Sold prices are the only honest benchmark.
A sale-value estimate should be evaluated against completed sales, with the cohort, date, metric and limitations published together.
The first public Pretira test result is pending. The required certificate must freeze predictions before a named release and publish the completed-sale comparison with its limitations.
There are two ways to grade a property model. You can check whether it agrees with a surveyor, or you can check whether it predicted the price the property achieved. Only one of those is a fact.
Pretira estimates sale value. The defensible evaluation is to freeze predictions before a named completion release and compare them with the prices achieved. The first public Pretira certificate is pending, so this article states the method required rather than claiming a result or cadence that is not yet evidenced.
A sold-price benchmark is useful only when the prediction cohort, date, metric and limitations are published with the result.
Why agreement with a surveyor is the softer test
A surveyor produces an opinion. A model that is tuned to agree with opinions inherits their biases and cannot be checked against anything harder than another opinion. Agreement feels like accuracy. It is not.
A completed sale is a number a buyer and a seller both accepted. It is the number your funding line cares about, the number a securitisation agency will ask you to defend. Predicting it is a harder benchmark, and that is exactly why it is the right one.
What a repeated test would show
A dated series can reveal changes that one snapshot hides, but frequency alone proves nothing. Each result should be segmented by region, property type, price band and confidence band, with cohort size and error distribution stated. The product release cadence must not be mistaken for the testing cadence.
- -Freeze predictions ahead of the named reference release.
- -Publish results per period and segment rather than one flattering figure.
- -State weak or inconclusive segments and the evidence limitations plainly.
The one number problem
One national accuracy number tells a lender little about its own book. The relevant evidence is performance in the intended segments and at the proposed confidence thresholds. Until a completed Pretira certificate publishes that working, ask for the evidence currently available and treat the open gaps as open.
The benchmark is the sold price. Everything else is a story you will have to tell later.