Auquan Blog

Adoption of AI in Investment Management: How can firms setup for success

June 16, 2020 by Chandini

Preface

I recently delivered a keynote at CogX’s AI festival on the above topic of AI’s (lack of) adoption in Investment Management. Since then, we received a few requests for slides and content of the talk, so I decided to put down some of my thoughts in writing.

As a brief introduction, I am the co-founder of a data science firm, Auquan, which works with clients in investment management to enhance their existing investment strategies using artificial Intelligence and machine learning solutions.

We regularly meet with investment teams at hedge funds and asset managers and I usually ask about their investment style, research process and use of data and data science at the start of the meeting. I have been highly surprised at most frequent answers to the above. Most teams are unaware of how data science and machine learning (which is different from quantitative finance) can add value to their existing investment process. Some have a deep distrust of any insights delivered by a machine.Some have tried to, unsuccessfully, set up and implement data science solutions. The bulk of this piece draws on insights from these conversations.

I have broken this blog into three short stages:

  1. Possible ways in which AI can add value in different styles of investment firms

  2. Current barriers to adoption of AI within investment management

  3. How firms can structure internal teams and integrate with third party vendors to set up for successful implementation

An obvious precursor to this is, why now? - This is a topic for another day, but some obvious reasons are:

  • The explosion of online data, to the point where even hundreds of analysts can not effectively process it all (at least in an economical way)

  • Recent advances in machine learning which have made it possible to make sense of this data at scale

  • Access to affordable compute power

  • Cost pressures and outflows into “passive” ETF strategies, forcing discretionary asset management to look for new sources of cost effective alpha or investment edge and upgrade their operational efficiency

Part 1: Use Cases of AI in investment management

Data and analytics can add dramatic value across the entire investment management value chain, as shown by this infographic from a recent BCG study.

Zooming in only on front office operations, where Auquan’s expertise lies, AI can add value across the entire investment process. Infact, in terms of raw impact to the bottom line, this is the most important business area that AI could impact.

This infographic shows various applications of AI to different parts of the investment process in different styles of investment firms, discretionary and quantitative. As well as these two traditional approaches, there is a new breed of funds in between.The quantamentals. They aim to integrate quantitative methods into the fundamental investment firm and benefit both from tools available to fundamental firms to advanced machine learning used by quant funds. These firms combine the best of both worlds Human analysts who can analyse how the future may look different than the past (like the current pandemic, or changing industry structures, emerging technologies etc). Combined with algorithms that are unmatched in automating tasks, analyzing how the past may predict the future and identifying anomalies. Integrating both presents an obvious opportunity.

I’m going to argue that any investment firm is essentially quantamental. A traditional manager is using a PE ratio screen to filter stocks or a quant overruling a model in an extreme market environment, both are using a blended, quantamental approach.

However, firms that fail to tilt on the side of increased data adoption are going to fall behind and see their returns and AUM suffer. The use of AI and data science has gone from being nice to have to a necessity within the front office to stay competitive. I’d like to emphasize that in this blended approach, the role discretionary portfolio managers is far from over, they are still the most important part of this process. MIT economics professor Richard Bookstaber put it simply in his book, “The End of Theory”: “No man is better than a machine, and no machine is better than a man with a machine.” You can find more detail on the use cases for each fund type and how a blended process works in the talk, here.

Part 2: Barriers to adoption of AI:

Despite this, adoption within investment management has still been low. According to a report by BCG in 2019, less than 30% of asset management firms are engaged in data and analytics.

In fact, a common answer at the investment team meetings is that their investment process involves analysts applying their “deep understanding” of each name to generate investment ideas and for the PM to decide which ideas to pursue.

When I ask if they measure the performance of their analysts' recommendation? No.

Do they arm their analysts with automation tools to make skimming through data faster? No.

Does a PM employ statistical analyses to identify confidence intervals around estimates and appropriately decide position sizing? No.

Do they have a data science team? Mostly, yes, a central team, quite often established recently.

