I got frustrated that every fuel price app just shows you what's cheap nearby. I wanted to know how stations actually behave: do prices go up faster than they come down, do supermarkets really save you much, how bad are motorway prices really?
So I built a scraper that hits the UK government's mandatory Fuel Finder API every 10 minutes and stores every price change. 90k records across 7,700 stations since January.
Some things I found that surprised me:
The rocket and feather effect is real and measurable. When stations raise prices the average move is 2.35p/litre. When they cut, it's 1.85p. There are also more up moves than down moves. I queried the raw history to check this rather than eyeballing a chart.
Motorway fuel is 28.4p/litre more expensive than everywhere else right now. That's about £14 extra on a 50L fill. Everyone knows motorways are expensive but I didn't expect the gap to be that wide.
The supermarket discount is only about 1.7p. I assumed it would be bigger.
Stack is Azure Functions, TimescaleDB, PostGIS, Next.js. The interesting thing about this project is the history. No public site shows how an individual station has priced over time or how a local cluster of stations react to each other. That's what I'm building towards.
Another insightful way to look at this is to include gasoline spot market data as a comparison.
I kept hearing about the vast profits of gas stations, so one day I started a spreadsheet of my gas purchases and kept it going over 10 years. When I tried lining up the graph of what I have actually paid per litre with a spot market graph, after converting for currency, units, taxes etc, they were almost identical, indicating extremely slim margins, if any. Yes there were differences, places in the graph where stations had likely made money on my purchase, but there were just as many where they likely lost money, unless I also stepped inside to but a snack.
You are correct. Non-chain gas stations often make only as little as one or two cents per liter, and that's before you look at pump maintenance, inspections, periodical tank replacements/upgrades/liners and other costs.
Manned stations really need that shop otherwise they'd go bankrupt.
Chains make a bit more money but mostly because they can play longer games with stock and options on much larger volume buys.
And yet I see in earnings that companies like BP and Shell make record profits over increased gas prices. How come that they do profit but the station holders not? Are shell/bp increasing the margin harder and eating the station’s lunch?
Opening a gas station is a lot easier than acquiring mineral rights, drilling an oil well, refining that oil, and getting it to market. Oh, and your customers can't just drive across the street to your competitor because they are 1c/litre cheaper.
There's naturally going to be a lot more friction and a lot less pop-up competition and therefore a lot more margin on the supply side of things.
The station has no power to raise margin - they are in tight competition with every other low-margin station around them. The suppliers, on the other hand... If they invested into wells that aren't affected by the war 10 years ago, and their competitors haven't (or have, but can't supply all the world's oil needs), and there's a global supply shortage - they have lots of room to raise prices.
Good shout. DESNZ publishes weekly wholesale rack prices and they are OGL, so there is no barrier there. The interesting bit isn't just showing the gap; it's the propagation lag. Wholesale spikes and pump prices follow within days. Wholesale drops and pump prices take their time. That asymmetry is basically what I built this dataset to measure. Adding the wholesale series as a reference line is on the list.
>The rocket and feather effect is real and measurable. When stations raise prices the average move is 2.35p/litre. When they cut, it's 1.85p. There are also more up moves than down moves. I queried the raw history to check this rather than eyeballing a chart.
Comparing the absolute size of price rises vs drops doesn't make sense, because it could very well be an issue with the underlying price (eg. crude oil or whatever). It seems hardly fair to blame gas stations for being slow to lower prices, when refineries are still also slow to lower prices. Same for blaming refineries when the global market is slow to lower prices.
For the "supermarket saving", did you include Asda in the supermarket pool, or as a general pool? They seem to be rather less price competitive than other supermarkets, I'd presumably because of the recentish private equity takeover involving petrol station operator Euro Garages meaning they've kinda opted out of the petrol price war (they're hardly likely to want to undercut their existing forecourts).
Although the other recent private equity takeover of Morrisons led to some sort of deal with Motor Fuels Group to operate their petrol stations (but no ownership stuff in this case?), but they're seemingly still being competitive with Sainsbury's and Tesco's.
