Guinevere Analytics
19 min read analyticssingaporemobilityurban-economicstransporte-hailing

The Quota Problem: Car-to-Ride-Hail Substitution in Singapore

The third market in the series, and the one where the model's divergence condition actually holds — driving a car you already own to the Singapore CBD costs more at the margin than the Grab fare for the same trip. Which raises a harder question: in a system where the fleet is a policy variable, what does individual substitution even do?

Two posts in this series have now run the same substitution model on two markets. Kuala Lumpur concluded that commuters are not switching, because parking is cheap and long car loans make retention rational. Jakarta concluded that they are not switching either, because a motorcycle costs Rp 222 a kilometre and no regulated tariff can compete with that.

Both posts identified the same theoretical escape hatch. The breakeven trip count is

N* = F / (g − m)

where F is the fixed monthly cost of the vehicle, m the marginal cost of one one-way trip in it, and g the ride-hail fare for the same trip. As m rises toward g the breakeven diverges, and once m > g there is no trip count at which driving is cheaper. The KL post noted that this happens somewhere above RM 40 a day of parking and observed that no Malaysian authority had gone near it.

Singapore went near it in 1975 and has not stopped since.

This post runs the model on Singapore, finds that the divergence condition holds, and then runs into the more interesting problem: Singapore fixes the size of its vehicle fleet by quota. In a quota system, an individual household's decision to give up its car has a fleet effect of exactly zero — the Certificate of Entitlement returns to the pool and someone else buys it. The entire question this series has been asking becomes, in Singapore, a question about prices rather than quantities.

Part 1: The Aggregate Picture

The fleet is a policy variable.

IndicatorValue
Vehicle growth rate0% for all categories except Cat C goods vehicles/buses (0.25%), since Feb 2018
Current policy horizonmaintained to 31 January 2028
Total motor vehicles (Oct 2025)~1,009,811
Cars566,527
Taxis (Dec 2025)12,161
Private hire cars (Dec 2025)62,092

COE premiums are at record highs.

CategoryRecent premium
Cat A (≤1,600cc, ≤97kW)S$128,501 (Aug 2026, 2nd bidding) — near the all-time high of S$129,000 set in July 2026
Cat B (>1,600cc or >97kW)S$131,001, having crossed S$130,000 for the first time in July 2026

Public transport is among the best in the world and its modal share is barely moving.

IndicatorValue
Average daily bus ridership (2025)~3,841,000
Average daily MRT ridership (2025)~3,490,000
Peak-period public transport modal share62% (2016) → 65% (2023)
Target75% by 2030
Walk-Cycle-Ride peak share71% (2016) → 74% (2022); target 9 in 10 by 2040
MRT networkexpanding to 360 km by the 2030s

That modal share line deserves a pause. Singapore has the region's most expensive car ownership, a functioning congestion charge, a dense and expanding rail network, and four competing ride-hail platforms. Peak public transport modal share moved three percentage points in seven years. Whatever the ceiling on voluntary modal shift is, Singapore is probably close to finding it.

Road pricing is being rebuilt.

ERP 2.0 replaces the gantry network with satellite-based on-board units. Installation is mandatory for all Singapore-registered vehicles by 1 January 2027, and as of May 2026 nearly all vehicles had transitioned. Critically, LTA has stated there will be no distance-based charging in the initial rollout — the capability exists, the policy does not yet. That distinction matters enormously for everything below, and I come back to it.

The taxi fleet was displaced, exactly as in Malaysia.

12,161 taxis against 62,092 private hire cars. Same pattern as Kuala Lumpur's 120,000 → 40,000 collapse: ride-hail substitutes near-perfectly for taxis and imperfectly for private cars. Three markets, three confirmations of the same asymmetry.

Part 2: Parameters

A mass-market Category A car bought at 2026 prices.

