The purpose of this study is to build up a comprehensive model for gold price forecast. The data including Gold Price, Gold Volume, Oil Price, S&P 500 Index, USD Index, and US CPI Index in the period January 2005 to May 2024 are considered as critical variables in the models. The study built up time-series forecasting models applied techniques of ARIMA, VARMA, and VECM; and overcame disadvantages of the techniques by proposing an ensemble learning model which is a subfield of machine learning that aims to combine multiple different models to create a new model that improves accuracy and minimizes error. Several stacking algorithms has been utilized, including unweighted averaging and weighted averaging based on Markowitz's optimization theory. As the results, applying separately forecasting methods indicates that ARIMA, VARMA, and VECM is most effective for short-term forecasting, medium-term forecasting, and long-term and overall trend forecasting of gold prices, respectively. The final model proposed is the ensemble learning model applied weighted averaging Markowitz theory which has been demonstrated as improving forecasting accuracy compared to other models.