Brasa DB Explorer

BrasaDB Explorer

Execute SQL queries against the brasa DuckDB database to explore, analyze, and combine Brazilian financial market data.

Setup: Creating Views

Views must exist before querying. If queries return empty results or "table not found" errors, create the views first:

uv run python -c "from brasa import create_all_views; create_all_views()"

This only needs to be done once per Python process. If you open a new terminal or restart the kernel, run it again.

If a query returns empty results but no error is raised, first verify your filter values by sampling the table without filters (for example, SELECT DISTINCT symbol FROM "staging.b3-cotahist" LIMIT 20) before assuming views are missing.

Connecting to DuckDB

Don't open the database file by hand. The BrasaDB class is the easy path: it owns a single shared connection to brasa.duckdb and exposes helper methods for running queries. Because every caller reuses that one connection — already opened with read_only=False and re-established automatically if it drops — you never construct the database path yourself, and you sidestep the error DuckDB raises when read-only and read-write connections are mixed in the same session (brasa holds a write connection internally).

The most convenient entry point is the module-level sql() helper, which delegates to BrasaDB.query():

uv run python -c "
from brasa import sql
print(sql('''
<SQL QUERY HERE>
''').df().to_string())
"

sql() returns a DuckDB relation. Call .df() for a pandas DataFrame — chain .to_string() so wide or long results print in full — or .fetchall() for a list of tuples.

The same functionality is available as methods on the class, which is handy when you want discovery and querying in one place:

• BrasaDB.query(sql) — run a query (this is what sql() calls)
• BrasaDB.list_tables() — list every layer.dataset view
• BrasaDB.create_all_views() — (re)create all views; see Setup above
• BrasaDB.get_connection() — the shared duckdb.DuckDBPyConnection, if you need raw access

BrasaDB resolves the database path through brasa's CacheManager, so it honors the BRASA_DATA_PATH environment variable (defaulting to .brasa-cache/) without any manual path handling.

Table Naming Convention

Tables follow the pattern "layer.dataset-name" and must be double-quoted in SQL because they contain dots and hyphens:

SELECT * FROM "staging.b3-cotahist" LIMIT 10

Datasets are organized in layers. Default to the staging layer — it holds the cleaned, enriched, analysis-ready tables intended for end users, and it's where you should look first:

• staging — transformed and enriched data ready for analysis (prices, returns, index compositions, macro series, corporate events). Start here.
• curated — a small set of fully unified, analytics-ready datasets built on top of staging (e.g., all returns, all prices).
• input — raw parsed data. This is an internal ingestion layer, not meant for end users. Only reach for an input.* table when a dataset genuinely has no staging equivalent (e.g. some B3 registries, PTAX FX, intraday trades); otherwise the staging version is cleaner and is what you should use.

Choosing the right table

Selection questions arrive as information needs, not table names. Match the need to the canonical table below, and heed the caveat — it's usually what makes the tempting alternative wrong. Prefer these over anything you'd find by keyword-matching a table name. For anything not listed, fall back to the Key Datasets Reference and the discovery queries. (This map mirrors the authoritative Status notes in docs/datasets.md.)

Equity / ETF prices & trading activity

• Daily stock OHLC (unadjusted) → staging.b3-cotahist. Longest history. Stock prices are not adjusted for corporate events.
• Adjusted stock OHLC → ⚠️ does not exist yet. Don't invent a table. Adjusted returns exist (see below); an adjusted price series still has to be built.
• ETF OHLC → staging.b3-cotahist. ETF close is fine unadjusted (most BR ETFs pay no dividends) except NDIV11, DIVD11, SPYI11.
• Equity-option OHLC → staging.b3-cotahist. Longest history.
• Volume / nº de negócios / nº de trades (deepest history) → staging.b3-cotahist.
• "What stocks/assets exist" (registry) → input.b3-bvbg028-equities. Daily registry of everything listed (stocks, fractional…); also market_capitalisation.
• Which stocks trade on the spot market (non-fractional) → staging.b3-equities-spot-market. Full-span since WIL-132: filter refdate = (SELECT max(refdate) FROM "staging.b3-equities-spot-market") for current members.
• Historical listing / registry as of a past date → staging.b3-equities-register (full-span mirror of bvbg028; same-day distribution_id duplicates kept — dedup with max(distribution_id) when you need one row per day).

Returns

• Adjusted stock returns → staging.b3-equities-returns. Built from b3-bvbg086's daily oscillation; history starts 2018. Pre-2018 you only have unadjusted returns computed from staging.b3-cotahist.
• ETF returns → staging.b3-equities-etfs-returns. Longer history — concatenates b3-cotahist (<2018) + b3-bvbg086 (>2018).

