Overview of AI in Investment Management

Hedge funds utilizing artificial intelligence (AI) have increasingly gained attention as technology continues to be a driving force behind large and fundamental changes in the investment management and financial industries. For most groups in this space, AI hedge funds represent a new way to process information and ultimately to use that information to execute various investment strategies. This post discusses the various structural, regulatory, and operational issues that arise for managers who utilize an AI strategy in their hedge fund.

Foundational Items – AI Definition in this Context

Many will liken AI funds to quant strategies which operate on algorithms without human intervention (and probably most AI funds will have programs to automate all trading) but AI is not necessarily quant – AI really is the process behind the selection of investments.  Artificial intelligence fund strategies cannot be grouped into one category and there are not specific AI investments – instead, managers utilize AI with respect to their strategy.  So an AI hedge fund may be focused on certain sectors, may be long/short, short only, etc.  Many AI programs are going to be based on long/short strategies in the large cap space because there is going to be the widest possible universe of data points and liquidity, but this is not a requirement.  We would imagine that over time as AI programs have more experience and have learned more, the programs would migrate into other investment universes and trading strategies.

Structural Considerations

AI programs are likely to be focused on liquid markets with large investment universes so the structure is likely to be basic and straightforward.

·  Hedge Fund or Private Equity Strategy.  At this point in time we have only seen AI deployed in the public markets space so most strategies are going to be utilizing the more liquid hedge fund structure.  Some AI inventors may have expertise in other areas related to technology and those areas may be ripe for early stage investments which might make for good side pocket investments (including cryptocurrencies / altcoins).  Given what we see as investor appetite for the AI itself, and not necessarily the manager’s specific expertise in other technological areas, we believe that side pocket type structures in an AI hedge fund strategy are, and will continue to be, rare in the near term.

·  Fund Terms.  Fund terms will be linked to the strategy.  As we expect most AI programs to be long/short, large cap strategies, the fund terms are likely to be basic and are likely to have favorable liquidity terms because of the liquidity profile of the strategy and the (potential) investor unease with a strategy being implemented with an AI paradigm.  Contributions will normally be accepted monthly as is standard with more standard trading programs.  Fee terms may be favorable, especially based on the recent trend toward lower fees for hedge fund products – low management fees can always be offset by higher performance fees in a tiered performance fee structure.  Right now AI strategies may utilize leverage and we have seen a number of groups do this.

·  Onshore / Offshore Structures.  There is nothing about an AI fund structure which would materially change any decision with respect to an onshore or offshore structure.  In general, a fund complex will only maintain only a U.S. fund if there are only U.S. investors; if there are non-U.S. investors (or U.S. tax exempt investors, if the manager is utilizing leverage) then the structure will be a master-feeder structure or a mini-master structure.  We currently have only had experience with AI in the liquid securities space, but if programs move to other instruments that are illiquid or have tax characteristics different from publicly traded securities, then the onshore / offshore structure should be reviewed.

Business, Regulatory and Other Considerations for AI Hedge Funds

Whereas other strategies may have instrument-related issues to consider, AI programs have a host of technological, oversight, regulatory and perhaps most important, intellectual property, issues to consider.

·  Intellectual Property.  Identification of and protection of intellectual property will be a central concern to the AI manager (as it would be with the quant manager) and we have discussed a number of these issues below. [Note: this section written in conjunction with Bill Samuels, an expert in IP issues and of counsel to Cole-Frieman & Mallon.]

Ownership of AI Code – many times the AI code will originally be developed by an individual (or individuals) and then tested on data sets and tweaked.  Therefore ownership of the code will reside in the individual who created the code.  Once in final form the individual may assign the AI code to an IP holding company that will then license the AI code to the management company and/or fund.

License Agreement – in the event the IP holding company licenses the code to the management company and/or the fund, terms of any license agreement will depend on the needs of the manager and the fund structure as a whole, but the following are common issues which will be addressed: exclusivity/non-exclusivity, ownership (including of derivative works), fees, term, termination.  Each of these issues has a number of sub-issues and other items to consider and a manager should discuss this license agreement with their attorney very carefully.

Copyright and Patent Considerations – while the actual code underlying the AI program cannot be patented, it can be copyrighted.  The copyright protects the actual code, but the conceptual framework of the code cannot necessarily be protected.  If the code interacts with the AI program in such a way that the implantation is somehow improved, then the implementation of the software may be able to be patented.  For these reasons, managers are very sensitive about protecting who can see their code, but may be able to protect themselves (potentially) through a patent.

