## Mending the Crystal Ball: Enhanced Inflation Forecasts with Machine Learning

_IMF Working Papers, September 27, 2024_

## Source details

**Canonical URL:** [Mending the Crystal Ball: Enhanced Inflation Forecasts with Machine Learning](https://www.imf.org/en/publications/wp/issues/2024/09/27/mending-the-crystal-ball-enhanced-inflation-forecasts-with-machine-learning-552848)

## Other formats

- [Markdown version](/en/publications/wp/issues/2024/09/27/mending-the-crystal-ball-enhanced-inflation-forecasts-with-machine-learning-552848/index.md)
- [Structured JSON version](/en/publications/wp/issues/2024/09/27/mending-the-crystal-ball-enhanced-inflation-forecasts-with-machine-learning-552848/index.json)
- [Bundle manifest](/en/publications/wp/issues/2024/09/27/mending-the-crystal-ball-enhanced-inflation-forecasts-with-machine-learning-552848/bundle-manifest.json)

## Bibliographic details
- Authors: Yang Liu, Ran Pan, Rui Xu
- Published: September 27, 2024
- Series: IMF Working Papers
- DOI: https://doi.org/10.5089/9798400285387.001

---

### Summary
- Forecasting inflation has become a major challenge for central banks since 2020, due to supply chain disruptions and economic uncertainty post-pandemic.
- Machine learning (ML) models can improve forecasting performance by:
  - incorporating a wider range of variables,
  - allowing for non-linear relationships,
  - focusing on out-of-sample performance.
- The paper applies ML models to forecast near-term core inflation in Japan post-pandemic. Japan is a challenging case because inflation had been muted until 2022 and has now risen to a level not seen in four decades.
- Four machine learning models are applied to a large set of predictors alongside two benchmark models.
- For 2023, the two penalized regression models systematically outperform the benchmark models, with LASSO providing the most accurate forecast.

### Methodology
- Models applied:
  - Four machine learning models (unspecified in summary text beyond penalized regression and LASSO).
  - Two benchmark models.
- Evaluation focus:
  - Near-term core inflation forecasts for Japan post-pandemic.
  - Emphasis on out-of-sample forecasting performance.
- Data scope:
  - A large set of predictors (details not specified in the summary text).

### Key Findings
- Penalized regression approaches systematically outperform benchmark models for 2023.
- LASSO provides the most accurate forecast among the evaluated models for 2023.
- Machine learning methods can enhance inflation forecasting by capturing richer predictor sets and non-linearities and by prioritizing out-of-sample accuracy.

### Useful Predictors Identified Post-2022
- Household inflation expectations
- Inbound tourism
- Exchange rates
- Output gap

### Policy and Analytical Implications
- Central banks and forecasters may benefit from incorporating machine learning techniques, particularly penalized regression methods like LASSO, when forecasting inflation in environments with large structural shifts.
- Monitoring non-traditional or higher-frequency indicators such as household inflation expectations and inbound tourism can improve near-term inflation signals.
- Using broader predictor sets and models that capture non-linear relationships can be valuable in periods of heightened uncertainty and structural change.

---

## Content in this bundle

- **Working Paper**
  - [Working Paper (Markdown version)](/-/media/files/publications/wp/2024/english/wpiea2024206-print-pdf.pdf.md){rel="alternate" type="text/markdown"}
  - [Working Paper (PDF)](/-/media/files/publications/wp/2024/english/wpiea2024206-print-pdf.pdf){rel="external" type="application/pdf"}

---

_Source: https://www.imf.org/en/publications/wp/issues/2024/09/27/mending-the-crystal-ball-enhanced-inflation-forecasts-with-machine-learning-552848_