Indexes

• What's in IBOV (membership) → staging.b3-indexes-composition (no weights).
• Index weights — official quarterly target → staging.b3-indexes-theoretical-portfolio (rebalance-date weights, valid for the quarter).
• Index weights — live/drifted as of a day → staging.b3-indexes-current-portfolio.
• Index level over time → staging.b3-indexes-historical-prices.

Futures

• Futures daily data (2006-05-22 onward) — prices/OHLC, settlement, open interest, contract metadata → staging.b3-futures. Union of the bvbg join (2016-02+) and the BD_Final bulletin (pre-2016). Settlement lives in adjusted_quote/adjusted_tax, not settlement_value; adjusted_tax is in decimal form (0.09005, not 9.005).
• Futures settlement before 2006-05-22 → not available yet (WIL-128 backlog). input.b3-futures-settlement-prices (frozen, starts 2010) is now only a cross-check source — staging.b3-futures covers its whole range with richer data.
• Contract registry (maturity, multiplier) → staging.b3-futures-register.
• Avoid staging.b3-futures-di1-consolidated (deprecated) and staging.b3-futures-settlement-prices (outdated, ships 2× duplicate rows).

Interest-rate & inflation curves

• DI nominal curve at contract maturities (+ CDI overnight) → staging.b3-curves-di1.
• DI nominal curve at standardized fixed-term vertices (e.g. DI1T252 = 252 business days) → staging.b3-curves-di1-standard.
• Returns of a fixed-tenor DI vertex → staging.b3-curves-di1-standard-returns.
• Real-rate (inflation) curve → staging.b3-futures-dap for now (standardized b3-curves-dap* are planned but not yet in the DB). The rate column is adjusted_tax.

Macro, indicators & FX

• Macro series (CDI, SELIC, IPCA, IGPM) → staging.bcb-sgs. Other SGS series → input.bcb-sgs (by numeric code).
• Indicators to use with B3 contracts/pricing/curves → staging.b3-economic-indicators (this is the pricing companion, not general macro).
• FX / PTAX → input.bcb-currency (input-only; currency ∈ {USD, EUR, GBP, JPY, CHF, CAD, AUD}).

Corporate events

• All events for a stock (one place) → staging.brasa-corporate-events (unified; event_family ∈ {CASH, STOCK, SUBSCRIPTION}).
• Cash dividends & JCP (with yield) → staging.b3-cash-dividends-events; split/bonus factors → staging.b3-stock-events; subscriptions → staging.b3-subscription-events.

Company info, identity & sectors

• Company-level info → staging.brasa-companies. No symbol column — join via code_cvm.
• Ticker ↔ company → staging.b3-companies-symbols.
• Sector / industry classification → staging.brasa-industry-sectors (best source; prefer over the sector columns on brasa-companies).

Funds, options, intraday

• Listed funds (ETF/FII) → staging.b3-listed-funds (fund_type ∈ {ETF, FII, Fixed Income ETF}).
• Option theoretical price / implied vol → input.b3-equity-options; vol surface → input.b3-equities-volatility-surface (both input-only).
• Intraday tick trades → input.b3-trades-intraday[-equities/-derivatives] for deepest history; staging.b3-trades-intraday if its 2023+ coverage suffices.

Discovery Workflow

Use the Key Datasets Reference for common tables listed below. Run discovery queries when the user asks about datasets not listed there, or when a query returns unexpected results.

1. List all tables

SELECT table_name
FROM information_schema.tables
WHERE table_type = 'VIEW'
ORDER BY table_name

2. Inspect a table's schema

DESCRIBE "staging.b3-cotahist"

3. Sample data

SELECT * FROM "staging.b3-cotahist" LIMIT 5

4. Check row counts and date ranges

SELECT
    COUNT(*) as rows,
    MIN(refdate) as first_date,
    MAX(refdate) as last_date
FROM "staging.b3-cotahist"

Key Datasets Reference

These are the most commonly used datasets, with their actual column names verified against the live database. This is a curated subset — for the complete catalog of every view grouped by topic, see docs/datasets.md in the repo, or run the discovery queries above to see every available table and its current schema.

Prefer staging.* tables — they are the cleaned, user-facing datasets. The input.* tables listed below appear only where they expose data that has no staging equivalent (e.g. PTAX FX, intraday trades, options/volatility, some B3 registries); when a staging table covers the same data, use it instead. Column names are case-sensitive: the raw input.b3-company-* tables use camelCase (e.g. issuingCompany, codeCVM) while staging tables use snake_case.

Prices & Trading

Indexes

Returns (staging layer)

Macro & Rates (BCB / ANBIMA)

Corporate Events

Company & Fund Data

Note: the registry views input.b3-listed-stock-etfs, input.b3-listed-fixed-income-etfs, input.b3-listed-cripto-etfs, and input.b3-listed-reits exist but may currently hold no parquet files (queries raise an IO Error: No files found). Use staging.b3-listed-funds for fund data instead.

