MCP servers
MCP server for the BCB Focus survey (market expectations)
The Focus survey is the closest thing Brazil has to a published consensus: a weekly poll of market forecasters covering inflation, policy rate, growth and exchange rate expectations over several horizons. Its analytical value comes almost entirely from its history — what the market expected before a surprise is the only way to measure the surprise.
That makes vintages non-negotiable. This server keeps every weekly release rather than a single current view, so an agent can ask what forecasters expected for a year's inflation as of a specific past week, and receive that week's numbers with their publication date.
Expectations are exposed alongside realized series, which is what turns the survey from a curiosity into a research input: the gap between what was expected and what printed is computable in one place, without stitching two sources together and hoping the dates line up.
hindcast — session
previewhindcast> what did forecasters expect for inflation as of a past week?
→ tool: bcb_focus { indicator:"...", horizon:"...", as_of:"..." }
✓ that week's median and dispersion with its publication date
source: BCB Focus survey, every weekly vintage retained
What's covered
- Tool: Focus expectations by indicator, horizon and as-of week
- Median, mean and dispersion where the survey publishes them
- Every weekly vintage retained rather than a single current table
- Publication date attached to each release
- Realized BCB series available for expectation-versus-outcome analysis
- English field names over the Portuguese source
Join the waitlist — this page's dataset is already selected.
Questions
Why keep old survey vintages at all?
Because a forecast is only meaningful next to the date it was made. Overwriting last month's expectation with this month's makes surprise unmeasurable, which removes the main reason to use the survey.
Does it cover the top-forecaster subsets?
Where BCB publishes them, they are treated as their own series rather than blended into the headline, since the whole point of the subset is that it is not the average.
Related
Join the waitlist. Vote the roadmap. First in, first served.