Portfolio optimization: a management or a computer job?
Do you recognize this recurring question in portfolio management: “Can your portfolio management tool calculate the best portfolio?”. Over the years we have been answering this question with a “No, but..” and a “Yes, but..” answer. The essential line of reasoning in both answers is very similar:
1. In mathematical terms, portfolio optimization is well-defined and thoroughly studied optimization problem, where project selection and resource allocation over time and under uncertainty can be accommodated. The associated optimization algorithms may be time-consuming to execute. In any real portfolio, with multiple dimensions, this quickly becomes infeasible to include in a management decision-making cycle.
2. In order for any tool to calculate the optimal portfolio, it needs very specific instruction of what you mean by “optimal”, as in gets the best result from the available means. Typically, this requires definition of constraints on these means: on funding and resource, on committed or mandatory projects, and so on. In practice, these constraints are never the strict constraints of mathematical optimization. A little over the expense budget in the 3rd quarter of next year and little under in the 4th quarter (at the plan level) could be no problem. A resource pool can be grown and shrunk in the long run but may be fixed in the short-term.
It is even more difficult to elicit the proper result optimization goals upfront. This requires the upfront answer to questions a such as: “how to combine financial versus strategic value?”, “how much risk are you willing to take for each level of value creation?”, and “what is the optimal build-up of results over time?”.
As an example of the latter, I have seen preference for steady and gradual growth over value-maximizing but discontinuous (jumpy) growth.
3. Now even assuming we could elicit all requirements and preferences and calculate the optimal portfolio, imagine a management team looking at the results. Will they have the confidence in the results of a complex mathematical optimization algorithm to implement the recommended portfolio choices? Can they convince their project teams to change course, to reallocate resources, and ultimately to stop their project?
In the more recent academic research in portfolio optimization, the way out of these issues is to develop an iterative and interactive portfolio optimization process (e.g. as described by Stummer and Heidenberger in their research).
This is the direction we are following in Flightmap, where different portfolio scenarios can be compiled manually. By choosing an optimization direction, the tool can propose a relatively simple step to further improve the portfolio. By going through this process in a number of iterations, the tool helps the portfolio manager understand the pro’s and con’s of the various optimizations. This can be seen as eliciting (or developing) the trade-off preferences based on the real data, instead of upfront.
One of the best portfolio reviews I have facilitated focused around the following portfolio alternatives:
This example shows how each of 3 portfolio scenarios can be the optimal portfolio depending on the selected optimization goal: the “business” portfolio optimizes mid-term revenue growth, the “financial” portfolio optimizes return on investment, and the “strategic” portfolio optimizes on value in Emerging Markets (EMA). Talking the portfolio board through these 3 portfolio scenarios (each composed of a few dozen projects) triggered valuable strategic discussions with a direct link to their execution. Each of these 3 portfolios consists of a different set of projects (with a limited overlap: the no-brainer projects that score well on all criteria).
This interactive portfolio optimization approach brings (just) enough trade-off discussions to the table, and builds the common rationale for the resulting decisions. What is your take on optimization?