This paper presents a systematic literature review of deep learning (DL) applications for financial time series forecasting, covering studies published between 2005 and 2019. It is motivated by the lack of a comprehensive survey focused solely on DL in this domain, despite the growing number of DL-based forecasting models that significantly outperform traditional machine learning (ML) approaches. The review categorizes 153 papers based on both the forecasting application area (e.g., stock, index, forex, commodity, volatility, cryptocurrency) and the DL model type (e.g., CNN, LSTM, DBN, RNN, autoencoders, deep reinforcement learning). Key findings include that LSTM is the most widely used model, followed by CNN, and that DL models consistently achieve superior performance compared to ML counterparts. The paper also provides a statistical analysis of publication trends, identifies open issues such as data scarcity, model interpretability, and hyperparameter tuning, and suggests future research directions including hybrid models, attention mechanisms, and the integration of alternative data sources. This survey serves as a valuable resource for researchers and practitioners seeking to understand the state-of-the-art and identify opportunities for further advancement in DL-based financial time series forecasting.