ashwingopalsamy/brazil-selic-rate-dataset

Brazil BCB Selic Target Rate History

An independent, reproducible dataset of the Banco Central do Brasil selic target, built from the BCB’s own published series and statements. Every row cites its source, every rate is an exact number of basis points, and the whole dataset is rebuilt automatically when a bank publishes a decision. It is part of Central Bank Policy Rates, which covers seven banks in one schema.

Current build: data observed through 7 Oct 2026; daily coverage from 1 Jan 2000; 151 rate changes and 9,777 daily rows.

At a glance

Country	Central bank	Instrument	Rate now	Since	Decisions	Series from
Brazil	Banco Central do Brasil	Selic target	13.75%	17 Sept 2026	being added	1 Jan 2000

Questions this dataset answers directly:

• What was the selic on a given date?
• How long did each tightening and easing cycle last, and how far did it go?
• How much did the rate change in a given year?
• Which day-level rate should I join to my daily or monthly data?

This is not an official product of any central bank. For anything that matters, follow the source_url or statement_url on the row to the bank’s own publication.

Configurations

Configuration	Row grain	Rows	What it holds
rates (default)	One row per change in the rate	151	Every change point of the policy rate, with the level before and after, its era and its source.
daily	One row per calendar day per country	9,777	The rate in force on every calendar day, from 1 Jan 2000 (or the first point of the series, if later) to the date the official series was last observed. Ready to join to any daily or monthly data.
cycles	One row per tightening or easing cycle	17	Each run of consecutive moves in one direction since the policy-rate era began, computed exactly as on the site: a cycle starts with the first move after a move the other way.
annual	One row per country per calendar year	27	Yearly summaries computed from rates and decisions: the rate at each end of the year, its range, the net and gross change, and how many hikes, cuts and holds there were.
countries	One row per country	1	What each country’s data covers: the bank, the instrument, the dates covered and the release it was built from.
eras	One row per instrument period	1	The periods in which one instrument and basis applied. A change of era is why change_bps is null at its first point.
sources	One row per source	2	Every source the rows cite: official series, statements, calendars and labelled secondary sources.
transmission	One row per loan product per country	2	How the policy rate reaches common loans: the benchmark each is priced off, how often it resets and how directly a change passes through. Curated reference notes.

Every configuration has one split, full, because this is a historical record rather than a train/test corpus. The Parquet files under data/ are the only files mapped to configurations. CSV copies of every table and JSONL copies of rates are under exports/, deliberately left out of the configurations so the viewer does not count them twice. Column types are in schema/ and every column is described in schema/data-dictionary.json.

How to read the fields

• Rates are exact. rate_bps is an integer number of basis points; rate_pct is the same value in percent. For a target range (the Fed since 16 December 2008), rate_pct is the upper bound and rate_low_pct/rate_high_pct give both ends. A midpoint is never computed.
• rates is the canonical history. One row per change, with the rate before and after. change_bps is the signed change against the previous upper bound; it is null for the first point and at an era boundary, where an instrument or its basis changed and the two numbers are not comparable.
• Eras say what the number is. era_basis is policy for the policy rate itself and observation for an earlier proxy kept for history. The eras configuration explains each boundary.
• Evidence is explicit. evidence is official for the bank’s own series or statement and secondary for a cited secondary source. Every row carries a source_id that joins to sources.
• Series-only countries. Brazil has the full official series of rate changes, but announced decisions (and so holds and votes) are still being added. Their rows in annual carry null hold_count and decision_count rather than zero.
• daily gives the rate in force on every calendar day, including weekends, so it joins directly to other daily or monthly data.
• cycles are runs of consecutive moves in one direction, computed the same way as on Policy Rate Atlas: a new cycle starts with the first move after a move the other way. Only policy-era moves count.
• annual uses the rate in force on 1 January (the last change before it) and on 31 December, or the last observed date for the current year. is_partial_year marks a first or current year that the data does not fully cover.
• record_text is one deterministic sentence built from the fields, for search, retrieval and agents. No text is written by a language model.

Coverage

• Brazil. Every change in the Selic target from BCB SGS series 432 (daily). Decision records are being added.

Load it

from datasets import load_dataset

repo = "ashwingopalsamy/brazil-selic-rate-dataset"
rates = load_dataset(repo, split="full")                 # every change, the default configuration
daily = load_dataset(repo, "daily", split="full")

import pandas as pd

daily = pd.read_parquet("hf://datasets/ashwingopalsamy/brazil-selic-rate-dataset/data/daily.parquet")
monthly = daily.set_index("date")["rate_pct"].resample("ME").last()

import polars as pl

cycles = pl.read_parquet("hf://datasets/ashwingopalsamy/brazil-selic-rate-dataset/data/cycles.parquet")
print(cycles.filter(pl.col("is_current")))

-- DuckDB
SELECT year, start_rate_pct, end_rate_pct, net_change_bps, hike_count, cut_count
FROM 'hf://datasets/ashwingopalsamy/brazil-selic-rate-dataset/data/annual.parquet'
ORDER BY year;

Provenance and verification

Each country’s data comes from one content-addressed release, also served by the site’s open API at https://rates.ashwingopalsamy.in/api/v1/:

Country	Release (SHA-256 content hash)	Observed through
Brazil	59c6949350fb38008d952ac1510c310429ae0f266d9f4524660a0ba9ff53155a	7 Oct 2026

provenance/build-manifest.json records those releases, the row count of every configuration and a checksum of every file; SHA256SUMS covers everything. The build is byte-for-byte reproducible from the source repository:

npm ci && pip install -r requirements-hf-dataset.txt
npm run build:hf && npm run test:hf-dataset

Attribution and terms

Each publisher’s own terms apply to its data. NOTICE.md lists them with the attribution each asks for. This dataset is offered under the ODbL 1.0, as its source is.

Limitations

• Coverage differs by country: announced decisions are not yet available for Brazil. See countries.coverage_note.
• Dates are the bank’s local dates. Historical announcement times for India before the published calendar use 10:00 IST as a convention.
• transmission and events are curated context, not exhaustive and not causal claims.
• Nothing here is a forecast or advice.

Citation

@dataset{gopalsamy_brazil_selic_rate_dataset_2026,
  author    = {Gopalsamy, Ashwin},
  title     = {Brazil BCB Selic Target Rate History},
  year      = {2026},
  version   = {1.0.0},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/ashwingopalsamy/brazil-selic-rate-dataset}
}

Changes between versions are in CHANGELOG.md. The live view of this data is Policy Rate Atlas.