Como a IA reage a cisnes negros?

Eventos de baixa probabilidade e alto impacto, por definição, não têm precedente suficiente nos dados históricos para serem previstos com precisão. O modelo não tenta prever o evento em si — ele mantém filtros de risco que reduzem exposição automaticamente quando a volatilidade observada excede os limites configurados, limitando o dano em vez de antecipar a causa.

Qual a frequência das atualizações dos dados?

A base de dados é atualizada continuamente conforme novas informações de mercado chegam às fontes integradas. Os modelos preditivos são recalculados em ciclos regulares, e qualquer recomendação ativa é revisada sempre que um novo ciclo é concluído.

Como a transparência é garantida?

Todas as recomendações emitidas são registradas em um log público, junto com os dados e premissas que as originaram. O histórico de acerto e erro permanece visível, incluindo os períodos de desempenho abaixo do esperado, na seção de logs de performance.

Data intelligence platform for investors

Smart Decisions in Real Time

Valitância processes large volumes of market data and converts it into actionable recommendations, with risk filters applied before each suggestion reaches you. No recommendations are published without going through the risk mitigation layer.

Explore Demo View Performance Logs

Predictive accuracy (90d)

73.2%

Max drawdown registered

-11.4%

Backtesting cycles

1,248

Simulated historical data, presented for illustrative purposes. See the performance logs section for the complete methodology.

Methodology

How the model arrives at a recommendation

The process is divided into four sequential steps. Each of them is registered, which allows you to retroactively audit any recommendation issued by the platform.

Data ingestion

Series of prices, volume, liquidity and macroeconomic indicators are collected from multiple sources and normalized on a single, continuously updated basis.

Predictive modeling

Stochastic analysis models generate probability scenarios for different time horizons rather than a single deterministic projection.

Risk filtering

Each scenario is compared to previously defined exposure and volatility limits. Scenarios outside the risk tolerance are discarded before reaching the user.

Final recommendation

The resulting recommendation includes the logic that originated it, allowing the investor to evaluate the assumptions before deciding.

Verifiability: every recommendation is accompanied by a record of data and assumptions that generated it.

Transparency in Numbers

Below, an excerpt from the public performance log. Each line corresponds to a closed evaluation cycle, with historical data — there is no projection of future results in this table.

Period	Strategy	Simulated return	Maximum drawdown	Predictive accuracy
Jan–Mar 2024	Conservative	+6.1%	-4.2%	71.8%
Apr–Jun 2024	Moderate	+9.4%	-7.9%	69.3%
Jul–Sep 2024	Moderate	-2.7%	-11.4%	64.1%
Oct–Dec 2024	Conservative	+5.0%	-3.6%	74.6%

Historical data simulated in a backtesting environment, presented for illustrative purposes on how the model works. Past performance, whether simulated or real, does not constitute a guarantee of future results.

Simulated return:: percentage change in the hypothetical portfolio in the period, before operating costs.
Maximum drawdown:: greatest accumulated loss observed between a peak and a valley within the evaluated period.
Predictive accuracy:: proportion of scenarios in which the direction predicted by the model coincided with the effective movement of the asset.

Risk management tools

What makes up the analysis

Three modules work together to reduce the distance between the volume of data available and the decision that needs to be made.

Predictive Volatility Analysis

Estimates of range of variation for each monitored asset, recalculated with each new batch of data received by the platform.

Portfolio Optimization via AI

Reallocation suggestions based on correlation between assets and exposure limits defined by the investor himself.

Real-Time Risk Alerts

Notifications issued when a monitored asset exceeds configured risk parameters, before the variation is consolidated.

Practical applications

Who uses this type of analysis

The same modules serve different profiles, with risk parameters and time horizon adjusted to each case.

Institutional Hedging Strategies

Management desks use real-time risk alerts to calibrate protective positions, reducing the portfolio's net exposure in high volatility scenarios identified by the predictive model.

Net exposure reduction observed in backtesting

-18.6%

Diversification for individual portfolio

Individual investors use portfolio optimization to distribute positions between assets with low correlation, replacing decisions made in reaction to short-term news with previously defined parameters, which reduces the weight of the emotional factor in trading.

Different assets considered per rebalancing cycle

12–20

Validity — data analytics team and infrastructure behind the platform

About the operation

A platform designed for those who are wary of black boxes

Valitância was structured on the premise that a cautious investor needs to understand the reasoning behind a recommendation, not just receive it. Therefore, each step of the model — from data ingestion to risk filtering — is recorded and available for consultation.

This does not eliminate the risk inherent in any investment decision. The goal is to make this risk visible and measurable, rather than assumed.

Know the complete methodology

FAQ

Common questions from cautious investors

The questions below reflect the issues most raised by those evaluating incorporating an AI model into the decision process.

How does AI react to black swans?

Low-probability, high-impact events, by definition, do not have enough precedent in historical data to be accurately predicted. The model does not attempt to predict the event itself — it maintains risk filters that automatically reduce exposure when observed volatility exceeds configured thresholds, limiting the damage rather than anticipating the cause.

How frequently are data updates?

The database is continually updated as new market information reaches the integrated sources. Predictive models are recalculated in regular cycles, and any active recommendations are revised each time a new cycle is completed.

How is transparency guaranteed?

All recommendations issued are recorded in a public log, along with the data and assumptions that gave rise to them. Hit and miss history remains visible, including periods of underperformance, in the performance logs section.