Why plural investing?

What is ‘plural’ investing?

Most investment funds follow one of two ‘singular’ approaches:

  • Systematic: automated computer systems that invest based on rules.
  • Discretionary: human investors who choose what to invest in.

A ‘plural’ approach combines systematic and discretionary elements into a process that is greater than the sum of its parts. It does this by having these two elements cooperating in a way that amplifies their strengths while reducing their weaknesses. This requires constant cooperation between them and the human investor to understand how and why the system is making its recommendations.

Lessons from chess

“Weak human + machine + superior process was greater than a strong computer and, remarkably, greater than a strong human + machine with an inferior process.” – Garry Kasparov

Chess was one of the first games where a ‘plural’ process combining machines and human players was shown to be superior to machine-only and human-only. It turns out the cooperation between man and machine was the key to this success:

In 1997 the world chess champion, Garry Kasparov, was defeated by a computer system, IBM’s Deep Blue. After his defeat, Kasparov introduced ‘freestyle chess’, in which competitors can enter as a human player, a computer program, or a ‘plural’ player – a human player working with a computer which suggests moves.

Chess computers were far superior to human players, but when the first major tournament was held in 2005 all four semi-finalists were ‘plurals’, not computer-only players. Three were teams of grandmasters and supercomputers, but the winners were two amateur chess players choosing moves from three consumer grade computers.

They won because they considered moves from several computer systems, understood why these systems recommended them, why the recommendations might be different, when these systems would be strongest, and the strengths and limitations of themselves. This allowed them to select better moves than a grandmaster using one supercomputer.

The two key conclusions are that not only does a ‘plural’ process get the best results, the best processes have systems and humans cooperating in a way that gets the best out of both.

‘Plural’ processes are even more useful in investing

Like in chess, in investing computer systems can download or recall data and process it exceptionally quickly and rationally. But they struggle to adapt, think or be creative. They are therefore strong at making a wide range of analysis but weaker in their depth of analysis.

On the other hand, human beings are very limited by the speed they can ‘download’ data, process slowly and are emotional. However, our ability to think often enables us to make deeper analyses. Unlike in chess, humans also have access to valuable data that machines don’t because we can consider more qualitative data and can proactively look for unique sources of data for every investment idea.

The aim of the ‘plural’ process is to maximise both the range of data and the depth of analysis.

How we maximise the cooperation effect

Our investment process involves an alpha model, portfolio construction model, and human investor cooperating in a way to bring out each other’s strengths rather than weaknesses:

1) The alpha model analyses every stock in our universe. It considers objective data, but also subjective data create by the human investor. This focuses the human on acquiring knowledge (data) and the model to weight it up. The human investor uses his understanding of the model to exclude long/short recommendations where the it was unlikely to have considered significant external variables, such as takeover speculation, changing politics, and the launch of a game-changing product.

2) The remaining ideas are placed into a portfolio construction model that uses machine learning techniques to ‘learn’ which are most attractive in a portfolio. This effectively recommends less correlated ideas. Once again, the human investor prunes these recommendations.

3) The human investor now conducts in-depth analysis. The primary aim of this is to do analysis that the machine cannot and to understand why the human’s conclusions are different to the machine’s. This results in the human investor’s estimated risk and reward values.

4) These are added to both the alpha model and portfolio construction model. As human investors tend to be too optimistic about expected reward and too pessimistic about expected risk, the models shrink the human’s estimated rewards and grows the estimated risk. They use simulation and machine learning techniques to arrive at several final recommended portfolios. Importantly, the human investor can now see why different recommendations disagree, and ask the model to look more closely at certain scenarios. Finally, he builds a plural portfolio.