Rethinking portfolio optimization for fixed income

Still using traditional stratified sampling for fixed income portfolios? This approach can be enhanced with optimization and factor risk models to handle thousands of constraints, reduce transaction costs and maintain tighter benchmark alignment.

While portfolio optimization has long been a staple for equity strategies, fixed income presents a unique set of challenges on an entirely different scale: indices with tens of thousands of securities, intricate risk dimensions, liquidity and practical trading constraints – just to name a few.

For a passive manager, full index replication is often impractical, if not impossible. So, what is the industry standard for approaching this challenge?

Enter stratified sampling 
This reality forces fixed income managers to use stratified sampling techniques, where managers divide the investment universe into buckets across multiple dimensions. For example:

• Countries and currencies
• Maturity
• Industries and sectors
• Credit ratings
• Seniority levels

Within these buckets, managers attempt to match key analytics against the benchmark: duration, key rate durations (KRDs), option-adjusted spreads (OAS), yield to maturity (YTM), and more. The goal is to control exposures and limit tracking error by keeping these analytics within tight ranges. This approach might sound straightforward, but in practice, it quickly becomes overwhelming. Consider a real-world example from our recent case study, Past and Future¹ where we analyzed a fixed income passive mandate which needed to control:

• Over 7,100 inequality conditions across various dimensions
• Portfolio-level metrics (duration, KRDs, DTS, yield)
• Country-level allocations and risk measures (70 countries)
• Sector-level metrics across three classification tiers
• Issuer-level allocations and durations (over 3,000 issuers)
• Security-level trading constraints

Faced with this type of scenario, portfolio managers tend to add more buckets thereby creating more complex optimization problems. Ultimately, if the individual bucket characteristic targets can’t be matched then the manager is forced to manually prioritize the measures that matter most. The obvious situation where this occurs is a new fund launch (especially if you’re tracking something like the Global Agg with 30,000+ names) and there is a limit on the number of bonds that can be included.

While that may not happen every day, there are other situations that introduce a similar level of increasing complexity during the normal course of operations like restricted holding lists that include a specific country or a “brown” issuer.

Enhancing stratified sampling with optimization
Portfolio managers find themselves facing impossible trade-offs between competing objectives. How do you prioritize matching country weights versus sector exposures? Is duration matching more important than credit spread alignment?

Portfolio optimization offers a systematic approach to implementing stratified sampling methodologies. Instead of manually selecting securities for each bucket, an optimizer can mathematically determine the optimal bond selection that satisfies the various constraints.

Transforming the approach with risk models
But stratified sampling can only take you so far, even with an optimizer. When constraints become too numerous or tight, there is a point where the problem becomes mathematically infeasible. For example, if you're launching a new fund with a 1,000-bond trade limit against a benchmark with 1,469 issuers weighted above 5bps, perfect issuer-level matching becomes impossible. 

The game-changer comes from integrating factor risk models with the optimization process. Rather than indirectly controlling risk through hundreds of bucket constraints, a parsimonious fixed income risk model can directly measure and manage tracking error. 

This approach offers several powerful advantages: 

• Instead of treating all deviations equally, risk models quantify their actual impact on portfolio behavior.
• Capture correlation effects: Risk models account for how different factors interact, something stratified sampling misses entirely. 
• With explicit risk control, many tight bucket constraints can be relaxed while still maintaining desired risk levels.

In practice, many managers prefer a middle ground, combining elements of risk-based optimization to overcome the practical limitations of traditional stratified sampling. This approach allows managers to maintain tight controls on critical exposure dimensions (like duration and key sectors) while providing risk-based flexibility in other areas.

The real power comes from being able to incorporate custom tracking errors for individual factors or specific categories of factors in the risk constraints – for instance, a manager with a separate currency hedging program can exclude currency factors from tracking error calculations, focusing optimization purely on security selection while maintaining their established hedging process.

Real-world applications
Referring back to those challenging scenarios, this is how the optimizer adds value.

• Cash flow management: When handling subscriptions or redemptions, the optimizer can find the optimal set of trades that maintain benchmark alignment while respecting minimum denomination and increment requirements.
• New fund setup: For initial portfolio construction with trade limitations, the risk model approach can identify which bonds will deliver the lowest tracking error despite impossibly tight issuer-level constraints.
• ESG integration: When excluding certain issuers or sectors for ESG reasons, the risk model can help identify compensating positions that minimize the tracking impact of these exclusions.

The future of fixed income portfolio construction
As fixed income markets grow more complex with additional securities, risk dimensions, and ESG considerations, the limitations of traditional stratified sampling become increasingly apparent. Portfolio optimization with integrated risk models offers a sophisticated yet practical path forward.

By embracing these advancements, fixed income managers can build more efficient portfolios, reduce transaction costs, and deliver more consistent benchmark-relative performance – all while maintaining the flexibility to incorporate their unique investment views and constraints.


Reference/footnote
¹ Fill out the form below to request the more detailed case study.