Raw vs clean DVF: bulk sales, price outliers and off-plan sales

Why not use raw DVF directly?

The “Demandes de valeurs foncières” files and Etalab's DVF API are free, complete and faithful to the source. That is exactly their job: to return transactions as they were declared, without interpreting them. To display the sales on a parcel, that is what you want.

To compute a price per m², a median or a trend, though, every line first has to answer three questions. Is it a sale? Does the price refer to this property alone? Does the surface allow it to be expressed per m²? The file doesn't say. The answer has to be computed, and Normi does that work once, at ingestion.

What is dropped at ingestion

Three categories of line cannot produce a reliable price per m². They are not loaded into the database, so they cannot be retrieved through the API.

What is kept, but flagged

Two further categories carry a real price, but one that doesn't describe the resale market for a home. Normi keeps and flags them, so you decide.

• Bulk sales: 6.7% of published lines (1,208,430)

  A transaction covering a building, a block of parking spaces or a portfolio of flats has a single overall price, repeated on each of its lines. A line is flagged is_bulk_sale when it has more than 3 lots, or when its summed Carrez areas exceed 1.8 times its built area, a sign that several properties were aggregated.

• Off-plan (VEFA) sales: 2.6% of published lines (468,970)

  A VEFA sale (vente en l'état futur d'achèvement) is a new-build purchased off-plan, at a developer's price. Mixed with resales, it pushes medians up in neighbourhoods under construction. The line is flagged is_vefa from its transaction type.

In total, 1,648,774 lines (9.1%) are bulk or off-plan. Separately, 99.7% of published lines have coordinates, which makes them queryable by radius.

How Normi applies these filters

The valuation and comparison tools (estimate_property_value, compare_locations, get_zonal_price_distribution) exclude off-plan sales with no option. Two tools don't exclude them yet: analyze_market_activity, which measures volumes, and score_market_health. The exclude_outliers parameter is deprecated: price outliers are never loaded, so it has nothing to exclude. It is still accepted so existing calls don't break, but it is ignored.

To compare both versions yourself, run the same request with and without the filters. The API limits commune-level queries to the last two years, so its figures cover a shorter window than the table below and differ slightly from it.

# Median with the default filters
curl -H "X-API-Key: $NORMI_API_KEY" \
  "https://mcp.normi.fr/v1/stats/market?commune=MONTPELLIER&type_local=Appartement"

# The same, bulk and off-plan sales included
curl -H "X-API-Key: $NORMI_API_KEY" \
  "https://mcp.normi.fr/v1/stats/market?commune=MONTPELLIER&type_local=Appartement&exclude_bulk_sales=false&exclude_vefa=false"

Before and after: the median price of flats, 2024-2025

Six large communes, every one we measured. “Before” counts every published line (so price outliers are already absent from both), “after” excludes bulk and off-plan sales.

SELECT commune,
  count(*),
  percentile_cont(0.5) WITHIN GROUP (ORDER BY prix_m2),
  count(*) FILTER (WHERE NOT is_bulk_sale AND NOT is_vefa),
  percentile_cont(0.5) WITHIN GROUP (ORDER BY prix_m2)
    FILTER (WHERE NOT is_bulk_sale AND NOT is_vefa)
FROM dvf_transactions
WHERE commune IN ('BORDEAUX','LILLE','MONTPELLIER','NANTES','PARIS 15','TOULOUSE')
  AND computed_type_local = 'Appartement' AND prix_m2 IS NOT NULL
  AND date_mutation >= '2024-01-01' AND date_mutation < '2026-01-01'
GROUP BY commune;