Forecast Standard Deviation (FSD) Explained
The industry-standard measure of how much uncertainty surrounds an AVM’s estimate — and why it matters more than most people think.
What is forecast standard deviation?
Forecast standard deviation — universally abbreviated to FSD — is the single most widely used measure of uncertainty in automated property valuation. It tells you how confident the model is in its estimate for a specific property, expressed as a percentage of the estimated value.
A low FSD means the model had plenty of relevant evidence and the property is straightforward to value. A high FSD means there is greater uncertainty — perhaps because few comparable transactions exist, the property is unusual, or the local market is volatile.
FSD is not an error measure. It does not tell you how wrong a specific valuation is — nobody knows that until the property actually sells. Instead, it is a forward-looking estimate of how much the true value might reasonably differ from the model’s point estimate. Think of it as a radius of uncertainty around the central figure.
The concept was formalised by the European AVM Alliance and adopted by the European Systemic Risk Board and the European Banking Authority as part of their property valuation standards. In the UK, it is referenced by lenders, regulators, and the RICS as the primary measure of AVM confidence.
How Meridian calculates FSD
The principle is consistent across the industry: FSD is derived from the distribution of valuation errors observed in back-testing — comparing the model’s estimates against actual transaction prices for properties whose sale price is known. The differences between providers lie in how finely that error distribution is segmented.
Meridian uses 49 property and location features — the number and recency of local comparables, the homogeneity of the surrounding market, the property type and size relative to the local norm, and the completeness of the available data — to compute a per-property confidence score. That score, together with the property’s type and price band, places the property in a granular FSD lookup segmented by confidence decile, property type, and price band.
Each cell of that lookup holds the standard deviation of the percentage errors observed in back-testing for properties in the same segment, calibrated against 295,026 actual Land Registry transactions. Where a segment has too few observations to be reliable, the lookup falls back to broader segmentation — type-level, then decile-level — so every valuation receives an empirically grounded FSD.
The same lookup, with the same fallback cascade, runs at request time in the live API and in our published back-tests. The FSD a lender sees on a valuation is the FSD we measured — not a separate marketing calculation.
The result is a property-level confidence measure, not just a model-level average. Two properties valued by the same AVM on the same day can have very different FSDs, because the model recognises that it has more evidence for one than the other.
Interpreting FSD: what the numbers mean
FSD follows a normal-distribution logic. If a property has an estimated value of £300,000 and an FSD of 8%, the model is saying that the standard deviation of its error distribution for similar properties is £24,000. Under a normal distribution, approximately:
| Confidence interval | Range (FSD = 8%, value = £300k) | Probability |
|---|---|---|
| ±1 FSD (±8%) | £276,000 – £324,000 | ~68% |
| ±1.65 FSD (±13.2%) | £260,400 – £339,600 | ~90% |
| ±2 FSD (±16%) | £252,000 – £348,000 | ~95% |
In plain terms: an FSD of 8% means the model expects to be within 8% of the true value about two-thirds
of the time, and within 16% about 95% of the time. Lower FSD means tighter prediction intervals and
more reliable estimates. The 90% prediction interval shown on every Meridian valuation is derived
directly from the FSD: value × (1 ± 1.645 × FSD).
The appropriate FSD tolerance depends on the use case. A lender approving a low-LTV remortgage may accept a higher FSD than one underwriting a 95% LTV purchase, because the financial exposure to valuation error is very different. The Fitch bands below give that tolerance a standard structure.
FSD and the Fitch confidence bands
Meridian maps each property’s FSD to one of the four classification bands used by Fitch Ratings in RMBS analysis. The band — A, B, C, or D — is the confidence expression on every valuation, alongside the raw FSD value and a European AVM Alliance confidence level on the 0–7 scale.
| Band | FSD range |
|---|---|
| Band A | FSD ≤ 0.05 |
| Band B | ≤ 0.10 |
| Band C | ≤ 0.20 |
| Band D | > 0.20 |
Band performance on the latest bulk test of 295,026 Land Registry transactions (H2-2025):
| Band | n | % of test | MdAPE | PE10 | Bias |
|---|---|---|---|---|---|
| Band A | 0 | 0.0% | — | — | — |
| Band B | 143,546 | 48.7% | 6.0% | 70.4% | -0.7% |
| Band C | 123,787 | 42.0% | 7.0% | 63.9% | +1.7% |
| Band D | 27,693 | 9.4% | 11.2% | 46.4% | +16.5% |
No valuations in the current test cohort carry Band A. The current FSD lookup does not produce FSDs at or below 0.05 for any segment; the band is shown for completeness of the Fitch scale.
