Editorial

Machine learning superiority? Not yet …

Machine learning is among the buzzwords asset managers are increasingly paying attention to. For one thing, machine learning tends to make processes across the global value chain more efficient. Investing in companies that generate machine-learning and artificial-intelligence (AI) solutions could hence mean investing in a potential megatrend of the next decade. Furthermore, like any industry, the financial industry and not least asset management will need to adopt the new technology. Consequently, the pressing question for many is: will machine learning eventually replace the human touch in stock selection and asset allocation decisions?


Estimating the amounts managed using machine learning is close to impossible as firms are reluctant to share this information. Even if they were more forthcoming, there is no single, unique understanding of machine learning. Anecdotal evidence points to some USD 10 billion being managed using pure AI or machine learning in 2017, with this number growing rapidly. 


Access to AI stocks via thematic indices …


Index providers have recently started to publish a number of indices to gain exposure to companies that provide machine learning and AI solutions. STOXX, for instance, has created the Global Artificial Intelligence Index (STOXX Global AI), Nasdaq the CTA Artificial Intelligence and Robotics Index (NQROBO), ROBO Global the Robotics and Automation Index (ROBO) and INDXX the Global Artificial Intelligence & Big Data Thematic Index (IAIQ). These equity indices comprise companies that are ‘positively exposed to Artificial Intelligence’ (STOXX), ‘engaged in the artificial intelligence and robotics segment of the technology, industrial, medical and other economic sectors’ (Nasdaq), ‘are positioned to benefit from the development and utilisation of Artificial Intelligence (‘AI’) technology in their products and services, as well as companies that produce hardware used in Artificial Intelligence applied for the analysis of Big Data’ (IAIQ). The definitions are broad enough to include companies such as Nvidia, Intel, and Facebook which have significant operations outside the AI space – representing one of the major problems for fund managers these days: pure plays in AI or machine learning are in many cases unlisted start-ups requiring venture capital, which only specialised funds are usually able to provide. The rather short ‘live’ history of most of the indices shows diverging developments since the beginning of the year. The performance of the IAIQ sticks out – and may provide an indication that AI was indeed the cherry on the cake of the recent tech performance – as it introduces a cap on securities with AI exposure of less than 20%. However, the IAIQ and the other indices have all shared a common fate in recent months: they were not spared from the tech sell-offs in February, March, and October this year (Chart 1).


Chart 1: Indices focussing on AI companies have not been spared from the recent tech sell-off
Chart 1: Indices focussing on AI companies have not been spared from the recent tech sell-off
Source: Bloomberg, Nasdaq, STOXX, Commerzbank Research
… and via AI-powered asset allocation


Funds based purely on a machine’s allocation decision remain a rarity. While machine learning is likely employed in one way or another by many asset managers’ quant teams, these teams are unlikely to execute an AI’s allocation decision without first trying to determine how the AI arrived at its decision. One index claiming to leave this completely to the computer is the Rogers AI Global Macro Index. While the AI model itself is a well-kept secret, it is said to ‘identify likely changes in market directions in individual countries and within the global economy’ by analysing macroeconomic data1. 


Simply speaking: the AI applies a regional equity allocation plus an allocation in US cash. The jury is still out as to how the AI index will fare in an environment changing from an upwards trending equity market and falling to stable US Treasury (UTS)y ields. Compared to a monthly rebalanced 75%/­25% global equity/short-term US Treasury portfolio, the index generated an outperformance of roughly 22% (Chart 2). However, in terms of reward to risk, the 75/25 portfolio offers a better reward-to-risk ratio2 due to the significantly lower volatility. This also holds true for a pure global equity portfolio. A plain vanilla 50/50 portfolio would achieve the same risk-to-reward ratio as the index.


Chart 2: Leaving asset allocation completely to AI has yet to prove consistent outperformance
Chart 2: Leaving asset allocation completely to AI has yet to prove consistent outperformance
Source: Bloomberg, Commerzbank Research
Hedge funds are deemed among the first movers in AI asset selection

As mentioned above, a number of asset managers are likely to employ machine learning at some point in time in their stock selection and asset allocation process – at least as an input. A 2017 report of the Financial Stability Board3 states that ‘hedge funds, broker-dealers, and other firms are using AI and machine learning to find signals for higher (and uncorrelated) returns’. The same report points out that primarily one category of funds – systematic (‘quant’) funds – is using machine learning, and that most of these funds are hedge funds. How have these funds fared? HFRX provides two indices of hedge funds engaging in systematic trading: Quantitative Directional funds primarily investing in equity markets and Systematic Diversified applying strategies on a macro-level. The performance picture of these indices has diverged strongly since the financial crisis. While both managed to perform nicely during the 2005 to 2007 bull market, systematic macro failed to deliver sustainable positive returns, while quantitative directional benefited from the strong trends in equity markets for most of the past decade. However, since 2005 none of them have beaten the equity market after costs (Chart 3).