The biggest barrier to adoption isn't hiring a data science team that can mine insights from datasets for you, or getting access to alternative datasets that you can mine unique insights from, or setting up the infrastructure up to store or analyse that data. These are necessary, but not sufficient.

The true problem is more deeply embedded in the culture of the discretionary firms. PMs need to fundamentally buy into the notion that they need to be making investment decisions systematically.

In fact, investing in technology and pushing for AI adoption before solving for this will only make things worse. When performance doesn’t improve because the output of AI models was not followed, the quant and data teams are going to be the first thing to be axed when reviewing resource allocation.

There are numerous firms that have built big data science teams and bought a lot of data only to have the firm at large claim it all useless. At these firms, the data science teams they’ve hired sit almost entirely outside the process itself. The CIOs don’t know where to put these people, and the PMs don't know how to deal with them, so they just sit in a standalone team in another room. I have met multiple portfolio management teams stating that their data science team is like a large supercomputer to them - they know it has some serious capabilities but they don’t know how to operate it, what questions to ask of it. Business requirements are often lost in translation because PMs and data scientists don’t speak the same language.

At the same time, I have met heads of data sciences and innovation frustrated with zero power to implement anything, because the PMs who run each book are still in charge.

Most data science projects in these firms are push rather than pull, with the data science team trying to figure out how to incorporate any of the stuff they’ve built into the PMs decision making.

That happens because the firm didn’t rethink the core investment process from the ground up. A successful implementation would see most data science projects being driven by a business need supplied by the PM and then co-developed by the data science team.

Part 3: How to set up for success

Internal Teams:

Each firm should set up a central data acquisition team that acquires new datasets, cleans them and does basic descriptive work to determine whether it’s reliable enough for further use. That’s it. Data scientists should not be placed here because they will have no impact on anything used by the PMs if they are not part of the PM’s actual group. And each PM’s desk should have data scientists included directly. These data scientists can then develop a deep understanding of the names that PM trades. These people can rapidly prototype ideas for the PM for proof of concept and then pass them to the central data and IT teams to be developed in a production-ready way.

Software:

The tendency of most investment management firms is to attempt to build all software in-house, which is a big mistake. Software required for core operations and stemming from core finance expertise should be built in-house. However, other data analysis and automation tools, which leverage diverse ML domains like language processing, big data processing or image processing, should not be built in-house. Building in-house systems is expensive and time consuming, requires hiring for a range of skills that are otherwise not needed in the firm and the systems outlive themselves quickly, unless there is a large and active development force maintaining them. In any case, your edge doesn’t stem from having this tool inhouse, other funds will have access to some variation of it as well. Just like an alternative dataset, where the edge isn’t necessarily in access to data but the answers you draw from data, the edge is in how well a PM uses the tool to draw insights. For much lower cost, you can find a third party vendor that will:

  • build the feature set you’re looking for, ensure that you are on the latest version of the software

  • that your standards are consistent with your peers

  • that your system is stable and secure.

Integrating with third party vendors maximizes the use of your resources, and ensures that you can focus on your core business.

See further details on the benefits of a decentralised team implementation, what a PM needs to do and why building non core software inhouse isn’t the best idea here.

Summary

To conclude, AI is just another tool, albeit a powerful one, to enable investment management firms to enhance their decision making - if you can have a new tool at your disposal, that you can take advantage of to provide you an investment edge, why would you not use it?

For example, a mechanic attempting to fix a car with a half empty toolkit will only do a shoddy job or worse, fail at it. He will always want to use the latest tools that all his peers are using. And the outcome still depends on the skill of the mechanic using all the tools at his disposal in the best possible way.

Similarly, your portfolio’s final performance, irrespective of using AI, depends on the portfolio manager's skill to combine his experience, intuition and insights delivered by his data team to know when to double down on a position and when to deviate from the model.

Auquan is partnering with multiple discretionary firms looking to adopt more data driven processes in their decision making. We’re excited to be able to enable this transformation to AI first for the firms - if you would like further thoughts on how you can drive adoption of AI in your firm, reach out to us via our site www.auquan.com

Share this Article: