Hong Kong petrol, 30 April 2020 to October 2026
Hong Kong's five petrol brands posted the same Standard pump price on 87.6% of the last 2,373 days. When one changes its price, the others follow within about 2 days. Increases outnumber cuts 110 to 33, and of the HK$12.94 a litre that the posted price rose between the first and last year of the record, HK$10.27 went to margin. Cost rises reached the pump faster than cost falls (30% against 7% in the first week, p = 0.009). This write-up runs the standard screens for coordinated pricing on that record, with Perth and Los Angeles beside it.
of 143 price changes were matched by all five brands within a week, usually within 2 days. All five moved on the same day in 1.4%.
increases to cuts. In 2023 and 2024 there were 17 increases and no cuts, while import costs fell.
of the HK$12.94 rise in the posted price went to margin over import cost and duty (79%). The petrol itself accounts for HK$2.67.
The question
In Perth or Los Angeles, stations a few hundred meters apart often post prices tens of cents a litre apart. In Hong Kong the five brands that run nearly every station post one price for days or weeks at a time. A small market where every seller pays the same import cost and every price is public can settle there with no agreement at all, because any cut would be matched at once. Coordination produces the same picture. Competition law treats the two very differently, and the tools for telling them apart are screens: tests of whether prices move the way costs and independent decisions would move them.
Parallel pricing, in the Commission's words, “reveals very little about whether firms are competing or coordinating.”
Competition Commission, Hong Kong, auto-fuel market study, 2017
So the identical prices are the starting observation, and the screens look at what comes with them: who moves first, which way prices move, what happens to the margin over cost, and how fast cost changes reach the pump. Each screen is also run on a market with many sellers, to see what the same test looks like there.
Finding 1
A change episode starts with the first brand to change its price and collects every other brand's first change in the same direction within seven days. Of 143 episodes since April 2020, all five brands moved in 90%, and by the same amount in 83% of those. They moved on the same day in only 1.4%. One brand goes first and the rest catch up over the next day or two.
Exhibit 1
Shell's line moves first at most steps; the other four catch up within days
The order is stable. When one brand moved alone, it was Shell 37 times, Esso 17 and Caltex 12. PetroChina and Sinopec, the two brands owned by Chinese state oil companies, almost never go first; they usually match one or two days later.
Exhibit 2
Caltex, Shell and Esso tend to move on day 0; PetroChina and Sinopec on days 1 and 2
Not every move sticks. 15 episodes were not joined by every brand, and in 8 of them a single brand changed its price and nobody followed within the week. 2 of those lone moves were taken back within two weeks: the brand returned to the price everyone else was still posting.
The premium grade shows the same pattern over a longer span. Premium sells at a fixed surcharge over Standard, and in January 2024 Shell widened its surcharge while the others held theirs for a full year.
Exhibit 3
Shell raised the premium surcharge, waited a year, then raised it again three days after the others caught up
| Date | Who moved | Surcharge over Standard |
|---|---|---|
| Before 2024 | All five brands | HK$1.19 |
| 8 Jan 2024 | Shell | HK$1.60 |
| 25 to 31 Dec 2024 | Sinopec, Esso, PetroChina | HK$1.39 |
| 7 to 11 Jan 2025 | Esso, Sinopec, PetroChina, Caltex | HK$1.60 |
| 14 Jan 2025 | Shell | HK$1.80 |
| 15 to 21 Jan 2025 | Esso, PetroChina, Sinopec, Caltex | HK$1.80 |
Finding 2
The median increase was HK$0.20 a litre and the median cut HK$0.10. The import cost of petrol did rise over the whole period, so more increases than cuts is no surprise by itself. The test is the years when that cost fell. In 2023 and 2024 the posted price rose 17 times and was never cut once.
