. The reinforcement learning-enhanced synthetic data generation achieves the least Mean Absolute Error of 0.103, demonstrating its effectiveness in replicating real-world bond market dynamics. We not only enhance data-driven trading strategies but also provides a scalable, high-fidelity synthetic financial data pipeline for risk & volatility management and investment decision-making. This work establishes a bridge between synthetic data generation, LLM driven financial forecasting, and language model evaluation, contributing to AI-driven financial decision-making.
Complexity vs Empirical Score
- Math Complexity: 7.5/10
- Empirical Rigor: 8.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper employs advanced mathematics including Causal GANs and Soft Actor-Critic RL, but provides extensive empirical validation with specific metrics like MAE (0.103%), profit rates (60%), LLM evaluation scores, and a clear experimental pipeline, indicating high implementation readiness.
Research Flowchart
flowchart TD A["Research Goal: Predict Liquidity-Aware Bond Yields via CausalGAN & DRL"] --> B["Data Input: 4 Bond Categories & 12 Macroeconomic Variables"] B --> C["Synthetic Data Generation: CausalGAN & Soft Actor-Critic SAC"] C --> D["LLM Processing: Qwen2.5-7B Finetuning for Signals/Risk/Volatility"] D --> E["Automated & Human Evaluation"] E --> F["Key Outcomes: 0.103% MAE, 60% Profit, 4.67/5 Expert Score"]
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper proposes a novel framework that combines CausalGANs, Soft Actor-Critic reinforcement learning, and a fine‑tuned LLM to generate synthetic bond‑yield data and produce trading signals.
Source sentence
we propose a novel framework that leverages Causal Generative Adversarial Networks (CausalGANs) and Soft Actor-Critic (SAC) reinforcement learning (RL) to generate high-fidelity synthetic bond yield data for four major bond categories (AAA, BAA, US10Y, Junk).
Assumes
No explicit assumptions found in the text.
Data
- Synthetic bond yield data (four categories) synthetic unknown
Source sentence
generate high-fidelity synthetic bond yield data for four major bond categories (AAA, BAA, US10Y, Junk).
- Macroeconomic variables macro unknown
Source sentence
By incorporating 12 key macroeconomic variables, we ensure statistical fidelity by preserving essential market properties.
Costs
Not modeled
Finds 4
- The RL‑based approach yields a 60% profit rate for AAA bonds and 54% for BAA bonds. performance
Source sentence
the RL- based approach achieves the most number of profit months which was 60% for AAA and 54% for BAA Bonds.
- LLM evaluation scores: RL method obtains an average LLM judge score of 3.37, higher than GAN (2.87) and actual data (2.58). predictive-accuracy
Source sentence
Method Avg. LLM Judge Score↑ RL 3.37 GAN 2.87 Actual 2.58
- Expert evaluation gives the RL method an average score of 4.67, compared with 4.17 for GAN and 3.5 for actual data. predictive-accuracy
Source sentence
Average↑ RL 4.5 5.0 4.5 4.67 GAN 4.0 4.0 4.5 4.17
- Ablation study shows removing the RL component increases MAE by 0.3 for US10Y and 0.1 for Junk bonds. statistical
Source sentence
removing the RL component increases the MAE a value of 0.3 on US 10 year bond yields and 0.1 on Junk Bond Yields.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 17 related papers). Coverage is partial, so this is neither replication nor refutation.
4 papers in the archive
Code
No repository by the authors found in the paper.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.
