Rafet Eriskin of AP4: Simple vs. Complex Models [5 of 5]

With massive amounts of data available at our fingertips, and analysis techniques made easy by recent open source software development, it’s become possible to deploy an array of analytical models to portfolio analysis. This is true whether you’re running money yourself, or investing in someone else’s portfolios (and performing due diligence on a manager). But not all models are created equally. Some are overly simple, some are overly complicated, some are just right. A few unearth signals, many just generate more noise out of noise. How do we make sense of it all? I had a discussion with Rafet Eriskin, Senior Portfolio Manager at AP4, about how he chooses analytical models for his portfolio analysis.

A Map the Size of the World

If a map includes too much information, it loses some of its value. Readers of the map may get confused. They may lose track of which information will guide them toward their planned destination.

The Argentine writer Jorge Luis Borges illustrated this paradox in a short story, On Exactitude in Science. An empire’s cartographers build a map of the empire so sophisticated that it’s the size of a province. Unsatisfied, they build an even more complex map, this one the size of the entire empire, coinciding point for point with the actual empire.

Borges explains the outcome—

“The following generations, who were not so fond of the study of cartography as their forebears had been, saw that that vast map was useless, and not without some pitilessness was it, that they delivered it up to the inclemencies of sun and winters.”

Perhaps this type of map was what Rafet had in mind when he said to me, in our recent conversation, “It’s easy to drown in the models.” With Rafet, as with other allocators and managers we partner with, we strive to balance straightforward data models with more sophisticated ones—but not so complicated that anyone runs the risk of drowning.

‍

Sophistication in Simplicity 

Simple models have the virtues of being intuitive, immediate, and easily actionable. They’re good for getting a sense of the big picture, and for making quick decisions. Novus’ classic platform leverages simple data models that provide snapshots of a handful of key variables.

One such model is the Novus Selection model. This model helps investors quantify stock picking or, more broadly, security selection skill (and the persistence thereof). “Picking” means being discerning—choosing the few from a wealth of possibilities in the same sector, liquidity, geography, or benchmark.

To measure the effectiveness of these choices, we have to compare them to alternative outcomes which, in the case of security selection, may be as simple as asking, what is the performance I would have obtained if I had invested in a benchmark (or ETF) which represents the original selection universe?

And that’s precisely what the Novus Selection framework does. Figure 1 depicts an example Novus analysis where security contributions are decomposed into a selection component (orange), and a relative, or benchmark, component (blue).

‍

‍

Consider, for example, a security which was held in the portfolio with an initial weight of w = 10% on AUM of USD 100 mn and yielded a P&L of USD 2 mn, equivalent to a contribution of 200 bps. Further, assume that the reference benchmark the security was chosen from (say mid cap German stocks) increased by 5% during the same period.

Had we invested 10% in the benchmark, we would have obtained a P&L of USD 500 k or, equivalently, a contribution of 50 bps. Since the total contribution of the security chosen was USD 200 bps, we conclude that the portion of the contribution coming from security selection was 150 bps (total contribution of 200 bps less benchmark contribution of 50 bps). Hence, this was a good pick.  

The model is simple, in that it measures the impact of a real decision—was I right picking this security or should I have picked another one? A few Novus clients who frequently rebalance their portfolios and trade around names more actively are using a related model, called the Novus Framework, where the impact of additional degrees of freedom (e.g., trading acumen) is included. While containing a few more parameters, the Novus Framework too is straightforward and actionable.

Another variation of this model further removes a stock’s inherent volatility from the benchmark component (the relative portion), also known as “stripping out the Beta of a security.” In the example above, assume that the security chosen had a Beta of 1.5 (meaning the security typically increases or decreases by 1.5 times compared to the benchmark movement over the same period). In this case, selection is defined as 200 – (1.5 * 50) = 125 bps, and relative as 1.5 * 50 = 75 bps.

