The fuel-cost adjustment — when it settles, and when it is announced
A Japanese electricity bill carries a fuel-cost adjustment line: a unit rate in yen per kWh, multiplied by consumption. The rate carries a sign — a negative rate deducts from the bill, a positive one adds to it. Suppliers announce the rate monthly, but the values it is computed from are customs import prices for crude, LNG and steam coal, published by the Ministry of Finance ahead of that announcement.
The formula averages three consecutive months of those prices, three to five months back from the billing month. That three-month span is the window. Once the window's last month prints, the rate for that billing month is settled arithmetic. Measured over the archive of past customs releases stored with this product — the billing months it actually covers, rather than a nominal calendar — that happens 37–46 days before the billing month begins, and 5–11 days before the supplier announces the rate, reading the announcement as the 28th of the month two before. Both figures move with the convention they are measured against, so the convention travels with them.
Only months whose window has fully printed can be computed — today that is the nearest one or two. This page draws those as columns, and past them the same series continues in the same figure as a band: where the rate lands once the import prices behind it are allowed to move. Months past that boundary are simulated rather than computed.
What is simulated is that unit rate itself, one distribution per billing month: ten thousand paths of the four traded instruments the customs prices are estimated from — Brent crude, Henry Hub natural gas, Newcastle steam coal and the dollar-yen rate — each stepped forward a month at a time and then put through the tariff's own formula. The instruments are stepped with zero drift, so each one's expected level in every future month is its last observed monthly average; the width around that is the volatility and correlation fitted over the calibration window, which is printed in full below the figure. Nothing forward-looking is read — not a futures curve, not an implied volatility — so a quantile on this page is how often an outcome occurs under the fitted historical dynamics, not what the market charges for it.
Loading…
Why it is known before it is announced
Only one input to the formula comes from outside the tariff schedule — the 約款, the supply terms a Japanese supplier publishes and files, and where α, β, γ, the base fuel price and the base unit rate are fixed. Those do not move month to month. The one input that does is three months of customs prices. The rate is therefore not a forecast — it is arithmetic whose answer is determined the moment its inputs exist.
Customs does not publish a reference month once. It publishes it four times, each release revising the last: a 速報 (preliminary) 17–23 days after the month ends, a 確速 (revised preliminary) about a week after that, a 確報 (confirmed) a month later again, and a 確々報 (final) that arrives irregularly and in batches, anywhere from three months to two years on. Each of those releases is a vintage — the series as it stood on the day that release went out. Reading the series as of a chosen past date rather than as it stands today is what point-in-time means, and it is how a settled month is reproduced: a supplier computed its rate from whichever vintages existed on its own announcement date, so reproducing that number means reading the same ones back.
The window's last month sits three months before the billing month, so the 速報 that completes the window necessarily lands ahead of the supplier's announcement. The interval this page fills is the few days between the two.
Past the boundary — how to read the band
One dashed rule crosses the figure. Everything to its left is the formula evaluated on customs releases that have printed. Everything to its right is what the rate does when the import prices behind it are allowed to move: simulated rather than computed. The rule marks where one becomes the other.
The move happens in three stages. First the traded instruments — Brent crude, Henry Hub natural gas, Newcastle steam coal and the dollar-yen rate — are stepped forward month by month, each with its own annualised volatility and with the correlations between them. Then the customs CIF import price is estimated from those instruments. An instrument and a CIF price are not the same thing: freight, destination and contract structure keep them apart, and the gap between them is the basis. Finally the estimated CIF prices go through the formula and come out as a unit rate.
The basis does not reset to zero each month. A month that is far off tends to be followed by a month that is still somewhat off, and how much of last month's gap carries into this one is itself fitted. The number of months for that carry-over to decay to half its size is the half-life. The instruments also feed through with a delay rather than in the same month, so the lag is fitted per leg too — a leg being one of the three fuels the formula prices: crude, LNG and steam coal. What was fitted is below.
The band is those paths, sorted within each month and cut at fixed positions from the bottom. A quantile is one of those positions: the 5% quantile is the level only 5% of paths fall below. By default there are 10,000 paths, the outer band runs from the 5% to the 95% quantile so nine paths in ten land inside it, and the inner band runs 25% to 75% so one in two does. The line is the 50% quantile, the middle. Lowering the path count leaves too few observations in the tails to support those cuts, so the band falls back to inner quantiles; the line under the figure always names the ones being drawn.
