Multivariate and Univariate Prediction of Stock Prices using an
Optimized Gated Recurrent Unit with a Time Lag Proportional to the
Wavelet Approximation Coefficient
The advancement of precise prediction models is still very helpful
across a wide range of fields. Deep learning models have demonstrated
strong performance and great accuracy in stock price prediction.
However, the vanishing gradient problem, which some activation functions
have exacerbated, has a significant impact on these models. In order to
combat poor convergence, disappearing gradients, and significant error
metrics, this study suggests using the Optimized Gated Recurrent Unit
(OGRU) model with a scaled mean Approximation Coefficient (AC) time lag.
This study employed the Rectified Linear Unit (ReLU), Hyperbolic Tangent
(Tanh), Sigmoid and Exponential Linear Unit (ELU) activation functions.
Real-life datasets were used including the daily Apple and 5-minute
Netflix closing stock prices, decomposed using the Stationary Wavelet
Transform (SWT). The decomposed series formed a multivariate model which
was compared to a univariate model with similar hyperparameters and
different default lags. The Apple daily dataset performed well with a
Default_1 lag, using a univariate model and the ReLU, attaining
0.01312, 0.00854 and 3.67 minutes for RMSE, MAE and runtime. The Netflix
data performed best with the MeanAC_42 lag, using a multivariate model
and the ELU achieving 0.00620, 0.00487 and 3.01 minutes for the same
metrics. The study concluded that the OGRU is made resilient to the
vanishing gradient problem by avoiding the Sigmoid activation function
and applying the proposed lag on high frequency data with the ELU
activation function using decomposed data.