1[2 {"id":"C1","text":"In the estimation sample of 1,800 firm-years (300 firms), a one-unit increase in end-of-year leverage is associated with a 0.264 higher probability of default in the next year, holding cash ratio and log sales fixed.","results_keys":["beta_leverage","n_analysed","n_firms"],"status":"supported","caveat":"Associational linear probability model on simulated data."},3 {"id":"C2","text":"The association is statistically distinguishable from zero: firm-clustered SE 0.041, 95% CI [0.184, 0.344].","results_keys":["se_leverage_cluster","ci95_leverage","p_leverage"],"status":"supported","caveat":"Clustered SE uses statsmodels' default small-sample correction with 300 clusters."},4 {"id":"C3","text":"rating_2025 was excluded because it is known only in 2025 and is built from realised defaults, so it leaks the outcome.","results_keys":["variables_excluded","predictors_used"],"status":"supported","caveat":"Every default in the data has rating D, which is consistent with leakage."},5 {"id":"C4","text":"The estimation sample comes from removing 40 exact duplicate rows and 300 origin-year-2022 rows whose outcome was not observed, with no imputation.","results_keys":["n_input_rows","n_analysed","data_issues"],"status":"supported","caveat":"See out/exclusions.csv."},6 {"id":"C5","text":"The model explains little of the variation in default (R² = 0.035); the baseline default rate is 12.1%.","results_keys":["r2","default_rate"],"status":"supported","caveat":"Low R² is typical for a binary outcome with an LPM."}7]8