run-regression

name	run-regression
description	Stage 2. Run a CAPM (and optionally 3-factor) regression on the fetched data. Write coefficients to state. Triggers when state.stage == "fetched".

Run regression

Pre

Read state/state.json. Confirm state.stage == "fetched" and state.artifacts.data_path exists.

Act

Load data/returns.csv. For each permno (or for the equal-weighted portfolio if only one cross-section), run:

ret_excess = ret - rf
ret_excess ~ mkt_rf            (CAPM)

Use statsmodels.OLS with HAC (Newey-West, 3 lags) standard errors. Capture: alpha, beta, alpha_se, beta_se, n_obs, r2, converged.

Verify

• converged == True
• n_obs >= 24
• abs(alpha) < 0.10 (sanity bound on monthly alpha)
• 0 < beta < 3 (sanity bound)

If any check fails: halt with reason.

Post

state.artifacts.regression = {alpha, beta, alpha_se, beta_se,
                              n_obs, r2, t_alpha, t_beta}
state.stage = "regressed"
state.last_action = {...}

Run python scripts/verify_state.py.