Time series forecasting is a well-established research domain, particularly in finance and econometrics, with a multitude of methods and algorithms proposed to achieve accurate future trend predictions. This study aims to examine the effectiveness of two popular models, ARIMA and LSTM, for predicting trends in gold prices in finance and econometrics. Monthly global gold prices from January 2010 to December 2022 are analyzed, with a training set from January 2010 to December 2020, a validation set of 12 months randomly selected from the training set, and a test set from January 2021 to December 2022. The results show that the LSTM model with a forget gate cell at 600 epochs yields the highest accuracy in term of RMSE, MAPE and SMAPE, surpassing all other models, including the ARIMA model. The study also suggests that increasing the number of epochs beyond 600 does not lead to significant improvements in the LSTM model's performance. While the ARIMA model is simpler to implement and requires less time for parameter tuning and training, it is less accurate than the LSTM model. Incorporating a peephole connection to the LSTM cell does not improve the model's accuracy or training speed. The study's outcomes provide valuable insights into optimal practices for gold price prediction, with implications for decision-making and risk management processes.