Forecasting India's Bilateral Trade Flow: A Modified Neural Network Approach

Bilateral trade flow is a significant economic indicator used by government and policy makers in making decisions on country's economic growth and forming trade policies to enhance global economy development. To evaluate and enhance bilateral trade flow, this paper proposes a modified neural network approach to predict India' bilateral trade flow with India's 45 major trading partner countries and assess the potential for expanding trade. To avoid exponential and multiplicative correlations among features like GDP, distance and population, logarithmic features are incorporated in a feed forward fully connected neural network architecture. The study also uses gravity model as baseline model for comparison. The gravity model is a conventional linear regression model which is used to assess international trade by policy makers. The study reveals that the utilization of a fully connected, feed-forward neural network leads to a successful improvement in prediction accuracy compared to the Gravity Model when applied to India's trade dataset. This outcome holds significance as it emphasizes the efficacy of neural network, subset of machine learning methodologies, in forecasting economic variables using solely time-agnostic features.