Yeah Asda is in there. I match on brand name directly rather than the API's is_supermarket flag because that flag is all over the place (loads of Asda stations don't have it set). So it explicitly checks for Asda, Tesco, Morrisons, Sainsbury's, Co-op, Costco.
Your point about post-PE Asda is interesting, I've noticed it too. If you want to see how they compare individually you can check the brands page on the site, shows each supermarket chain as its own line. Pretty easy to split the supermarket aggregate out per brand too, would probably show Asda creeping back towards the independents since the takeover. Might add that.
> So I built a scraper that hits the UK government's mandatory Fuel Finder API every 10 minutes and stores every price change. 90k records across 7,700 stations since January.
Only 1 change per station per week on average? Fewer than I expected. Not sure I'd call it a scraper, myself.
It is, up to the point where payment limits hit, or things are not set to auto-scale for that reason, or the project just isn't designed in a way to actually take advantage of the scaling (in which case you might be better off on your own/rented servers which will be much cheaper than one of the big clouds, unless you don't want the faf of managing backups & other admin and freedom from that is worth the extra cost).
In Germany fuel processes must be reported to the anti trust authority (Bundeskartellamt) the data is than published to providers of apps and websites. Unfortunately there isn't a free public data stream from them.
checked the db, tesco e5 is 163.8p average across 292 stations in the last 21 days. Worth spot checking a few specific stations if it still looks off to you.
I got frustrated that every fuel price app just shows you what's cheap nearby. I wanted to know how stations actually behave: do prices go up faster than they come down, do supermarkets really save you much, how bad are motorway prices really?
So I built a scraper that hits the UK government's mandatory Fuel Finder API every 10 minutes and stores every price change. 90k records across 7,700 stations since January.
Some things I found that surprised me:
The rocket and feather effect is real and measurable. When stations raise prices the average move is 2.35p/litre. When they cut, it's 1.85p. There are also more up moves than down moves. I queried the raw history to check this rather than eyeballing a chart.
Motorway fuel is 28.4p/litre more expensive than everywhere else right now. That's about £14 extra on a 50L fill. Everyone knows motorways are expensive but I didn't expect the gap to be that wide.
The supermarket discount is only about 1.7p. I assumed it would be bigger.
Stack is Azure Functions, TimescaleDB, PostGIS, Next.js. The interesting thing about this project is the history. No public site shows how an individual station has priced over time or how a local cluster of stations react to each other. That's what I'm building towards.
Site: https://fuelinsight.co.uk
Happy to talk through the architecture or the data if anyone's interested.
Another insightful way to look at this is to include gasoline spot market data as a comparison.
I kept hearing about the vast profits of gas stations, so one day I started a spreadsheet of my gas purchases and kept it going over 10 years. When I tried lining up the graph of what I have actually paid per litre with a spot market graph, after converting for currency, units, taxes etc, they were almost identical, indicating extremely slim margins, if any. Yes there were differences, places in the graph where stations had likely made money on my purchase, but there were just as many where they likely lost money, unless I also stepped inside to but a snack.
You are correct. Non-chain gas stations often make only as little as one or two cents per liter, and that's before you look at pump maintenance, inspections, periodical tank replacements/upgrades/liners and other costs.
Manned stations really need that shop otherwise they'd go bankrupt.
Chains make a bit more money but mostly because they can play longer games with stock and options on much larger volume buys.
Source: former gas station owner.
And yet I see in earnings that companies like BP and Shell make record profits over increased gas prices. How come that they do profit but the station holders not? Are shell/bp increasing the margin harder and eating the station’s lunch?
Opening a gas station is a lot easier than acquiring mineral rights, drilling an oil well, refining that oil, and getting it to market. Oh, and your customers can't just drive across the street to your competitor because they are 1c/litre cheaper.
There's naturally going to be a lot more friction and a lot less pop-up competition and therefore a lot more margin on the supply side of things.
The station has no power to raise margin - they are in tight competition with every other low-margin station around them. The suppliers, on the other hand... If they invested into wells that aren't affected by the war 10 years ago, and their competitors haven't (or have, but can't supply all the world's oil needs), and there's a global supply shortage - they have lots of room to raise prices.
