Mon 31-August-2020

I spent a better part of the past 18 months researching Machine Learning in equity investment decision-making for my PhD. During that time two high-profile industry surveys and one not-so-high-profile were published (CFA / BoE, CFA Institute, and Cambridge Judge Business School respectively). They provided a valuable insight into the degree of adoption / utilisation of Artificial Intelligence in general and Machine Learning in particular in the investment management industry.

Below you will find a brief summary of their findings as well as some critique and discussion of individual surveys.

My research into ML in the investment management industry delivered some unobvious conclusions:

  • The *actual* level of ML utilisation in the industry is (as of mid-2020) low (if not very low).
  • There are some areas where ML is uncontroversial and essentially a win/win for everyone – chief among them anti-money laundering (AML), which I discussed a number of times in meetups and workshops like this one [link]. Other areas include chatbots, sales / CRM support systems, legal document analysis software, or advanced Cybersecurity.
  • There are some areas where using ML could do more harm than good: recruitment or personalised pricing (the latter arguably not being very relevant in investment management).
  • There is curiosity, openness, and appreciation of AI in the industry. Practicalities such as operational and strategic inertia on one hand and regulatory concerns on the other stand in the way. It’s not particularly surprising nor unexpected, and attitude towards this situation is stoical. Investment management has once been referred to as “glacial” in its adoption of new technologies – I think the industry has made huge progress in the past decade or so. I think that AI / ML adoption will accelerate, much like the adoption of the cloud had in recent years.
  • COVID-19 may (oddly) accelerate the adoption of ML, driven by competitive pressure, thinning margins (which started years before COVID-19), and overall push towards operational (and thus financial) efficiencies.

I was confident about my findings and conclusions, but I welcome three industry publications, which surveyed hundreds of investment managers among them. These reports were in the position to corroborate (or disprove) my conclusions from a more statistically significant perspective.

So… Was I right or was I wrong?

The joint FCA / BoE survey (conducted in Apr-2019, with the summary report[1] published in Oct-2019) covered the entirety of UK financial services industry, including but not limited to investment management. It was the first (chronologically) comprehensive publication concluding that:

  • Investment management industry as a subsector of financial services industry has generally low adoption of AI compared to, for example, banking;
  • The predominant uses of AI investment management are areas outside of investment decision making (e.g. AML). Consequently, many investment management firms may say “we use AI in our organisation” and be entirely truthful in saying so. What the market and general public infer from such general statements may be much wider and more sophisticated applications of the technology than they really are.

The CFA Institute survey was conducted around April and May 2019 and published[2] in Sep-2019. It was more investment-management centric than the FCA / BoE publication. Its introduction states unambiguously: “We found that relatively few investment professionals are currently exploiting AI and big data applications in their investment processes”.

I consider one of its statistics particularly relevant: of the 230 respondents who answered the question “Which of these [techniques] have you used in the past 12 months for investment strategy and process?” only 10% chose “AI / ML to find nonlinear relationship or estimate”. I believe that even the low 10% figure represented a self-selected group of respondents, who were more likely to employ AI / ML in their investment functions than those who decided not to complete the survey.

Please note that for any of the respondents who confirm that their firms use AI / ML in investment decision-making (or even broader investment process) it doesn’t mean that *all* of their firm’s AUM will be subject to this process. It just means that some fraction of the AUM will be. My educated presumption is that this fraction is likely to be low.

Please also note that both FCA / BoE and CFA Institute reports relied on *self-selected* groups of respondents. The former is based on 106 firms’ responses out of 287 the survey was sent to. 230 respondents answered the particular question of interest to me in the CFA Institute report – out of 734 total respondents the survey was sent to.

The Cambridge Judge Business School survey report[3] (published in Jan-2020) strongly disagrees with the two reports above. It concludes that “AI is widely adopted in the Investment Management sector, where it is becoming a fundamental driver for revenue generation”. It also reads that “59% of all surveyed investment managers are currently using AI in their investment process [out of which] portfolio risk management is currently the most active area of AI implementation at an adoption rate of 61%, followed by portfolio structuring (58%) and asset price forecasting (55%)”. I believe that Cambridge results are driven by the fact that the survey combined both FinTech startups and incumbents, without revealing the % weights of each in the investment management category. In my experience within the investment management industry, the quotes above make sense only in sample dominated by FinTechs (particularly the first statement, which I strongly disagree with on the basis of my professional experience and industry observations). I consider lumping FinTech’s and incumbents’ results into one survey as unfortunate due to extreme differences between the types of organisations.

That Cambridge Judge Business School publishes a report containing odd findings does not strike me as particularly surprising. It is, frankly, not uncommon for academics to get so detached from the underlying industry that their conclusions stand at odds with observable reality. However, the CJBS report has been co-authored by Invesco and EY, which I find quite baffling. Invesco is a brand-name investment management firm with USD 1+ Tn in AUM, which puts it in the Tier 1 / “superjumbo” category size-wise. I am not aware of it being on the forefront of cutting-edge technologies, but as is the case with US-centric firms, I may simply lack sufficiently detailed insight; Invesco’s AUM seem sufficient to support active research/implementation of AI. One way or another, Invesco should know better than to sign off on a report with questionable conclusions. EY is well on the forefront of the cutting-edge technologies (I know that from personal experience), so for them to sign off on the report is even more baffling.

Frankly, the Cambridge Judge report fails to impress and fails to convince (me). My academic research and industry experience (including extensive networking) are fully in line with FCA / BoE’s and CFA Institute’s reports’ findings.

The fact that AI adoption in investment management stands at a much more modest level than the hype would have us believe may be slightly disappointing, but isn’t that surprising. It just goes to show that AI as a powerful, disruptive technology is being adopted with caution, which isn’t a bad thing. There are questions regarding regulation applicable to AI which need to be addressed. Lastly, business strategies take time (particularly for larger investment managers), and at times the technology is developing faster than business can keep up. Based on my experiences and observation with cloud adoption (and lessons seemingly learned by the industry), I am (uncharacteristically) optimistic.