Gravity Model

What It Does

The Gravity Model estimates the total potential flow of diners toward this location from population rings at increasing distances. Think of it as: how many person-visits per day could realistically arrive here given population, income, and distance?

Unlike the Huff model (which calculates our market share), the Gravity model measures the raw demand pipeline. Multiply flow by Huff captureProb to get actual covers.

Core Formula

flow_zone = k × Oᵢ × Dⱼ / distM^β

Where:

• Oᵢ = origin mass (zone population × income match × dining frequency)
• Dⱼ = destination attractiveness (effective area × cuisine score × income match)
• distM = midpoint distance of zone in metres
• k = scaling constant (varies by floor area)
• β = distance decay exponent (varies by service model / price band)

Current Implementation

Scaling Constant

k = max(0.0008, 0.003 - (floorArea - 300) × 0.000002)

Examples: 300 sqft → k=0.003; 800 sqft → k=0.002; 1200 sqft → k=0.0014.

Larger venues get a lower k to prevent unrealistic flow numbers — physical capacity caps actual throughput.

Destination Attractiveness (Dⱼ)

effectiveArea = max(400, floorArea)
cuisineScore = min(2, effectiveArea / 500)
Dⱼ = effectiveArea × cuisineScore × incomeMatch

The effective area floor of 400 means even small budget restaurants (e.g. a 200 sqft cha chaan teng) get reasonable gravity pull — they attract as much foot traffic as a larger one; they just can’t serve the overflow.

Income Match Factor

Price band	Low income (<18–22K)	Mid income	High income
Budget	1.2	1.0	0.8
Mid	0.8	1.0	1.0
Premium	0.7	1.0	1.2
High-end	0.5	1.0	1.3

Gravity Beta (Distance Decay within Zones)

Separate from the Huff beta — this controls how steeply flow falls off within each ring:

Service / price	gravityBeta
Delivery	0.8
Takeaway	0.7
High-end	0.6
Premium	0.7
Mid	0.8
Budget	0.9

For delivery and takeaway, zones within 1400 m (delivery) or 600 m (takeaway) get an additional effectiveBeta × 0.7 boost.

Population Zones

Zone	Midpoint used	Primary data source
0–400 m (5 min walk)	200 m	STPU census zone cumulative ring
400–800 m (10 min walk)	600 m	STPU ring differential
800 m–2 km (transit)	1400 m	STPU ring differential
2 km+ (destination)	3500 m	STPU 5 km ring minus 2 km ring

When STPU data is unavailable, falls back to density × π × r² × coverage factor.

Zero-population zones: If a ring has pop = 0, flow = 0 is returned (previously returned flow = 1).

Minimum floor: Zones with pop > 0 always return at least flow = 1.

Data Sources

• Population: Census 2021 STPU (211 zones), aggregated into distance rings
• Income: STPU weighted median household income
• Dining frequency: constant 15% daily dine-out rate (HK average)

Tip

Competition is not applied in the Gravity model — that’s handled entirely by Huff. Gravity measures the total demand pool; Huff determines our slice of it.

Interpretation

estimatedFlow	Meaning
5,000+	Dense urban core with high foot traffic potential
1,000–5,000	Typical urban neighbourhood
200–1,000	Suburban or low-density area
< 200	Remote or very low-density site

Known Limitations

• Assumes uniform population distribution within each ring (STPU zones reduce this error significantly)
• Dining frequency (15%) is a Hong Kong average — varies by day of week and season
• No distinction between lunch and dinner demand patterns
• Worker/commuter population not separately modelled here (handled in KPI estimates)

Changelog

Date	Change
2026-03-25	Zero-population rings now return flow=0 instead of flow=1
2026-02-10	Initial implementation with STPU ring data integration