Build vs. buy: A guide to portfolio optimizers

When evaluating optimization technology for financial applications, decision-makers often focus solely on algorithmic specifications and theoretical performance, overlooking crucial factors that can make or break real-world implementation. The key differentiators lie in how well an optimizer understands the unique constraints of financial problems, handles real-world complexity, and most importantly, its track record of delivering reliable solutions – at scale.

Through our experience working with asset managers, hedge funds, asset owners and wealth managers, we've identified several critical characteristics that separate truly effective financial optimization engines from general-purpose solvers that have been retrofitted for financial use. We put together a few points to consider when selecting an optimizer for your strategies.
 

Can your optimizer handle real-world portfolio management problems?  

The true test of an optimizer is being able to handle the diverse, complex scenarios that arise in actual portfolio management. Leading financial optimizers have been battle-tested against extensive libraries of real-world problems compiled through years of working with portfolio managers. This practical experience enables the development of specialized heuristics and fine-tuned branch-and-bound algorithms to find good, feasible solutions quickly.  

On the other hand, general-purpose optimizers require users to manually translate financial constraints into mathematical problem statements. This translation process is not only time-consuming but also introduces potential for errors and inefficiencies. Purpose-built financial optimizers allow portfolio managers to express constraints naturally in financial terms, handling the complex mathematical transformations automatically. Examples include:

• Long-Short Holdings constraints with automatic safeguards against simultaneous long-short positions
• Relative Marginal Contribution to Risk constraints handled internally without explicit reformulation
• Combinatorial/Discrete/Integer objectives and constraints implemented without requiring explicit binary variable definitions
• Tax-Aware optimization with built-in IRS rule encoding
• Advanced trading cost models including fixed charge costs and various market impact specifications
   

Can your optimizer handle more than simple mean-variance?  

For an institutional investor running even mildly complex strategies, an optimizer will need to be able to incorporate more sophisticated capabilities such as:

• Robust optimization techniques that account for uncertainty in alpha predictions
• Advanced risk model integration with automatic bias reduction
• Multi-portfolio optimization for managing separately managed accounts with both aggregate and account-specific constraints
• Multi-portfolio optimization supporting fairness in the allocation of trading liquidity
   

Does your optimizer increase workflow efficiencies?  

The real value of an optimizer emerges in day-to-day portfolio management operations. Purpose-built solutions offer extensive libraries of pre-configured objectives and constraints that make model specification faster and easier to maintain. They provide comprehensive portfolio analytics integration, allowing managers to seamlessly analyze characteristics like risk decomposition, transfer coefficients, and implied alphas. And speaking of risk, purpose-built optimizers typically incorporate risk models directly, eliminating the need to plug in (and perhaps build) a proprietary model.  

For some investors, it is also important to find an optimizer that can easily move between API and UI interfaces and different languages, perhaps to facilitate collaboration between quantitative and fundamental teams. For example, a rebalancing session in the Python API should be exportable to formats compatible with the Java API or the UI.
 

What are the hidden costs of building an optimizer in-house?  

Some institutions consider developing their own optimizer or building on top of an existing one. However, both approaches often underestimate the complexity involved. Beyond the core algorithmic implementation, a production-grade optimizer requires:

• Extensive testing across diverse real-world scenarios
• Sophisticated reformulation techniques for handling complex constraints
• Ongoing research to incorporate new financial innovations
• Robust error handling and diagnostic capabilities
• Comprehensive documentation and support infrastructure

This all adds up to a commitment that extends far beyond initial development, which could take months or even years depending on the functionality typically needed. Once built, an in-house optimizer demands continuous maintenance, ongoing enhancements and constant monitoring in optimization advancements – requiring not just one developer, but a team of specialized experts. Without a team, there exists significant key person risk, should someone leave behind software that no one truly understands. And, when technical issues do arise, troubleshooting the code can lead to extended downtime or even undetected errors, potentially creating material reputational and financial risk. While organizations may perceive building in-house as cost effective, they often overlook the substantial long-term investment that falls outside their core business focus.
 

What happens when you have questions?  

Perhaps most importantly, leading financial optimizers come backed by teams of specialists who understand both the technical and financial aspects of portfolio optimization. This includes dedicated support staff for day-to-day questions, Ph.D.-level experts for advanced modeling issues, and research teams continuously working on innovations. When complex questions arise – as they inevitably do in portfolio management – having access to this expertise can be invaluable.
 

Successful portfolio optimization isn't just about the mathematics

Having a package of software and people that truly understand the nuances of financial portfolio management is key. Not only does the Axioma Portfolio Optimizer use state-of-the-art algorithms for optimization like Second-Order Cone Optimization (SOCP) with Branch-and-Bound, it is backed by optimization experts with years of experience working with portfolio managers.