Practical analysis for investment professionals
10 May 2018

Manager Selection and Due Diligence in Autonomous Learning Investment Strategies (ALIS)

We are in the midst of a technological revolution. The increased use of artificial intelligence (AI), machine learning, blockchain, and big data technologies is rapidly changing the asset management industry.

A recent survey on the future of the hedge fund industry by the Alternative Investment Management Association (AIMA) found that new “statistical and computational tools, including advanced quantitative techniques and artificial intelligence is forcing hedge fund firms to re-evaluate how they operate and invest.”

So how are manager selection and due diligence changing as a result of these new technologies?

For insight on this question, we met with Michael Oliver Weinberg, CFA, chief investment officer of MOV37 and Protégé Partners, to discuss the evolution of the manager selection and due diligence processes.

Protégé and MOV37 are specialized asset management firms that invest in smaller hedge funds and selected emerging managers. Founded in 2002, Protégé has been a key market player in supporting the growth of the hedge fund industry through seed deals and capital allocations to discretionary managers. MOV37 was recently established and invests exclusively in managers that specialize in autonomous learning investment strategies (ALIS), which leverage recent advances in AI and machine learning.

Weinberg will be speaking on “Autonomous Learning Investment Strategies (ALIS) — An Investor’s Perspective” at the Annual Quantitative Trading Symposium of the Arison School of Business on 14 May 2018 at IDC Herzliya, Israel.

CFA Institute: Your legacy firm Protégé Partners has been a leader in the hedge fund industry by investing in emerging managers since its establishment in 2002. Can you walk us through your manager-selection process in a traditional discretionary framework at Protégé? How do you “source” your managers?

Michael Oliver Weinberg, CFA: The reason we named our firm Protégé is because we like managers that are “protégés” of the world’s best managers. We look for managers that have received superior training while working with the world’s best managers, and are ready to bring their own perspectives to the strategies that they have worked on during their formative years to develop their own investment processes.

My partners and I have a very large network of referral sources on Wall Street. We are able to cover a good swath of the hedge fund universe just through our personal connections. In addition to our network, there are also prime brokers, capital introduction firms, and hedge fund databases that can help in sourcing.

How do you proceed once you have identified a manager for your “sourcing” pool?

We initiate our investment due diligence process. In our first meeting, we typically sit down with the portfolio manager and look to understand the manager’s strategy and how the manager differentiates itself from other managers. What can the manager contribute to our portfolio of fund investments? Does the manager have a clear, differentiated approach to investing and specific expertise to implement it? Also, the manager’s experience is very important for us. Where did the manager work? How does the manager’s strategy compare to that of its mentor? Is it similar, is it the same, do we believe there are differences, ideally improvements?

For managers that pass this first phase, the following meetings are likely to dig deeper into portfolio construction, risk management, and the research process. We want to see the manager’s analytical models, and get a good understanding of their research edge. Is the manager trying to earn outsized returns through access to higher-quality research data, better analysis, or a combination of both? Are they using public databases or do they rely on a proprietary data set?

We will also look at the manager’s team. Is the manager able to engage and retain top talent? Does the manager have the ability to run a business? In a large asset management firm, you can just focus on portfolio management and be successful, but if you want to be successful at launching and growing a new asset management business, you must be able to run a business and develop a team.

In the final stage, we’ll do reference checks. We will ask the manager to provide a list of former colleagues and social relationships. We want to get a sense of the manager’s personality and personal and business dealings. We’ll use our networks to obtain references that were not recommended by the manager, and which are potentially more impartial. We look for controversial references. If a manager left a firm under unclear circumstances, we’ll try to find the manager’s prior direct supervisor, colleagues, and people that worked for the manager at that firm.

Overall, our due diligence process typically takes six months to a year. It can also be longer.

How relevant is a manager’s historical track record in your selection process?

Our preference is for managers with established track records of at least a year. They don’t necessarily have to a present a separate track record for their new fund. For instance, a manager may be coming from a large firm where it was managing a piece of the overall portfolio and may present to us a record of carved-out performance.