Exhibit 4
Increases outnumber cuts in every year; in 2023 and 2024 there were no cuts at all
Finding 3
Hong Kong has no refinery and imports all of its petrol, and the duty has been HK$6.06 a litre since 1998. Whatever is left of the pump price after the import cost and the duty pays for storage, delivery, stations, land and profit. The posted price averaged HK$17.55 in the first twelve months of the record and HK$30.49 in the last twelve. The import cost rose by HK$2.67 over that span and the margin by HK$10.27. From January 2023 to December 2025 the posted price went from HK$23.09 to HK$28.89 while the import cost went from HK$5.23 to HK$4.29.
Exhibit 5
Most of the rise in the pump price since 2020 is margin, before and after discounts
The posted price is not what most drivers pay. Every brand advertises a walk-in discount, and the discounts grew as posted prices rose: the average across the five brands went from HK$2.44 a litre in the first year of archived pages to HK$5.52 in the last. Measured on the price after that discount, the margin still went from HK$6.74 to HK$13.11. Card and loyalty discounts go deeper, and they are not in any public data.
Finding 4
Economists call this pattern rockets and feathers: retail prices shoot up when costs rise and drift down when they fall. The standard test is an error-correction model that lets rises and falls in cost pass through at different speeds (Borenstein, Cameron and Gilbert 1997). It was run on weekly data for all three markets with the same specification.
Exhibit 6
Hong Kong passes on rises faster and further than falls; Perth shows no lasting difference
| Market | Rise, week 0 | Fall, week 0 | Rise, week 4 | Fall, week 4 | Rise = fall (p) |
|---|---|---|---|---|---|
| Hong Kong | 30% | 7% | 67% | 39% | 0.009 |
| Los Angeles | 53% | 23% | 86% | 85% | 0.028 |
| Perth | 35% | 3% | 95% | 104% | 0.280 |
The gap between rises and falls is significant in Hong Kong and Los Angeles with four, six or eight weeks of lags, and Perth's is not significant at the 5% level in any version. Part of Hong Kong's slowness in both directions is mechanical: its posted price was unchanged in 66% of weeks, so any delay shows up as slow pass-through. The difference between rises and falls is the part that is not mechanical.
The comparison
Perth's retailers must tell the state's FuelWatch service their next day's price by 2pm and hold it for 24 hours, so every price is public a day ahead, as in Hong Kong. What differs is the number of sellers. Perth's prices run a weekly cycle: the median is lowest on Tuesday (155.9 cents a litre) and jumps on Wednesday (187.1 cents), the day prices are restored. Byrne and de Roos (2019) traced that cycle to price leadership by the dominant firms, so Perth is no picture of pure competition either. But its coordination ends in a price war every week, and on any given day its stations are spread wide.
Exhibit 7
Perth's prices jump every Wednesday and slide for the rest of the week, with a wide spread between stations
What it means
None of this is evidence of an agreement, and it is not offered as one. In the United States, conscious parallelism “has not yet read conspiracy out of the Sherman Act entirely” (Theatre Enterprises, 1954), but parallel conduct alone does not support an inference of conspiracy (Bell Atlantic v. Twombly, 2007). Hong Kong's First Conduct Rule, in force since December 2015, also targets agreements and concerted practices, and identical prices on their own are neither.
Each pattern here has an innocent reading on its own. Identical prices follow from identical costs; a stable leader is what price leadership looks like with or without an agreement; asymmetric pass-through appears in competitive markets with search costs; rents and wages rise. What the screens add is that the patterns arrive together, last six years, and are strongest where cost explanations are weakest: two years of increases and no cuts while the import cost fell.
Three kinds of data would separate cost from conduct: station operating costs and rents, the timing of each brand's price notices to the Consumer Council, and dated discount terms by brand. The Council collects some of these. None is published as data.
Limits: the open-data record starts in April 2020; day-level timing depends on when the archive copied the file; the import price is a monthly average of consignments; the margin is gross of storage, delivery, station costs and rent; card and loyalty discounts are not observed.
Data and method
Every number on this page, on the explorer and in the paper is written from the analysis outputs by one script, so a rerun updates all three. Each data decision, from the archive's label change to Perth's Tuesday price, is listed in docs/decisions.md.