Abstract:
Financial service providers are subjected to stringent regulations and their investment
decisions undergo continuous scrutiny by regulatory committees. Despite the potential of Artificial Intelligence (AI) to transform financial services, its implementation in
complex market analysis is limited by its opaque nature. To address this challenge, we
apply Explainable AI (XAI) to elucidate the underlying complexities of the Pakistan
stock market, enhancing transparency and interpretability. Therefore, in this study, we
address the challenge of time series forecasting to predict market stability by utilizing
a comprehensive dataset from the Pakistan Stock Exchange. We performed an experimental comparative analysis aimed at classifying time series data, improving model
interpretability, and identifying the most impactful economic indicators. An XAI framework catering the explanation needs of investors and short-term market participants is
proposed, using post-hoc XAI techniques such as Local Interpretable Model-agnostic Explanations (LIME), SHapley Additive exPlanations (SHAP) and Layer-wise Relevance
Propagation (LRP). In addition to achieving high accuracy, our AI models combined
with XAI techniques revealed that moving averages were the main contributors to the
model’s predictions, alongside the impact of foreign and local investor contributions
in the volatile environment of the PSX. This framework clarifies feature contributions,
ensuring transparency and interpretability of model decisions for stock market participants.
