This content is from: Portfolio

Why Are Allocators in the Manager Selection Stone Age?

Imagining the future of the manager cull.

  • By Angelo Calvello & Christopher Schelling

Selecting asset managers is an essential part of every allocator’s job. 

It’s an admittedly imperfect process, especially in asset classes with broad dispersion of manager returns and higher risks, and one that has remained largely the same since 1969, when Russell Investments hired its first manager analyst. Improving the process — improving the accuracy of the selection and increasing the efficiency of the process — is quite literally a fiduciary duty. 

To discuss — and debate — Institutional Investor columnist Angelo Calvello spoke with Christopher Schelling, director of private equity at the Texas Municipal Retirement System.




Calvello: Chris, how does the manager selection process start for you?  

Schelling: It typically starts with reverse inquiries from managers seeking to get on our radar. We generally get about 500 unsolicited pitch decks a year — in private equity alone. 

Calvello: How do you cull the herd? 

Schelling: We usually start with a cursory review of the manager materials applying the traditional Five Ps framework: people, philosophy, process, portfolio, and performance. This might lead to a quick intro phone call — quick because we don’t have the time to commit to anything longer. We are looking to see if the manager meets some minimum standards (e.g., team tenure, resources, alignment) and presents an interesting proof of concept.

Calvello: Let me offer an alternative scenario; tell me if it could improve either your accuracy or efficiency. Imagine your introductory call with a prospective manager went like this:

A proprietary Texas Municipal Retirement System artificial intelligence bot arranges a time for a video call (which can be conducted with any webcam and is recorded with permission) with the inquiring manager and his or her team. 

However, unlike the customer services bot we regularly encounter, this bot sounds so natural that the manager will likely not even know it is a bot (although it would be prudent to disclose this to the manager). Importantly, this bot has been trained for the explicit purpose of extracting the information you, the allocator, deem critical to properly assess whether the manager meets your minimum standards and your proof-of-concept criteria. The bot’s interrogation process does not follow a predetermined, rigid, linear Q&A format — but because it is guided by a deep learning program, it is capable of understanding the manager’s answers and comments within context and adroitly pivoting to dig deeper into these comments as they occur.

Simultaneously, and in real time, another AI-based technology is automatically capturing and analyzing the team’s emotional valence by reading their linguistic (e.g., word choice) and paralinguistic expressions (including accent, pitch, volume, speech rate, and modulation) and over facial micro-expressions, including their vascular responses. 

Post-interview, a verbatim transcript of the call is robotically produced, highlighting any strengths or weaknesses the manager might have exhibited. Both the video and the transcript are searchable by specific emotions (e.g., when did the manager show contempt? Disgust?).

How does this sound? 

Schelling: It definitely sounds like this emotion recognition technology would be a big leap forward — it would provide me with quantifiable data on qualitative attributes. We could use the output from this introductory meeting to prioritize any potential next steps. It would also allow us to take more initial introductions, as our time would be spent more efficiently reviewing analysis as opposed to sitting in meetings. 

Calvello: Let’s assume the manager passes this autonomous introductory interview. What would you do next?

Schelling: The next step in the review process is what we refer to as desk work. This is the traditional collection and analysis of specific requested materials (historical return data, ownership structure, team biographies, ethics policies, sample investment memorandums, quarterly letters, SEC ADVs, etc.) and the manager’s standardized due diligence questionnaire and private placement memorandum. This usually requires a review of 1,000 or more pages of printed materials. It is a very laborious and time-consuming process, and, in general, TMRS, like all allocators, lacks the time and human resources to fully vet all the materials in detail. 

Calvello: Let me relieve you of some of the tedium of this desk work by using natural language processing technology that would review the DDQ and other printed materials and turn this text into structured data that could be flagged and analyzed for insights.

Humor me and assume that the manager also passes these traditional screens; what would you typically do next?

Schelling: I think most allocators would agree with me that the qualitative aspects of manager selection are the most critical considerations, but often the most difficult to investigate. 

While historical returns are often a necessary but insufficient condition for picking managers, research supports the impact of qualitative manager information on future performance. Here is where I would dig deeper into what many allocators call the intangibles: Is the team committed? Are they aligned with the investors? Are they honest, hardworking, and accountable? 

At TMRS we try to supplement our qualitative opinions about these issues with personality inventories and aptitude tests, which allow us to at least put metrics around some of these otherwise subjective items. And while we realize the process isn’t perfect, it’s the best we’ve got. 

Calvello: The emotion detection technology we used above could certainly help discern some of the intangibles. And we could use other AI-based technology to supplement this and your current personality tests. 

For example, you could request information derived from the manager’s internal application of ubiquitous computing, specifically, data passively collected from employees’ sociometric ID badges. These badges are embedded with microphones and sensors and automatically quantify individual and collective signals at the millisecond level. These signals, integrated with information from the employees’ emails and calendars, provide a comprehensive picture of how employees spend their time at work, a picture that scientists say is highly predictive of group performance. 

We also could review data from the company’s real-time, automatic compliance monitoring software of landline and cell phones. And let’s complement this granular data with a report from the company’s happiness meter — AI-based technology that uses employee cell phones to measure the collective happiness of the organization.

Let’s top this off with technology that screens social media to determine if a manager’s public posts and profiles reveal a possible risk to the company that could adversely impact your investment — like racist or misogynistic behavior.

Schelling: All of these technologies would be very helpful — but it almost doesn’t seem possible, at least today.

Calvello: These technologies might seem futuristic but all are available today and, in many cases, they are being used for non-investment commercial purposes.

The bot used for the screening call is an appropriation of Google’s new Duplex, “a new technology for conducting natural conversations to carry out ‘real world’ tasks over the phone.”   

