Investigate the causal effect of diversification strategy on risk-adjusted performance using Bayesian additive regression trees

This paper aims to develop a hybrid causal model based on the inverse probability weighting (IPW) and Bayesian additive regression trees (BART) as an advanced machine learning technique. IPW relies on parametric logistic regression model to estimate the propensity scores, however, the required assumptions and pre-specified relationships degrade its application for many real-world problems. We use BART to model the propensity scores to mitigate the limitations of standard IPW model. In addition, we apply Bayesian model to estimate the average treatment effect (ATE) in the pseudo-population instead of simple regression to provide posterior predictive distribution of ATE. Using a simulation study, we show that our model corrects the bias and RMSE introduced by the original IPW model and can recover the true ATE. Lastly, we apply the new IPW model to investigate the causal effect of diversification strategy on risk-adjusted performance for US public firms. The results show diversification can help firms to improve their performance even after considering the associated risk.