PETROL_95   = 3.05        # S$/litre
COE_CAT_A   = 128_501     # Aug 2026 2nd bidding
CAR_PRICE   = 190_000     # all-in: OMV + ARF + COE + dealer margin
PARF_REBATE = 10_000      # ~50% of ARF, recovered on 10-year deregistration
COE_YEARS   = 10

depreciation = (CAR_PRICE - PARF_REBATE) / (COE_YEARS * 12)   # S$1,500/month
ComponentMonthly
Depreciation (S$180,000 over 10 years)S$1,500.00
Road tax + insurance + maintenanceS$295.00
Home season parking (HDB, sheltered, 1st car)S$110.00
Fixed cost FS$1,905.00
Marginal cost, per one-way 12 km CBD trip
Petrol (15 km/l at S$3.05)S$2.44
ERP (~S$3/day)S$1.50
CBD parking (~S$25/day)S$12.50
Marginal cost mS$16.44

Ride-hail, from the GrabCar rate card plus the platform fee that rose to S$1.20 a trip in January 2026:

def grab_fare(km, surge=1.0):
    return (4.00 + 0.83 * km) * surge + 1.20     # 12 km -> S$15.16

S$15.16 for 12 km sits at the bottom of the S$15–22 band reported for 8–12 km trips in early 2026, which is the right place for a no-surge estimate.

Part 3: The Divergence Condition Holds

Marginal cost of driving a car you already own, 12 km to the CBDS$16.44
Ride-hail fare for the identical tripS$15.16
Difference+S$1.28
Ratio (fare ÷ own-car marginal cost)0.92×

m > g. Not by much, but the direction is what matters. In Singapore, a person who already owns a car, has already paid the COE, and has the vehicle sitting in the carpark downstairs is still better off leaving it there and booking a Grab — before counting the depreciation, road tax, insurance and season parking they are paying regardless.

This is the condition the KL model said would end the argument, and no Southeast Asian city except Singapore has produced it. It is not an accident of fuel prices. It is ERP plus CBD parking, which together contribute S$14.00 of the S$16.44 — 85% of the marginal cost of driving into the Singapore CBD is policy, not petrol.

Compare the three markets on exactly this ratio:

MarketFare ÷ own-vehicle marginal cost
Jakarta — ojol vs own motorcycle8.27×driving wins overwhelmingly
Kuala Lumpur — Grab vs own car2.05×driving wins
Jakarta — taksi online vs own car1.96×driving wins
Singapore — Grab vs own car0.92×ride-hail wins

Part 4: Scenarios

At N = 74 one-way trips a month, 12 km each, carried over from the previous two posts for comparability:

ScenarioFmfareN*CarRide-hailGap
A CBD commuter, ERP + paid parking1,90516.4415.16never3,1221,122−2,000
B free workplace parking1,9053.9415.16169.82,1971,122−1,075
C no ERP, no paid parking1,9052.4415.16149.82,0861,122−964
D CBD commuter, 1.3× surge1,90516.4419.35655.13,1221,432−1,690
E car fully depreciated (year 10)40516.4415.16never1,6221,122−500

Every single row is negative. Ride-hail is cheaper in all of them, and in the two rows where the driver actually pays for ERP and parking there is no trip count that reverses it.

Scenario E is the one that separates Singapore from the other two markets most sharply. In Kuala Lumpur, a paid-off car was unbeatable — N* fell to 21.5 and the car won by RM 673 a month. In Jakarta, a paid-off motorcycle was unbeatable by a factor of four. In Singapore, a car at the end of its COE life, with the depreciation fully absorbed, still loses by S$500 a month, because the marginal cost of using it exceeds the fare.

Put the other way round: the S$3,121.56 a month that a Singapore CBD commuter spends on a car buys 206 Grab trips a month — 6.9 rides a day, every single day, including weekends.

The car-lite basket

ComponentMonthly
44 MRT/bus commute legs at ~S$1.75S$77.00
20 Grab tripsS$303.20
4 GetGo car-sharing bookings (5 h + 60 km each)S$230.40
TotalS$610.60

Against the car's S$3,121.56, that is a saving of S$2,510.96 a month — S$30,132 a year.