Chart 3: Quant hedge funds – minimised losses during the financial crisis, but trailed equities since 2010
Chart 3: Quant hedge funds – minimised losses during the financial crisis, but trailed equities since 2010
Source: Bloomberg, Commerzbank Research
Diversification benefits of AI and hedge funds


The story to date: with an admittedly short history, stock indices focussing on AI companies have delivered a mixed performance compared to tech stocks in general, and neither pure AI indices nor hedge funds likely to be using machine learning have outperformed the market.


It might still make sense to consider these investments in a multi-asset context though. The reward to risk of the two hedge fund indices and the AI is for instance higher than that of commodities and REITs. In addition, they are uncorrelated to one of the traditional asset classes and only moderately positively correlated to the other (Table 1). Adding the hedge fund indices and the AI index to a diversified portfolio of US equities, bonds, credit, commodities, and REITs highlights the benefits of diversification from the hedge fund indices: while the overall portfolio return declines, the volatility declines even more, increasing the reward-to-risk ratio of the overall portfolio significantly (see Chart 4 and Table 2). Adding the AI index increases the performance of the portfolio, but also the volatility, leaving the reward-to-risk ratio unchanged.

Table 1: Decent reward to risk for Quantitative Directional


Diagonal shows reward to risk (return/volatility), bottom left triangle shows correlation since 2005, bottom right beta since 2005


Rogers AI

Systematic CTA

Quant Directional

Equities

Sovereigns

Credit

Commodities

REITs

Rogers AI

0.53

0.03

1.16

0.59

–0.69

0.76

0.44

0.28

Systematic CTA

0.02

0.35

0.17

–0.10

0.64

0.05

0.02

–0.05

Quant Directional

0.53

0.26

0.82

0.16

–0.24

0.04

0.10

0.06

Equities

0.73

–0.18

0.43

0.52

–1.23

1.11

1.11

0.53

Sovereigns

–0.21

0.27

–0.16

–0.30

0.85

0.34

–0.04

–0.01

Credit

0.32

0.03

0.04

0.38

0.47

0.86

0.09

0.10

Commodities

0.57

0.03

0.29

0.50

–0.17

0.27

–0.15

0.09

REITs

0.50

–0.13

0.25

0.78

–0.04

0.42

0.27

0.31

Source: Commerzbank Research

Chart 4: Hedge funds (HF) using quantitative directional (QD) trading strategies do not increase the performance of a multi-asset (MA) portfolio substantially, but …
Chart 4: Hedge funds (HF) using quantitative directional (QD) trading strategies do not increase the performance of a multi-asset (MA) portfolio substantially, but …
Source: Commerzbank Research
Table 2: …. increase the reward-to-risk ratio



Multi-asset ­portfolio 
(40% equities, 
30% bonds, 10% credit, commodities, REITs)

Incl. 
HF QD ­position (10%)

Incl. ­systematic CTA ­position (10%)

Incl. Rogers AI (10%)

Return p.a.

5.25%

5.25%

5.14%

5.46%

Volatility

9.35%

8.65%

8.33%

9.39%

Return-risk ratio

0.56

0.61

0.62

0.58

Source: Commerzbank Research

Summary

AI is an evolving field for the asset management industry, in particular with regard to the decision-making process for stock selection and asset allocation. A number of start-ups are offering AI tools for this purpose; funds based purely on AI are only emerging at the moment. From a multi-asset perspective, a pure AI index has not fully convinced as a means of diversification, while adding systematic hedge funds improved the reward to risk.

1 See: https://www.etfmg.com/assets/ETFMG-BIKR-FactSheet.pdf.
2 Reward to risk is computed as continuously compounded annual growth rate divided by annualised volatility.

3 www.fsb.org/wp-content/uploads/P011117.pdf.