Time series data prediction is an essential area of research in finance, and economics, among others. It involves analyzing and modeling data collected over time to make future predictions or forecast future trends. With the increasing availability of historical data and advancements in machine learning and deep learning techniques, time series data prediction has become an increasingly popular research topic in recent years. In this work, we investigate the application of machine and deep learning models to time-series data for predicting the optimal time for trading stocks and options. Time-series data is defined as a collection of historical data points ordered by time, commonly used in predicting stock prices, stock indexes, and cryptocurrency prices. We use a publicly available dataset of Japanese stocks and options to train and test Support Vector Regression (SVR) and Long Short-Term Memory (LSTM) models. The goal is to improve trading strategies by identifying the best times to buy and sell assets based on predictive models. The performance of the model is compared using three accuracy measurements: Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE). The study has shown that The LSTM with dropout technique provided the best possible results with MSE 0.000124763, RMSE 0.011169727, and MAE 0.009058733.
The time-series data is considered numerical data points. Financial data is an example of this type. It consists of numeric variables. Time series data is usually fractional numbers (also known as decimal numbers), which takes any value between two numbers, known as continuous data. This data contains historical data for a variety of Japanese stocks and options. The data used in this study were obtained from the Kaggle website for data scientists and machine learning practitioners. The dataset was created by the Japan Exchange Group, which operates one of the largest stock exchanges in the world. It contains historic data for various Japanese stocks and options starting from 4/1/2017 and ending in 27/5/2022.
The dataset we used shown in this link: https://www.kaggle.com/competitions/jpx-tokyo-stock-exchange-prediction/data
Each time series dataset should be tested to determine if it exhibits stationarity, meaning that its statistical properties such as mean and variance are constant over time. Summary statistics are one of several methods used to assess the non-stationarity of a time series. This involves splitting the time series into two or more partitions and comparing the mean and variance of each group. If the means and variances differ significantly between the groups, the time series is likely non-stationary.
A big difference between the mean and variance as shown in the original data indicates that the data is non-stationary. To make it stationary, we applied the natural logarithm to the data. Fig. 2 and Fig. 3 "Dataset Distribution" and "Transformed Dataset" present the behavior of the data before and after the stationary test. Fig. 3 shows the suitable format that is accepted by the model.
To identify the optimal hyperparameters for the Support Vector Regression (SVR) model. Various experiments were conducted to determine the optimal statistical model with the lowest error. The values of ”C” and ”ϵ” were investigated by experimenting with different values of ”C” and ”ϵ” while the kernel function remained fixed at the RBF function. Based on the evaluation metrics we got, it was found that ”C” = 100 and,”ϵ” = 0.0005 provided the smallest error. To confirm that these values were indeed optimal, additional experiments were conducted with different values of ”C” while ”ϵ” was fixed at 0.0005.
Two experiments were conducted in this study, one with and one without the dropout technique, using different numbers of neurons, namely 4, 45, 50, 55, 123, 128, and 133. The LSTM model with dropout technique outperformed the LSTM without dropout, as evidenced by the lower MSE, RMSE, and MAE metrics with values of 0.000124763, 0.011169727, and 0.009058733. A comparison between the The LSTM model outperformed the SVR model by producing the best results, with an MSE of 0.000124763, an RMSE of 0.011169727, and an MAE of 0.009058733.
