Quantamental investing is the new active. When combined with machine learning, what is old is new again. The marketing machine does much of the rest. Thats how we roll.
Quantamental managers are said to combine bottoms-up analysis of companies with quantitative approaches to the alpha prediction problem. Fueled by new data sets, the idea dates back to the early 1970s, pioneered by luminaries such as Jack Treynor and Fischer Black. The same data availability breathes new life into machine learning cum data-mining techniques developed in the 1940s.
As with any industry fad, opinions vary. New data do not imply new insight. Statistical tools designed for servomechanisms do not blindly lend themselves to human behavior. As Steven Einhorn of Omega Advisors points out, Big data and quantitative analysis can help identify [predictive power], but so can, and so will, fundamental security analysis. The emphasis is his, and he makes a case that there will always be a place for the traditional active manager, while complementing fundamental insights with quantitative approaches will happen over time.
Data extraction does not mean idea inception. Value the question first, before invoking the tools.
I have such a question, born of numerous conversations with active managers against the backdrop of big data: How does one incorporate the benefits of data-driven portfolio mathematics without disturbing the core beliefs of the fundamental manager? The query strikes directly at the intersection of Einhorns perspective with quantamental thinking, exhibited by data research firms such as M Science and Qineqt. To be credible, an answer must fit into the multiplicity of portfolio strategies in the market place.
Active managers are in a difficult position. They have an understandably limited number of views on the world, but they can pick tradable instruments with alpha opportunities that express these views. If they use an optimization process and alternative data in portfolio construction, the mathematics rearrange their priorities and even discard some views in which they strongly believe. They may need diversification, given the small number of views, but they cant let that upset their core portfolio. Liquidity considerations, manifested by substantial transaction costs, exacerbate the issue, making it even more difficult to implement an investment viewpoint.
One answer lies in the simple image of a donut. The hole in the donut is the core portfolio, the fundamental managers expression of views in terms of tradable instruments. Any additions to the core, represented by the body of the donut, must not disturb the composition of the fundamental portfolio. We neatly sidestep the issue of alpha generation; that belongs to the manager. Completely agnostic with respect to the nature and origin of the core portfolio, the remainder of the donut need only be composed of tradable instruments, chosen through a mathematical optimization process which, while respecting the core, takes its properties into account in the overall donut construction.
The donut embeds a variety of existing strategies. In portable alpha, the core is composed of areas which theoretically have no correlation with the broader market, which in turn is represented by the donut. Separation of alpha from beta permits some correlation of the core with the donut and the dough need not be the market. Enhanced indexing amplifies returns while minimizing tracking error; the core is the amplification engine, and the donut factory tunes the donut to minimize mistakes. This sometimes is called hybrid investing. In passive investing, tracking error is the key, and the core disappears. The opposite is true for mean-variance portfolios—the core expands to include the entire donut. Even smart beta strategies can be described by the construct.
The performance of donut portfolios owes a great deal to the inclusion of implementation costs and the notion of net, as opposed to paper, returns. Transaction costs are large relative to alpha even for large cap securities, growing more important as managers search for yield in less liquid markets. I quote Bruce Jacobs and Kenneth Levy, who coined the term smart alpha, on the big picture here: a smart alpha strategy combines numerous factors in an integrated framework that allow for optimal tradeoffs among expected return, risk and transaction costs. Factor exposures, as well as exposures to other sources of tracking error, are typically controlled by portfolio optimization. The entire donut, including its core, is one representation of a smart alpha process.
Simplicity and transparency are virtues in any portfolio setting. In the donut paradigm, one can completely characterize portfolio performance along only two dimensions, the percentage of the portfolio residing in the core and the return of the core itself. Both are easily measurable, and only closet indexers need fear the transparency of the first. Quantamentalists should feel right at home, while the fundamental manager retains a seat at the table. Donuts simplicity and transparency are attractive, if only because investors should have a clear understanding of an active management strategy.
In research with Ameya Moghe, we explore the potential benefits of relying on a donut factory for portfolio construction. Two conclusions stand out. Risk-adjusted returns are essentially unaffected by the diversification offered to the core portfolio from the donut. When the core produces net returns lower than the benchmark, risk-adjusted returns dramatically rise with donut construction. The donut acts as a hedge against fundamental performance, while minimizing drag when the core performs. Second, overall implementation costs can be cut almost by half. Even if risk-adjusted returns are unchanged, the donut construction permits extra fund capacity in terms of assets under management, while maintaining the fundamental-driven investment process. The converse also is true; liquidation costs fall, should that be necessary.
Portfolio theory is full of tradeoffs and belief systems. Risk versus return is the classic. Minimization of transaction costs may increase capacity, but at the expense of large positions in alpha-generating instruments. Stability may be preferred over dynamism, and even the number of positions may require attention, offsetting return characteristics. Ask the right question, have a donut, and dont wait on the next fad.
Ian Domowitz is a Managing Director and Head of Analytics at ITG. He remains a quantamental fundamentalist. The research cited here is available on Analyticsincubator.itginc.com, entitled Doughnuts: A Picture of Optimization Applied to Smart Alpha.
