where Liquidationpi are the predicted values from the first-stage regression. The second-stage regression mirrors the OLS regression in equation (1), except that it relies on the exogenous componentof Liquidation—i.e., the component that is induced by the randomization of bankruptcy judges.In all regressions, we cluster standard errors at the division by year level. Doing so accountsfor any arbitrary correlation of the error terms within bankruptcy courts. Lastly, we weight allregressions by the inverse of the number of establishments operated by the bankrupt firm to ensurethat each firm receives the same weight and hence avoid overweighting large bankruptcy cases.If the conditions for a valid instrumental variable are met, captures the causal effect of Chapter7 liquidation on local employment and other outcomes of interest, relative to reorganization. It isimportant to note that the estimates in the instrumental variables analysis are coming only from thesensitive firms—i.e., firms that switch bankruptcy regimes because they were randomly assigned to ajudge who commonly converts cases to Chapter 7 (Imbens and Angrist (1994)). Clearly, some firmswill stay in Chapter 11 no matter the judge, and other firms will convert to Chapter 7 regardlessof the judge. Thus, the instrumental variables estimates only capture the local average treatmenteffect on the sensitive firms, and should be interpreted as such.