Key Investment Findings
Last Updated: 2026-03-31 Data Coverage: 911,797 transactions (2017-2026) Primary Analysis Window: 2021-2026
Key Takeaways
The clearest finding
The market premium often attributed to MRT access is mostly a city-access premium. CBD distance explains 22.6% of price variation on its own, while adding MRT access increases explanatory power by only 0.78 percentage points.
What matters most in practice
- HDB buyers should not overpay for MRT proximity alone. Lease remaining, affordability, and neighborhood quality matter more.
- Condo buyers and investors should still care about MRT access. Condo prices are roughly 15x more MRT-sensitive than HDB prices.
- Timing and segment matter. Policy effects, lease decay, and forecast reliability differ materially across HDB, condo, and EC markets.
- Amenity mix matters. Hawker proximity outranks MRT as a pricing feature for HDB. Parks are underpriced. Multi-amenity hubs command synergistic premiums.
- Macro conditions matter. Interest rates (SORA) are the most actionable macro signal for housing timing. Real returns are modestly positive across all segments.
- Yield and growth are separate strategies. High-rental-yield areas tend to have lower appreciation. Pick one or balance deliberately.
Recommended decision order
- Start with behavioral segment, not property type label. See
market-segments.md. - Check city access before station access.
- For HDB, focus on lease, affordability, and daily convenience (hawker, park). See
amenity-impact.md. - For condos, treat MRT proximity as a real pricing factor and check for facility differentiators. See
facility-premiums.md. - Use forecasts selectively: HDB and mass-market condo models are more reliable than luxury condo models.
- Check macro conditions before timing a purchase. SORA trends matter more than GDP headlines. See
macro-sensitivity.md. - Decide your yield vs growth strategy before picking an area. See
rental-yields.md. - Verify that premiums are still valid. COVID shifted amenity weights. See
temporal-evolution.md.
The Findings That Matter Most
1. CBD access explains more than MRT access
The most decision-useful location result is simple: centrality matters more than station distance.
| Model | R² | Main interpretation |
|---|---|---|
| CBD only | 0.2263 | City access explains a meaningful share of pricing |
| CBD + MRT | 0.2341 | MRT adds only a modest incremental lift |
| Full model | 0.4977 | Broader housing features still matter materially |
Impact
- For HDB, paying a large premium purely for being a few hundred meters closer to MRT is usually hard to justify.
- For condos, MRT still matters, but as part of a broader accessibility package rather than as a standalone rule.
2. MRT sensitivity is highly segment-specific
The headline average hides a sharp split by property type.
| Property Type | MRT premium per 100m | Takeaway |
|---|---|---|
| HDB | about negative 1 to 5 dollars | Small effect |
| Condominium | about negative 19 to 46 dollars | Large effect |
| EC | roughly plus 6 to negative 37 dollars | Unstable effect |
Impact
- Investors targeting rental demand should prioritize MRT proximity for condos, not HDB.
- Owner-occupiers in HDB should treat MRT access as a convenience factor, not the main valuation anchor.
- EC buyers should be cautious about using historical MRT premiums as a stable guide.
3. Lease decay is real, but not linear
Remaining lease affects HDB prices, but not in a smooth straight line. The sharpest pricing pressure appears in the 70-80 year lease band.
| Lease Band | Discount vs 90+ yrs | Transactions |
|---|---|---|
| 80-90 years | -13.2% | 29,562 |
| 70-80 years | -21.9% | 47,044 |
| 60-70 years | -23.8% | 54,521 |
| <60 years | -15.0% | 41,595 |
Impact
- Buyers can often find value in the 60-70 year range if financing and holding period fit.
- Sellers of 90+ year leases should recognize they are selling into the market’s most favored band.
- Straight-line depreciation assumptions are not reliable enough for pricing decisions.
4. Forecast quality varies more by segment than by model brand
The important question is not “is there a model?” but “which market is being modeled?”
| Segment | R² | Directional accuracy | Practical use |
|---|---|---|---|
| HDB | 79.8% | 99.4% | Useful for timing and trend direction |
| EC | 98.5% | 97.1% | Strong, but sample is smaller |
| Mass market condo | 85.6% | 96.4% | Useful with caution |
| Luxury condo | 30.1% | 92.3% | Trend only, magnitude unreliable |
Impact
- HDB and mass-market condo forecasts can inform planning.
- Luxury condo forecasts should not be used as precise valuation tools.
5. Policy shocks do not transmit evenly across the market
Cooling measures appear to have worked in prime condo markets, but not in HDB.
| Policy question | Observed effect |
|---|---|
| Sep 2022 CCR condo cooling | about -$137,743 relative effect vs OCR |
| Dec 2023 HDB cooling | about +$13,118 immediate jump |
Impact
- Policy headlines are not enough. Buyers need to ask which segment is exposed.
- HDB buyers should be careful about delaying purchases based on a generic “prices may cool” narrative.
