Prediction of Bank NEO Commerce Stock Prices Using a Multiple Linear Regression Algorithm

Abstract

This study addresses the challenge of forecasting Bank Neo Commerce (BBYB) daily prices, where volatility and data leakage often bias results. We build a leakage-free pipeline to predict next-day Adjusted Close using multiple linear regression (OLS) with t−1 predictor lags and technical indicators: AdjClose_t−1​, AdjClose_t−5, Vol_t−1​, SMA5_t−1​, EMA5_t−1​, and RSI 14_t−1. Daily BBYB.JK data from Yahoo Finance (12 March 2019–12 March 2025) are evaluated with a 5-fold rolling time-series split, and metrics are reported as mean ± SD. The goal is to assess OLS accuracy and its practical value against a persistence baseline. OLS attains MAE 25.15 ± 18.24, MSE 2,344.41 ± 2,956.78, R² 0.94 ± 0.06, while the baseline AdjClose_t−1 yields MAE 22.40 ± 17.21, MSE 1,894.99 ± 2,420.42, R² 0.96 ± 0.04. A walk-forward long-only backtest (0.1% fee) delivers a final value of 1.04 versus 0.72 for buy-and-hold, with lower volatility and drawdown. The approach is interpretable, reproducible, and ready for extensions (feature reduction/regularization, non-linear models, and return/volatility features)