Research Grounding
Academic foundationsPaper onlyAcademic literature grounding - adaptations stay unvalidated hypotheses until tested.
About our research methodology
We test quantitative strategies under realistic constraints: delayed data, no volume, and full Indian transaction costs.
Strategies from the literature (like ORB or residual momentum) are simplified adaptations. When a paper falsifies a method (e.g. arXiv:2605.04004 on ORB), we treat it strictly as an unvalidated hypothesis.
Built here
Walk-forward logistic model (ml_logit)
Logistic regression on returns/RSI/EMA features refit at day boundaries on prior days only. Starts UNVALIDATED. Deliberately simple per arXiv:2603.01820 findings.
News sentiment lexicon
Rule-based keyword scoring over RSS headlines. Transparent heuristic veto layer, not machine learning.
Papers
Event History Over Scale: Compact Transformers for Low-Latency Limit Order Book Forecasting
Not implementable on free dataarXiv:2610.02917
Paper: Short-horizon order book forecasting from level-3 (L3) limit order book events.
Here: Requires L3 order book data. No L2/L3 brokerage feed is connected in this deployment; documented as a future adapter.
Tabular Deep Learning for Algorithmic Trading: Cross-Regime Bayesian Optimisation for Equity Signal Generation
Adapted as hypothesisarXiv:2608.27076
Paper: Regime-robust parameter selection; gradient-boosted trees beat deep tabular models.
Here: Intraday regime classification (volatility percentile + index drift) gates signals. Parameter optimization is pending.
How do standard time series forecasting benchmarks translate to finance? A broad study of risk-adjusted performance
Adapted as hypothesisarXiv:2603.01820
Paper: Volatility-targeted Sharpe evaluation; simple models often beat complex architectures.
Here: Position sizing uses EWMA vol-targeting; evaluation prioritizes risk-adjusted returns over raw accuracy.
Factor-Driven Network Informed Restricted Vector Autoregression (FNIRVAR)
Adapted as hypothesisarXiv:2508.02198
Paper: PCA common factors + network VAR on idiosyncratic components for co-movement.
Here: Single market factor with rolling beta residuals vs NIFTY. Simplified adaptation hypothesis.
Structural Limits of OHLCV-Based Intraday Momentum Signals in MNQ Futures: A Systematic Falsification Study
Core design inputarXiv:2605.04004
Paper: Falsification of 14 momentum families after costs; five validation gates including friction floor.
Here: Implemented 5 validation gates and friction floor. ORB was falsified on futures in the paper; re-tested here as an unvalidated hypothesis.
Empirical analysis of Q learning, DQN and DDQN for intraday trading
Future workSpringer DOI: 10.1007/s44163-026-01138-x
Paper: Deep reinforcement learning for intraday trading decisions.
Here: Deep RL requires extensive historical data and an offline training pipeline; not implemented in v1.
Interpretable Hypothesis-Driven Trading: A Rigorous Walk-Forward Validation Framework for Market Microstructure Signals
Core design inputarXiv:2512.12924
Paper: Expanding-window walk-forward, strict information discipline, interpretable signals, modeled costs.
Here: Expanding walk-forward evaluation, causal features only, full statutory costs, and human-readable signal reasons.