Inflation Forecasting with AI and LLMs: RBI Perspective
Why in the news
Central banks face a harder job forecasting inflation, and the RBI Governor stressed advanced tools. The discussion centres on whether LLM-based AI can sharpen forecasts.
Key facts
- Before the Global Financial Crisis, central banks struggled to contain inflation; in 2022-2025 they struggled to bring it down.
- India’s CPI inflation is shaped by food prices, global commodity trends and central banks’ gold purchases.
- Nowcasting means short-term inflation estimates using real-time text data.
- AI agents can simulate household responses, a cheaper route to inflation expectation surveys.
Evidence on AI forecasting
| Study / user | Finding |
|---|---|
| Faria-e-Castro & Leibovici (2024), Fed | LLMs gave lower mean-squared errors and beat traditional professional-forecaster surveys |
| Bybee (2023) | GPT-3.5 simulated economic expectations successfully |
| Bick et al. (2024) | 40% of US adults use AI by Aug 2024; 28% at work |
| Aldasoro et al. (2024) | Half of US households use AI tools |
| Kcore Analytics (2024 Indian elections) | AI analysis of social media, with economic sentiment including inflation, outperformed traditional exit polls |
Concerns
- Central banks lack control over the external data LLMs are trained on.
- Training data is not timestamped, so real-time retraining is hard.
- Public LLMs are retrained periodically, making results hard to replicate.
Way forward
- Faster, more accurate forecasts from real-time data.
- Cheaper alternatives to household surveys.
- Central banks need to adapt AI-driven models as AI-assisted decisions spread.
Exam angle
- Term: nowcasting.
- Benchmark method beaten: Survey of Professional Forecasters.
- Source of inflation pressure: food prices and commodities.