Monthly trend forecasting performance of convolutional neural networks in the Nasdaq market

Accurate short-term forecasting in stock markets is essential for agile trading strategies, yet the volatile, nonstationary nature of financial data complicates predictive modeling. This study examines the performance of one-dimensional convolutional neural networks (CNNs) in identifying monthly directional movements for NASDAQ-listed equities. Although CNNs have proven effective in pattern recognition, many stock forecasting studies still rely on recurrent architectures. In this paper, by concentrating on CNNs and a sliding-window technique to generate lagged input features, we streamline the computational process for stock trend forecasting. Historical closing prices from multiple NASDAQ companies serve as both training and out-of-sample test sets. Pearson correlation, emphasizing directional alignment between actual and predicted data, is our primary evaluation metric, while mean squared error (MSE) is used to measure predictive accuracy. Through varying hyperparameters, such as network depth, batch size, and window segmentation, we show that CNNs remain robust under diverse conditions. Most scenarios yield strong positive correlations, indicating these networks can effectively capture local price dynamics with minimal hyperparameter tuning. This paper contributes to the field by confirming CNNs’ viability for stock trading and offers a reproducible framework. Our findings support the use of CNN-based pipelines for practitioners seeking rapid, directionally accurate stock trend insights.