Harnessing Machine Learning for Real-Time Inflation Nowcasting
Published: 08 March 2024
By: Richard Schnorrenberger Aishameriane Schmidt Guilherme Valle Moura
We investigate the predictive ability of machine learning methods to produce weekly inflation nowcasts using high-frequency macro-financial indicators and a survey of professional forecasters. Within an unrestricted mixed-frequency ML framework, we provide clear guidelines to improve inflation nowcasts upon forecasts made by specialists. First, we find that variable selection performed via the LASSO is fundamental for crafting an effective ML model for inflation nowcasting. Second, we underscore the relevance of timely data on price indicators and SPF expectations to better discipline our model-based nowcasts, especially during the inflationary surge following the COVID-19 crisis. Third, we show that predictive accuracy substantially increases when the model specification is free of ragged edges and guided by the real-time data release of price indicators. Finally, incorporating the most recent high-frequency signal is already sufficient for real-time updates of the nowcast, eliminating the need to account for lagged high-frequency information.
Keywords: inflation nowcasting; machine learning; mixed-frequency data; survey of professional forecasters;
JEL codes E31; E37; C53; C55;
Working paper no. 806
806 - Harnessing Machine Learning for Real-Time Inflation Nowcasting
Research highlights
- The study evaluates the effectiveness of machine learning in generating weekly inflation nowcasts for the Brazilian economy, utilizing high-frequency macro-financial data and a daily survey from professional forecasts.
- Using an unrestricted mixed-frequency ML structure, we provide clear guidelines to improve inflation nowcasts upon forecasts made by SPF specialists.
- The best-performing nowcasts depend on variable selection performed via the LASSO combined with accurate timely signals from price indicators and informed judgment entailed in SPF expectations.
- Major nowcasting gains are achieved during the COVID-19 crisis where professional forecasters underestimated the rapidly evolving inflationary environment.
- The most recent high-frequency signal alone suffices for updating nowcasts in real-time, making it unnecessary to consider lagged high-frequency information.
Discover related articles
DNB uses cookies
We use cookies to optimise the user-friendliness of our website.
Read more about the cookies we use and the data they collect in our cookie notice.