Predicting Future PTP LBO Announcements

Abstract

This thesis examines how accurately future public-to-private leveraged-buyout (PTP LBO) announcements can be predicted using an ensemble of machine learning models. The models are trained on a novel dataset of 1,805 PTP announcements across more than 40 countries from 2000 to 2023 and tested out-of-sample on 782 PTP LBO announcements in the same period. Specifically, the thesis investigates i) which machine learning model and associated hyperparameters best predict PTP LBO announcements, and ii) which firm-specific features are most important in predicting the probability of a PTP LBO announcement. To guide the model design and feature selection, a theoretical examination of the key motivations behind PTP transactions, an empirical review of the LBO determinants and takeover prediction literature, and industry expert interviews were conducted. The final ensemble model, consisting of random forest and XGBoost models selected via two-level cross-validation, achieves strong out-of-sample recall performance, which measures how well the model identifies actual PTP LBO announcements, particularly in the year preceding LBO announcements. While precision, the proportion of predicted PTP LBOs that are correct, remains low due to a high number of false positives, the model significantly outperforms random classification, providing a meaningful signal on likely targets. Feature importance analyses, drawing from regression coefficients, tree-based splits, and permutation methods, consistently highlight four predictive constructs: Capital Structure (Total Debt/Equity, Net Debt/EBITDA), Valuation (EV/EBITDA, EV/EBIT), Firm Size (Log of Assets, Log of EV, Market Share-%), and Agency Conflicts (Dividend Payout Ratio, Change in Dividend Payout Ratio). The findings suggest that while PTP LBO activity is inherently difficult to forecast with high precision, firm-level financial indicators contain predictive information that can be systematically exploited. The results contribute to the literature by demonstrating the feasibility and limitations of forecasting LBO activity internationally and offer practical implications for investors engaged in deal origination and target screening.