When we look at a manager’s track record, we drill deeply into the assumptions that flow into the performance metrics. Is the return net or gross of fees? How are fees calculated? Does it include trading costs and commissions? Does it include financing costs? We often make adjustments to generate a pro-forma record that is comparable to what an actual hedge fund track record would look like. Sometimes managers tell us that they expect their performance to be different going forward because, for instance, they are making changes to their leverage, risk management or other key areas of their strategy.

Overall, a manager’s track record will provide us with a good starting point to evaluate performance. Having said that, we will make exceptions and consider also managers that don’t have an actual track record. We had a case recently of a young manager that did not have an identifiable track record, but worked at a very prominent asset management firm, and was immensely impressive, smart and eager to build a successful business. We asked him to set up a paper portfolio, trade the paper portfolio for several months (six to nine would be a good reference point) with full transparency, and finally seeded him with actual capital. During our “trial” period, we had a chance to go through the paper portfolio, ask questions, see his strategy being implemented, and observe his interaction with his team. That’s a good example of how we could engage a new manager that offers great potential but without a substantive track record.

Let’s now shift to your new venture MOV37. How do you source your managers at MOV37 and how is the process different from the process in place at Protégé?

MOV37 was founded to invest in emerging managers that specialize in automated, non-discretionary machine learning strategies which have been made possible by the confluence of the “Five ALIS Factors”:

  1. Availability of large amounts of data.
  2. Development of data science on how such data can be used.
  3. Development of machine learning techniques that automate data use.
  4. Low data processing and storage costs.
  5. Playing too close to the information edge.

We like to compare the development of ALIS to that of Uber. For Uber, you also needed a confluence of factors: iPhone technology, satellite technology, and a robust GPS system. Same for ALIS: If you don’t have a confluence of all five ALIS factors, ALIS will not work.

For MOV37, we typically look for PhDs from world-leading academic programs in science and technology, in areas such as computer science, particle physics, statistics, applied mathematics, electrical engineering, epidemiology. Some managers come to us from the top quant funds, such as Two Sigma, DE Shaw, Renaissance. Others come from technology firms such as Facebook, Amazon, Google, Netflix. Gamers and hackers are also fertile ground for ALIS portfolio managers.

At MOV37, we constantly look for ways to expand our network beyond Wall Street and into the broader scientific community. We are actively involved in the scientific community through thought leadership, articles, conference attendance, and meetings with scientists and potential managers. Also, MOV37 has an advisory board which includes world leaders in the development and application of AI and machine learning to a broad range of areas, including applied mathematics, contemporary art, agriculture, software engineering, sports. Our advisers help us in sourcing managers and identifying promising trends in artificial intelligence. Sometimes scientists come to us directly when they are about to launch a fund or know someone who is doing so and is looking for capital and advisory support.

What are they key features of your due diligence process as it applies to ALIS?

At MOV37, it is critical for us to have strong insight into the science and the technology behind a manager’s strategy. Our analyst team includes a number of MIT PhDs. In addition to financial knowledge, our analysts typically have a very strong scientific background.

Our due diligence meetings typically involve developers, coders, and programmers as opposed to MBAs or business graduates. We look to understand what are the basic rules for the portfolio construction and how they are using machine learning techniques. We consider how they maintain their algorithms and the process for algorithm adjustments. We also consider questions such as: What computer language are they using? What is their hardware? Are they using the cloud? Are they developing their processing and storage capacity internally? Are they aware of the available commercially applicable solutions? How robust are their statistical models? Are they overfitting?

Also, ALIS managers are supposed to be fully systematic and nondiscretionary. Typically, these managers have one or two PhDs that program the systems and define the algorithms for the trading strategy, including system constraints. Once these are set, we expect that there will be no material overriding of the automated specifications. Even if the manager’s personal view at a certain point in time is that a specific investment is not suitable, the system must prevail. In our analysis we consider how systematic the manager’s strategy truly is. Some managers incorporate machine learning techniques into their systems but still leave a significant amount of discretion for the portfolio manager to determine whether to buy or sell a particular investment at a certain point time. This hybrid model is not what we are looking for.

We are looking for funds that are fully systematic and where the order execution is automated. The managers of these funds don’t have research analysts as you find in traditional discretionary funds. You also don’t find risk management as a separate function, since the risk management component is built into the systems by PhD strategists, coders, and developers. The roles of traditional portfolio and risk managers are effectively systematized. We do believe that ALIS managers may be able to provide a strong potential source of alpha and excess return at a cost that is significantly lower for investors than that of most traditional discretionary managers.

