How to read this

What a forecast here is

Governments publish a house price index — a single number, updated monthly, that tracks how much prices have moved in a place. It is not the price of a house. It is closer to a thermometer reading for a whole market.

When we say +3% over 12 months, we mean: we think that index will be about 3% higher when the figure for that month is finally published. If it reads 100 today, we are guessing roughly 103. Nothing more specific than that is being claimed — and nothing vaguer, either.

What "locked" means

Every forecast is written into a store that cannot be edited or deleted afterwards. Not by us. The timestamp is applied by the database at the instant of writing, so it cannot be moved backwards to make a late prediction look early.

This matters because the easiest way to look like a good forecaster is to quietly adjust what you said once you know the answer — or to publish ten predictions and later mention only the two that worked. Both are invisible to a reader. Neither is possible here, and not because we promise: because the record is built so that it cannot happen.

Each forecast also records exactly which published data it was built from, and when that data was released. A prediction that turned out to have used information from after it was made is automatically refused a grade — it was never a prediction.

The range, and why it is there

Every forecast carries a range as well as a single number. Read it as: if our method is working as intended, the truth should land inside this range about 95% of the time.

A wide range is not evasion. It is the honest width of what we know, and it is calculated from how wrong this method has been in the past rather than chosen to look reassuring. Ranges widen the further ahead we forecast, because uncertainty genuinely compounds — a forecaster whose twelve-month range is as tight as their three-month one is telling you something false about time.

A range that is right 95% of the time is not the same as a forecast that is right 95% of the time. The single number will usually be a little off. The question the range answers is by how much, at most, most of the time.

When a forecast gets its verdict

On a date we publish at the same time as the forecast — not whenever the result looks good. A twelve-month forecast waits roughly fifteen months for its final verdict, because the statistics agency publishes a month or two in arrears and then revises the figure for up to a year afterwards.

That wait is inconvenient and we are not going to shorten it. Grading against an early estimate that later gets revised would let us count a win that reality withdrew.

So a new forecast reads PENDING for a long time, and that is the system working. A site that showed a verdict immediately would be grading itself against its own homework.

The scorecard

The scorecard totals up how the forecasts have done. Two things about it are worth knowing before you read a number off it.

Counts before rates. "We were right 8 times out of 10" only means something if you can see the 10. Where we cannot yet establish the denominator honestly, we show the count and withhold the percentage. A percentage with an unknown bottom half is not a measurement.

Repeated forecasts of the same thing are not separate tests. If we forecast the same target month several times as new data arrives, those are related attempts at one question, not independent successes. Where that happens, the scorecard shows how many distinct outcomes are actually behind the count.

The seam: tested-on-history vs predicted-in-advance

There are two very different things a forecaster can show you, and most sites blur them.

Tested on history means: we ran the method over the past and checked how it would have done. This is useful, and it is also the easy one — the method was built by people who already knew how that history turned out.

Predicted in advance means: we said it before the answer existed, and you can check the timestamp.

We keep these two completely apart, on separate pages and in separate tables, and we never average them together. Mixing them produces an impressive number that means nothing, which is exactly why it is a common thing to do.

Why we show the misses

Because a record with the failures removed tells you nothing about the next forecast. If we hid our misses, our hits would be worthless to you — you would have no way to tell our record from one that had been curated.

Every miss stays on the page, permanently, beside the hits, with the same detail. When we find a mistake in our own methods we publish that too, including mistakes found after the fact.

The claim this site is making is not "we are right". It is "you can check". Those are very different claims, and only the second one survives contact with a bad year.

What this is not

Not advice. Not a valuation of your home. Not a forecast of any individual property, and not a claim about anywhere we do not publish a forecast for.

Each arena page lists what its model has never been tested against. Those limits are part of the record, not disclaimers appended to it.