Emotion and sentiment speech recognition technology is offered by numerous firms and widely used in several verticals. For example, The Economist reports that Chinese insurer Ping An Insurance (Group) Co. allows prospective borrowers to apply for a loan by answering “questions about their income and plans for repayment by video, which monitors around 50 tiny facial expressions to determine whether they are telling the truth. The program, enabled by artificial intelligence (AI), helps pinpoint customers who require further scrutiny.”

Natural language processing is certainly being used by some investors. For example, one company, Prattle, says it uses it to automate “investment research by quantifying language” and to “produce analytics that predict the market impact of central bank and corporate communications.”  

Humanyze, a people analytics firm, offers employers the sociometric badges described above. The company claims that several Fortune 500 companies use the badges and underlying technology. 

Numerous companies offer landline and cell phone compliance technology (for example, Behavox). According to Bloomberg, some banks already use this technology to monitor trading desk activity by scanning “petabytes of data, flagging anything that deviate[s] from the norm for further investigation. That could be something as seemingly innocuous as shouting on a phone call, accessing a work computer in the middle of the night, or visiting the restroom more than colleagues.”

More generally, employee monitoring software is widely available. These tools, according to one vendor, allow companies to monitor “almost 100 percent of employee activity and communication,” including internet and app usage, email, phone use, and vehicle location.

Such monitoring software is also widely used. According to a survey of 1,627 large and midsize firms by the American Management Association — whose clients and members employ more than a quarter of the U.S. workforce — “nearly 80 percent of major companies now monitor employees’ use of email, internet, or phone,” up from 35 percent in 1997. 

Hitachi has developed and deployed AI that “utilizes data from the accelerometer embedded in smartphones to measure the collective happiness level of an organization.” 

And while Hitachi’s happiness gauge might be a bit Orwellian, technology that compiles and reports a person’s social media profile is amply available

So Chris, as my prototypical allocator, the big question is: Would you use these technologies?

Schelling: They’re pretty cool and they would clearly aid us in our due diligence. 

But I want to be clear that the final decision — whether to hire the manager — would still rest with humans. However, there would still be a few issues that must be dealt with before I, or probably any other allocator, would use them. 

Let’s put aside the issue of cost and simply assume we could afford them.

A big issue would be institutional buy-in. Public pensions — and most fiduciaries — are averse to change, and without definitive proof of the benefit, it could be hard to convince the relevant constituents to support the process.

I’m also concerned about the issue of employee privacy. I understand the information we would receive (for example, from the sociometric badges or monitoring software review) would be anonymized, but employers’ use of surveillance software in general seems like a violation of employee privacy.

Calvello: I was concerned about this, too, until I learned that this monitoring software does not appear to violate privacy laws. 

For example, I read this comment in the Chicago Tribune by a privacy attorney: “‘Generally in the workplace, there isn’t a right to privacy.’” The Tribune says “management can look at most anything a worker creates on the job or with company equipment. That means emails, social media posts, internet searches, text or instant messages and GPS devices that track employee whereabouts.” 

Schelling: Still, there could be some concerns about Freedom of Information Act issues, although in most states, laws support the stance that such sensitive and confidential material acquired during the diligence process is non-disclosable. 

Calvello: What about another hot-button issue for asset owners — their fiduciary duty. Do you see any concerns there?

Schelling: Certainly, all asset owners take their fiduciary duty seriously. Delegation of decision-making authority to staff is sometimes considered unacceptable because it is viewed as abdication of fiduciary duty, and relying on technology for decision-making, even if it is merely prioritization or ranking, could be viewed similarly. 

However, because the decision to invest/not invest ultimately still rests with human beings, all this technology is doing is giving us new and different data and streamlining the entire process — so I’m not sure the potential violation-of-fiduciary-duty argument would hold water. Moreover, fiduciary case law has held that the process, not the outcome, is the measure of the standard of care, and I would argue that a process utilizing these tools is one that is more robust, with more data, and less subjective. It would arguably be a tighter fiduciary standard, not weaker. But I would defer to counsel for a legal opinion.

Calvello: I’d also argue that the technologies create an auditable diligence record to support that a prudent level of research led to the investment decision. 

Schelling: The last potential difficulty is manager participation. Managers might be reluctant to participate in monitored and taped video calls or to share even anonymized HR data, assuming they even have this data. 

Allocators need sufficient leverage to ensure compliance. I can see this being applied in asset classes where the investors have all the power, or by someone with more competitive strength with the hard-to-access managers. In general, investment consultants might be best positioned to both benefit from the improved accuracy and increased efficiency these technologies offer and to require managers to comply with their requests. 

Calvello: I understand these objections — but they relate more to cultural or behavioral issues, not to the technologies themselves. As I said before, companies are using these technologies but, after speaking with a mix of allocators, I could not find one that was using the available technologies mentioned above or really any other AI-based technologies as part of their manager selection process. 

Given the universal opinion that the process could be improved, I found this odd. Screening a manager’s online presence with Google Alerts does not qualify, and planning to develop an AI tool to scrape a manager’s PDF documents does not meet my bar. However, I did discover a multi-manager shop (Weiss Multi-Strategy Advisers) that is using machine learning to vet the performance of its internal managers.

Schelling: In an ideal world, I’d be exploring how we could use these tools as soon as practically possible. 

In the real world, however, while the potential is clear, it seems unlikely we’ll see broad adoption — even from the consultants who would most benefit from them.

However, I do believe that our fiduciary duty includes the obligation to always try to do better for our beneficiaries, and if tools and technologies exist that can improve the likelihood of successful manager selection, I, and we collectively as allocators, need to be considering them. 

Calvello: From your mouth to God’s ear . . . .