The following is a guest article written by Estimize Founder and CEO Leigh Drogen. This is the last of a three-part series examining the rise of quantitative investing and how it is transforming discretionary asset management. The previous articles can be found here and here.
We are in the beginning stages of a massive transformation taking place within the discretionary institutional management industry. Systematic investing has been eating the lunch of discretionary asset managers. It doesn’t have to be this way, though. With the right discipline, discretionary PMs have advantages that quants can’t duplicate. But to succeed will require radical changes to the way talent gets hired and how investment teams are organized. Unfortunately, most discretionary managers won’t be up to the challenge.
- Algorithms + Human Experience = Optimal Trading
A passage from Michael Lewis’ latest book, “The Undoing Project,” speaks directly to the issue discretionary firms face today. Lewis writes about a behavioral experiment performed on a set of first year residents and accomplished oncologists. To begin the experiment, the scientists asked the accomplished doctors to tell them how they make a decision regarding whether a patient has cancer from looking at an x-ray. The doctors gave the scientists a 10 point checklist with a 1-10 rating for each item on the list. By adding up the points you could accurately determine whether it’s cancer or benign. Then scientists proceeded to give the checklist and a set of x-rays (the outcomes of which are known only to them) to both the doctors and the first year residents, asking them to determine whether each x-ray evidenced cancer or not.
I think you can guess what happens next. The oncologists who supplied the rubric in the first place showed almost zero ability to accurately determine whether the x-ray was cancer or not. They didn’t follow their own checklist, relied mainly on representativeness heuristics, and failed to do their job well. Meanwhile, the first year residents were able to score far higher accuracy rates on average because they were simply acting as the human measurement component of an algorithm.
Similarly, most discretionary PMs would likely supply a rubric for how they make decisions, but when it comes down to it, they don’t actually adhere to it. No set of new data or analytical tools thrown into the “mosaic of information” that the PM is supposed to be using will matter unless they are disciplined enough to remove their ego from the equation and discipline themselves to follow their rubric systematically.
There’s an inevitable question that arises from the above, what’s the point of the human PM if we’re going to ask humans to basically be algorithms? Why not just run a fully systematic strategy and remove the human altogether? Could a first year analyst and some good portfolio construction software more faithfully execute the signals than a PM with 20 years of experience? Science would seem to say yes. That said, there’s obviously a more optimal scenario where that 20 years of experience alongside the discipline to execute the rubric faithfully results in better outcomes due to the ability to see regime changes in the market, something quantitative strategies built on linear analysis have a hard time doing.
It’s my belief that good quantamental / systamental / factor-aware PMs can crush the systematic quants if they are disciplined. Systematic strategies are designed to make small bets across a lot of names using half a dozen or more different signals that each have a weighting in the stock selection and exposure model. A lot of them hit for singles, consistently. But that also means that when a really fat pitch comes down the plate based on all the data, they can’t swing for the fences. This is the advantage of discretionary managers. With the right discipline, they can take a big cut with a 7% position in their book when all the data lines up, and reap the rewards of the hard work.
While it’s been tough recently, there are reasons to believe this is a great time to enter the market with a solid quantitative approach to discretionary trading. The chart below shows that while there may be many secular headwinds for the discretionary investing world, the cyclical nature of this industry is extremely strong, and we’re certainly at the deepest part of the trough regarding performance, with only one direction to go.
- There’s Plenty of Talent, You’re Just Hiring the Wrong People
The last part of this puzzle is obviously the people. And here’s the sad truth: the way that discretionary hedge funds have staffed themselves historically is almost criminal (there were actually some real criminals in there too!).
Picture the normal funnel to becoming a PM running a $500M long/short equity book. You grew up in a wealthy family in a wealthy town, usually in the New York metropolitan area, parts of Silicon Valley, Chicago or Michigan. You went to Harvard, Yale or Princeton. You took an IB analyst position at Goldman or another bulge bracket. You spent a few years learning how to build a financial model before a hedge fund picked you up for an analyst spot. You made friends with your PM, who if you were lucky did well, and 5 years later when the firm had more capital than it knew what to do with, your PM told the firm to give you $200M to play with.
At no point in this process did you ever have to exhibit a lick of skill for the job that you’ve been given. Yes, you are probably a very smart individual, and you worked hard, but we all know that smart does not equal good in the investment world. Every step along the way you were selected not for the trait which would make you the best qualified to do that job, you were selected because you jumped through the hoops which lead to the correct selection bias. The sad truth is that hedge funds are run by white dudes who grew up in Greenwich, and they like (and trust) working with white dudes who grew up in Greenwich and look like them.