The car-sharing line is worth noting on its own. GetGo runs 3,000+ vehicles across 1,700+ locations, mostly in HDB carparks, at S$3–10/hour depending on band plus S$0.44/km. That is a genuinely different competitive landscape from Kuala Lumpur or Jakarta, where the equivalent line item was a conventional daily car rental. It fills exactly the gap that made pure ride-hail substitution unattractive in the other two markets: the weekend trip with luggage, children, or an IKEA flat-pack.

Part 5: The Quota Problem

Here is where Singapore breaks the analytical frame the previous two posts used.

In Malaysia and Indonesia, the chain runs: household decides → buys or doesn't buy a car → fleet grows or doesn't. Registration data is a noisy but real measurement of aggregated household decisions.

In Singapore that chain is severed. The vehicle growth rate has been 0% since February 2018. The number of cars is not the sum of household decisions; it is a number the Land Transport Authority sets, and the COE auction allocates the fixed supply to whoever bids most. So:

A Singapore household that gives up its car has a fleet effect of exactly zero. The COE returns to the pool, the quota is unchanged, and the next-highest bidder — who was previously priced out — takes it.

Everything this series has been trying to measure in Malaysia and Indonesia is, in Singapore, structurally unmeasurable in the same units. You cannot detect substitution by watching the car population, because the car population is pinned.

What substitution does instead is change the price. If ride-hail and rail genuinely erode the willingness to pay for a car, that shows up as a falling COE premium, because fewer bidders chase the same fixed supply. The COE premium is the market-clearing price of residual car demand after every substitute has been considered. It is, in principle, the cleanest measurement of modal preference that exists anywhere in the region — a continuously-quoted, publicly-auctioned price for the marginal car.

And it is at a record high. Cat A at S$128,501, Cat B at S$131,001, both setting all-time peaks in July 2026.

The honest caveat, which is a large one: the COE premium is jointly determined by residual demand and by quota supply, and the quota is mechanically driven by the deregistration cycle roughly ten years earlier. A registration trough a decade ago produces a quota trough now, and a quota trough produces high premiums regardless of what demand is doing. Reading demand strength directly off the premium is exactly the error the previous two posts warned about in a different guise.

So the correct statement is narrower and still interesting: premiums at record highs are inconsistent with a collapse in car demand, and consistent either with resilient demand or with a tight quota, and the public data separates those only if you model the quota supply explicitly. That is a tractable piece of work — LTA publishes quota and premium series — and it is the single most valuable analysis anyone could do on Singaporean modal shift. It does not appear to have been done publicly.

Part 6: What the Model Says People Are Actually Buying

If ride-hail is cheaper in every scenario, cheaper than a fully-depreciated car, and cheaper by S$2,000–2,500 a month, then 566,527 Singaporean car owners are not making a cost error at that scale. They are buying something the model does not price.

S$/monthS$/year
Premium over Grab-for-everything1,999.7223,997
Premium over the car-lite basket2,510.9630,132

S$24,000–30,000 a year is the revealed price of car ownership as a product rather than as transport: guaranteed availability with no waiting and no surge, a boot that stays loaded, child seats that stay installed, air conditioning that is already cold, no driver, no conversation, and — in a market where a COE costs more than most countries' cars — a positional good of considerable clarity.

The KL post called this the optionality premium and put it at RM 432 a month. Jakarta's was Rp 3.5 million. Singapore's is S$2,000–2,500. In every market the premium is real; what differs is that Singapore's is large enough, and priced clearly enough by the COE auction, that nobody can pretend the purchase is about getting to work.

This is also why the 65% modal share number has barely moved. The households still buying cars in Singapore are, by construction, those for whom S$2,000 a month is worth paying. Policy that raises the cost of driving prices out the marginal buyer — but the COE auction was already doing that, and the quota means the seat simply passes to the next bidder. In a binding quota system, cost-side policy redistributes car ownership up the income distribution; it does not reduce car use. Only the quota does that, and the quota has been flat since 2018.

Part 7: The Supply Side

Singapore's platform-labour settlement is the most developed of the three markets, and it moves in the opposite direction from Indonesia's.

The Platform Workers Act has been in force since 1 January 2025. It gives ride-hail and delivery platform workers CPF contributions, work-injury compensation, and representation rights. On 1 January 2026 the platform operator's CPF share doubled from 3.5% to 7%, with government transition support covering 75% of the worker's own increase in 2026, tapering to 50% in 2027 and 25% in 2028.

