The Structural Ânima Index: Measuring Persistent One-Sided Market Behaviour in Brazilian Equities

Radar Perene Working Paper Series, No. RPWP-002 · Version 2.3 · July 2026 Investor sentiment is among the oldest intuitions in finance and among the hardest to measure with discipline. Market participants, journalists, and policymakers routinely speak of "euphoria" and "capitulation" as if these states were directly observable, yet most references to market mood in Brazil rest on anecdote, on survey instruments imported from other markets, or on single indicators read in isolation. The problem this paper addresses is the absence of a systematic, daily, multi-input measure of persistent one-sided market behaviour — the standing disposition that sentiment language gestures at — constructed specifically for the Brazilian equity market. The Structural Ânima Index summarizes that behaviour — the standing risk disposition participants carry into each session, not what they did today. It is a composite index computed daily over the Brazilian trading calendar and scaled from 0 to 100, where high readings indicate a persistently risk-seeking disposition and low readings a risk-averse one. Four categories of market-derived inputs — options-market positioning, credit conditions, market breadth, and a foreign-flow proxy — are each transformed into rolling percentiles against their own trailing 252 trading days, roughly one year of market memory, and aggregated into a single reading. Version 2.0 revised the construct: a former price-trend component, which by construction restated the market's own price, was replaced with the credit-conditions component, and the index's dependence on price fell materially — a reduction that survives formal inference, a monthly-frequency replication, and an out-of-sample split (Section 7.1a). This version also reports the tests that qualify that result: placebo replacements — white noise, and, new in this version, slow monthly-cadence series matched to the credit component's own temporal structure — showing that the raw reduction is available in full to any slow non-price replacement; the credit component's specific merit is exclusively the information it adds about financial stress, which the placebos add none of. A race against trivial single-input benchmarks remains mixed and is reported as such, with its one dated miss now shown to be materially an artifact of a documented data-suppression window (Sections 7.1b, 4.1), and the band thresholds and episode rule behind the conditional statistics are stress-tested over a full sensitivity grid (Section 7.2b). Two of the four inputs remain price-derived, so the index still does not fully separate disposition from the conditions that produce it — a boundary Section 3.2 states plainly. The vocabulary of "mood" is retained as interpretive shorthand for that behavioural object, distinct from flow (daily action), which the index does not measure. Because every input is expressed relative to its own history, the index reads position within regime rather than raw level. Public interpretation bands are extreme optimism above 65, a neutral zone between 35 and 65, and extreme pessimism below 35; readings at extremes are interpreted contrarily, but always conditioned on the prevailing long-term trend, since an extreme means different things above and below the market's long-term average. The index is designed for regime description — situating today's collective disposition within the market's own emotional history and contextualizing episodes such as 2016, 2020, and the 2024 fiscal scare. It is not a forecasting device, not a trading signal, and not investment advice; it does not predict the direction of the Ibovespa, the timing of reversals, or the resolution of any episode. Where historical statistics are reported, they are conditional distributions with documented failure cases, verified against the production database as of July 2026 and printed beside the corresponding unconditional base rate. An episode-level analysis — introduced in Version 1.1 and re-verified here on the revised series — collapses consecutive extreme days into independent episodes and finds that neither tail robustly exceeds the base rate at any horizon; the apparent day-level exceedance is a clustering artifact. The index's documented value is measurement, not prediction of what follows it. ---