Investment Solutions – Portfolio Construction Experts | PortfolioWizards

A Glimpse at Morningstar Direct’s Asset Allocation Tool Enhancements

 
As a Chicagoan I spent much of my professional career with peripheral awareness of Morningstar, but not too much first-hand familiarity. Since the firm began as a mere collator of publicly available data on mutual funds Morningstar has come a long way and so has my appreciation of its offerings. These days I rely on aspects of Morningstar and follow the firm more closely. In particular I subscribe to Morningstar Premium and listen to the firm’s podcasts regularly. I am also aware that the firm’s data – especially since it now includes Ibbotson data – and technology drive many analytical software packages that advisors, wealth managers, and family offices might use.

Morningstar Direct Asset Allocation Tool Enhancements
One of their tools I’ve been eager to test drive is their Morningstar Direct asset allocation tool, to which I am NOT a subscriber (yet). Teased in the spring with the promise suggested by labels like “Markowitz 2.0” – which alludes to multi-period optimization along with consideration of downside risk – I have been increasingly anticipating trying or witnessing this tool first hand.

Morningstar seems to have changed the phrase “Markowitz 2.0” to “Asset Allocation in a Non-Normal World,” and the firm has begun presenting on this to appropriate audiences. I was disappointed that their road show would not be traveling to southern California any time soon (the closest they’ll be getting is Las Vegas), so several weeks ago I contacted the firm and requested more detail on the technical aspects.

The response was wonderful. Morningstar connected me to Cindy Tsai, CFA, CAIA, Senior Research Analyst. Cindy was quite generous with her time; she described the tools over the phone and sent me some supporting materials. When I mentioned I would be in Chicago for the Society of Actuaries annual meeting, she invited me to visit her and a few of her colleagues in person. Of course I had to take her up on it, so after the SOA meeting ended I stuck around town to visit Cindy and some of her teammates, Mike Laske and Joey Zhang.

First Distinction: Multi-period Optimization
If you download sample workbooks from PortfolioWizards’ Spreadsheet page, you’ll find examples of single period optimization. Enter some assets’ volatilities and correlation coefficients (in many cases, not even expected returns) and these workbooks will identify different “optimal” portfolios, depending on your definition of optimal. Several of these portfolios can be found analytically, meaning if you understand the math you could identify optimal portfolio weights simply, even with paper and pencil.

These portfolios are optimal ONLY for a single period, because the inputs do not take into account how the portfolios’ return might be distributed in the long run1. Nothing wrong with this as long as you’re making a decision for only one period!

However, what’s optimal for a single period is probably different from what’s optimal over multiple periods, and most people are investing for multiple periods. Keep in mind that an arithmetic mean of historical returns is a nearly unbiased best estimate of a single period return, but over longer horizons the geometric mean, which is always lower, plays a larger role in the unbiased estimate of cumulative return. This means that your typical single period mean-variance optimizer, while probably accurate for assessing your choices for a single period, could be exaggerating what you could expect out of your more volatile assets over longer time horizons.

Unlike single period optimizers, Morningstar Direct’s allocation tool simulates assets’ future return paths and identifies an efficient frontier from the simulation results. In so doing Morningstar is helping insulate the user from basing expectations on unduly optimistic expectations.

Next Distinction: Downside Risk Metrics
Multiperiod optimization aside, this tool boasts of a few other notable differences from the way most mean-variance optimizers work, including:

  1. It allows for skewed and fat tail distributions
  2. Morningstar’s new tool accommodates non-linear correlations among assets
  3. It offers asymmetric risk measures such as conditional tail expectation (CTE), aka conditional value at risk (CVaR)

In order to accommodate skewed and fat tail distributions, Morningstar Direct includes a library of alternative probability distributions, some of which will be unfamiliar to most investment professionals, although this will probably change in the future. I wouldn’t be surprised if Tsai’s Ibbotson colleagues’ recent FAJ article, The Impact of Skewness and Fat Tails on the Asset Allocation Decision, makes it into future CFA syllabus requirements. The authors suggest that the combination of skewness and fat tails, not just fat tails alone, make Mean-CVaR optimization superior to Mean-Variance optimization. Skewness requires a distribution other than Normal, on which most mean-variance optimizers rely. Hence, it would not be as useful to model downside risk without using alternative distributions to the Normal distribution.

Ever since reading Sam Savage’s “The Flaw of Averages” I have been a fan and advocate of using correlation scatter plots instead of correlation coefficients. For audiences seeing this for the first time, Morningstar has some missionary work to do. However I expect users will get comfortable thinking of correlations as scatterplots instead of as linear approximations much sooner than they’ll feel comfortable toggling parameters of a Johnson or a truncated Levy flight distribution. It will take a while for users to feel confident selecting among a menu of distributions they previously have never heard of.

The downside risk metrics include the usual suspects, semi-variance, and semi-deviation. But those have been around a few decades. Morningstar also includes CTE/CVaR. CVaR is generally regarded as more robust than the more common VaR (Value at Risk). Unlike VaR, which reports a point estimate of adverse return, CVaR reports the expected adverse result conditional on the outcome residing in the tail below a certain probability threshold, which is why it’s also known as conditional tail expectation.

Morningstar is on Another Mission
For all the sophisticated math and programming behind the tool, its design is comfortably understated and friendly. I think that’s great – I used to manage assets at a place where we had such dizzying GUIs and dashboards that for the first two weeks on the job I literally went home every night with a headache. Flashy graphics might make for more interesting demonstrations, but they don’t facilitate thoughtful application by users.

The minimalist interface and muted color scheme will be crucial for adoption, because in order for this tool’s demand to grow it will have to appeal to a wider audience than only to quants; its display and interface will have to avoid appearing overwhelming, and I believe it succeeds.

It’s easy to imagine lots of quant geeks eating this up, becoming early adopters if they work at places that subscribe to Morningstar Direct. However for your average family office or wealth advisor this will be a stretch; this will involve a steep learning curve. Non-quants will not only need help running it, they’ll need help having someone peer review their work until such time as they recalibrate how they apply the sniff test.

Morningstar, which introduced the investing public to the idea of style boxes, has decided to go on another mission.

TA: I have corrected this post after receiving feedback from Ms. Tsai. Previously I had incorrectly written that Ms. Tsai had been an Ibbotson consultant, and that the Morningstar Direct allocation tool included VaR among its risk metrics. I have updated the post accordingly.
 


1I wrote earlier about Jacquier, Kane, and Marcus’ work modeling expected returns over future time horizons, and Pastor and Stambaugh’s work demonstrating how to think about predictive variance. These topics all relate to the importance of adjusting future expectations to take into account sampling error and uncertainty.

Exit mobile version