Compare the three approaches:

MarketMechanismWho pays
Malaysia20% commission cap (GSF pilot to 30%); no fare regulationNobody — the market clears below sustainable driver earnings
IndonesiaCommission cut to 8% from 1 July 2026; regulated per-km tariff, held flatThe platform
SingaporeCPF contributions phased in; operator share 7% in 2026The platform and the worker, with the state subsidising the transition

Singapore is the only one of the three that put public money into the transition rather than simply reassigning the cost by decree. It is also the only one where the platform passed the cost straight through to passengers in a visible line item: Grab's per-trip platform fee rose 30 cents to S$1.20 in January 2026, Gojek's rose 20 cents to S$1.10–1.70 in February, while TADA held its fee flat for 2026 and said so publicly.

That pass-through is the thing to watch, and it is directly measurable — platform fees are posted. Every increase raises g and therefore erodes the 0.92× ratio that makes the Singapore result work. At the current margin, a further S$1.30 on the platform fee would push g back above m and restore driving as the cheaper marginal option for a CBD commuter. The Singapore substitution case is real but it is thin, and it is being eroded from the supply side by roughly the same mechanism that is meant to protect drivers.

Part 8: What the Data Can and Cannot Tell Us

Singapore has, by a distance, the best public transport data of the three markets.

What exists: LTA's vehicle population series by type, make, cc rating and fuel; monthly and annual public transport ridership; COE quota and premium series by category and bidding exercise, going back decades; taxi and PHV fleet counts; all published on data.gov.sg or LTA's statistics pages. The COE series in particular is a continuously-quoted price for the marginal car, which no other market in the region has.

What does not exist: trip-level platform data; any link between a household's vehicle and its travel; and — the specific gap that matters here — a public decomposition of the COE premium into quota supply and residual demand.

Three designs, in ascending order of usefulness:

  1. Decompose the COE premium. Model premium as a function of quota supply, interest rates, income, and a modal-substitution proxy (ride-hail and rail supply). LTA publishes quota and premium; the deregistration cycle is mechanical and forecastable a decade out. This is the analysis that would settle whether Singaporean car demand is actually softening underneath a tightening quota. It is runnable today on published data.

  2. ERP 2.0 as a natural experiment — but only if the policy changes. The on-board units are in place and mandatory from 1 January 2027, but LTA has said there is no distance-based charging in the initial rollout. That means ERP 2.0 as currently scoped is an infrastructure swap, not a price shock, and it will identify nothing. If distance-based charging is ever introduced, it would be the cleanest marginal-cost shock any of these three markets could offer, because it raises m continuously rather than at gantries. Worth having the pre-period data ready.

  3. The CPF step-ups as a fare shock. The operator CPF share doubled on 1 January 2026 and the government's transition support tapers on a published schedule through 2028. Each step is a dated, exogenous cost increase to platforms. Tracking platform fees and realised fares against those dates identifies the pass-through rate — how much of a platform-labour cost lands on passengers — which is a number every regulator in the region currently guesses at.

Part 9: Three Markets, One Table

Kuala LumpurJakartaSingapore
Dominant private modeCarMotorcycleCar
Fleet controlNoneNoneQuota, 0% growth since 2018
New car sales 2025820,752, record803,687, −7.2%Fixed by quota
Fare ÷ own-vehicle marginal cost2.05×8.27× (bike)0.92×
Does driving ever win?UsuallyAlmost alwaysNever, with ERP + parking
E-hailing pricingMarket-set, time-heavyState-regulated, per-kmMarket-set + posted platform fee
Congestion chargeNoneERP targeted 2028–29Operating since 1998; ERP 2.0 from 2027
CBD parkingRM 1.50–2.50/hourRp 4,000 first hour~S$25/day
Platform labour20% cap, no fare rules8% cap from July 2026CPF, operator 7% in 2026
Optionality premiumRM ~432/monthRp ~3.5m/monthS$2,000–2,500/month
Binding constraintCheap parking, long loansMotorcycles are just cheapThe quota, not the price

Read across the "fare ÷ marginal cost" row and the series has a clean spine. The economics of car-to-ride-hail substitution are determined almost entirely by how much a city charges its drivers for the marginal trip. Jakarta charges almost nothing to a motorcyclist and gets 19.5 million motorcycles. Kuala Lumpur charges a little and gets record car sales. Singapore charges S$14 in ERP and parking on a 12 km CBD trip and gets the only ratio below 1.0 in the region.

