Prof. Dr. Gustavo Henrique Valente: Five Signals an AI Model Should Watch in Brazil
There is a popular fantasy about artificial intelligence in financial markets:
Feed enough data into a machine, and it will reveal tomorrow’s price.
That is not how uncertainty works.
Brazil offers a useful case study. The Selic rate currently stands at 14.25% after the central bank’s June reduction. The latest IPCA-15 showed consumer prices increasing 0.41% in June and 4.80% over the previous twelve months.
Inflation slowed on a monthly basis, but the annual picture remained uncomfortable. At the same time, the Brazilian yield curve delivered different messages across maturities: shorter yields declined, while longer-dated yields rose.
One economy. Several messages.
An intelligent model should not force these signals into one bullish or bearish label. It should ask which forces are dominant, how stable they are and what could cause them to change.
Here are five signals I would place on the screen.
1. The shape of the yield curve
The policy rate receives most of the headlines, but the curve often contains more information than the overnight rate itself.
If short rates fall while longer rates rise, the market may be accepting near-term monetary easing while demanding greater compensation for future inflation, fiscal uncertainty or policy risk.
That distinction matters for banks, utilities, infrastructure companies, real estate businesses and any asset whose value depends heavily on distant cash flows.
An AI system should not treat “rates down” as one uniform event.
It should identify where rates moved, how quickly they moved and which maturities are carrying the greatest risk premium.
2. The composition of inflation
June’s IPCA-15 was not one uniform price increase.
Food and beverages and housing were important contributors. Residential electricity produced the largest individual positive impact. Some food products rose sharply, while gasoline and ethanol declined. tails matter because different sources of inflation have different levels of persistence.
A temporary food shock does not carry the same policy meaning as accelerating service inflation. An electricity tariff adjustment is different from broad wage pressure. Imported inflation is different from domestic demand pressure.
A good model should separate regulated prices, tradable goods, services, food and energy rather than compressing everything into one headline number.
3. The real’s dominant risk driver
The Brazilian real can strengthen because domestic interest rates are attractive.
It can weaken because international investors reduce emerging-market exposure.
It can react to commodity prices, fiscal expectations, the global dollar cycle, hedging demand or changes in market liquidity.
A quantitative system should estimate which driver currently explains the largest share of currency movement.
This is more useful than a fixed exchange-rate target.
A target gives one number.
A driver map provides a structure for understanding why that number may change.
4. Correlation during stress
Many portfolios are diversified only in appearance.
A local equity fund, a long-duration bond position and exposure to the Brazilian real may sit in three different sections of an investment report.
Yet all three can be vulnerable to the same shock: a rise in long-term inflation expectations, a reduction in policy credibility or a sudden decline in global risk appetite.
Artificial intelligence can help monitor rolling correlations, but simple historical averages are not enough.
Correlations behave differently in comfortable markets and stressed markets. A model should place greater emphasis on how assets behaved when liquidity became scarce, not only when trading conditions were calm.
An asset that appears independent during normal periods may become highly correlated with the rest of the portfolio when protection is needed most.
Liquidity is often treated as a background condition until it disappears.
Bid-ask spreads, trading volume, futures positioning, market breadth and the relationship between cash and derivatives markets can reveal whether a price movement is supported by broad participation or driven by a narrow group of investors.
A market can rise while its internal liquidity becomes weaker.
It can also fall while risk is being transferred in an orderly and transparent way.
The direction of the index alone cannot tell us which situation we are observing.
It should not manufacture confidence.
It should not convert a probability into a promise.
It should not hide uncertainty behind a precise decimal point.
It should not produce a buy or sell instruction without explaining the risks, assumptions and limitations behind the result.
And it should not encourage an investor to take a risk that the investor does not understand.
The real strength of quantitative research is not that it eliminates uncertainty.
It makes uncertainty more visible.
For Brazilian investors, this means building systems that recognize changing relationships among inflation, interest rates, the currency, liquidity and asset prices.
The output should not be a dramatic trading command.
It should be a clearer map of the risks already present in the portfolio.
A model that says “I am less certain” may be more valuable than a model that always gives an answer.
It is intellectual discipline.
https://www.profdrgustavohenriquevalente.com/
