QuackStats

Time series forecasting and statistics as a DuckDB extension. Run forecasts, detect seasonality, and perform statistical analysis directly in SQL.

Features

• Multiple forecasting models - ETS (Error-Trend-Seasonality), linear regression, exponential growth, logistic (S-curve), and automatic model selection via cross-validation
• Prediction intervals - Configurable confidence bounds on all forecasts (default 95%)
• Seasonality detection - Identify periodic patterns in your time series data
• Multi-group support - Forecast multiple series independently in a single query using group_by
• Automatic interval detection - Handles daily, weekly, and monthly timestamp spacing

Installation

INSTALL quackstats FROM community;
LOAD quackstats;

Usage

Forecasting

-- Basic forecast
SELECT * FROM forecast(
    'sales',
    timestamp := 'date',
    value := 'revenue',
    horizon := 30
);

-- Grouped forecast with model selection
SELECT * FROM forecast(
    'sales',
    timestamp := 'date',
    value := 'revenue',
    horizon := 30,
    group_by := ['region', 'product'],
    model := 'auto',
    confidence_level := 0.95
);

Output columns: group columns (if specified), forecast_timestamp, forecast, lower_bound, upper_bound

Available models:

Model	Description
ets	Error-Trend-Seasonality (default)
linear	Linear regression with prediction intervals
exponential	Exponential curve fitting (y = ae^bx)
logistic	S-curve fitting for bounded growth
auto	Cross-validation across all models, picks the best

Seasonality Detection

SELECT * FROM detect_seasonality(
    'sales',
    timestamp := 'date',
    value := 'revenue',
    group_by := ['region']
);

Output columns: group columns (if specified), period, strength

Building from Source

make configure
make debug

Testing

make test_debug

License

MIT © 2026 Alytic Pty Ltd

QuackStats is maintained by the team at Kyomi and is the forecasting engine that powers the platform.