6. Hawker proximity outranks MRT for HDB pricing
The strongest non-structural pricing feature for HDB is not MRT access but hawker centre proximity (27.4% feature importance vs 5.5% for MRT). Parks are the most underpriced amenity driver (7.2% importance).
Impact
- HDB buyers should prioritize daily convenience (hawker, park, supermarket) over transit labels.
- Multi-amenity hubs (MRT + mall + hawker nearby) command synergistic premiums that exceed individual amenity effects.
See amenity-impact.md for full analysis.
7. Behavioral segments cut across property type labels
Clustering reveals five behavioral segments (High-Growth HDB, Stable Mid-Tier, Premium Condo with Low Yield, High-Yield Apartment, Luxury) that are more decision-useful than the HDB/Condo/EC split.
Impact
- A “Growth Play” HDB in a developing town can outperform a “Yield Play” condo.
- Investors should pick segments based on risk-return profile, not just property type.
See market-segments.md for full analysis.
8. Premiums are not static — COVID created a structural break
MRT premiums declined during COVID and have only partially recovered. School premiums proved more resilient. Pre-COVID premium estimates are unreliable for current decisions.
Impact
- Buyers should check whether a premium is strengthening or eroding before paying for it.
- Areas on accelerating trajectories (developing towns) may offer better future appreciation than currently expensive areas.
See temporal-evolution.md for full analysis.
9. Interest rates are the most actionable macro signal
SORA changes have an immediate correlation with housing affordability and a 1-quarter lag with price trends. GDP growth correlates more with transaction volume than price level.
Impact
- Rate stabilization or decline is a favorable entry signal. Rising rates compress yields and reduce transaction volume.
- Real (inflation-adjusted) returns are approximately 2% above CPI across all segments — housing preserves wealth but is not a high-growth real asset.
See macro-sensitivity.md for full analysis.
10. Yield and appreciation are separate strategies
Rental yield and price appreciation are weakly negatively correlated. High-yield areas tend to have lower capital growth.
Impact
- Income-focused investors should target 1-2 room flats in non-central areas for yields above 5%.
- Growth-focused investors should accept lower yields (3-4%) in high-appreciation areas.
- A balanced approach (moderate yield + moderate growth) often delivers the best risk-adjusted total return.
See rental-yields.md for full analysis.
11. Most condo facilities are table stakes, not differentiators
Swimming pools, gyms, and security are near-universal and do not command pricing premiums. Tennis courts and sky gardens are the genuine differentiators.
Impact
- Condo buyers should not pay extra for standard facility lists.
- Properties with at least one strong differentiator retain value better than those with many standard facilities.
See facility-premiums.md for full analysis.
So What Should Different Buyers Do?
Investors
- Use segment-specific rules, not market-wide rules.
- Prioritize MRT access for condos, hawker and park proximity for HDB.
- Treat forecast outputs as stronger in HDB and mass-market condos than in luxury condos.
- Decide your yield vs growth strategy before picking a property. The two are weakly negatively correlated.
- Track SORA trends as your primary macro timing signal.
First-time buyers
- Avoid stretching budget mainly for an MRT label on HDB listings.
- Compare lease, affordability, and town-level context before accessibility premiums.
- Be careful with “school premium” or “future MRT premium” claims unless the unit trade-offs are also clear.
- Check that premiums you are paying for are still valid post-COVID. Many have shifted.
- Use rental yields as a downside floor. Strong rental demand = lower risk even if prices correct.
Upgraders
- OCR and selected RCR locations often provide the best balance between access and price.
- When selling HDB and buying condo, remember that the two segments price accessibility differently.
- Check your current HDB’s rental yield before selling. It may be worth keeping as a rental property.
- For condo upgrades, look for at least one facility differentiator (tennis court, sky garden), not just a long facility list.
Technical Appendix
Data Used
- Full dataset: 911,797 transactions (2017-2026) across HDB, condo, and EC segments
- Primary analysis window: 2021-2026
- Key inputs:
data/parquets/L3/housing_unified.parquet,data/parquets/L1/housing_hdb_transaction.parquet,data/parquets/L1/housing_hdb_rental.parquet - Detailed data sources: see individual analysis documents for per-topic data specifics
Methodology
Each finding draws from a distinct analytical pipeline. The table below cross-references the detailed analysis document and the underlying scripts.
