Aeschbacher, Thomas PhilippThomas PhilippAeschbacher2023-04-132023-04-132022-09-19https://www.alexandria.unisg.ch/handle/20.500.14171/108230In the first paper, I analyze fallen angel bonds' returns before and after their downgrade to high-yield. Fallen angel bonds experience a sharp price decline prior and a sharp recovery after the rerating announcement by the rating agency. I introduce a novel benchmark that should more closely mirror the price fallen angel bonds would have had, had they not experienced a fire sale prior to their downgrade. This allows me to estimate the total preannouncement sell-off and I then use this in order to decide on which fallen angel bonds should be bought after their downgrade. There exists a strong negative relationship between the size of the preannouncement sell-off and future postannouncement returns. I can then show, that an investor fares better, if only fallen angels are bought that trade at a discount to their respective estimated benchmark return had they not experienced the sell-off. Using the introduced innovative, data-driven, novel benchmark allows an investor to generate higher returns. This outperformance gets more pronounced as one focuses on the fallen angel bonds that experienced the highest preannouncement sell-off. In the second paper, we study the cross-section of corporate bonds utilizing a large set of financial statements, equity and bond characteristics. We use a predictive regression framework and the adaptive Lasso to choose the most relevant characteristics for the cross-section of corporate bonds. Applying the adaptive Lasso to the full dataset, we find a ten-factor model, with value, bond reversal, and equity momentum spillover being the dominant factors. Contrary to equity studies, financial variables from Compustat do not appear to have strong power in predicting corporate bond returns. We validate our initial results by running an out-of-sample exercise using an expanding window approach. Out of the 60 months utilized in the out-of-sample, the adaptive Lasso consistently chooses value, bond reversal, and equity momentum spillover. Finally, we evaluate the economic benefits of investing according to the predictions of the adaptive Lasso and find significant benefits in terms of absolute and risk-adjusted returns. In the third paper, we evaluate the ability of U.S. corporate bond fund managers to generate alpha. We apply the False Discovery Rate (FDR) to distinguish between «skill» and «luck.» We find that long-term outperformance remains elusive, with only 1% of the funds able to generate significant alpha over their life. However, fund managers are able to generate alpha over the short-term with the proportion of skilled funds increasing to 13.5% when we examine three-year sub-periods. To confirm these findings, we design an out-of-sample investment strategy where we invest in funds according to their estimated «skill» from past returns. Our strategy generates positive and significant alpha, which confirms the persistence in outperformance over the short-run. Our results are economically meaningful for investors suggesting that dynamic and active manager selection pays off.enIndustrieobligationMaschinelles LernenÖkonometrieKausalitätKapitalanlageEDIS-5262LassoFalse discovery rateEvent studyCausalityMachine learningFactor selectionCorporate bond fundsCapital investmentCorporate bondsSynthetic control methodEconometricsFallen angeldoctoral thesis