The Tactical Ânima Index: Short-Horizon Sentiment Dynamics in Brazilian Equities

Radar Perene Working Paper Series, No. RPWP-003 · Version 2.0 · July 2026 Sentiment in equity markets does not operate on a single clock. The slow accumulation and erosion of conviction that defines a market's structural mood — the subject of the companion paper RPWP-002 — coexists with a much faster cycle of fear and relief that plays out over days and weeks rather than quarters. Brazil's equity market, characterized by concentrated ownership, episodic foreign participation, and recurrent macro-political shocks, exhibits this fast cycle with unusual amplitude: washouts and euphorias that would take months to develop in larger markets frequently complete within weeks. Yet no public, systematically constructed, daily instrument has existed to measure this short-horizon disposition for Brazilian equities. This paper documents the Tactical Ânima Index, a daily 0–100 composite that measures the short-horizon sentiment position of the Brazilian equity market relative to its own recent history. The index aggregates five categories of observable market behaviour: short-dated options positioning, volatility pricing (constructed as an implied-volatility overlay anchored on a realized-volatility base), price momentum and oscillation, short-horizon market breadth, and participation. Each category is normalized through a rolling-percentile transformation over a short window and combined into a weighted composite renormalized over available components, with a light smoothing pass — a span-5 exponential moving average applied at computation time — carried by the persisted, published series, on which all validation statistics are computed. The index is the short-horizon sibling of the Structural Ânima Index (RPWP-002): the two share a component philosophy but read the market on deliberately different clocks, and their agreement or disagreement defines a configuration vocabulary that neither publishes alone. The index is designed to answer one question: where does the market's short-term disposition sit within its own recent distribution — stretched toward fear, stretched toward greed, or unremarkable? It is explicitly not a forecasting device, not a timing signal, and not a recommendation engine. The name "tactical" refers to the construction window and reading cadence, not to a validated short-horizon return edge: no validated tendency exists at horizons of one week to two months. Version 1.1 adds a cluster-corrected, episode-level analysis in which consecutive and near-consecutive extreme readings are collapsed into independent episodes. At episode level, the only fear-side statistic that exceeds the unconditional baseline of the same decade is an isolated six-month median after extreme pessimism (+11.0% against +8.4%); direction-confirmation rates do not beat the unconditional frequency of positive windows robustly at any horizon, and extreme-optimism episodes carried above-base medians at three and twelve months — momentum-flavoured outcomes, so that nothing in either tail supports a contrarian fade. The record is accordingly reported as what it is: the index measures short-window one-sided behaviour in the Brazilian equity market — where the crowd's fast disposition sits within its own recent range — and its conditional forward-return information at episode level is not distinguishable from the unconditional baseline of the 2016–2026 sample. The failure cases are shown alongside the hits; the drift of the structurally rising sample discounts both tails symmetrically, which is precisely why the reading is conditioned on trend and on the structural sibling. Version 2.0 subjects the record to a full adversarial verification cycle: the smoothing disclosure is corrected (all validation statistics are computed on the published production series, which carries the display smoothing; an unsmoothed vintage is not persisted); an episode-level bootstrap quantifies the master sentence (the isolated six-month excess carries P(median>baseline)=0.77, with confidence intervals that mutually contain the baseline); the structural×tactical grid is reconditioned on the current production vintage of the structural sibling and found materially unstable under that change; the benchmark comparison committed at Version 1.1 is delivered (the composite wins its claimed cell descriptively against trivial single-input gauges but does not dominate outside it, and no series separates from the baseline); band stability is documented as a measurement property; and one legacy registry count is retracted as unreproducible. The paper presents the conceptual framework, construction methodology at the disclosure level appropriate for a public methodology document, interpretation doctrine including the contrarian-conditioned-on-trend rule and the required comparison between structural and tactical readings, historical behaviour across the 2016–2026 window, validation statistics including the episode-level analysis introduced at Version 1.1 and the inferential and benchmark analyses introduced at Version 2.0, and candid limitations.