Band D valuations show a mean bias of +16.5% on the current test — the model overvalues these properties.
Band assignment is property-specific. The same model produces a Band B result for a standard three-bedroom semi in a well-transacted suburb and a Band D result for an unusual conversion flat in a rural area with few comparables. The model is the same — the confidence in its output differs because the evidence differs. An AVM that claims equal confidence for every property is not being honest about its limitations.
Why FSD matters for lenders
For mortgage lenders, FSD is the link between AVM output and credit risk. A valuation with a low FSD carries less risk of the true value being significantly below the estimate. A valuation with a high FSD carries more. The lender can use this to calibrate their acceptance criteria.
In practice, this means lenders can set policy rules such as:
- Accept AVM-only for Band B results at LTV ≤ 75%
- Require a desktop review for Band C results or higher LTV
- Escalate to a physical valuation for Band D or any result above a defined FSD threshold
This band-based approach reflects how regulators expect lenders to use AVMs. PRA expectations, RICS guidance, and (in the EU) the EBA’s loan-origination guidelines all emphasise that AVM results should be subject to confidence-based acceptance criteria — not used indiscriminately. FSD provides the mechanism to do this in a consistent, auditable way.
Under Basel 3.1, which takes effect in the UK from 1 January 2027, lenders using AVMs for property revaluation will need to demonstrate that their AVM meets defined accuracy and confidence standards. FSD is expected to be a central component of that demonstration.
For more on the regulatory context, see our articles on Basel 3.1 and property revaluation and RICS and UK regulation.
What FSD cannot see
FSD quantifies evidence-based uncertainty: how much the available data supports the estimate. It cannot price what no dataset records. The clearest category is current condition — renovation, extension, or deterioration that has happened since the property’s data was last captured.
A real example from our own testing: a recently extended detached house whose model estimate came in at roughly half the inspected value established after the works. No register had yet caught up with the renovation — the EPC, the floor area, and the transaction history all described the house as it stood years earlier. The model valued the property the data described, not the property that now stands.
Critically, FSD does not flag such cases. The comparables look fine, the data looks complete, and the confidence score is unexceptional — because the uncertainty lies in the gap between the records and reality, which no statistical measure of the records themselves can detect. This is the category of error confidence scoring cannot see.
This is why lenders pair AVMs with condition checks at high LTV, and why every Meridian valuation lists the comparable evidence it relied on. An AVM is a measure of the recorded property; where the recorded property and the physical property have diverged, a physical inspection is the only corrective.
FSD compared to other accuracy metrics
FSD is one of several metrics used to evaluate AVM performance. It is important to understand what makes it different and why it is preferred for property-level confidence assessment.
FSD vs MdAPE (Median Absolute Percentage Error)
MdAPE is a model-level aggregate — the median error across all back-tested properties. It tells you how the model performs on average, but says nothing about any individual property. FSD is property-specific: it varies from one valuation to the next depending on the model’s confidence in that particular case.
FSD vs PE10 (Percentage within ±10%)
PE10 tells you what proportion of valuations fall within 10% of the sale price. Like MdAPE, it is an aggregate metric — useful for evaluating the model as a whole, but not for assessing confidence in a single valuation. FSD fills that gap.
FSD vs R² (Coefficient of determination)
R² measures how well the model explains variation in property values overall. A high R² means the model captures the main drivers of value well, but it does not tell you whether any individual estimate is reliable. FSD provides that property-level granularity.
In summary: aggregate metrics like MdAPE, PE10, and R² answer the question “is this a good model?”. FSD answers the question “how much should I trust this specific valuation?”. Both questions matter, but for operational decisions on individual properties, FSD is what you need.
See FSD in action
Every Meridian valuation includes an FSD, a Fitch confidence band, and a prediction interval. Try a valuation or review our sample report.