| Finding Area | Analysis Doc | Key Scripts |
|---|---|---|
| CBD vs MRT decomposition | mrt-impact.md | analyze_mrt_impact.py, analyze_cbd_mrt_decomposition.py |
| MRT segment heterogeneity | mrt-impact.md | analyze_mrt_heterogeneous.py, analyze_mrt_by_property_type.py |
| Lease decay and band pricing | lease-decay.md | analyze_lease_decay.py, analyze_lease_decay_advanced.py |
| Price forecast reliability | price-forecasts.md | forecast_prices.py, train_by_property_type.py, create_smart_ensemble.py |
| School quality premium | school-quality.md | analyze_school_impact.py, analyze_school_rdd.py, analyze_school_spatial_cv.py |
| Spatial autocorrelation | spatial-autocorrelation.md | analyze_spatial_autocorrelation.py, analyze_h3_clusters.py |
| Rental hotspots | spatial-hotspots.md | analyze_spatial_hotspots.py |
| Policy causal effects | causal-inference-overview.md | analyze_causal_did_enhanced.py, analyze_rd_policy_timing.py |
| Amenity impact and feature importance | amenity-impact.md | analyze_amenity_impact.py, analyze_feature_importance.py |
| Market segmentation and investment profiles | market-segments.md | market_segmentation.py, market_segmentation_advanced.py, analyze_investment_eda.py |
| Temporal evolution of premiums | temporal-evolution.md | analyze_mrt_temporal_evolution.py, analyze_school_temporal_evolution.py, analyze_appreciation_patterns.py |
| Macro-economic sensitivity | macro-sensitivity.md | fetch_macro_data.py, prepare_timeseries_data.py |
| Rental yields and affordability | rental-yields.md | analyze_hdb_rental_market.py, residual_analysis.py |
| Private property facility premiums | facility-premiums.md | L3 pipeline facility processing |
Technical Findings (Consolidated)
| Topic | Key Metric | Value | Confidence |
|---|---|---|---|
| CBD effect | R² from CBD-only model | 0.2263 | High — robust across specifications |
| MRT incremental lift | ΔR² after adding MRT to CBD model | +0.0078 | High — hierarchical regression |
| Condo vs HDB MRT sensitivity | Relative magnitude | ~15× | High — consistent across OLS and XGBoost |
| Lease steepest penalty | 70-80 yr band discount | -21.9% vs 90+ yr | High — 47,044 transactions |
| Lease deepest discount | 60-70 yr band discount | -23.8% vs 90+ yr | High — 54,521 transactions |
| Pure lease effect | Per extra year (after controls) | +$54.75 PSF | Moderate — hedonic regression |
| HDB forecast accuracy | R² / directional accuracy | 79.8% / 99.4% | High — segment-specific XGBoost |
| Luxury condo forecast | R² / directional accuracy | 30.1% / 92.3% | Low — magnitude unusable |
| Ensemble vs unified | Accuracy improvement | 74% vs 47% | High — out-of-sample |
| School quality OLS | Coefficient | +$9.66 PSF | High — predictive |
| School RDD | Treatment effect at 1 km | -$79.47 PSF | Low — covariate balance failed |
| Spatial clustering | Moran’s I / z-score | 0.766 / 9.91 | High — p < 0.001 |
| Rental hotspot selectivity | 99% confidence cells | 12 of 847 | High — Gi* statistic |
| CCR condo policy effect | DiD estimate (Sep 2022) | ~-$137,743 | Moderate — regime-specific |
| HDB policy response | RDiT jump (Dec 2023) | ~+$13,118 | Moderate — bandwidth-sensitive |
| Hawker feature importance | XGBoost importance for HDB | 27.4% | High — consistent across models |
| Park feature importance | XGBoost importance for HDB | 7.2% | High — underpriced by market |
| Yield vs appreciation correlation | Pearson r by planning area | ~-0.3 | Moderate — weakly negative |
| Median HDB rental yield | Annual rent / resale price | ~4.2% | Moderate — median across types |
| Real CAGR spread over CPI | Nominal CAGR minus inflation | ~2% | Moderate — all segments |
| MRT premium COVID decline | Pre vs post-COVID coefficient | Reduced | Moderate — structural break |
| School premium resilience | Pre vs post-COVID stability | Stable | Moderate — maintained through COVID |
Conclusion
Across all thirteen analysis domains, the most decision-useful findings share a common pattern: segment-level specificity matters far more than model sophistication. HDB and condo markets respond differently to MRT access, lease decay, policy shocks, and forecasting signals. The strongest technical evidence supports the CBD-over-MRT finding (R² decomposition), the non-linear lease decay curve (223K transactions), the forecast reliability gradient across segments (74% ensemble accuracy), and the hawker-over-MRT amenity ranking (27.4% XGBoost importance). The weakest causal claims are around school quality premiums (RDD covariate balance failed) and luxury condo forecasting (R²=30.1%). Key additions from the six new analysis domains: macro sensitivity confirms interest rates as the most actionable timing signal, rental yield analysis reveals the yield-growth trade-off, market segmentation shows behavioral clusters outperform property-type labels, temporal evolution confirms COVID created a structural break in premium structure, and facility analysis identifies tennis courts and sky gardens as the genuine condo differentiators. All findings are strongest for the 2021-2026 market regime. Forecasts are best used as decision support, not standalone valuation.
Scripts
All scripts referenced above are located under scripts/analytics/analysis/. See individual analysis documents for full script paths and methodology details.