This isn’t some bullshit idealistic push for equality, it’s about results. If you are hiring these people exclusively, you are not selecting for skill and you will not be able to make the shift to a more data driven quantitative approach, I guarantee it. If I were starting a fund from scratch, I’d rather have a more racially, socioeconomically diverse group of kids from schools other than the Ivy’s than those from Yale who studied political science.
And don’t get me started on the lack of women running money. Every single study ever done says that they are more successful than men due to a range of behavioral and psychological factors. Yet firms tend to overlook women for PM positions due to their inability to play the game that gets them the capital allocation. And of course, we come back to the fact that the entire industry is designed to hire for people that look like the people who are currently in charge.
Firms need to start incorporating measurement of variables pre-hiring that actually correlate to success as a PM. They need to start selecting for skill, not just smarts. Our Forcerank platform is beginning to be used for this purpose, and I expect others will pop up over time. I also expect some kind of psychometric testing firm to be created which has done the research to identify certain skills and traits that correspond to success in different strategies. You don’t want the same kind of people running momentum models as the ones running deep value.
There isn’t a lack of talent. You just need to look in the right places and be willing to elevate people who might not look, talk, or act like you.
- Building the Right Team
The other major personnel issue firms grapple with is the question of how to structure their teams to incorporate the quantitative research and data science capability. Some approaches have been successful, and others have failed.
Each firm, whether quant or discretionary, is going to need a centralized infrastructure that is capable of imbibing a new data set and making it available across the firm. Many systematic multi-manager funds and large centralized managers are already setting up data teams to search for, ingest, clean, and quickly analyze new data sets to test for alpha in their multi-factor models. The heads of these teams are getting paid big dollars, upwards of $2M a year to run this process that feeds the heart of the machine – and there aren’t many good ones out there.
The imbalance of supply and demand for this position is causing some funds to make poor hiring decisions in order to simply get someone in the door. The role itself is incredibly multidisciplinary in nature and requires a strong understanding of the quantitative research process, a decent technical background, the ability to travel across the globe to conferences meeting with hundreds of potential vendors, sniffing out what’s real from what’s bullshit, determining what startups will be around tomorrow and which won’t, and then haggling over price. Please tell me which previous role prepares you for all of that?
The firms that don’t hire well are going to fall behind and see their returns suffer as data sets more quickly move from being alphas to betas as they get arbed. This doesn’t happen overnight, it takes years for alpha to get arbitraged from a data set, but some data sets won’t have as much capacity as previous ones, and as there becomes more systematic analysis, things will speed up.
The centralized infrastructure and data acquisition team is going to also house engineers, a product manager, and optimally a quant who can do basic descriptive work on a data set to determine whether it’s clean and reliable enough to have PMs use.
And that’s where the centralized team should end. Each PM or “pod” should then have a quant, an engineer or two, and a data analyst placed on their desk directly.
Here’s why. Each PM is going to be trading different names, and will need to access different sets of information. Fighting over centralized quantitative research capacity with other pods is a disaster. And then receiving some kind of report that doesn’t fit into your actual process is useless. Each PM is going to have a different checklist or rubric with different signals.
And the key are the data analysts, who need to have a deep understanding of the industries the PM is trading so that they can work in coordination with the PM and the quant to build a process that can be effectively utilized. I’ve seen people in this role who also have some coding experience so that they can rapidly prototype stuff for the quant before the centralized team goes out and does the job in a production-ready way. The quant, of course, will be testing different data sets for efficacy, and handing them over to the engineers to build factor models.
A quantitative approach and a commitment to data science by firms is not a thing you do in some other room. The only way this is going to work is if you build cross functional teams on the PM’s desk and support them with a data and infrastructure team at the top.
How Far Down the Rabbit Hole?
So if you’re a PM, do you need to take that data science class at night? Yes, but not for the reason you think. PMs aren’t going to be writing python code and working in R to do quantitative research, that’s not their job. But in order to effectively communicate and run their teams they are going to have to understand all the pieces to the process. And most of all, if they aren’t educated as to how all of this works, how are they ever going to trust the data and signals coming out of the process when the time comes to make buy and sell decisions?
On June 20th the L2Q conference hosted by Estimize is going to give discretionary PMs, analysts, and traders a one day overview of the different pieces they need to get up to speed on in order to effectively build and run their teams. The goal of the conference is not to have everyone walking away knowing everything. It’s meant as a jumping off point, to give a sense of perspective for where managers need to go next. We’ll have the vendors there that can help them take the next steps to getting educated. We’ll also have a number of heavily vetted data vendors which can fit into this process and add alpha generating signals, including our own Estimize and Forcerank data sets.
Hope to see you there.