The global metals market, particularly the gold market, plays a critical role in ensuring the overall stability an‎d performance of international financial systems. Due to its well-established position as a safe-haven asset, gold is widely regarded by investors as a strategic hedge against periods of economic uncertainty, elevated inflation levels, geopolitical tensions, an‎d fluctuations in global currency valuations. However, the price of gold is inherently volatile an‎d affected by a diverse range of macroeconomic variables, including supply-deman‎d dynamics, inflation expectations, interest rate adjustments, currency index movements, an‎d global equity an‎d commodity market behavior. Consequently, predicting trends in gold price movements remains a highly challenging an‎d strategically significant topic within financial analysis, investment planning, an‎d risk management. Traditional econometric approaches, such as autoregressive moving average (ARMA) an‎d generalized autoregressive conditional heteroskedasticity (GARCH) models, have been widely used to model the statistical properties of gold price time series. Nevertheless, these models often struggle to effectively capture nonlinear dependencies, highdimensional interactions, an‎d long-memory characteristics inherent in real-world financial markets. To overcome these limitations, recent studies have increasingly turned to deep learning methodologies, particularly recurrent neural networks such as Long Short-Term Memory (LSTM) structures an‎d Convolutional Neural Networks (CNN), due to their superior capacity to extract hidden, complex patterns from sequential financial data. In this research, a hybrid deep learning architecture based on the integration of CNN an‎d LSTM (CNN–LSTM) networks is proposed to predict short-term directional movements in gold prices. The CNN layer is utilized to extract spatial an‎d structural features from historical pricing an‎d technical indicator data, while the LSTM layer captures temporal dependencies an‎d long-term behavioral trends present in the underlying financial series. Furthermore, the model incorporates several influential macroeconomic an‎d financial market indicators, such as the U.S. Dollar Index (DXY), West Texas Intermediate (WTI) crude oil prices, the S&P 500 index, the USD/JPY exchange rate, an‎d the price of Bitcoin, each of which has been empirically demonstrated to contribute to volatility patterns in the gold market. The dataset for this study was collected from reliable international financial databases an‎d underwent a series of preprocessing steps, including normalization, resampling, an‎d feature-scaling operations, before being partitioned into training an‎d testing subsets. Model training was conducted using backpropagation through time, with mean squared error (MSE) employed as the primary loss function an‎d adaptive learning-rate optimization techniques applied to enhance convergence stability an‎d computational efficiency.

ساره گلي فروشاني

زهرا صابري , حامد لروند