Challenges of Artificial Intelligence in Foreign Exchange Forecasting : A Comparison with ARIMA Models

The forex market in India has grown significantly since 1978, especially with the introduction of currency derivatives, leading to a substantial increase in trading volumes. Over the past five years, trading in forex instruments like forwards and swaps has risen by over 25%, with interbank trading making up 77% of this activity and new AI technologies like neural networks and machine learning are being explored to improve forecasting. This growth has pushed market participants to use more advanced forecasting methods to stay ahead. This research paper examines the use of the ARIMA model for predicting the USD/INR exchange rate, a method that combines autoregressive and moving average components. We reviewed existing research to compare ARIMA with these AI methods and used data from the Reserve Bank of India from 1998 to 2024 for data analysis. The ARIMA (2, 2, 0) model worked well, explaining 36% of the changes in the exchange rate and predicting a steady rise in the USD/INR rate through 2029. However, we also found challenges with AI, such as the complexity of models, the need for high-quality data, and the ability to adjust to sudden market changes. To improve forecasting, we suggest combining AI with traditional methods, focusing on better data quality, simplifying AI models, and ensuring they can adapt to new market conditions while maintaining ethical standards.