Statistics Canada

Statistics Canada connector: the CPI, monthly, back to January 1914.

Canada held 40 series before this -- the IMF's headline and twelve COICOP divisions, from 1949. StatCan's own table 18-10-0004 carries 358 products for the nation, seven levels deep, back to 1914, which is the longest consumer price history in this database.

ONE REQUEST, EVERYTHING. StatCan publishes a table as a zip holding two CSVs: the whole cube, and a metadata file describing it. The second is what makes the first usable -- it lists every dimension member with its id AND ITS PARENT -- so the product hierarchy arrives in the same snapshot as the data and parse() never has to ask the network what a code means. 15 MB, 1.15 million rows, 30 geographies.

Five things worth knowing before changing anything here.

Not COICOP, and it does not nest into COICOP. Canada has eight top-level groups where COICOP has twelve divisions, and the boundaries differ rather than merely aggregating: there is no communication group at all (telephone services sit inside Household operations), no restaurants group (restaurant meals sit inside Food), and recreation and education are one group. Checked against StatCan's own catalogue on 2026-09-06: NONE of its 8,269 cubes mentions COICOP, so this is not a case of choosing the wrong release. The tree is kept verbatim under taxonomy: 'statcan' and linked to the shared vocabulary by handle and relation, which is the rule item 6 settled. Canada's COICOP view continues to come from the IMF, beside this one.

The rates are DERIVED. This cube publishes an index level and nothing else, so month-over-month and year-over-year are computed here -- permitted because the source publishes no rate for these concepts, declared per series in derived_from/derived_by, and validated at the aggregate against the IMF's Canadian headline by validate_statcan.py.

Each product carries its own base. Most are 2002=100, but ten other references appear because a product introduced in 2023 is referenced to 2023. The UOM column says which, per row; it is never assumed.

One old-base rerun is excluded. 'All-items (1992=100)' is 'All-items' on the previous reference, ending 2018-12. Of the 358 products, 84 carry a base in their name and exactly one of those has a plain twin also present -- that one. The other 83 are genuinely their own series.

Only Canada, not the provinces. The cube also carries 9 provinces and 20 cities. That is a filter and nothing else, the way EUROSTAT_GEO is.