The band widens with distance because the nearer months have part of their window already printed and have correspondingly less room to move. A month whose window has printed in full has no width left at all, and is drawn as a column.
The period total above the figure is the median of the whole period summed path by path, which differs slightly from adding up the monthly medians in the table: a median is not additive. Settled months enter that total at the settled figure the table shows, not at the simulator's own value for them, because a settled rate is computed from the customs vintage the supplier read on its announcement date.
Hedging, and where the variance comes from
Enter a hedge quantity in the assumptions and the page reports, month by month, how much of the rate's variance that position removes. Quantities start at zero and nothing fills in an "optimal" one: sizing a position is the reader's decision.
What variance is left is split five ways: the basis (instrument against CIF), the timing lag (how many months back each instrument is read), window averaging (the smoothing that averaging three months produces), FX, and the cap kink (a capped rate stops moving once the cap binds). The split is a Shapley value — all 32 on/off combinations of the five are run and averaged — so the answer does not depend on the order they are written in. A share comes out negative when switching that source on lowers the variance rather than raising it, which happens: the cap, for one, takes variance out by truncating the upper tail.
A sixth bar, the remainder, carries what is left when all five are idealised away — the instruments' own price move, the position you sized against it, and the rounding the formula applies. The six sum to the total variance exactly.
The calculation
average fuel price = round100( α·crude + β·LNG + γ·steam coal ) unit rate = round0.01( (average − base fuel price) × base unit rate ÷ 1,000 ) − government rebate
Crude is yen per kilolitre; LNG and steam coal are yen per tonne. α, β and γ are each fuel's heat share multiplied by a crude-equivalent conversion factor — the reciprocal of its calorific value relative to crude — and are published in the tariff schedule alongside the base fuel price and base unit rate. The divisor is 1,000 and the rounding is half-up; either one, taken wrongly, moves the rate by a sen — 銭, one hundredth of a yen, and the unit the rate is quoted to.
The implementation reproduces 19 announced months — 14 from Kansai Electric Power (関西電力) and 5 from TEPCO Energy Partner (東京電力エナジーパートナー) — to the last sen, or 33 tariff-months once extra-high-voltage menus are counted separately. Coefficients and the verification procedure are in methodology; endpoints are in the API reference.
The assumptions panel makes α, β, γ, the base fuel price and the base unit rate editable in place, so a retailer running its own fuel-adjustment clause can compute against its own coefficients rather than against a registered tariff. Edit any of them and the implied heat shares are summed on the spot, which is where a mistyped digit shows up.
The cap
Regulated tariffs cap the average fuel price at 1.5× the base. All ten incumbents removed that cap from their liberalised menus between October 2022 and May 2023. The cap replaces the average fuel price rather than clipping the unit rate, and a month where it binds is tagged in the table below.
On a capped tariff the unit rate stops rising past a fixed level however far fuel prices run. That is why the same supplier can charge two different rates for the same month depending on which menu a customer is on, and why comparing headline rates alone does not settle which menu is cheaper.
Data
Ministry of Finance customs releases (crude 30301, LNG 3050103, steam coal 3010105) and the published parameters in each tariff schedule. Steam coal is taken from its own commodity series rather than the coal aggregate, which blends in coking coal and runs about 11% high.
Each of the four customs releases — 速報, 確速, 確報, 確々報 — is stored here as its own vintage rather than overwriting the one before it, which is what makes a settled month reproducible from what was visible on the day its supplier announced. Point-in-time coverage — the stored vintages themselves — begins January 2024; the final revised series runs from January 1988 to the most recent release, with no gaps.
The simulation's instruments are EIA spot prices and Federal Reserve Board exchange rates, both US Government works. The steam-coal instrument is the exception. Its prices reach this model through the World Bank Pink Sheet — the monthly commodity price table the Bank publishes — whose licence is open except for the third-party series inside it, and coal is one of those, so the levels cannot be redistributed. Only a single origin level is used, inverted from the published CIF and the fitted residual. One residual recovers one observation, which is why steam coal has an origin and no history.
Suppliers are named on this page by their own English name, with the Japanese legal name after it. The tariff schedule is published under the Japanese name only, so that is the one to search on. A supplier whose English name is not held here is shown as the registry serves it.
Computed output, not advice — see Terms. Confirm billed amounts against your supplier's own figures. Months past the boundary are a model's estimate rather than a figure computed from published inputs.