Authors :
Shradha Ranjan; Chhavi Saini; Saumya Samir; Akshita Goel; Ela Kumar
Volume/Issue :
Volume 9 - 2024, Issue 4 - April
Google Scholar :
https://tinyurl.com/am7kem6m
Scribd :
https://tinyurl.com/yvwt8jbx
DOI :
https://doi.org/10.38124/ijisrt/IJISRT24APR664
Abstract :
The efficacy of Long Short-Term Memory
(LSTM) neural networks and attention-based models in
predicting next-day closing prices of the MSFT 500index
is meticulously examined. A comprehensive suite of nine
carefully chosen predictors spanning fundamental
market data, macroeconomic indicators, and technical
metrics is amalgamated, fostering a holistic
comprehension of market behavior. Through rigorous
analysis, the research evaluates single-layer and
multilayer LSTM architectures alongside attention-
based LSTM variants, juxtaposed against traditional
ARIMA models. Surprisingly, the single-layer LSTM
consistently outperforms its multilayer counterpart,
demonstrating superior accuracy and model fit. The
integration of corporate accounting statistics augments
predictive capabilities, enriching the models' efficacy.
Notably, attention-based LSTM models, particularly the
Attention-LSTM variant, exhibit markedly lower
prediction errors and higherreturns in trading strategies
compared to other methodologies. However, the
heightened complexity of stacked-LSTM structures fails
to surpass the predictive acumen of simpler LSTM
architectures.This inquiry underscores the paramount
importance of leveraging advanced AI techniques and
comprehensive datasets in navigating the intricate
nuances of modern financialmarkets, offering invaluable
insights for both researchers and practitioners engaged
in stock priceforecasting endeavors.
The efficacy of Long Short-Term Memory
(LSTM) neural networks and attention-based models in
predicting next-day closing prices of the MSFT 500index
is meticulously examined. A comprehensive suite of nine
carefully chosen predictors spanning fundamental
market data, macroeconomic indicators, and technical
metrics is amalgamated, fostering a holistic
comprehension of market behavior. Through rigorous
analysis, the research evaluates single-layer and
multilayer LSTM architectures alongside attention-
based LSTM variants, juxtaposed against traditional
ARIMA models. Surprisingly, the single-layer LSTM
consistently outperforms its multilayer counterpart,
demonstrating superior accuracy and model fit. The
integration of corporate accounting statistics augments
predictive capabilities, enriching the models' efficacy.
Notably, attention-based LSTM models, particularly the
Attention-LSTM variant, exhibit markedly lower
prediction errors and higherreturns in trading strategies
compared to other methodologies. However, the
heightened complexity of stacked-LSTM structures fails
to surpass the predictive acumen of simpler LSTM
architectures.This inquiry underscores the paramount
importance of leveraging advanced AI techniques and
comprehensive datasets in navigating the intricate
nuances of modern financialmarkets, offering invaluable
insights for both researchers and practitioners engaged
in stock priceforecasting endeavors.
