Liquidt: stock market analysis using liquid time-constant neural networks

Accurate and efficient predictions concerning stock prices are an intriguing and sought-after task in the field of computational financial analysis. This paper aims to leverage and validate novel deep learning pipelines for predicting NSE stock prices of Adani Ports, Reliance, and Tata Steel using recurrent neural architectures. The scope of this paper pertains to the utility of Liquid Time-Constant Networks (LTCs) and the development of a novel enhanced computing paradigm centric on LTCs. By assessing the use of bi-directionality for time-series prediction and the existing modality of architectures with LSTM and BiLSTM implications, the authors functioned on developing an LTC-based architecture that is capable of predicting stock market trends in a superlative fashion. The paper thoroughly explains the temporal characteristics and predictive trends and completes a comparative analysis with simple RNNs, GRUs, LSTMs, and their bidirectional counterparts, emphasizing the potential substitutes of the frequently used LSTMs. After sufficient experiments, the paper presents a comparative study for gauging a computational efficiency and accuracy trade-off for the aforementioned task.