Employee or Contractor Considerations – managers will want to protect their AI code and will need to be careful with employees and independent contractors and therefore most managers will enter into written agreements with anyone involved in the development or improvement or implementation of the code.  These agreements will normally specify that any code produced (as well as any derivative and resulting code) belongs to the manager (or the IP holding company).

Data Set Terms – when developing AI, many managers will use large data sets to begin the learning process.  Managers should make sure that they understand the terms of the license and the rights of the data owner with respect to anything derived from use of the data sets.  The big point is to make sure that the manager has the rights to any resulting manipulation and development of the data and that the manager is aware of any other person’s right to the resulting information.

Safeguarding of Code – some firms will choose to safeguard their code in some way.  Although safeguarding is not strictly necessary, there are software escrow companies that can hold code specifically for licenses and demonstrating ownership. As mentioned above, managers may choose to secure copyright registration on source code, redacting any sections that are trade secrets.

·  Technology and the Prime Broker – there are a number of issues with respect to the implementation of the AI program with the prime broker.  The manager will work with the broker’s API to integrate their trading system with the prime – managers should be aware of any triggering events (drawdown, leverage, etc) that could affect normal trading of the AI, and the manager should create infrastructure for monitoring such events and perhaps such events should be integrated into the code.  The manager should also examine what kind of human overrides the program will have if the program is an automated trading program.  Many managers also are concerned with reverse engineering by a prime broker.

·  Reverse Engineering – this has traditionally been an issue for large quant managers so many decided to use multiple prime brokers to try to hide how their quant algorithms work.  AI managers, likewise, could be susceptible to reverse engineering and may want to think about multiple prime brokers.  The confidential information provisions of any prime brokerage agreement (PBA) then become very important.  At a minimum, AI fund managers should include language in the PBA specifically noting that the broker will not reverse engineer or create derivate works on the clients confidential information.

·  Regulation of Management Company – management companies implementing AI programs are subject to the normal forms of regulation for management companies investing in securities and futures/commodities.  Generally, if the AI hedge fund trades securities and has less than $150M in AUM, the management company will be subject to state-level securities regulations – in general the management company will need to register as an investment adviser with the state or claim an exemption from registration.  If the AI hedge fund trades securities and has more than $150M in AUM, the management company will be subject registration with the SEC.  If the AI hedge fund trades futures/commodities, the standard CPO/CTA exemptions are in place.

·  Future Regulation of Use of AI?  Both the SEC and CFTC have made minor mention of artificial intelligence when discussing technology and the investment markets.  FINRA has begun to look into artificial intelligence (see here) and broadly puts this under its FinTech focus.  We believe that these regulatory bodies will continue to explore how AI technologies work in the various marketplaces and we believe that there will eventually be specifically regulations about the use of AI in trading.  Managers should note, that while there are not specifically AI regulations, manager using AI are still subject to the same regulations as managers utlitizing only human intelligence.

·  Compliance Considerations for AI Managers – managers utilizing AI should have robust compliance systems in place.  Managers will either have in-house personnel devoted to implementing their compliance program or should think about utilizing outside compliance consultants.  In addition to normal investment advisor regulatory considerations, manager will also want to have trading level compliance systems in place – for example, if the manager trades futures, the manager should have position limit systems in place.

·  Other Items.  Other items for an AI fund manager to consider include the specific risks to be disclosed with respect to the AI; as mentioned above, most risks are related to the strategy and with respect to the technology in general.  There may be specific risks associated with a certain AI program though.  With respect to other fund service providers to the AI fund, there should not be any issues.

Conclusion

As some of the world’s largest asset managers are beginning to utilize artificial intelligence with respect to their investing (see here and here), and some of the largest tech companies in the world are placing a focus on developing AI (see here about Google’s “AI-first” world), we are entering the very beginning phase of a new world where AI is an integral part of our lives and the financial markets.

****

Bart Mallon is a founding partner of Cole-Frieman & Mallon LLP. Cole-Frieman & Mallon has been instrumental in structuring the launches of some of the first AI hedge funds. For more information on this topic, please contact Mr. Mallon directly at 415-868-5345.