This paper proposes a study on the efficacy of using non-stock-related news, such as political events, global incidents, and social trends, to predict market timing for investment portfolios. The research aims to extend the current understanding of market dynamics by incorporating a broader spectrum of news sources, moving beyond the conventional focus on stock-specific news. By doing so, the study seeks to understand how these external factors contribute to market movements, influencing investor behavior and decision-making processes.Unlike the traditional word-based methods, such as bag-of-words or word vectors, the contextualized representation captures both the syntax and semantics of text, thus providing a more comprehensive un- derstanding of its meaning. The research employs advanced machine learning algorithms to parse non-traditional news data. The paper included data from from different sources like Bloomberg, Reuters, and The Wall Street Journal. The datasets cover comprehensive financial market coverage, including corporate news, earnings reports, and market analysis. We found that different types of embeddings from non-stock related news can have significant predictive power for market returns, especially under models such as Convolutional-LSTM and sentiment analysis. This find offers novel insights into market behavior, potentially providing investors and portfolio managers with innovative tools for predicting market shifts and making more informed investment decisions.