Factor-Based Basket Optimization in Credit Trading

Trading credit portfolios has become an important source of liquidity for asset managers, but the viability and cost of a trade depend on the nature of the basket. One important characteristic of a basket is its tracking error relative to a relevant broad-market index. Our analysis shows that by optimizing the basket with a factor risk model, tracking error can be reduced by a factor of two or more, even when the target basket differs materially from the broad-market index.

By letting asset managers trade an entire bond basket in a single transaction, portfolio trading delivers liquidity quickly and cost-effectively. In addition to other hedging mechanisms, broker-dealers acting as authorized participants (APs) can buy a portfolio from the asset manager and subsequently deliver the bonds into an ETF via the creation/redemption mechanism. For this mechanism to work, it is important that the basket tracks the ETF as closely as possible.

Tracking error is a natural measure of similarity between a candidate basket and the ETF. A basket with low tracking error may be more attractive to the AP, since it can be delivered into creation/redemption with minimal additional trading and less basis risk. It also poses less risk of disturbing the ETF's tracking error, which is a key dimension investors use to judge ETF managers. Once asset managers have a candidate basket, they can not only measure its tracking error, but proactively minimize it using a factor model and optimizer to make it more attractive to the broker-dealer and, ultimately, to the ETF sponsor.

To quantify the benefit of this factor-based approach, we built baskets to reflect a variety of realistic goals of a basket trade: reducing broad market exposure, adjusting exposure to quality and taking an active sector view.¹ We added a randomly constructed basket to serve as a baseline for comparison. We measured tracking error relative to the MSCI USD High Yield Corporate Bond Index using the MSCI Fixed Income Factor Model and performed optimization using the Barra Optimizer. We also designed the optimization to reduce tracking error while maintaining the original goal of the trade. Once we identify a candidate basket trade, we give the optimizer varying degrees of freedom to trade, ranging from 10% to 30% of the basket's base value. We also tested the results across market regimes.

As shown in the chart below, results are consistent across every tested portfolio strategy and market regime. A 10% reallocation alone can cut tracking error by more than a factor of two, and that reduction grows consistently as the size of the reallocation increases. Once it reaches 30%, the basket's tracking error is similar across all trade goals. The number of bonds allowed does not significantly impact these results: Most of the benefit may be achieved with only 15 additional bonds.²

Trading credit portfolios is ultimately a complex negotiation among three parties: the asset manager, AP and ETF sponsor. Our analysis has shown that by optimizing a portfolio trade with a factor model, asset managers may be able to put themselves at a stronger starting point in the negotiation by anticipating and addressing the likely needs of the AP and ETF sponsor. This proactivity can in turn lead to lower transaction costs, faster negotiations and smoother trading.