One of the key features of ALIS managers is the ability to acquire and process a large amount of data in a cost-efficient manner. How do you evaluate the type and quality of data used by a manager in your due diligence process?

The data market currently is highly inefficient. There are large commercial data sets that are heavily marketed, such as satellite images and credit card databases. Our view is that these large commercial data sets may have value for certain players, but in terms of ALIS, they are largely already potentially commoditized.

The market for large commercial data sets makes me think of an analog Marcos Lopez de Prado, an AI professor and portfolio manager, uses regarding the late Gold Rush: You can look for a mine with big nuggets of gold, but chances are all the large mines have been detected and you may be better off finding your gold by sifting through a lot of water with high-technology tools. It is very hard to extract value just by trading on credit card data or satellite data. However, you can find ways to put your data sources together, and generate alpha from the synergies that come out of it.

Also, you can search for data sources that are not easy to find and that can be potentially very valuable: This is what ALIS managers typically do. There are massive amounts of data on the internet that we are just starting to tap into. We can classify data in four categories: structured/non-structured and financial/non-financial, and organize the aggregate of the available data visually in four quadrants: At the top left you have structured, financial data, including the data available on Bloomberg, Reuters, Capital IQ, FactSet. On the bottom right, you have the non-structured, non-financial data, such as data from Twitter, social media, video and news data, corporate exhaust data. The data in the bottom right quadrant is typically available at low cost.

Low processing and storage costs can be used to create whole new alpha sources for unstructured and non-financial data. Most of our ALIS managers are millennials who don’t like to pay for data access. Many of them do not subscribe to financial data providers like Bloomberg: They’ll go online and set up their own program to extract trading and pricing information directly from the exchanges. They also have an incentive to keep their costs down to increase the profitability of their funds and returns to investors.

Still, there are a number of challenges in using proprietary databases and generally non-structured data (financial and non-financial): All the data has to be anonymized, it has to be obtained legally, with permission, and not be material non-public information. Also, the data has to be of a quality that can be relied upon as part of model building. These considerations are at the core of our due diligence process for ALIS managers.

Can you talk about your seed process for emerging managers? Do you engage in seeding for your ALIS managers?

In a seeding arrangement, we take a revenue share with a manager that also allows us to participate in the manager’s own investment return through a share of the asset-based management fee and performance-based compensation. This is in addition to any return that we may receive based on our ordinary share of a fund’s profits as a limited partner.

Protégé has been seeding managers for over 15 years and right now we have four seeds. We have not yet done seeds in MOV 37, but we have the intention to do so in the near future. In a seeding deal, we look for managers that show promise to be able to run a commercially viable business. In seeding, there are two sources of return: the return on the fund and the return from investments made by third parties, namely your share of revenue coming from the management of third-party assets. In order to raise capital from third parties, you need to have a strong business, and you need to have marketing.

If you are investing in a fund at arm’s length, without seeding it, you may not care if the manager is a good marketer. We have multiple situations where we have managers that are not good marketers but are still generating excess returns for us and their investors. On the other hand, if we are seeding, we very much care that our manager be good at marketing and can raise capital, so that we can have the incremental return for our clients from third-party revenues. Another significant difference is that in arm’s length investing (non-seeding) our capital can be redeemed typically on a monthly basis, or also more frequently if the investment is through a managed account.

In seeds, on the other hand, we typically make a two or three year commitment: As long as the manager does not violate its seed arrangement terms, our capital is locked. When we consider a seed deal, we pay special attention to the terms of the seed arrangement and the clauses that can allow us to exit from the investment if certain circumstances occur: For instance, the agreement may allow for an exit if the manager is down more than a certain percentage of a pre-defined period of time, or if the manager violates certain leverage constraints, limitations on illiquid investments, or other risk constraints. Also, our reputation risk is much higher in a seed deal, as we associate our name to that of the manager.

What is the fee model you typically negotiate with your ALIS managers?

