Public / Private SAA

Private asset optimization - to model or not to model, that is the question

I’ve always had two conflicting sides to my personality. On the one hand...

February 28, 2021

I’ve always had two conflicting sides to my personality. On the one hand, I am quite mathematical and enjoy the quantitative nature of investment strategy and developing models that assist with that. However, on the other hand, I enjoy using intuition and reasoning to come to an answer without modelling the outcomes.

These two conflicting sides drive this week’s post which relates to how to set the optimal allocation to private assets. 

Private assets have a number of characteristics that make them more difficult to include in standard optimization frameworks. For example, smoothed valuations, lack of data, illiquidity and the requirement to have a strategy for how capital is committed on an ongoing basis to maintain an allocation.

Investors therefore need a different approach to evaluate the optimal allocation of private assets within a strategic asset allocation.

The two most common approaches are:

  • Run a constrained mean-variance optimization with a subjective limit on the private asset allocation
  • Create a more sophisticated stochastic model that attempts to capture both the unique characteristics of private assets and their cashflow profiles

Under the first approach, adjustments can be made to try to more accurately measure the volatility of private assets, for example using unsmoothing techniques. However the models also tend to assume a return premium of private assets over public assets. Therefore, even after unsmoothing the volatility, optimization tends to favour allocations to private assets (particularly at higher return targets).

So under these techniques it is the subjective limit that tends to drive the optimal allocation to private assets. These limits can be set based on liquidity stress testing or may arise from other constraints such as governance or cost constraints. 

The second approach requires more sophisticated modelling techniques to capture the dynamics. These models calculate not only the optimal strategic allocation, but also the required contribution strategy to get there. The model output is a range of potential outcomes for the investment based on potential returns and the contribution / distribution dynamics.

This output can be helpful, as it provides confidence levels for different strategic asset allocations of meeting both return hurdles and liquidity requirements. There is also some potential comfort, that this approach may “feel” more rigorous.

However, a model is only as good as it’s assumptions and the data used to form and test those. This is where the big challenge lies for private assets - there simply isn’t much data available in the right format.

BlackRock has created a sophisticated model and their approach is overlaid by what they call the “Alternatives Turing Test”. Essentially this means that they pass the model output to an experienced investor to sense check the validity. 

The belief is this blend of judgement and robust analytical methods leads to better portfolio construction decisions. However, this demonstrates a degree of subjectivity to the process and we have just gotten there with a more complex model.

If you fundamentally believe that private assets outperform public assets then no matter how complex the model is, it will tend to allocate more to them up to some limit. 

This is revealed in the framework that PGIM have created. PGIM proposes a simulation-based asset allocation framework to "Maximize horizon portfolio value provided I am sufficiently confident of meeting my cash flow obligations".

They show optimal allocations for different levels of confidence in meeting liquidity requirements. However, the split between public and private assets doesn’t really change for confidence levels of 90% of more. The model instead chooses less risky assets within the public and private asset classes in order to meet the higher confidence level.

Said differently, this result says that for higher confidence levels it’s not optimal to change the public / private asset mix. Investors are better off changing the allocation between riskier and more defensive assets.

The result is certainly interesting and demonstrates the dynamics between the amount of risk being taken and the confidence level in meeting liquidity requirements. However, it is perhaps less satisfying that the more complex probabilistic model also seems to settle on a limit for the private asset allocation.

I believe that given the degree of subjectivity required in the modelling techniques, it is better that the subjectivity is applied to a more simple model. 

The question should not be how to get the best model, but what is the simplest model that can support and guide the decision where I can clearly see how my assumptions have impacted the result.

Therefore, at least for now, I favour the (relatively) simple approach to public / private asset allocation of a constrained mean / variance optimisation. However, this could very well change as more data is accumulated on private asset investments.

Sign up for more



Zermelo is a platform for institutional investors to research, request and develop custom investment solutions. We make it quicker and easier to find the best solution that exactly meets your needs.

Ed Studd

CEO, Zermelo

Ed is passionate about helping institutional investors meet their goals through better custom solutions. Prior to Zermelo, he gained extensive experience across asset management, investment banking and investment consulting.