Brazilian Intramarket Relationships: Relative Strength as Regime Information in an Emerging Market

Radar Perene Working Paper Series, No. RPWP-101 · Version 2.0 · July 2026 A large practitioner and academic literature studies intermarket relationships — the co-movement of stocks, bonds, commodities, and currencies — and intramarket relationships — the relative strength of one market segment against another — as sources of information about the prevailing market regime. Almost all of that literature is written about the United States, and almost all of it is silently assumed to generalize. This paper documents that the assumption fails in Brazil, and it fails in an instructive way. The canonical United States playbook treats a financial sector outperforming the broad index as a signal of easy money and expanding risk appetite. In Brazil the logic runs in the opposite direction: the profitability of the large listed banks is materially supported by high policy rates, which widen spreads and remunerate large low-risk asset books, so financial-sector relative strength has historically accompanied tightening cycles and defensive conditions rather than easing cycles and risk expansion. The financials-versus-index ratio is, in the terminology of this paper, risk-inverted relative to the logic built around XLF — the flagship United States financial-sector exchange-traded fund — that dominates the imported literature. Building on this observation, the paper proposes a taxonomy of Brazilian relative (ratio) relationships organized by economic category — financials, utilities, commodity producers, listed real-estate income vehicles, consumer cyclicality, and cross-asset ratios expressed in local currency — and describes how the Radar Perene research program aggregates a curated set of such relationships into a daily composite indicator of domestic capital flow, mapped to a 0–100 scale with published regime bands. The paper describes the construction philosophy at the level of categories and normalization families; proprietary weights, exact component lists, and tickers are not disclosed. The measure is descriptive: a high reading indicates that capital inside the Brazilian market is rotating toward historically risk-seeking segments relative to the index's own history; a low reading indicates rotation toward historically defensive segments. The indicator does not forecast index direction, does not identify tops or bottoms, and is not investment advice. Validation is reported honestly, with all statistics verified against the program's production database as of July 2026, regenerated in July 2026 on chained-return sector baskets after documented basket-construction defects — the July 2026 methodological correction under which member entries and exits are level-neutral by construction — printed with overlap-adjusted effective sample sizes alongside raw event counts, and read against the index's unconditional forward baseline: hit rates carry their failure complements and asymmetries, and three documented data-integrity incidents and their corrections are disclosed in full. Version 2.0 records a factual setback and retracts results. The episode-level discriminating test first reported at version 1.1 — computed, it is now clear, on a basket vintage since superseded — was recomputed at this version on the operative chained-return baskets, and its published pattern did not survive: the cell structure of the rate-cycle and market-state splits inverts under the operative construction, and no episode-level directional reading of the financials relationship is stable to basket construction, extreme threshold, or episode-separation rule. Both episode-level conclusions of version 1.1 are accordingly retracted, and the day-state financials headline is downgraded: under moving-block bootstrap inference against the family's own baseline, its frequency edge is not distinguishable from zero, and its subsample record concentrates entirely in the first half of the sample. Under the same inference, the utilities relationship is the file's only individually robust family, stable across sample halves — which means the framework's most robust validated relationship is the one that replicates the imported Gayed–Bilello channel, not the inverted signature. The taxonomy's sign observation stands as historical description; the paper claims no validated transmission mechanism, and it now also claims no robust episode-level forward-return pattern for the financials relationship. The paper's central contribution is the taxonomy and its transmission-channel analysis, which rest on documented institutional facts; the backtest statistics are supporting descriptive evidence, not proof of the inversion.