## Lessons from Recent Experiences in Macroeconomic Forecasting

## Source details

**Canonical URL:** [Lessons from Recent Experiences in Macroeconomic Forecasting](https://www.imf.org/-/media/files/news/speech/2023/lessons-from-recent-experiences-in-macroeconomic-forecasting.pdf)

## Other formats

- [Markdown version](/-/media/files/news/speech/2023/lessons-from-recent-experiences-in-macroeconomic-forecasting.pdf.md)
- [Structured JSON version](/-/media/files/news/speech/2023/lessons-from-recent-experiences-in-macroeconomic-forecasting.pdf.json)

---

### Forecast accuracy and horizon
- Forecast accuracy falls rapidly as horizon lengthens (chart: Median Root Mean Squared Errors of WEO Real GDP Growth Forecasts, Advanced Economies).
- Forecast categories referenced: "Current year, Autumn"; "Current year, Spring"; "1-year ahead, Autumn"; "1-year ahead, Spring"; "2-year ahead, Autumn"; "3-year ahead, Autumn"; "4-year ahead, Autumn"; "5-year ahead, Autumn".

### Breakdown of standard macro relationships in 2020–22
- Household income and expenditure decoupled: series shown as Real disposable incomes (index, 2015=100) and Real final consumption expenditure (index, 2015=100) covering 2007Q1 through 2022Q4.
- Energy markets decoupled: Brent Crude oil (index, 2015=100) and TTF natural gas (index, 2015=100) series shown from 2006M1 through 2023M1.
- Period highlighted: 2020-22 characterized by extreme volatility and breakdowns of usual macro relationships.

### IMF forecasting process and aggregation
- The IMF’s World and Regional (including European) forecasts are aggregated up from individual country team forecasts.

### Co-movement and large shocks
- Co-movement of forecast errors increases with large shocks (based on IMF World Economic Outlook Database).
- Forecast errors defined as: the annual GDP forecast made in autumn of year t-1 for year t, minus the actual GDP growth outturn in year t.
- Box constructed around the 25th to 75th percentile of the country forecast error distribution.

### The 2020 experience and a more centralized approach
- The April 2020 WEO captured 2020 GDP dynamics reasonably well when based on a more centralized approach.
- Average forecast errors over 2011-19 are reported as the average of the absolute value of the difference between forecast and actual turnout (no numeric value provided in source).

### Inflation forecast errors and model limitations
- Stark forecast errors for inflation occurred amidst a sequence of shocks (IMF World Economic Outlook Database).
- Forecast errors for inflation defined as: the annual inflation forecast made in autumn of year t-1 for year t, minus the actual inflation outturn in year t.
- Model granularity and non-linearities issues:
  - The inflation model lacked some granularity (e.g., for energy) and did not capture certain non-linearities.
  - A pseudo out-of-sample projection over 2020Q1-2023Q1 of McGregor and Toscani (2022) estimated on data until 2019Q4 is compared with an alternative projection removing wholesale electricity and gas from the inflation model (so energy prices enter projections purely through Brent crude oil prices).
  - Time span shown for actual and model projections: 2018Q1 through 2023Q1.
  - Lines shown: Actual headline inflation; Model with electricity and gas; Model without electricity and gas.

### Key lessons and policy recommendations
- 1) Balanced – need to get the balance between top down and bottom up right. An increased role for central guidance will likely stay with us at the IMF.
- 2) Nimble – continuously monitor and enhance tools, exploit underused data sources including big data, and have a constant awareness of what information the models do and do not observe and be ready to apply well-grounded judgement.
- 3) Modest – forecasting is a humbling task; point estimates are necessary but cannot be the only focus, scenario and risk analysis is important, the objective should be to avoid forecasts that would yield gross policy mistakes.

*Source: Lessons from Recent Experiences in Macroeconomic Forecasting — JUNE 2023, Alfred Kammer, IMF (PDF).*

---


_Source: https://www.imf.org/-/media/files/news/speech/2023/lessons-from-recent-experiences-in-macroeconomic-forecasting.pdf_