Cited References (68)
- Attention-Based LSTM for Bond Yield Forecasting (2023)
- Using Barbells to Lift Risk-Adjusted Return (2020)
- Riding the yield curve: a variety of strategies (2005)
- Portfolio Optimization using Deep Reinforcement Learning (2024)
- Adaptive learning by genetic algorithms: Analytical results and applications to economic models (2011)
- The FRB/US Model: A Tool for Macroeconomic Policy Analysis (2013)
- DeepSeek R1: A Large Language Model for Robust Decision Evaluation (2024) — link
- Regulatory Constrained GANs for Fixed-Income Markets (2023)
and 60 more
- Qwen2. 5 technical report (2024) — link
- Statistical Validation of Synthetic Time Series (2023)
- RAG for Low-Data Financial Forecasting (2023)
- Time-series Generative Adversarial Networks (2019) — link
- Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor (2018)
- Generative Model for Financial Time Series Trained with MMD Using a Signature Kernel — → 2407.19848
- Deep Learning Based on GANs and CNNs for Financial Time Series Predictions (2020) — link
- Time-FFM: Towards LM-Empowered Federated Foundation Model for Time Series Forecasting (2024)
- TEMPO: Prompt-based Generative Pre-trained Transformer for Time Series Forecasting (2024)
- GPT4MTS: Prompt-based Large Language Model for Multimodal Time-series Forecasting (2024)
- MOMENT: A Family of Open Time-series Foundation Models (2024)
- TIME-LLM: Time Series Forecasting by Reprogramming Large Language Models (2024)
- Timer: Generative Pre-trained Transformers Are Large Time Series Models (2024)
- Unified Training of Universal Time Series Forecasting Transformers (2024)
- TradingAgents: Multi-Agents LLM Financial Trading Framework — → 2412.20138
- FinCon: A Synthesized LLM Multi-Agent System for Enhanced Financial Decision Making (2024)
- Term-Structure Models: A Graduate Course (2009)
- Portfolio Optimization Using Deep Reinforcement Learning (2021)
- FinBERT: Financial Sentiment Analysis with Pre-trained Language Models (2019) — link
- ‘Finance Wizard’ at the FinLLM Challenge Task: Financial Text Summarization (2024) — link
- Generative Adversarial Nets (2014)
- Wasserstein GAN (2017)
- Human-level control through deep reinforcement learning (2015)
- Deep Learning with Long Short-Term Memory Networks for Financial Market Predictions (2018)
- Forecasting the term structures of Treasury and corporate yields using dynamic Nelson-Siegel models (2011)
- FEB-stacking and FEB-DNN models for stock trend prediction: a performance analysis for pre and post covid-19 periods (2021)
- Forecasting government bond yields with large Bayesian vector autoregressions (2012)
- HuggingFace’s Transformers: State-of-the-art Natural Language Processing (2020) — link
- PyTorch: An Imperative Style, High-Performance Deep Learning Library (2019) — link
- Integrating Stock Features and Global Information via Large Language Models for Enhanced Stock Return Prediction — → 2310.05627
- PIXIU: A Large Language Model, Instruction Data and Evaluation Benchmark for Finance (2023) — link
- LLMFactor: Extracting Profitable Factors through Prompts for Explainable Stock Movement Prediction (2024) — link
- Can ChatGPT Forecast Stock Price Movements? Return Predictability and Large Language Models — → 2304.07619
- Can Large Language Models beat wall street? Evaluating GPT-4’s impact on financial decision-making with MarketSenseAI (2024) — link
- Fine-Tuning Large Language Models for Stock Return Prediction Using Newsflow — → 2407.18103
- Financial Statement Analysis with Large Language Models — → 2407.17866
- Sentiment trading with large language models — → 2412.19245
- Using generative adversarial networks to synthesize artificial financial datasets (2020) — link
- Quant GANs: Deep Generation of Financial Time Series (2019) — link
- Generating Realistic Stock Market Order Streams (2020) — link
- Generative Adversarial Networks Applied to Synthetic Financial Scenarios Generation (2022) — link
- Market-GAN: Adding Control to Financial Market Data Generation with Semantic Context — → 2309.07708
- Simulating financial time series using attention (2022) — link
- Deep Hedging: Learning to Simulate Equity Option Markets (2019) — link
- Deep Reinforcement Learning for Trading (2019) — link
- Optimal Market Making by Reinforcement Learning (2021) — link
- Financial Trading as a Game: A Deep Reinforcement Learning Approach (2018) — link
- Model-based Deep Reinforcement Learning for Dynamic Portfolio Optimization (2019) — link
- FinRL-Podracer: High Performance and Scalable Deep Reinforcement Learning for Quantitative Finance (2021) — link
- An Application of Deep Reinforcement Learning to Algorithmic Trading (2020) — link
- The Evolution of Reinforcement Learning in Quantitative Finance: A Survey (2024) — link
- Deep reinforcement learning for portfolio management (2022) — link
- Recent advances in reinforcement learning in finance (2023)
- SuperCLUE-Fin: Graded Fine-Grained Analysis of Chinese LLMs on Diverse Financial Tasks and Applications (2024) — link
- Empowering Many, Biasing a Few: Generalist Credit Scoring through Large Language Models (2024) — link
- Evaluating Financial Relational Graphs: Interpretation Before Prediction — → 2410.07216
- FinBen: A Holistic Financial Benchmark for Large Language Models (2024) — link
- LOB-Bench: Benchmarking Generative AI for Finance – an Application to Limit Order Book Data — → 2502.09172
- Qwen/Qwen2.5-7B-Instruct-1M · Hugging Face (2025) — link
- deepseek-ai/DeepSeek-R1-Distill-Qwen-32B · Hugging Face (2025) — link
10 of 68 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.