1{2 "task_id": "F1-conf-03",3 "n_analysed": 1800,4 "n_firms": 300,5 "predictors_used": [6 "leverage",7 "cash_ratio",8 "log_sales"9 ],10 "variables_excluded": [11 {12 "variable": "rating_2025",13 "reason": "Final-vintage field assigned in 2025 using realised default history through 2022 and sample-average leverage; not known at end of origin year t (look-ahead / outcome leakage)."14 },15 {16 "variable": "default_next",17 "reason": "Outcome variable (known only at end of year t+1); used as dependent variable, not as a predictor."18 },19 {20 "variable": "firm_id",21 "reason": "Identifier; used only for clustering standard errors, not as a predictor."22 },23 {24 "variable": "year",25 "reason": "Origin-year index; kept in analytic.csv for identification but not entered as a regressor (task model specifies leverage, cash_ratio, log_sales only)."26 }27 ],28 "beta_leverage": 0.263784,29 "se_leverage_cluster": 0.040757,30 "ci95_leverage_cluster": [31 0.183901,32 0.34366633 ],34 "r2": 0.034906,35 "default_rate": 0.121111,36 "coef_all": {37 "const": 0.089979,38 "leverage": 0.263784,39 "cash_ratio": -0.185948,40 "log_sales": -0.00734241 },42 "se_all_cluster": {43 "const": 0.043516,44 "leverage": 0.040757,45 "cash_ratio": 0.058886,46 "log_sales": 0.00762647 },48 "se_leverage_nonrobust": 0.035821,49 "cluster_se_type": "statsmodels CR1 (small-sample corrected), G = number of firms",50 "feasibility": "complete",51 "feasibility_reason": null,52 "data_issues": [53 {54 "issue": "40 exact duplicate rows (80 rows involved in duplicated firm_id-year keys); all duplicates identical across all columns",55 "handling": "kept first occurrence, dropped the copies (ledger step 3)"56 },57 {58 "issue": "default_next missing for 300 unique firm-years (origin years [2022]), because year t+1 is after the data cut",59 "handling": "excluded from estimation sample, not imputed (ledger step 4)"60 },61 {62 "issue": "leverage has a point mass of 54 estimation-sample rows at its minimum 0.02 (apparent floor/censoring in the extract)",63 "handling": "retained as recorded; noted as limitation"64 },65 {66 "issue": "rating_2025 is a final-vintage field built from realised outcomes (perfectly separates defaulters: only rating D firms ever default)",67 "handling": "excluded from the analytic dataset (ledger step 2, variables_excluded)"68 }69 ],70 "n_input_rows": {71 "firm_panel.csv": 214072 },73 "random_seed": null74}