Insurers' and banks' market connectedness: generalized event study estimates from random forest residuals regression

This paper proposes a new methodology to examine spillover effects using insurer/bank connectedness as an example. Using large standard deviation jumps in daily returns to measure market shocks, our generalized event study approach can address the issues of multiple events happening in one day, positive and negative responses, asymmetries in the connected responses and endogeneity issues. We employ a random forest residuals regression approach that offers more flexibility in the connectedness relationships than standard regression models, yielding results that are more consistent with the literature (i.e., the predominance of contagion over competitive effects in connectedness). We show that standard linear regression models do not appropriately capture these relationships because they do not account for the endogeneity of the shocks to the daily returns. We find evidence that insurers' and banks' returns move in the same direction after shocks, indicating market contagion effects, rather than in the opposite direction, which would indicate market competitive effects. We also find that the event shocks of January, April, July, and October are larger and more statistically significant than other months. The evidence shows that bank shocks are relatively more destabilizing than insurer shocks.