Price Appreciation Prediction
Analysis Date: 2026-02-17
Data Period: 2021-2026
Coverage: HDB, condo, and EC segments
Key Takeaways
The clearest finding
Forecast quality depends far more on market segment than on having a single “best model.” HDB, EC, and mass-market condos are reasonably forecastable; luxury condos are not forecastable with enough precision for valuation use.
What this means in practice
- HDB buyers and upgraders can use forecasts as a timing aid.
- Mass-market condo buyers can use them as directional support, but should still expect wider error bands.
- Luxury condo buyers and investors should treat forecast outputs as weak signals, not pricing anchors.
Core Findings
1. Reliability differs sharply across segments
| Segment | Accuracy (R²) | 95% CI Width | Directional Acc | Practical read |
|---|---|---|---|---|
| HDB flats | 79.8% | ±18.58% | 99.4% | Strong |
| Executive Condos | 98.5% | ±50.15% | 97.1% | Strong but wider uncertainty |
| Mass Market Condos | 85.6% | ±40.48% | 96.4% | Useful with caution |
| Mid Market Condos | 72.6% | ±1877.88% | 94.2% | Direction only |
| Luxury Condos | 30.1% | ±1076.24% | 92.3% | Magnitude unusable |
2. Direction is often easier to predict than magnitude
The directional hit rate remains high even when confidence intervals are too wide for precise action.
Impact
- If the question is “up or down?”, some segments are usable.
- If the question is “how much?”, only the stronger segments deserve much trust.
3. Momentum is the dominant forecast driver
| Segment | Top feature | Importance |
|---|---|---|
| HDB | 2-year YoY appreciation | 51.14% |
| Mass market condo | 2-year YoY appreciation | 65.50% |
Impact
- Short- to medium-term price forecasting is still driven mainly by recent trend persistence.
- Amenities and local features matter, but they are not the main short-horizon forecasting signal.
How To Use These Forecasts
For investors
- Use HDB and mass-market condo forecasts to support entry and exit timing.
- Treat wide confidence intervals as a stop signal, not a footnote.
- Ignore precise appreciation percentages in luxury segments.
For first-time buyers
- HDB forecasts can help with timing if your move window is flexible.
- Avoid making affordability decisions based on aggressive upside forecasts.
For upgraders
- A forecast is most useful when both the sell-side and buy-side segments are forecastable.
- If upgrading into a noisy condo tier, do not assume your next purchase has the same predictability as your current HDB.
Technical Appendix
Data Used
- Primary input:
data/parquets/L3/housing_unified.parquet(2021-2026) - Train/test split:
L5_price_appreciation_train.parquet/L5_price_appreciation_test.parquet - Forecast horizon: 6 months (ARIMA), annual (XGBoost ensemble)
- Feature selection: numeric only, max 20% missing values allowed
Methodology
- XGBoost models trained per property type (HDB, Condo, EC) in
scripts/analytics/price_appreciation_modeling/train_by_property_type.py- Parameters: n_estimators=100, max_depth=6, learning_rate=0.1, subsample=0.8
- Target variable:
yoy_change_pct(year-over-year appreciation) - Smart ensemble: stacked model combining segment-specific predictions via
scripts/analytics/price_appreciation_modeling/create_smart_ensemble.py - Confidence intervals: calibrated per segment using residual distribution in
scripts/analytics/price_appreciation_modeling/generate_confidence_intervals.py - ARIMA(1,1,1) for time-series forecasting per planning area via
scripts/analytics/analysis/market/forecast_prices.py - Condo sub-segmentation: mass market, mid market, luxury in
scripts/analytics/price_appreciation_modeling/train_condo_by_segment.py
Technical Findings
| Segment | R² | Directional Accuracy | 95% CI Width |
|---|---|---|---|
| HDB | 79.8% | 99.4% | ±18.58% |
| Executive Condos | 98.5% | 97.1% | ±50.15% |
| Mass Market Condos | 85.6% | 96.4% | ±40.48% |
| Mid Market Condos | 72.6% | 94.2% | ±1877.88% |
| Luxury Condos | 30.1% | 92.3% | ±1076.24% |
- Top feature: 2-year YoY appreciation (51.14% importance for HDB, 65.50% for mass-market condo)
- Ensemble accuracy: 74% vs 47% for a unified one-size-fits-all model
- ARIMA: 20 planning areas with ≥12 months of data, 6-month forecast horizon
Conclusion
Segment-specific modeling dramatically outperforms unified approaches (74% vs 47% accuracy). HDB and EC segments are the most forecastable; luxury condos are essentially unpredictable in magnitude (R²=30.1%, CI width ±1076%). Momentum (recent 2-year appreciation) dominates feature importance, which means these models are vulnerable to regime changes and policy shocks. High directional accuracy can coexist with impractically wide error bands, especially in mid/luxury condo segments. Key limitations: momentum-heavy models assume trend persistence, and the luxury segment lacks sufficient signal for reliable magnitude forecasts.
Scripts
scripts/analytics/analysis/market/forecast_prices.py— ARIMA(1,1,1) per planning areascripts/analytics/price_appreciation_modeling/train_by_property_type.py— Segmented XGBoostscripts/analytics/price_appreciation_modeling/create_smart_ensemble.py— Ensemble stackingscripts/analytics/price_appreciation_modeling/generate_confidence_intervals.py— CI calibration