Spatial Autocorrelation

Analysis Date: 2026-02-06
Data Period: 2021-2026
Coverage: HDB, EC, and condominium appreciation patterns

Key Takeaways

The clearest finding

Appreciation is spatially clustered. Nearby neighborhoods tend to move together strongly enough that location context becomes a major part of short- to medium-term performance.

What this means in practice

• Investors should treat neighborhood cluster status as a real risk and return factor.
• Homebuyers should not evaluate units in isolation from their surrounding submarket.
• Upsiders and catch-up plays exist, but not every lagging area converts into a hotspot.

Core Findings

1. Neighborhood effects are statistically strong

This means appreciation is not randomly distributed. Nearby areas tend to share price momentum.

2. Hotspot and lagging clusters differ in performance

Impact

• Hotspot neighborhoods have a measurable appreciation edge.
• LH areas deserve case-by-case analysis; some are opportunities, some are structural laggards.

3. Spatial dependence is stronger in some segments than others

Impact

• Condo buyers are especially exposed to neighborhood momentum and peer pricing effects.
• HDB still shows strong neighborhood dependence, though somewhat less than condo.

Decision Guide

For investors

• Start with neighborhood classification, then evaluate the unit.
• Use HH clusters for stability and LH clusters only where there is a clear catalyst or discount.

For first-time buyers

• A good unit in a weak micro-market can still underperform a slightly less attractive unit in a stronger surrounding cluster.

For upgraders

• Cluster effects matter most when your hold period depends on future resale strength.

Technical Appendix

Data Used

• Primary input: data/parquets/L1/housing_hdb_rental.parquet
• Spatial aggregation: H3 hexagonal grid at resolution 8 (~0.5 km² cells)
• Aggregation method: median monthly_rent per H3 cell
• Minimum threshold: ≥10 records per cell for analysis

Methodology

• Global Moran’s I: KNN weights (k=8), row-standardized, 99 permutations
• Local LISA (Local Indicators of Spatial Association): Moran_Local with 99 permutations
• Cluster classification: HH (high-high), LL (low-low), HL (high-low), LH (low-high), NS (not significant)
• Significance threshold: p ≤ 0.05
• Spatial lag correlation: computed per property type (Condo, HDB, EC) to measure neighborhood dependence

Technical Findings

• Global Moran’s I: 0.766 (strong positive spatial clustering)
• Z-score: 9.91, p-value < 0.001 — highly significant
• Cluster distribution:
  • HH (hotspot): 16 cells, 12.7% YoY appreciation
  • LH (lagging in strong neighborhood): 17 cells, 11.3% YoY
  • LL (coldspot): 1 cell, ~10% YoY
  • Not significant: 8 cells, 12.0% YoY
• Spatial lag correlation by segment: Condo 78%, HDB 71%, EC 65%
• Transition analysis: moderate persistence — hotspot and coldspot status are sticky but not permanent

Conclusion

The Moran’s I of 0.766 with z=9.91 confirms that appreciation is far from randomly distributed; nearby areas share price momentum strongly. The HH cluster premium (12.7% YoY vs ~10-12% elsewhere) is modest but statistically robust. LH clusters (17 cells) represent the most analytically interesting group — potential catch-up opportunities, but also potential structural laggards requiring case-by-case assessment. The stronger spatial dependence in condos (78%) vs HDB (71%) vs EC (65%) suggests that condo pricing is more influenced by peer neighborhood performance. Key limitations: cluster labels summarize local context and do not override unit-specific factors; spatial patterns can shift with supply, policy, or infrastructure changes.

Scripts

• scripts/analytics/analysis/spatial/analyze_spatial_autocorrelation.py — Moran’s I + LISA
• scripts/analytics/analysis/spatial/analyze_h3_clusters.py — H3 cluster classification