Market matrix

What a factor is, and is not. The model behind these scales assumes a small number of unobserved influences generating the co-movement of the underlying indicators. Assuming is not finding: the factors are covariance patterns with names attached. The readings below are consistent with the loadings and with outside evidence, and they could be wrong. The structure is also regime-dependent: it is essentially the post-2021 structure, and positions are 2016–2025 averages, not current states.

S1 · Headline-cycle exposure. Food, drinks, home upkeep, furnishings, transport and hobby-goods inflation rising and falling together. This scale tracks headline inflation almost one-for-one (r≈0.94), so the most defensible reading is that it IS the 2021–23 cost shock spread across the categories it travelled through: energy, import and commodity costs passing into goods prices. A market's position here is exposure to that common cycle, not a hidden trait.

S2 · Sticky-services & excise. Rents, health, education and tobacco. The first three are the classic institutionally priced categories: contracts, indexation and administered tariffs that reprice slowly and catch up late. Tobacco moves on excise decisions. This cluster reproduces, from our data alone, the a-priori “stickiest categories” of the euro-area price-setting literature, which is the strongest outside corroboration any of these factors has. Reliability is nonetheless marginal (α 0.67) and tobacco is the weak member.

S3 · Tech-durables prices. Audio-visual gear and computers: globally traded goods with a persistent, quality-driven price decline. Their prices are set on world markets more than domestic ones, which reads as a separate influence, and part of the commonality may be shared measurement practice (hedonic adjustment). Two items only: an index, not a scale, but the only piece of this structure stable across the 2021 regime change.

S4 · Retail activity. Volume growth stands apart from every price factor in every encoding we tried. The honest reading is thinner than a latent influence: one real-activity indicator in a battery of prices cannot form a factor, so its separation partly reflects the instrument. What it does establish is that the price cycle and the demand cycle are different things: a market's inflation exposure tells you little about its demand cycle.

Why the items view exists. Items inside one factor do not sort markets identically: rents and health, for example, rank countries almost independently. The factor columns are the denoised summary; the item columns are the same encoding without the compression. Item positions are country means of winsorized year-on-year values over the same window, normalized the same way, so the two views are directly comparable.

Your category column. The selector adds one retail sub-category (Eurostat sts_trtu_m, volume of sales, automotive fuel excluded) as an extra column: country means of winsorized year-on-year growth over the same window, normalized the same way, so its positions read on the same scale and its header sorts like any other. It is a user overlay for ranking markets by a category you care about, and that is all it is: the sub-categories were never inputs to the factor estimation, and a market's position on your column says nothing about the validated structure in the other columns. Markets whose category series is unpublished or too short show “·”.