What changed when Brazil stopped checking inflation only in December

For 25 years, Brazil checked its inflation target once a year. From 1999 to 2024, the question was whether inflation over the calendar year, measured by the IPCA, ended inside the tolerance band. If December was fine, the year was fine, whatever had happened in between.

Since January 2025 the target is continuous. The center is still 3%, with a tolerance of 1.5 percentage points either way, but the test now looks at 12-month inflation every month. The target counts as missed if that number stays outside the band for six consecutive months. When that happens, the Central Bank has to explain itself in an open letter to the Minister of Finance, with the causes, the measures it is taking, and how long they should take to work.

It sounds like a technical detail. I think it is a good example of how the way you check a rule changes what the rule means.

Once a year is a sample of one

Under the old system, each year produced a single observation. Inflation could be above the ceiling from March to October and still count as a success if it came back down by December. The opposite could also happen: a shock in the last months of the year could turn an otherwise reasonable year into a breach.

So the result depended partly on the calendar. December carried all the weight, and the other eleven months carried none. If you were designing a control from scratch, you would probably not choose to look at the system on one fixed day a year.

Every month, but not every blip

The obvious alternative is to check every month and call a breach whenever 12-month inflation crosses the line. The new rule does not do that, and I think the six-month condition is the interesting part.

Twelve-month inflation moves slowly, because each new reading shares eleven months with the previous one. A single bad month, after a drought pushes up food or a jump in oil prices, can push it over the edge for a while and then drop out. A one-month trigger would turn many of these episodes into formal failures, even though monetary policy could have done little about them, since interest rates take a long time to affect prices.

Requiring six consecutive months is a persistence filter. It is the same idea used in monitoring systems everywhere: an alarm that fires on every spike gets ignored, so you ask the condition to hold for a while before acting. The cost is that a real problem takes longer to be declared.

The rule was tested quickly. In June 2025, 12-month inflation reached 5.35%, the sixth consecutive reading above the 4.5% ceiling, so the target counted as missed under the new framework in its first year.

What the rule is really for

None of this changes the interest rate directly. The target, the band and the letter are about accountability. They define when the Central Bank owes an explanation, and what that explanation has to contain.

That is why the design matters. A target checked once a year can be met by luck and missed by luck. A target checked every month without any filter would produce noise. The rolling version with a persistence condition tries to sit between the two, and its result says more about whether inflation has really been out of control for a while.

I find the parallel with risk management useful. When a bank tests a risk model, counting how many days it was breached is only the start, and the timing of the breaches matters as much as their number. A set of rules is always also a set of measurements. Before deciding whether a target was met, it is worth asking how often anyone looked, and what would have counted as an alarm.