Dueling Double Deep Q-Networks for Regime-Aware and Risk-Conscious Stock Trading: Evidence from NIFTY 50
Bibliographic record
Abstract
Reinforcement learning (RL) has emerged as a promising approach for automated stock trading, but standard deep Q-networks (DQNs) often struggle with volatile market dynamics and risk management. This paper presents a comprehensive study of trading the NIFTY 50 index using an RL agent based on DQN, enhanced with multiple novel improvements. We propose an adaptive action space that flexibly adjusts to market volatility, market regime detection via unsupervised learning (e.g., Hidden Markov Models or K-Means clustering) to inform the agent of bull/bear phases, and a composite reward function that balances profit with a Sharpe ratio proxy, drawdown penalties, and transaction costs. The DQN algorithm is further refined through risk-aware Q-value updates using soft Q-learning principles to encourage prudent exploration, and a multi-objective dueling network architecture that separately estimates profit and drawdown-centric value functions. To improve generalization, the agent is trained with synthetic data augmentation using noise injection and generative modeling, and its state representation is compressed via an autoencoder bottleneck to denoise and reduce dimensionality. The trading environment includes budget constraints and leverage mechanics to simulate margin trading, and incorporates macroeconomic indicators (USD/INR exchange rate, inflation, oil prices) into the state space for a richer market context. Using a modified Chainer DQN framework, we evaluate variants of DQN, Double DQN, and Dueling Double DQN on historical NIFTY 50 data. The results show that our improved Dueling Double DQN agent achieves superior performance, with higher profitability and risk-adjusted returns (Sharpe proxy) on the test period, compared to baseline approaches. These findings illustrate how integrating domain-specific insights (volatility, regimes, macro factors) and advanced RL techniques can yield a robust trading agent for emerging market indices like NIFTY 50.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.020 | 0.030 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.002 | 0.001 |
| Bibliometrics | 0.001 | 0.003 |
| Science and technology studies | 0.002 | 0.001 |
| Scholarly communication | 0.002 | 0.001 |
| Open science | 0.002 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".