Growing up in Greece before we could afford a microwave oven, I still remember what an ordeal it was for my mother to warm our milk on the stovetop. It took a long time, she had to wash the pot afterwards, and losing track of time would often spill boiling milk and cause a mess. When we finally got a microwave oven, it was life-changing; it was the right tool for the job.
Interest is on the rise in direct indexing, ESG and tax loss harvesting. Offering such custom portfolios to clients requires many technology components. The central “brain” is the investing logic, which finds the best possible portfolio subject to goals, often conflicting ones. The most sophisticated firms use portfolio optimizers, which excel at handling such tradeoffs. Historically, portfolio optimizers have been used for institutional trading, and repurposing them for custom portfolios is reasonable. An alternative is to take into consideration the different requirements of custom portfolios, and build from scratch the right tool for the job.
This article will discuss those different requirements, and—in certain cases—how they can be addressed.
Thousands or millions of accounts
This means human review may not always be possible. In technology, “99% automated” is very different from “100% automated”; moreover, going from 99% to 100% is not just an extra 1% of work. For example, say that ESG ratings change suddenly, and AMZN goes from “good” to “bad”. A large institution may manually update and review its target portfolio, possibly removing or underweighting AMZN, with some manual discretion. However, if 100,000 client accounts are holding AMZN, it’s not possible to have any such manual discretion.
This also affects technology choices significantly. Scalability and parallelization are important. This means avoiding slow optimization approaches, such as quadratic optimization (which gives slightly more precision but takes longer to solve) and mixed integer programming (which, very roughly, means “try all permutations”).
Service & support
Handling many accounts also means that more information should be recorded for customer service, compliance, etc. Reproducing what happened to a client’s portfolio in the past is useful, especially if the client had a question, or if there was unexpected behavior to investigate. This becomes much easier if the API (the “communication language” with the system) is “stateless”, i.e. does not “remember” anything from previous invocations, which means all necessary information must be packaged in the request. For example, if client holdings are not recorded as of the time of the request, then they must be recreated later. This will be difficult if the client holdings database only shows a current snapshot of holdings without any historical information.
Most institutions don’t need to worry about wash sales, short- vs. long-term capital gains, tax loss carryovers, held-away net gains, etc. A simplistic approach is to decide how much of each security to buy or sell, and then—as an afterthought—choose the best tax lots to sell. However, the correct behavior is more complicated: if selling an overweight security would realize a lot of tax gains, perhaps we should be selling less (or not at all) in the first place. This requires a change in the optimization approach, because the optimizer needs to “know” about the different tax lots, and assign them each a different penalty for selling.
Client perception considerations
Institutions mostly care about tracking error. Individuals, however, may dislike uninvested cash dividends sitting in an account, even if it does not affect tracking error by much. Or, perhaps some clients do not like seeing many trades. A larger range of requests means there must be many ‘knobs to tune’ to customize the behavior.
Less perfection needed around risk
For example, a large bank will normally want its factor exposures to be close to zero overnight, to avoid being subject to market movements. Individual accounts, however, are supposed to be holding a target portfolio. This means their factor targets are also non-zero. An (imperfect) analogy is the difference between the effect of an extra cup of milk on a lactose-intolerant person (whose target milk consumption is zero) versus a lactose-tolerant person (who would just exceed their normal consumption).
An individual may hold large executive stock grants, legacy positions, or even illiquid assets, whereas an institutional account is typically restricted as to what it can hold. As a simple example, a client who has large executive stock grants from AAPL should not only hold less (or no) AAPL in their portfolio, but should also have large-cap and technology stock exposure. A factor model is (roughly speaking) a set of numbers that describe how similar (or different) any two stocks are. However, those numbers are somewhat subjective and depend on the data provider. It is important to be able to support multiple such factor models simultaneously, including proprietary ones, e.g. if an advisor firm wants to combine its own factor model, i.e. its opinion on what drives stock returns, with an ‘industry standard’ factor model.
Variations across accounts (ESG preferences, tax brackets, model portfolios, etc.)
This implies a much wider range of outcomes, which brings a higher chance of undesired behavior. For example, if a client’s ESG preferences are too strong, a large change in ESG ratings can result in too much trading. There are several similar parameters that need to be “tuned” to achieve desired behavior. By contrast, an institution may have settled on parameters that work; they only need to do it once, and perhaps update them periodically. Therefore, simulating future outcomes, especially if tax considerations can be accurately accounted for, becomes more important with custom portfolios: it helps advisors tune a strategy correctly by looking at realistic after-tax returns for clients.
There is another reason speed matters: if simulating one day takes 15 seconds, simulating a 10-year run would take about 10 hours. However, if it takes 0.1 seconds, the entire simulation would take under 5 minutes. Such speed would facilitate trying out many scenarios to tune a strategy for a client.
The shift towards direct indexing requires rethinking many aspects of this business. We believe institutional trading and wealth management are sufficiently different enough to merit a fresh approach.
Finally, if you’re still wondering “Who warms up milk?” it’s very common in Greece; just think of it as an ESG preference in a Greek breakfast portfolio.
Iraklis Kourtidis is the founder and CEO of Rowboat Advisors, which builds portfolio optimization software with a focus on tax efficiency and direct indexing. He also built the first fully automated version of direct indexing in 2013 for automated investment service Wealthfront.