Equities (staging layer)

Futures & Options

Intraday Trades (input only — no staging equivalent)

Common Query Patterns

Filter by symbol and date range

SELECT refdate, symbol, close, volume
FROM "staging.b3-cotahist"
WHERE symbol = 'PETR4'
  AND refdate >= '2024-01-01'
ORDER BY refdate

Aggregate across symbols

SELECT symbol, COUNT(*) as trading_days, AVG(close) as avg_price
FROM "staging.b3-cotahist"
WHERE refdate >= '2024-01-01'
GROUP BY symbol
ORDER BY avg_price DESC
LIMIT 20

Join datasets

SELECT
    p.refdate,
    p.symbol,
    p.close,
    p.volume,
    r.log_return
FROM "staging.b3-cotahist" p
JOIN "staging.b3-equities-returns" r
    ON p.refdate = r.refdate AND p.symbol = r.symbol
WHERE p.symbol = 'VALE3'
  AND p.refdate >= '2024-01-01'
ORDER BY p.refdate

Find stocks in an index

SELECT symbol, corporation_name, specification_code
FROM "staging.b3-indexes-composition"
WHERE indexes = 'IBOV'
  AND refdate = (SELECT MAX(refdate) FROM "staging.b3-indexes-composition")
ORDER BY symbol

Get a macro series (CDI, SELIC, IPCA, ...)

SELECT refdate, value
FROM "staging.bcb-sgs"
WHERE symbol = 'CDI'
  AND refdate >= '2024-01-01'
ORDER BY refdate

Get FX rates (PTAX)

SELECT refdate, bid, ask
FROM "input.bcb-currency"
WHERE currency = 'USD'
  AND refdate >= '2024-01-01'
ORDER BY refdate

Find dividends/corporate events for a stock

staging.brasa-corporate-events has no price/refdate of trading — use ex_date for timing. Filter by event_family (CASH, STOCK, SUBSCRIPTION) or a specific event_type (e.g. CASH_DIVIDEND, INTEREST_ON_CAPITAL, SPLIT, BONUS_SHARES).

SELECT ex_date, payment_date, event_type, value_cash, yield_pct
FROM "staging.brasa-corporate-events"
WHERE symbol = 'PETR4'
  AND event_family = 'CASH'
ORDER BY ex_date DESC

Creating Derived Views

When the user wants to combine datasets for a specific application, create a persistent view:

CREATE OR REPLACE VIEW "custom.my-analysis" AS
SELECT
    p.refdate,
    p.symbol,
    p.close,
    p.volume,
    r.log_return,
    ic.indexes
FROM "staging.b3-cotahist" p
LEFT JOIN "staging.b3-equities-returns" r
    ON p.refdate = r.refdate AND p.symbol = r.symbol
LEFT JOIN "staging.b3-indexes-composition" ic
    ON p.symbol = ic.symbol AND p.refdate = ic.refdate
WHERE p.refdate >= '2024-01-01'

View names like "custom.my-analysis" use a dot as part of the quoted name, not as a schema separator. DuckDB stores this as a quoted object name in the default schema, so no schema creation is needed.

DuckDB SQL Features

DuckDB supports advanced SQL that is useful for financial analysis:

• Window functions: LAG, LEAD, ROW_NUMBER, RANK, running averages
• CTEs: WITH clauses for readable multi-step queries
• PIVOT/UNPIVOT: reshape data between wide and long formats
• Date functions: date_trunc, date_part, date_diff, interval arithmetic
• Statistical aggregates: stddev, variance, corr, percentile_cont
• List/array functions: list_agg, array_agg for grouping values
• QUALIFY clause: filter window function results directly

Example: Rolling volatility

SELECT
    refdate,
    symbol,
    log_return,
    STDDEV(log_return) OVER (
        PARTITION BY symbol
        ORDER BY refdate
        ROWS BETWEEN 20 PRECEDING AND CURRENT ROW
    ) as vol_21d
FROM "staging.b3-equities-returns"
WHERE symbol = 'PETR4'
  AND refdate >= '2024-01-01'
ORDER BY refdate

Guidelines

• Start with discovery (list tables, describe schemas, sample rows) before writing complex queries
• Always quote table names with double quotes due to dots and hyphens
• Use LIMIT 100 or fewer when exploring unfamiliar tables. Only omit LIMIT when the user explicitly requests full output or when row count is known to be small
• Prefer CTEs over subqueries for readability
• When the user's question is ambiguous, explore the data first and show what's available before making assumptions
• For date filtering, refdate is the standard date column across most datasets
• When creating views for the user, explain what each join and filter does
• Do not execute INSERT, UPDATE, or DELETE statements against existing brasa-managed tables. Only CREATE OR REPLACE VIEW and SELECT operations are safe; if the user requests data modification, explain that brasa manages data ingestion separately