In our ALIS investments, we favor a 1-10-20 structure, namely an annual management fee of 1% plus a 10% performance fee up to 10% return, which increases to 20% on the entire performance for returns above a 10% net return, i.e., a catch-up provision. We believe that such a fee structure favors the alignment of interest between managers and investors. The 1-10-20 fee structure acts as an incentive for ALIS managers to absorb their data costs at the management company level (or minimize them), rather than passing them through to the fund, otherwise risk not hitting the 20% incentive fee.

Fintech and ALIS are rapidly changing the environment in which asset management firms operate. Is there a place for traditional value investing principles in your vision of the future for the asset management industry?

I love value investing. I studied at Columbia University, and had exposure to all the top value investors there: Warren Buffett, Michael Price, Mario Gabelli, Glenn Greenberg, Seth Klarman. I expect that ALIS managers will be able to systematize value investing theories in their strategies. The principles of value investing will still be valid, but they will become the basis for automated trading execution.

ALIS managers will ask value investors what they like: high margins, high market shares, low cost, management that has a history of creating value, ESG. They will be able to use the financial data sets and factors that value investors are interested in and process them through their machines using machine learning algorithms. A lot of what value investors are currently doing can be systematized and done faster and more efficiently by machines.

Having said that, we still see opportunities for discretionary value investing managers in complex situations where the five ALIS factors do not yet apply: activism, illiquid securities, and frontier and emerging markets in Asia, Latin America. and perhaps Africa. Today ALIS strategies are primarily in US equities, futures, and ETFs: liquid strategies. You need a lot of data to be able to design and implement ALIS strategies. A market that is illiquid will not lend itself to ALIS.

Our industry, however, is evolving rapidly. We are already starting to see ALIS strategies in investment grade credit. There is now a municipal bond fund that uses clustering techniques and extracts data from the exchanges. We believe that ALIS will be a disruptive agent of change for the asset management industry as we know it.

If you liked this post, don’t forget to subscribe to the Enterprising Investor.


All posts are the opinion of the author. As such, they should not be construed as investment advice, nor do the opinions expressed necessarily reflect the views of CFA Institute or the author’s employer.

Image credit: ©Getty Images/ Pixtum

About the Author(s)
Antonella Puca, CFA, CIPM, CPA

Antonella Puca, CFA, CIPM, CPA, has more than 15 years of professional experience in the alternative investment industry. She has served as a director of Global Investment Performance Standards (GIPS) at CFA Institute, a director of the board and member of the executive committee at the CFA Society of New York, and a member of the Alternative Investment Management Association (AIMA)’s research committee. She has been part of the alternative investment group at KPMG/Rothstein Kass, where she helped launch RK’s Bay Area practice, the global hedge fund practice of EY in San Francisco and New York, and the financial services team at RSM US LLP in New York. Antonella has been an adjunct faculty member at New York University and is a frequent speaker at conferences on alternative investments. She is licensed as a CPA in California and New York. She is also a Certified Valuation Analyst and a member of the Italian Professional Association of Journalists. She holds a degree in economics with honors from the University “Federico II” of Naples, Italy, and a master of law studies in taxation from NYU Law School. She has been a research fellow at the Hebrew University of Jerusalem and a member of the 420 Italian National Sailing Team.

2 thoughts on “Manager Selection and Due Diligence in Autonomous Learning Investment Strategies (ALIS)”

  1. Martin says:

    Machine learning and AI models are often labelled as “black box” approaches and given the requirement to not leave discretion, how do we interrogate the product and allow transparency?

    1. Antonella Puca says:

      Hi Marin, thanks for your attention and for your question. Michael notes that they don’t consider AI models as black boxes because they are generally understandable, i.e. either fundamentally driven or technically driven, just as discretionary funds are. One needs to do attribution analysis on them just as one would for discretionary funds. As part of its process, MOV37 has a team of experts (mostly MIT PhDs) that has developed tools suitable for an attribution analysis of this kind. To protect proprietary information, MOV37 typically would sign an NDA. The managers give MOV34 their portfolios real time or on a lagged basis and MOV34 has very good transparency into their trades and attribution. I hope this helps.

Leave a Reply

Your email address will not be published. Required fields are marked *



By continuing to use the site, you agree to the use of cookies. more information

The cookie settings on this website are set to "allow cookies" to give you the best browsing experience possible. If you continue to use this website without changing your cookie settings or you click "Accept" below then you are consenting to this.

Close