STOCK MOVEMENT ANALYSIS OF ONLINE MARKETPLACE IN INDONESIA USING LONG-SHORT TERM MEMORY MODEL WITH HYPERPARAMETER TUNING

• Main Author: Miranda, Shereva
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/83269
• Institution: Institut Teknologi Bandung

Physics is the science that studies the structure of matter and the interactions between the fundamental elements of the observable universe. Physics can offer new perspectives in addressing economic issues. The stock market, categorized as a complex system with elements interacting in an irregular manner, can benefit from a physics-based approach in its analysis. E-commerce businesses or marketplaces that have achieved billion-dollar valuations have become major components of the stock market and an increasingly popular investment option. Due to the complexity of market movements, various methods are used for predicting market trends, including technical analysis, machine learning models, and fundamental analysis using internal company data. The model used is Long- Short Term Memory (LSTM) due to its ability to retain information from long- term loops, and Bidirectional Long-Short Term Memory (BLSTM) because of its capability to process information from both forward and backward sequences, for performance comparison. Model optimization is carried out using hyperparameter tuning with Grid Search to produce the best predictions. A comparison is then made between the model’s predictions and fundamental analysis. The model with the best performance is BLSTM-HT, with a training RMSE of 3,3346 and a testing RMSE of 3,8512, and maximum errors of 15,3361 USD and 9,0531 USD respectively. The model’s predictions indicate that the company's stock is generally rising, which aligns with the predictions from fundamental analysis. Despite some losses experienced by the company, the potential of a good business model and the growing sector suggest that the stock is expected to increase.