And yet the modal share needle has moved three points in seven years, because Singapore solved the pricing problem and then discovered the second problem: once you have priced driving correctly, the people still driving are the ones who can afford it, and there is no further price that moves them. Pricing gets you to the households that are cost-sensitive. It does not get you past them.

That is not an argument against the pricing. Singapore's 65% peak modal share is extraordinary by any comparative standard, and it was bought with exactly these instruments. It is an argument that Kuala Lumpur and Jakarta should expect the same ceiling — and that the last stretch toward 75% will not come from making cars more expensive, because in Singapore cars are already about as expensive as it is possible to make them.

Part 10: Limitations

The car price is assumed; only the COE is observed. S$190,000 all-in for a Cat A car in 2026 is a reasonable figure given a S$128,501 COE, but OMV, ARF and dealer margin vary. A S$170,000 car drops F to about S$1,738 and changes none of the signs.

Straight-line depreciation is a simplification. Real depreciation is front-loaded, and the PARF/COE rebate structure at deregistration is more complex than a flat S$10,000. The model's F is best read as an average over the COE life, not a year-one figure.

The 0.92× result depends on the driver actually paying ERP and market-rate CBD parking. Scenario B — free workplace parking — takes m to S$3.94 and restores a finite breakeven at N* = 169.8. Ride-hail still wins at 74 trips, but the "never" disappears. Employer-provided parking is as much of an anti-transition subsidy in Singapore as the KL post found it to be in Malaysia; it is simply rarer.

Grab fares come from a rate card, not from realised transactions. Surge is real, frequent, and unmodelled outside scenario D. Realised average fares are above S$15.16.

Trip counts are assumed. N = 74 is carried over for comparability across the three posts. No published Singapore travel survey validates it, and Singapore's high public transport share suggests car owners here may make fewer car trips than their KL counterparts, which would push further in ride-hail's favour.

The quota argument is about the fleet, not about usage. Individual substitution has zero effect on the number of cars. It does reduce vehicle-kilometres if the substituting household drives less before selling, and a ride-hail vehicle is more intensively used than a private one. Nothing here claims substitution is pointless — only that its effect cannot be read off the car population.

Closing

Singapore is the market where the arithmetic finally favours giving up the car, and it is the market where giving up the car changes the fleet least.

The marginal cost of driving a car you already own into the Singapore CBD is S$16.44 against a S$15.16 Grab fare, and 85% of that S$16.44 is ERP and parking — policy, not petrol. There is no trip count at which driving is cheaper. A fully depreciated car still loses by S$500 a month. The money a CBD commuter spends on a car buys nearly seven Grab rides a day.

Yet 566,527 cars remain, the COE premium is at a record, and peak public transport modal share has moved three points since 2016. Two things explain that, and only one of them is about behaviour. The households still buying are paying S$24,000–30,000 a year for availability, boot space and position, and have decided it is worth it. And the quota means that even when one of them changes their mind, the certificate simply moves to whoever was next in line.

For Kuala Lumpur and Jakarta, the lesson runs both ways. Singapore proves the pricing works: charge properly for the marginal trip and the substitution case flips, which no amount of app convenience achieved in the other two cities. It also shows where the pricing stops working, and that the remaining distance to a car-lite city has to be covered by something other than a higher price.

The cost model, fare calibration and premium estimates are the author's own, built from published prices, COE results and reported figures as of September 2026. Scenario outputs are model results, not survey findings; the car price, trip frequency and parking assumptions are labelled above. Reported figures are attributed in-line. This is analysis, not financial or policy advice.

Sources