Q2-2021 EXAM: EARLY REGISTRATION OPENS ON DECEMBER 7, 2020

Resources

Events  | News | Resources | Webinars |

Do More With Data: Modern Asset Management with AI and Machine Learning 


August 21, 2020 
By Mehrzad Mahdavi, Executive Director, FDP Institute

Recent conversations with FDP advisory board members, Tony Guida, Joseph Simonian and Ganesh Mani have highlighted modern quant trends implementing AI/Machine Learning in asset management.  Also, Marcos López de Prado has written about ten applications in finance ranging from asset pricing to credit ratings & analyst recommendations. He also addresses some of the well-known challenges: e.g., overfitting and interpretability of neural network models. 

I was also reminded by another FDP advisory board member and a thought leader in multiple industries, Nicolaus Henke, about the importance of scaling AI projects, auto coding, and data pipelines. 

AI-based asset management got a significant boost roughly three years ago, when BlackRock, one of the largest asset managers in the world, started transforming its entire business around AI. It currently uses AI - acting on diverse traditional and alternative data sources - to support many internal functions including portfolio management.

Morgan Stanley has successfully developed AI systems for wealth management, institutional investing and other businesses.

Recent surveys on hedge fund performance from consulting and research firm Cerulli, shows that the AI-powered funds produced cumulative returns of 34 percent in the three years through May, compared with a 12 percent gain for the general global hedge fund industry over the same period.

In the above, we presented plenty of evidence in successful applications of AI/ML in asset management.  So, what are the best practices and lessons learned?

Joe Simonian notes: “The AI and ML algorithms that are generally the most useful in asset management are those that are relatively transparent and those that have relatively low data costs. That said, practitioners must also be sensitive to interpretability issues and do their best to ensure that their ML frameworks can be reconciled with the basic assumptions of economics and finance.”

Tony Guida points out that in order to come up with the best practice AI/ML techniques in asset management, “you need to clearly frame the problem and determine the KPIs before coming up with appropriate data sets and models”.  In another words, one size does not fit all. 

Another key component in successful implementation of AI/ML, is efficient delivery of knowledge modules and tools tuned to the particular problem being solved.  This requirement gives rise to specialized training customized to the problem at hand.  More on this subject in my next writings…

In the following, I synthesize these conversations, readings and observations around AI/ML applications in asset management, in the form of few takeaways:

  • Focus on creating an AI platform that can grow to absorb different traditional and nontraditional data sources as inputs. Such a system will help in creating scale for the deployment of AI by providing workflows and services to ingest a variety of inputs, including data pipelines, to enable decisions under different regimes or circumstances.
  • Creating such a platform requires skills and resources many asset managers currently lack.  These resources include: chief science officers; machine-learning experts; data scientists/data engineers fluent in finding, vetting, and wrangling large amounts of real-time, unstructured, noisy data; and engineers to set up an efficient computing environment.
  • Avoid total outsourcing of the AI projects.  Instead, build cross-functional teams fluent in both AI and investing.  A key (new) role would be that of “translators” to bridge the gap between technologists and business managers; and, to serve as AI advocates internally and as external evangelists with clients, prospects and consultants.
  • Establish focused AI/ML training programs to align employees or new hires to excel in the above multi-functional teams.  Training programs should be modular, have different layers (think Six Sigma!) and practical to get the practitioners to their points of interest quickly and efficiently. 

References:

file:///Users/mehrzadmahdavi/Downloads/SSRN-id3365271.pdf

https://www.institutionalinvestor.com/article/b1505p7qq0b511/with-blackrocks-artificial-intelligence-pivot-the-rubicon-has-been-crossed

https://www.wsj.com/articles/ai-project-failure-rates-near-50-but-it-doesnt-have-to-be-that-way-say-experts-11596810601?mod=djemAIPro

https://www.institutionalinvestor.com/article/b1mssrswn1mpr0/AI-Powered-Hedge-Funds-Vastly-Outperformed-Research-Shows





  Brought to you by:

 

2020 Financial Data Professional Institute©

Contact Us: info@FDPinstitute.org 

Privacy | Policies | Terms of Use | FAQ

Powered by Wild Apricot Membership Software