Investment companies employ a variety of techniques to recommend mutual funds to their customers.
WealthTech company Kidbrooke recently delved into how fund managers can leverage machine learning to optimise mutual fund categorisation.
Frequently, these recommendation strategies rely on the metadata of mutual funds, including aspects such as region, category, or investment objective. This method of classification allows investors to peruse groups of similar funds, aiding in selections aligned with their preferences and investment strategy.
However, this recommendation system is not devoid of risks and limitations, Kidbrooke explained. For instance, it could oversimplify the funds, leading to potential misconceptions about portfolio diversity. Despite sharing common aspects like region or category, funds within a group could exhibit significant variations in performance, risk, and management style. This could also result in investors overestimating the diversity of their portfolios due to heavy reliance on selecting mutual funds from different regions or categories.
A more sophisticated alternative is clustering, an unsupervised machine learning method. Clustering divides a dataset into smaller groups based on some measure of similarity or dissimilarity. Using mutual funds’ historical performance data, this technique groups similar funds together and segregates dissimilar ones into separate clusters. Consequently, investors gain valuable insights into fund relationships and dependencies. The clusters consist of funds exhibiting similar return patterns over time, assisting investors in creating a diversified portfolio based on historical performance rather than manual classification.
Using this technique, investors can scrutinise their existing portfolios to identify and select mutual funds that augment their current holdings. Moreover, investors can find comparable funds within these clusters that may offer superior performance or lower fees.
Clustering also exposes the inadequacies of partitioning mutual funds using metadata. For example, North American funds and Global funds often show high correlation due to a significant proportion of stocks in Global funds being North American. Without this understanding, investors might falsely assume that owning both types of funds increases their portfolio diversity.
The task of selecting an optimal clustering method is not trivial, requiring careful evaluation of different methods. An effective method should account for both the quality and stability of the clusters. The trade-off between these two factors is crucial when deciding on the desired balance between stability, quality, or a mixture of both.
In conclusion, while choosing an appropriate clustering method might necessitate additional steps for achieving optimal results, the benefits are manifold. Using machine learning for mutual fund classification provides a more nuanced understanding of fund performance and diversity, reduces the risk of over-reliance on metadata, and aids in managing portfolio risk.
Read the full report here.
Keep up with all the latest FinTech news here
Copyright © 2023 FinTech Global