EVALUASI KINERJA PORTOFOLIO SAHAM OPTIMAL DENGAN LONG SHORT-TERM MEMORY BERBASIS MODEL MARKOWITZ PADA INDEKS SMINFRA18

Abstract

This study aims to determine the optimal stock portfolio within the SMinfra18 index using the Markowitz model and forecast the dominant stock price using the Long Short-Term Memory (LSTM) method. Monthly closing prices of 18 SMinfra18 stocks from January 2024 to December 2025 were analyzed. The study consisted of three main stages: (1) calculating stock returns and selecting stocks with positive expected returns; (2) optimizing the portfolio using the Markowitz model to construct an Equal Weight Portfolio, a Minimum Variance Portfolio, and an Optimal Portfolio based on the maximum Sharpe Ratio; and (3) forecasting the dominant stock price using LSTM. Seven stocks met the selection criteria: WIFI, SSIA, PGAS, UNTR, ELSA, MEDC, and PGEO. The optimal portfolio achieved a monthly expected return of 5.04% with a risk of 6.26% and a Sharpe Ratio of 0.8044, comprising UNTR (37.50%), PGAS (29.85%), SSIA (13.92%), WIFI (12.51%), and ELSA (6.21%). The LSTM model achieved a Mean Absolute Percentage Error (MAPE) of 1.36%, while a 5-day forecast for UNTR indicated stable prices between Rp 29,421.87 and Rp 29,371.21. These results demonstrate the effectiveness of integrating the Markowitz model and LSTM for portfolio optimization and stock price forecasting.