Skip to main content
method · 16 June 2026 6 min
Part of: How AI Grades

M-Value AVM under the hood: how marocain.investments values a listing

The M-Value chip on every listing is an Automated Valuation Model — the median price/m² of district + typology + ±20%-surface comparables, with a ±1.5σ confidence band, requiring ≥3 comparables or it doesn't render at all. Here's exactly how the comparable pool is built, why we use the median, what the vs-asking delta tells you, and — honestly — what an AVM is NOT.

M-Value AVM under the hood: how marocain.investments values a listing

How the comparable pool is built

Every M-Value starts by filtering the live catalogue down to listings that share three attributes with the subject property: same district, same typology (apartment, villa, riad, etc.), and surface area within ±20% of the subject. District granularity is non-negotiable — intra-city per-m² dispersion in Casablanca, Marrakech and Tangier is wide enough that a city-level median is uninformative for a specific neighbourhood. The ±20% surface bracket exists because price per m² is not linear with surface: studios and large villas trade at structurally different per-m² rates, and mixing them poisons the median.

The ≥3 comparables floor

If the filter returns fewer than three comparables, no M-Value chip renders at all. A median over one or two asking prices is statistically meaningless and worse than silence — so the surface stays blank rather than fabricate a number. (You will see this on rare typologies or off-the-grid districts.)

Central estimate and confidence band

  • Central estimate: the median MAD/m² of the comparable pool — chosen over the mean for robustness against outlier asking prices (one mispriced villa shouldn't move the read).
  • Confidence band: median ±1.5 standard deviations of the comparable per-m² distribution. Under normality, 1.5σ covers roughly the middle ~87% of the distribution — deliberately wider than the ~68% of ±1σ, because asking-price comparables are noisier than transaction comparables.

What the vs-asking delta actually means

The M-Value chip expresses the gap between asking and the district-typology-surface median as a % delta. For a foreign buyer who has no local price intuition, the delta reframes the asking price in district-relative terms: it surfaces whether a number reflects local comparables or a foreigner-premium markup. A listing 25% above the comparable median is a different conversation than one at the median — that's the conversation we want you to have with the agent.

What an AVM is NOT

M-Value is an indicative statistical signal, not a formal valuation. It is **not an *expertise immobilière*, not an appraisal, and does not substitute for a licensed expert assessment for mortgage underwriting, tax base, or legal proceedings. It also draws from asking prices of currently live listings, not closed transaction prices** — Morocco has no public sold-price registry comparable to the UK Land Registry or US MLS, so the input is structurally noisier than an Anglo-Saxon AVM.

The honest ledger

What it proves:

  • Whether a given asking price sits above, at or below the district median for the same typology and surface bracket.
  • The dispersion around that median is wide or tight (the ±1.5σ band).

What it does NOT prove:

  • What the property would actually transact at (asking ≠ sold; Morocco has no public sold-price registry).
  • What a bank's expert would value it at for a mortgage (that needs a formal *expertise immobilière*).
  • What a notaire would accept as the declared tax base (a separate regulatory process).

*Internal methodology document — this describes how marocain.investments' own AVM is built. Sourced to the platform's scoring logic; the structural-noise caveat (asking ≠ sold) is a known limitation of the Moroccan market, not of M-Value. Decision-support, not a formal valuation.*

How is the M-Value comparable pool built?

By filtering live listings to those sharing the subject's district, property typology (apartment, villa, riad, etc.) and a surface area within ±20% of the subject. District granularity matters because intra-city per-m² dispersion in Casablanca, Marrakech and Tangier makes a city median uninformative for a specific neighbourhood.

Why does the M-Value chip sometimes not appear on a listing?

Because the comparable pool returned fewer than three listings after district + typology + ±20% surface filtering. Below the ≥3 comparables floor we render no AVM — a median over one or two prices is statistically meaningless, and silence is better than a fabricated number.

Why the median and a ±1.5σ band, instead of the mean and ±1σ?

The median is robust to outlier asking prices — one mispriced villa cannot drag the central estimate. The ±1.5σ band covers ~87% of a normal distribution versus ~68% at ±1σ, deliberately wider because asking-price comparables are noisier than transaction comparables.

Is M-Value a formal property valuation?

No. It is an indicative statistical signal — not an expertise immobilière, not an appraisal, and not a substitute for a licensed expert assessment for mortgage underwriting, tax base or legal proceedings.

What is the biggest limitation?

M-Value draws from asking prices of currently live listings, not closed transaction prices. Morocco has no public sold-price registry comparable to the UK Land Registry or US MLS, so the input is structurally noisier than an Anglo-Saxon AVM. The vs-asking delta is therefore a relative signal — useful for negotiation, not a guarantee of transaction price.