We measured Korea's two oil prices in August. Ten weeks later the kerosene auction is almost gone.

The Korea Exchange runs a petroleum market, and for each fuel it publishes two weighted average prices per session: one for competitive trading, where orders meet on a book, and one for negotiated trading, where the two sides agree a price between themselves.

We counted those two prices once before. On 8 August we published a six-year read of them — 1,619 trading days to 6 August 2026 — and reported that the competitive auction printed nothing at all on 4.1 percent of diesel days, 5.1 percent of petrol days and 16.2 percent of kerosene days. One day in six for kerosene.

This is what the same measurement looks like on the 49 sessions to 9 September.

The negotiated figure never fails to appear. The competitive one is missing on more than four rows in ten — against roughly one in twenty across the six years to August. So on those days the number the exchange publishes as the price of that fuel is a price two parties agreed on, not one a market cleared.

Kerosene has essentially no competitive price

The absence is not spread evenly.

For petrol and diesel a missing competitive session used to be a rare exception and is now roughly a one-in-six event. For kerosene it has stopped being an exception at all: on 47 of 49 sessions there was no competitive weighted average, and the published figure was the negotiated one.

It is not one bad fortnight. Split by month, the level holds:

Our August figure of 16.2 percent was a six-year average, and July 2026 sits inside that window — so the average was already being lifted by this stretch while also hiding it. That is the point worth keeping: a rate computed over six years cannot tell you that the last three months were near-total.

Korea’s exchange kerosene price is now, for practical purposes, a negotiated price with a competitive price twice a quarter. Anyone treating the three fuels as three comparable exchange series is comparing two market prices with one that has largely stopped being one.

Where both exist, the negotiated price is the higher one

On rows carrying both figures we measured how far the negotiated average sits above the competitive one.

Both figures are small, both are positive, and both are steady enough across forty sessions to be worth stating: negotiated buyers in this market pay about three-tenths of a percent more than the book. That agrees with our August read, which put the median gap a little above the auction at 5.0, 1.6 and 0.9 won a litre while also finding the negotiated price below the auction on 37 to 46 percent of individual days. A small positive median and a near coin-flip day to day are the same finding stated two ways, and neither is a claim that one price is the right one.

The kerosene row is blank on purpose. Two comparable sessions cannot carry a median, and the tool behind this article will not emit one — it holds a floor of five and returns nothing below it. We would rather print an empty cell than a number that is really a coin flip.

A note on reading zeros

The dataset delivers a missing competitive price as 0. Read literally, that is a weighted average of zero won, and a tool that averages the column would report Korean kerosene trading at roughly four percent of its actual price — because 47 of its 49 values are zeros.

We treat a zero as absent and count how many there are, which is why the absence is the finding rather than a footnote. It is the same discipline as the rest of this desk: a gap gets counted and named, never filled.

One note on how this was built. The daily collection behind this piece did not exist yesterday: the collector was written on 4 August 2026 and had never been run, so its archive folder held no files. That is why our August article drew on a separate six-year pull and this one draws on a fresh daily archive — two collections, cited separately, not merged into one series. We have now pulled 1 July to 9 September and put the daily collection under the same watch as our bond and broker archives.

Forty-nine sessions is 49 sessions. What it can say is that the level in July, August and the first week of September is far above the six-year average we published a month ago. What it cannot say is when it changed, because the daily archive does not reach back before July.