ItemConference Paper

Large Language Models in Economic Forecasting: A Comprehensive Analysis of Predictive Performance and Benchmarking Against Traditional Methods for India FY 2025-26

Title
Large Language Models in Economic Forecasting: A Comprehensive Analysis of Predictive Performance and Benchmarking Against Traditional Methods for India FY 2025-26
Creator
Benny, Benison Jacob; James, C.K.; Gupta, Heena
Description
This study presents a comprehensive systematic evaluation of the performance of Large Language Models (LLMs) in economic forecasting, specifically examining their ability to predict key Indian macroeconomic indicators for the fiscal year 2025-26. Through a comparative analysis of ten prominent LLMs against traditional econometric models and expert forecasts from leading institutions, we assess the forecasting accuracy, reliability, and practical limitations of these models using a rigorous multistage validation framework. We validate predictions using actual quarterly data from Q1 and Q2 of FY 2025-26, providing a real-time assessment of forecasting capabilities with bootstrap confidence intervals and time series cross-validation techniques. Results reveal significant variations in LLM performance, with validation against Q1 2025-26 actual GDP growth of 6.7 per cent showing that several LLMs achieved superior accuracy (MAPE less than 3 per cent) compared to traditional ARIMA models (MAPE 13.58 per cent). Top-performing LLMs demonstrate forecasting capabilities that approach expert-level accuracy while maintaining computational efficiency and scalability. Statistical significance tests using the Diebold-Mariano framework confirm the superiority of ensemble LLM approaches over individual traditional methods. The findings demonstrate that leading LLMs can serve as valuable supplementary forecasting tools, positioning between conventional statistical methods and expert analysis in terms of accuracy, while offering advantages in processing qualitative information and adaptation to structural changes. 2025 IEEE.
Source
5th IEEE International Conference on Mobile Networks and Wireless Communications, ICMNWC 2025;
Date
01-01-2025
Publisher
Institute of Electrical and Electronics Engineers Inc.
Subject
ARIMA; Artificial Intelligence; Benchmarking Analysis; Bootstrap Methods; CrossValidation; Economic Forecasting; India Economy; Large Language Models; Quarterly Validation
Coverage
Benny B.J., Christ (Deemed to be University), Department of Statistics and Data Science, Bangalore, India; James C.K., Christ (Deemed to be University), Department of Statistics and Data Science, Bangalore, India; Gupta H., Christ (Deemed to be University), Department of Statistics and Data Science, Bangalore, India
Rights
Restricted Access; Hardcopy may be available in the library
Relation
ISBN: 979-833155664-8;
Format
online
Language
English
Type
Conference paper

Citation

Benny, Benison Jacob; James, C.K.; Gupta, Heena, “Large Language Models in Economic Forecasting: A Comprehensive Analysis of Predictive Performance and Benchmarking Against Traditional Methods for India FY 2025-26,” CHRIST (Deemed To Be University) Institutional Repository, accessed June 18, 2026, https://archives.christuniversity.in/items/show/26065.

Output Formats