## Figure 2.10. Out-of-Sample Forecasting Accuracy

## Source details

**Canonical URL:** [Figure 2.10. Out-of-Sample Forecasting Accuracy](https://www.imf.org/-/media/files/publications/gfsr/2019/april/english/ch2data/charts/figure2-10.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/gfsr/2019/april/english/ch2data/charts/figure2-10.pdf.md)
- [Structured JSON version](/-/media/files/publications/gfsr/2019/april/english/ch2data/charts/figure2-10.pdf.json)

---

### Scope and models compared
- Compares in-sample and out-of-sample forecasts for one-year-ahead real house price growth.
- Models:
  - Real house price baseline model from Figure 2.5 (green lines).
  - Out-of-sample model estimated using recursive estimation (red lines).
- Application:
  - Uses a country-specific model instead of the panel model for the out-of-sample analysis.
  - Shown cases:
    - United States: results reported for the 20th percentile.
    - China: results reported for the 25th percentile.

### Key quantitative markers (as shown on the chart axes)
- Axis tick values displayed: –20, –15, –5, –10, 0, 5, 10, 15
- Time labels shown: 1990 94 98 2002 06 10 14 18

### Main findings and interpretation
- Out-of-sample predictions closely track the in-sample estimates for one-year-ahead house prices at risk.
- Using a country-specific model reduces the degrees of freedom of the estimation and the robustness of the results when applied to low quantiles of the distribution; this motivated showing results for the 20th and 25th percentiles for the United States and China, respectively.

*Source: IMF staff calculations.*

---


_Source: https://www.imf.org/-/media/files/publications/gfsr/2019/april/english/ch2data/charts/figure2-10.pdf_