Yu, Teng-ChihTeng-ChihYuDing, Jian-JiunJian-JiunDing2026-04-242026-04-242025-11-28https://www.scopus.com/record/display.uri?eid=2-s2.0-105030462373&origin=resultslisthttps://scholars.lib.ntu.edu.tw/handle/123456789/737496Market forecasting plays a critical role in economic signal processing. In this study, a model that applies the moving average convergence divergence (MACD) indicator together with learning-based architectures is proposed for market forecasting. First, we propose a novel MACD decomposition method to preprocess the data. By employing the proposed decomposition approach, the high- and low-frequency components are effectively separated. These components correspond to the short-term and the long-term information in the market, respectively. Then, two well-known models in economic forecasting, LSTM and ARIMA models, are employed to perform next-day predictions using only past closing prices. The proposed algorithm well integrates the merits of learning-based techniques and classical regression model. Despite of relatively less complexity, the proposed approach achieves promising performance and demonstrates adaptability across a variety of economic signals.falseconference paper10.1109/apsipaasc65261.2025.112490542-s2.0-105030462373