Anomalies in financial markets—characterized by sudden shifts in returns or trading volumes—can signal systemic risk, structural breaks, or potential market manipulation. Timely detection of these events is crucial for maintaining the stability of trading systems, supporting early-warning tools, and enhancing regulatory oversight. However, real-world deployment of advanced anomaly detection models is often hindered by two challenges: the lack of labeled anomaly data and reliance on high-frequency inputs that may not be readily available.In this study, we propose a novel hybrid anomaly detection framework designed to work effectively with widely accessible daily return and volume data. Our approach combines a Long Short-Term Memory (LSTM) Autoencoder with a Generative Adversarial Network (GAN) to capture both temporal dependencies and distributional shifts. To improve precision, we integrate a One-Class SVM to identify subtle deviations in the LSTM’s latent representations. In the absence of ground truth labels, we introduce a synthetic anomaly injection mechanism that simulates realistic market irregularities, such as price shocks and volume spikes, enabling robust model evaluation.We evaluate the framework across six diverse stock categories—including indices, mega-cap, small-cap, high-volatility, low-volatility, and penny stocks—and under four major market regimes, including the Global Financial Crisis and the COVID-19 recovery period. The hybrid model consistently outperforms classical baselines such as GARCH, Z-Score, and One-Class SVM, particularly in recall and F4-score, demonstrating strong performance in both stable and volatile environments.Key contributions of this work include: 1. A scalable, interpretable LSTM-GAN hybrid architecture tailored for low-frequency financial data; 2. A novel anomaly injection protocol for unsupervised model validation; 3. A comprehensive evaluation across multiple asset types and historical financial regimes. This research offers a practical and generalizable solution to financial anomaly detection, narrowing the gap between academic advancements and real-world applications in data-constrained settings.