In diesem Forschungsprojekt untersuchen wir die Performance und das Trading-Verhalten von aktiv-verwalteten Investmentfonds mit Hilfe eines globalen Datensatzes im Zeitraum von 1980 bis 2020. Wir stellen die folgenden Forschungsfragen: (1) Zeigen Fonds die beste Anlageperformance, wenn es für die Anleger am wichtigsten ist? (2) Was treibt die Performance von Investmentfonds und wie handeln diese sogenannte Marktanomalien? (3) Sind Techniken des maschinellen Lernens hilfreich, um die zukünftige Performance von Investmentfonds vorherzusagen?
In Phase 1 des Projekts, evaluieren wir die Anlageperformance von Investmentfonds in verschiedenen Marktphasen und beantworten die Frage, ob aktive Fondsmanager/-innen in Krisenzeiten für Investoren Mehrwert schaffen. Unsere Analyse ist fokussiert auf die korrekte Messung von Performance und kann die Frage beantworten, ob es für Anleger lohnt in Zukunft in aktive Investmentprodukte zu investieren oder in passive Produkte zu wechseln.
Phase 2 des Projekts versucht die Treiber von Fondsperformance zu verstehen, indem das konkrete Handelsverhalten von Fondsmanager/-innen untersucht wird. Wir verwenden hierzu detaillierte Portfolioinformation von Fonds und evaluieren, wie sie spezielle Marktanomalien in ihren Handelsstrategien berücksichtigen. Wir können daraufhin bestimmen, welche Faktoren die Performance eines Fonds beeinflusst haben und ob Fondsmanager/-innen Anomalien profitabel ausnutzen.
In Phase 3 sagen wir zukünftige Fondsperformance mit Hilfe von individuellen Fonds- und Manager-Charakteristiken voraus. Wir benutzen hierzu in unseren Prognosemodellen Techniken des Maschinellen Lernens (z.B. Random Forests und Neurale Netzwerke) um eine bestmögliche Vorhersagekraft zu erzielen. Unsere Ergebnisse sind für Investoren höchst relevant und können für eine erfolgreiche Manager-Selektion benutzt werden.
Mutual funds have emerged as primary investment vehicles for households worldwide: Over the past decade, the number of regulated open-ended funds has increased from 86'301 in 2010 to 122'528 in 2019 with total assets under management of USD 54.9 trillion (Investment Company Factbook, 2020). Although only 8% of all mutual funds are domiciled in the United States (and, hence, 92% of all funds are located in other countries), academic literature on the topic has almost exclusively focused on investigating US funds. As a consequence, the evaluation, understanding, and prediction of mutual fund performance worldwide is not well understood.In this research project we empirically study the performance and trading behavior of actively-managed mutual funds on a global scale using various econometric techniques. Specifically, we ask the following research questions: 1. Do mutual funds outperform when it matters most to investors? 2. How do mutual funds trade cross-sectional stock market anomalies? 3. Are machine learning techniques helpful to predict future mutual fund performance?In Phase 1 of the project, we evaluate state-dependent performance of actively-managed mutual funds. While it is generally established that, unconditionally, active funds underperform their benchmark, we examine whether active funds deliver value for investors during periods of economic downturn. In a pre-test, we find that risk-adjusted performance of mutual funds is indeed positive with annualized 0.35% during recessions periods. We expect to confirm this result also on our full data sample and to answer research question 1 with 'yes'. Hence, our research can solve the puzzle why investors still prefer to invest in active mutual funds (instead of passive ones) worldwide.Phase 2 seeks to understand mutual fund performance by studying funds' trading behavior. To do so, we exploit detailed equity portfolio holding data and examine how funds engage in the trading of 241 cross-sectional stock market anomalies (such as, e.g., value, momentum, or reversal). We expect that successful mutual fund managers trade anomalies wisely, in a way that they are buying (selling) stocks that are undervalued (overvalued). Our results will answer research question 2 and shed light on the actual trading strategies of mutual fund managers on a global scale.In Phase 3, we predict mutual fund performance using 30 individual fund- and manager characteristics (such as, e.g., turnover, active share, or manager experience). For this purpose, we apply machine learning techniques, such as boosted regression trees and neural networks, which are particularly suited in such a high-dimenstional setting. In a pre-test, we already show that -- when combining eight common mutual fund characteristics to a joint predictor variable -- a standard deviation increase in the predictor leads to higher future fund performance of 2.5% p.a. Hence, machine learning is expected to be of great value in the prediction of mutual fund performance.To answer our research questions we will apply the most comprehensive global mutual fund database up-to-date. It is composed by merging data from (i) reported returns and characteristics, (ii) detailed equity portfolio holdings, and (iii) manager characteristics for individual mutual funds in more than 39 countries over the period from 1980 to 2020.
Last updated:19.01.2023