Good shout. DESNZ publishes weekly wholesale rack prices and they are OGL, so there is no barrier there. The interesting bit isn't just showing the gap; it's the propagation lag. Wholesale spikes and pump prices follow within days. Wholesale drops and pump prices take their time. That asymmetry is basically what I built this dataset to measure. Adding the wholesale series as a reference line is on the list.
>The rocket and feather effect is real and measurable. When stations raise prices the average move is 2.35p/litre. When they cut, it's 1.85p. There are also more up moves than down moves. I queried the raw history to check this rather than eyeballing a chart.
Comparing the absolute size of price rises vs drops doesn't make sense, because it could very well be an issue with the underlying price (eg. crude oil or whatever). It seems hardly fair to blame gas stations for being slow to lower prices, when refineries are still also slow to lower prices. Same for blaming refineries when the global market is slow to lower prices.
yeah that's fair, without wholesale rack prices as a baseline you can't really isolate station behaviour from what's happening upstream.
For the "supermarket saving", did you include Asda in the supermarket pool, or as a general pool? They seem to be rather less price competitive than other supermarkets, I'd presumably because of the recentish private equity takeover involving petrol station operator Euro Garages meaning they've kinda opted out of the petrol price war (they're hardly likely to want to undercut their existing forecourts).
Although the other recent private equity takeover of Morrisons led to some sort of deal with Motor Fuels Group to operate their petrol stations (but no ownership stuff in this case?), but they're seemingly still being competitive with Sainsbury's and Tesco's.
Yeah Asda is in there. I match on brand name directly rather than the API's is_supermarket flag because that flag is all over the place (loads of Asda stations don't have it set). So it explicitly checks for Asda, Tesco, Morrisons, Sainsbury's, Co-op, Costco.
Your point about post-PE Asda is interesting, I've noticed it too. If you want to see how they compare individually you can check the brands page on the site, shows each supermarket chain as its own line. Pretty easy to split the supermarket aggregate out per brand too, would probably show Asda creeping back towards the independents since the takeover. Might add that.
> So I built a scraper that hits the UK government's mandatory Fuel Finder API every 10 minutes and stores every price change. 90k records across 7,700 stations since January.
Only 1 change per station per week on average? Fewer than I expected. Not sure I'd call it a scraper, myself.
157p/L national average is about 8 USD/G.
If only this wasn't hosted on azure, we'd be able to actually look at the data
I thought cloud was supposed to be an answer to sudden surges in traffic. /s
It is, up to the point where payment limits hit, or things are not set to auto-scale for that reason, or the project just isn't designed in a way to actually take advantage of the scaling (in which case you might be better off on your own/rented servers which will be much cheaper than one of the big clouds, unless you don't want the faf of managing backups & other admin and freedom from that is worth the extra cost).
Microslopinator strikes again
In Germany fuel processes must be reported to the anti trust authority (Bundeskartellamt) the data is than published to providers of apps and websites. Unfortunately there isn't a free public data stream from them.
List of authorized places using the data: https://www.bundeskartellamt.de/DE/Aufgaben/Markttransparenz...
One of those vendors publishes it as creative commons data set, though. Including historic data. https://creativecommons.tankerkoenig.de/
The Tesco premium/super e5 is so much cheaper than everyone else I'm actually pretty sceptical of it.
checked the db, tesco e5 is 163.8p average across 292 stations in the last 21 days. Worth spot checking a few specific stations if it still looks off to you.
bros server crashed
Well they did call themselves the azure guy
haha not quite, db was struggling under the load for a few minutes. b1ms postgres doesn't love 6k requests in 30 mins.
A 30 second memcached/redis/etc goes a long way when sharing on HN. You go from 6k db hits to 60 hits in 30 minutes. Worked for me since 2013 [0].
[0]: https://idiallo.com/blog/handling-1-million-web-request
already on it, just shipped in-memory output caching while the thread was live. next step redis if it needs it
Those tiny b1ms VMs are absolutely pitiful (long time Azure sufferer here). It's crazy how little compute and memory Azure give you for so much money.
yeah scaling it up now, was hoping to keep costs down but HN had other ideas
Using azure managed PG is certainly an interesting choice if you were hoping to keep costs down