MUMBAI, India, Aug. 17 -- Intellectual Property India has published a patent application (202641072449 A) filed by Srm Trp Engineering College on June 11, 2026, for Predictive Financial Analytics: A Hybrid Prophet-Attention-Lstm Architecture For Non-Linear Stock Volatility Forecasting And Risk Management.
Inventors include Mr. M. Manoj; Mr. V. Murugudharshan; Mr. V. Madhesh; Mr. S. Mugesh; and Mrs. P. Jeyamedona.
The application for the patent was published on August 14, 2026, under issue no. 33/2026.
Abstract: The system uses Prophet to separate the deterministic components of a price series, covering the long-run trend, the weekly and annual seasonality, and the effects of known structural events, from the stochastic residuals that Prophet cannot explain. Those residuals, which carry the micro-volatility signal, are then fed into a two-layer LSTM network with a temporal attention mechanism. The attention layer allows the model to assign greater weight to past periods of high volatility when constructing its forecast, rather than treating all historical observations as equally informative. The final forecast combines both components: Prophet provides the structural baseline, and the LSTM provides the residual prediction. The system also generates a risk indicator from the predicted residual magnitude, which has been shown to correlate consistently with subsequent market drawdowns across different market regimes. Empirical testing across multiple equity market datasets shows that the hybrid architecture achieves approximately 12% lower RMSE than the best standalone baseline, with reliable performance in both quiet trending markets and high- volatility stress periods. The temporal attention weights also provide a degree of model interpretability that is absent from standard LSTM-based approaches, making the system more suitable for use in regulated financial environments.
Disclaimer: Curated by HT Syndication.
