Oversight Recommendations on AI and Quantitative Investment Strategies

With the growth of artificial intelligence (AI) and quantitative strategies such as smart beta, fund directors and wealth managers are increasingly asked to oversee funds that use AI and quantitative methods to manage investments. Fund directors especially must understand the basis of these technologies to properly govern the funds they oversee. 

A research paper developed by Brian Bruce, CEO & CIO of Hillcrest Asset Management, provides a framework for directors and wealth managers to understand how to oversee these funds. The paper discusses the concepts of artificial intelligence and quantitative investing, particularly focusing on key areas that directors should understand. The research poses vital questions and provides important perspectives on these increasingly employed investment approaches and tools that can be helpful for asset managers, advisors, and investors.

Hortz: How did you define and characterize Quantitative Investing and Artificial Intelligence in investment management for your research paper? 

Bruce: Quantitative investment management is simply the use of computers and data in portfolio management. AI is a distinct set of investment strategies in which the machine has taken over not only the acquisition of data and its processing, but also the judgment behind the decisions. 

Hortz: What do you feel are the main governance priorities in overseeing these strategies?

Bruce: The main priorities to determine for these strategies are:

Does the fund adviser have the expertise, knowledge, and resources necessary to carry out the intended strategy? Prior to any fund launch, directors should discuss whether the adviser has the expertise, knowledge, and resources to carry out the intended strategy of the new fund. However, boards may have a more difficult time assessing the adviser’s ability with respect to an AI strategy due to the newness of application and lack of substantial track records to be guided by. 

How is quality assurance monitored? When adding AI funds, the advisor may need to change approval procedures. The procedures are sound for current strategies, but AI is much more of a black box than most existing strategies. This means that directors will need to learn additional information and will need to ask different questions to make sure that the AI is properly tested and implemented. 

Hortz: Can you discuss with us some of your key findings in your research that you feel we need to consider in doing oversight of these strategies?

Bruce: A few key considerations would be: 

  1. Quantitative funds have been around for many years. They invest based on a predetermined set of rules created by the investment team. The difference with artificial intelligence funds is that there are no predetermined rules: AI looks at large amounts of data and creates its own rules.
  2. Quantitative and AI strategies involve some sort of backtesting to confirm that the strategy will work going forward. We provide questions in our research report to ask the investment team to ensure that the testing was done properly, as well as SEC rules for disclosure of backtesting.
  3. Directors should set up a framework for judging the AI process and its structure. They should also set criteria for expected outcomes in order to approve AI funds. Finally, directors should specify for management what they expect from the AI effort going forward and how that will be communicated. It is critically important that directors have a process that results in a consistent metric, which will enable them to better govern these new funds.

Hortz: What are the common problems you should be aware of that the SEC has with hypothetical backtesting?

Bruce: There are a quite a number of problems that the SEC has with hypothetical backtesting. We have also compiled an appendix to our report with specific questions you need to ask to evaluate a backtest in response to these SEC concerns: 

  1. Failure to disclose limitations. One common allegation is that firms fail to fully disclose the limitations on Hypothetical Backtested Performance (HBP).
  2. Insufficient backup data. The SEC will seek to verify that you have maintained adequate backup data to support your HBP claims. 
  3. Cherry-picking time periods. Many firms violate the SEC marketing rules when they cherry-pick a specific time period that makes their HBP look better.
  4. Misleading disclosures. Hidden or confusing HBP disclosure will draw the SEC’s enforcement interest.
  5. Retrospective model changes. Firms can’t keep tinkering with their models to improve the HBP results.
  6. Using incorrect historical market inputs. The SEC can verify market data from past time periods, so make sure to use the correct numbers.
  7. Applying different models. The SEC has raised red flags when HBP differs significantly from audited or live performance information applying the same models.
  8. Using the wrong model rules. Firms have gone astray by applying different model rules to the backtested data they use to manage real accounts.
  9. Investments didn’t exist. The SEC will call out HBP that includes investments that were not available at the time.
  10. Faulty algorithm. Faulty programming can result in inflated performance numbers.

Hortz: What do you consider the key Risks for fund boards? 

Bruce: Thinking that AI funds can be governed the same as quant funds. AI requires an understanding of a different set of complexities which can cause the computer to come up with the wrong answer. A famous example is an AI that was built to distinguish a dog from a wolf. It was provided a set of training images and was then able to very accurately identify the difference. Only when different photos were tested did it start to fail. The researchers realized that the AI was choosing “wolf” if it there was snow in the background since all the training photos of wolves were in snow. Boards need to be sure that the investing AI is not finding snow.

Hortz: What are your main conclusions? 

Bruce: Oversight of AI requires proper diligence and understanding of its limitations. Boards need at least one member that has experience or knowledge of AI if they hope to govern the AI funds properly.

Hortz: What other research or Additional Reading do you recommend for deeper understanding of these investment strategies or tools?

Bruce: I particularly like the following CFA Institute piece because it is comprehensive. Our study is the only one we know of that addresses these issues for boards.

Artificial Intelligence in Asset Management by the CFA Institute.

The link to my study we have been discussing was published on the Mays Business School Innovation Research Center: