Machine intelligence for robo strategies, factor investing, smart-beta and impact investing

Diversification pays - and AI can be used to establish a reliable and objective way of diversification. Factor investing is a wonderful example as it invests in factors that are supposed to be rewarding (they deliver a return for the risk taken) and the factors are supposed to behave differently given different market environments which creates the diversification. It is a set of good independent bets. So in theory this approach is very appealing as factors and factor outlooks can also be explained to asset owners.

In practice however, such an approach needs to be carried out very carefully and with sophistication to match investor preferences. Skilful portfolio construction and cost-effective execution are critical:

  1. The problem with factors is that is it not clear if they will be rewarding in the future and how stable this reward is. Hundreds of papers and hundreds of factors attempt to explain the cross-section of expected returns but most claimed research findings are likely false.
  2. A number of economic views and assumptions are required. We all know that human beings are not that good at forecasting, especially when it is about the future. In order to reduce model risk in such approaches it is required to first list all known assumptions and then (stress) test all of them, also in relation to each other. In contrast, data-driven diversification as explained in this article is a low-assumption approach which is a strong plus in terms of model risk.
  3. There might be hidden factors and in the worst case these are not rewarding and also strongly correlated to the known factors.

These three points are good reason to utilise machine-based, data-driven, objective diversification at an aggregation step in factor based investing to deal with the uncertainty. To elaborate further, in the following graph we are showing a correlation matrix of some styles:

 

It is a typical data set for developed Europe on 17 November 2016. The structure looks nice as very much uncorrelated (dark blue means high positive correlation while red is for negative correlations). A corresponding asset class correlation matrix would look much more correlated. However, some pairwise style correlations are as high as 0.65. We have reordered the matrix by the help of the same procedure to extract the hierarchical structure as in the HRP approach. It is obvious that Liquidity, Volatility, Value, Size, XR Sensitivity and Leverage build a correlation cluster as marked by the circles.

Some managers also apply the Risk Parity principle or the simpler inverse-variance principle at style level. The latter ignores the correlation structure as the styles are assumed to be independent. But it can be seen on the left hand side of the picture below that the inverse-variance weighting puts a decent amount of weight to the correlation cluster in the deeper nested tree structure, creating an unintended risk concentration:

If we now apply the HRP approach to this data, some of the weights ‘move’ away from the deeper cluster to the more separate clusters like Momentum and Growth as can be seen on the right hand side of the picture above. This allocation incorporates both the risk levels and the hierarchical structure of the styles. An asset manager creating factor/style portfolios could use this graph-based machine intelligence approach at an aggregation step to diversify the uncertainty of the identified styles/factors. It can be assumed that this allocation leads to much better results. 

Summary and Outlook

Computers, algorithms and data are available at almost zero cost. Many industries today rely on machine intelligence and we have shown in this article how it helps to diversify our investments. We have covered robo and factor investing in this article. In almost all investment problems these machine-based approaches can help. 

An example is impact/ESG investing. According to Hoepner there are three main drivers of portfolio diversification ((1) number of stocks, (2) correlation of stocks, (3) average specific risk of stocks) and when we restrict our universe to sustainable investments only, we do address the 3rd aspect through a reduction of the average stock’s specific risk due to more sustainable investments. Aspect 1 and 2 however are still causing trouble as the number of investments is reduced and their correlation is elevated. That is another example where machine based diversification is reasonable. 

This is just an article and no investment advice.


 This Blog is part of a series of blogs building up on each other: 

  1. Amplifying Investment Intelligence in Wealth & Asset Management by Machines
  2. Using AI to establish a reliable and objective way of diversification
  3. Machine intelligence for robo strategies, factor investing, smart-beta and impact investing

Über den Autor

  • Dr. Jochen Papenbrock

    Dr. Jochen Papenbrock

    CEO and founder of Firamis

    Dr. Jochen Papenbrock is CEO and founder of Firamis. He has more than 10 years’ experience of management and technology consulting as well as quantitative modeling in the financial industry. Also, he has invented, developed and operationalised several innovative financial technologies. He is author, consultant, data scientist, entrepreneur, financial engineer, fintech enthusiast, inventor, programmer, quant, researcher, risk management expert, and trainer/coach. He earned his doctorate and degree in business engineering at KIT (Karlsruhe Institute of Technology), Germany.

    RSS-Feed abonnieren