Rolling Regression

Use window functions to analyze trends over time.

Use Case

Detect when regression relationships change (e.g., price elasticity shifts).

Rolling Window Pattern

SELECT
    date,
    result.coefficients[2] as rolling_elasticity,
    result.r_squared
FROM daily_sales
ORDER BY date
WINDOW w AS (ROWS BETWEEN 60 PRECEDING AND CURRENT ROW)
APPLY anofox_statistics_ols_agg(
    demand,
    ARRAY[price]
) OVER w;

Example: Detecting Momentum Shifts

Stock beta changes over time:

SELECT
    date,
    result.coefficients[2] as rolling_beta,
    LAG(result.coefficients[2]) OVER (ORDER BY date) as prev_beta,
    (result.coefficients[2] - prev_beta) as beta_shift
FROM stock_returns
WINDOW w AS (ROWS BETWEEN 250 PRECEDING AND CURRENT ROW)
APPLY anofox_statistics_ols_agg(
    stock_return,
    ARRAY[market_return]
) OVER w
WHERE date >= '2023-01-01';

Next Steps

• Model Selection — Comparing models
• Reference: Aggregates — Window functions