Abstract:
Investing in stock market investments involves inherent risks. An accurate forecast of stock
market movement is important as it aids investors in identifying potentially profitable stocks,
or ones that might incur loss. However, training a model capable of performing this task with
reasonable accuracy remains a significant challenge. Each country’s stock market is different
in that the factors that influence it, such as certain economic factors, the laws that govern that
country, or the sentiments of the people, may be unique to that country. In Pakistan, where the
stock market is quite volatile due to the heavy influence of the country's politics on its economy,
building a good prediction model becomes even more of a challenge. Urdu being the national
language of Pakistan, is used by a majority of the country’s news agencies. This research study
aims to create structured datasets to help investigate the relationship between Urdu news events
and the performance of Pakistan’s stock market. Initially, this study focuses on the cement and
oil & gas sectors of Pakistan. The study uses Natural Language Processing (NLP) by leveraging
a combination of machine learning and deep learning techniques to extract insights from Urdu
news articles, encompassing sentiment analysis, named entity recognition, keyword extraction,
and event extraction. These insights are subsequently linked to corresponding data from the
Pakistan Stock Exchange (PSX) to create standardized datasets for analysis and prediction. The
findings of this study hold significant implications for investors, financial analysts and policy makers in Pakistan, as well as NLP enthusiasts for the Urdu language. The ability to quantify
the impact of news events on stock prices can guide investment strategies and risk management
decisions.