While possible in principle, not many investors think in such terms. A security is often chosen for its inherent outperformance potential. Volatility and risk are typically assessed ex-post at the portfolio level, and mitigated with portfolio hedges. So, in this case, the model would provide an unnecessary complication that gets the math further away from the reality of the decision-making process.  

In the book Simple Rules: How to Thrive in a Complex World, Donald Sull and Kathleen M. Eisenhardt write, “We often assume that the best way to make a decision is by considering all the factors that might influence our choice and weighing their relative importance. Psychologists have found, however, that people tend to overweigh peripheral variables at the expense of critical ones when they try to take all factors into account.” Simple models mute unhelpful peripheral variables while turning up the volume on the critical ones.

‍

Occasions That Require Complexity

Meteorology is an example of when it makes sense to apply a more complex model. Through the first half of the twentieth century, most weather prediction was made using linear models—that is, systems where the change in output is proportional to the change in input. The problem was linear models didn’t predict the weather very well; a lot of picnics got rained out. Then, in the 1960s, mathematician Edward Lorenz introduced nonlinear models into meteorology. These models took into account the chaotic interrelationships of the many variables that go into producing the weather. Lorenz’s insight led to significant improvements in weather forecasting (as well as the acknowledgement that it’s virtually impossible to predict the weather with much accuracy beyond a week or two).

As portfolio models increase in complexity, they have the potential to paint more nuanced pictures of an allocator or manager’s investing trajectory as well. And when implemented responsibly, more variables can give insight into investor performance that simpler models miss.  

As an example, imagine you are tasked with assessing whether securities tend to increase after a re-rating of any of their ESG scores. Figure 2 depicts security price trajectories (in excess of market returns) before (left of the y-axis) and after (right of a right axis) a negative re-rating.  

‍

‍

Taking an average of the price trajectories post-event and concluding that a negative price action ensues if the average trajectory is negative would be—in this case—overly simplistic, and could lead to erroneous conclusions. In this case, Novus’ proprietary methodology calculates a slightly more complex “average” trajectory and compares it to the breadth of the funnel.

A word of warning: when it comes to complex models, regressions are particularly pernicious, as Rafet and I discussed, “they can be applied to almost anything, anytime and may lead you into mistaking correlation for causality.” 

‍

Customizable Dashboards

Rafet tells me he prefers “a portfolio of many simple models rather than one complex model which tries to be all-encompassing.” While there is no such thing as a universal balance between simple and complex that is right for everyone, I’ll agree with Rafet that it’s better to lean simple. The competitive edge that Novus provides our clients lies in having all models at your fingertips and being able to deploy them as you see fit.

As a client of Novus, Rafet values this flexibility, “Unified frameworks are for physics” he says, “where a few laws govern most interactions among bodies.” In a multi-player game like finance where participants try to outsmart each other by playing on multiple chessboards at the same time, it’s most important “to find the model that’s right for your style, the one which most closely measures the strategy you’re putting to work.”

The combination of models that best suits an investor’s portfolio is as unique as the investor’s portfolio. That’s why Novus’ platform is powered by customizable dashboards, allowing investors to track investment components that are unique to them, and in a manner consistent with how they think about their portfolio. Our analytics team works with clients to visualize exactly what they need to see, while tuning out what is irrelevant.

Rafet compares the simple vs. complex question to the Apple vs. Android debate, “Do we go with Apple, which is refined and pure, but not very customizable? Or do we go with Android, which offers infinite configuration possibilities but runs the risk of being overly complicated?”

At Novus our stance is—YOU choose! With dashboard templates, users can start with a combination of models that have been chiseled over a decade’s time, through working with investors in similar roles, relying on comparable functionalities, and aiming at identical purposes. If investors are interested in fully owning the process of combining and visualizing their portfolio models, it’s possible with Novus.

‍

View Part 1 of this series, Selecting Managers

View Part 2 of this series, The New Transparency

View Part 3 of this series, The Digital Revolution in Asset Management

View Part 4 of this series, Investor Bias