BLOG. 3 min read
As the world of exchange-traded funds (ETFs) continues to expand into new investment strategies and asset types, SS&C Algorithmics has continued its in-depth research on sparse fixed-income portfolio construction. ETF managers who run passive investment strategies against popular market indices often use optimization techniques to control instrument selection, active exposures and relative risk/return indicators vs. their benchmark. These techniques may not all be part of core optimization books and can include intuitive stratified sampling approaches or advanced single-purpose heuristics as well.
Our previous "ETFs – Machine Learning-Enhanced Sparse Portfolio Construction" research report delved into the impact of using various machine learning (ML) techniques and/or “conventional” optimization algorithms on small dimensionality (sparse) corporate bond portfolios. Confined to market risk considerations, this initial study highlighted functional and technical benefits of adding an unsupervised learning transformation layer to our original tracking error (TE) optimization process. In the next phase of this research, the investment universe was increased to cover a broader set of corporate bond instruments. The modeling outcomes better reflect the large-scale nature of popular indices ETF bond managers choose to track. A new risk type was also included, with market liquidity considerations added to the sparse portfolio construction process (Figure 1).
Fig. 1 – Extended sparse portfolio construction framework with market liquidity inputs
For many fund managers, the need to integrate market liquidity data in the investment process goes beyond regulatory compliance. In the corporate bond segment in particular, rapid changes in market liquidity have a direct impact on the ability to trade large nominal amounts. To account for these complex effects, SS&C Algorithmics’ "Liquidity Stress Testing for Market Risk Management" research on granular market liquidity surface construction was considered.
The research report includes active exposures, ex-ante tracking error measures, TE contributions (Figure 2), effective duration bucket allocations and market liquidation profiles of optimal portfolios of various cardinality levels. It also covers position overlaps and expected time to liquidate under 2008-crisis-like conditions at fixed haircut levels.
Fig. 2 – Reporting matrix examples featuring active weights (sectors, positions) and TE contributions
In conclusion, the functional and technical benefits of using selected machine learning techniques alongside “conventional” portfolio optimization approaches are confirmed. This echoes SS&C Algorithmics’ previous findings from the September 2023-published research on market risk dimensions. In this more recent research, the functional benefits of ML techniques are even broader than in the previous study, as the new cases covered a greater variety of data sources and risk types.
You can download the "ETFs – Sparse Bond Portfolio Optimization and Market Liquidity" research report to read the detailed findings.
Contact us to learn how you can use SS&C Algorithmics’ risk analytics and scenario-based optimization solutions in your own portfolio construction processes.
