Hello Quantopian Community!
I want to share with you good news and important updates for our asset management business.
There are five key points:
- We have pivoted our investment strategy to use “signal combination”. Our community is fully focused on finding alpha in data, while Quantopian is responsible for global portfolio management and risk. The benefits of this approach are legion and spelled out in the long version below.
- The royalty arrangement for licensing your algorithm has changed from a payment on your algorithm in isolation, to a weighted share of the total portfolio performance.
- Our investment team leadership is transitioning from the incomparable vision of Jess Stauth to the deep quant portfolio management experience of David Sargent.
- Our contest and underlying risk model have been valuable tools for the community. ~6% of contest entrants have become licensed authors with the potential to receive royalties. This is a massive increase in our community members’ graduation rate!
- Over the last nine months, capital allocations have steadily risen. We now have over $200M allocated to community algorithms and we have paid royalties totalling $270,000 since August of 2018.
To fully explain these points, I want to share some history and some of the thinking behind these changes.
We launched our first crowd-sourced investment strategy in April 2017. Through a challenging few months with ups and downs along the way, we ended the summer of 2017 with $88M post-leverage capital allocated to 14 community members. Our investment strategy resembled a multi-manager fund, where each algorithm managed an independent portfolio and Quantopian managed the global portfolio by applying constraints to both the individual algorithms and to the global portfolio.
The multi-manager approach put incredibly high demands on each algorithm. For this model to work, we needed ~75% of algorithms to be profitable each period. We also wanted to execute trades for each algorithm as closely as possible as the source algorithms’ intent, out of a justified paranoia that we could drift away from crowdsourcing if we meddled with the algorithms’ behavior.
These constraints required a rigorous, extremely time-consuming, and onerous selection process that led to an extremely small number of viable algorithms. By the summer of 2017, we knew we needed to adapt.
We started by revisiting a crucial question: in our investment process, how should responsibilities be divided between our community of researchers and our internal investment team? Rather than expecting the community algorithms to handle everything, we concluded that the community excelled in scouring data for alpha. The breadth and diversity of the community is our unique strength in alpha research. However, we also concluded that our internal centralized team was best positioned to manage trading, global portfolio construction, and risk management. This thinking led to a key realization: we needed a new division of responsibilities to reflect the respective strengths of our community and our investment professionals.
With this new division, we had two deep questions to answer:
- How could we better guide the community to focus on alpha research?
- If we pivoted from the multi-manager model, what would our new model be?
To answer the first question, we decided to define an algorithm’s alpha as the portion of its returns that wasn’t explained by common risk factors. Our product team honed in on how to communicate our definition of alpha to the community. The answer came in two parts: defining what alpha meant by creating and publishing a multi-factor risk model, and updating our contest.
The multi-factor risk model let us define the common risk factors, make them available to the community, and let the community measure their ideas against these factors. We redefined our contest to require minimal risk exposure, forcing returns to be more driven by stock selection than by exposure to the market, industry, or style. The results have been an astounding increase in success rate for community authors — 6% of contestants have had their algos funded! This graduation rate is several orders of magnitude higher than our previous model.
The second question, about our new model, proved far more difficult. Jess Stauth led us to a completely new way of thinking by saying (often): “Think of the community algorithms as data sources. Instead of forcing an investment strategy onto our data, let’s listen to the data first and then design a strategy to match.” Her first-principles approach transformed the design of our investment process into a data science problem. Our conclusion is that we have algorithms producing alpha, but we need to combine them in a way that will allow our model to work when 51% of algorithms are up in a given period.
Rather than try to match every order from an algorithm exactly, we instead treat the algorithm’s positions as a signal and our job is to manage exposure to that signal. This “signal combination” approach allows us to incorporate many more algorithms, provides higher dollar capacity, and results in lower trading costs. To operate our signal combination investment strategy, we need to take on greater responsibility for portfolio construction and management.
We also needed to adapt our royalty structure for licensed algorithms. The multi-manager model provided a separate return for each algorithm. In signal combination, we have a single return for the full combination. As a result, we are now paying royalties calculated by:
Royalty = (weight of algorithm in signal combination) * (total net profit of the combination)
The weight of your signal will be based on the quality of the alpha. We have signed approximately 25 authors from 12 countries including the United States. These authors have licensed approximately 40 algorithms under the new agreement. As with our platform, we solicited feedback on our new model and agreement from our licensed authors and incorporated their feedback. Listening worked well: 100% of authors who were offered our signal combination package accepted the new payout structure.
The feedback from authors has been overwhelmingly positive -- they recognize that success comes from diversification so they appreciate their royalty being calculated on the performance of the entire strategy, but they also like that their reward is based on their contribution (via the weight).
We believe deeply in the power that comes from a community built on common interests. We have always seen collaboration in our forums, at our events, and in the teams that form to work on Quantopian together. We have always wanted that same style of community for our signed authors. Among signed authors, we’re seeing direct collaboration on algorithms as well as collegial activities like video meetups to review papers. Our goal is to continue to sign more authors, to spread this “multiplayer mode” to more and more of the community, and to foster virtual teams all over the world. Quantopian is the derivative of our community’s diverse talents and motivation. I’m delighted that our new model rewards a team-based approach.
One of the most challenging aspects of multi-manager model was strategy capacity -- just as a chain is only as strong as its weakest link, the multi-manager model was only as scalable as the most constrained algorithm. In signal combination, the individual capacity of an algorithm is no longer a constraint, since we are simply weighting the algorithm as a signal. As a result, we estimate a much larger total capacity, which means significantly more earning potential for authors. We believe in an empirical validation of capacity, so we are systematically increasing exposure and risk over time, observing market impact as we ramp. Since starting to convert to the signal combination approach in August of 2018, we have steadily added exposure/risk to our investment portfolios. We have currently allocated a total of over $200M of post-leverage capital to authors.
Community has always been central to the Quantopian mission and is a huge part of our history. Five years ago, Jess came to a meetup in our first office to hear Wes McKinney talk about his new pandas library. She introduced herself to me and we talked about the Quantopian platform, her background at StarMine and Thomson Reuters, and the future of investing. Since then, Jess has been leading us toward that future, culminating in the launch of our signal combination strategy last year. Jess, Quantopian can’t thank you enough.
Jess has reached her lofty goals for our company, her team, and our investment strategy. Having provided the Quantopian Community with leadership for so many years, Jess has decided to tackle a new challenge. She will always be a friend of Quantopian!
Under Jess’ leadership, we also added a talented quant portfolio manager to our team. In August 2018, David Sargent joined the Quantopian investment team. A seasoned quant portfolio manager and researcher, David is our new MD, Portfolio Management and Research. He has been instrumental in the design and implementation of our signal combination strategy, as well as a mentor to our signed authors.
With our signal combination strategy running and the right internal team in place, we are working on increasing capital allocations and licensing more community algorithms to diversify our signal sources. To that end, our top priority is to provide the community at large with more data.
Several new datasets will be released in the coming months via our relationship with FactSet. We recently released a FactSet earnings estimates dataset to the community. After just 48 hours, the first user submitted an estimates algorithm that met our initial criteria. After a week, we observed 8 algorithms meeting our initial criteria using this dataset!
The speed of the community in analyzing and extracting signals from data is incredible and is reinforcing our urgency to ship more and more data to you. Even more astounding is the ability of the community to produce signals with low correlation to our existing portfolio. We are very happy with the effect of the contest on community research, so we are exploring ways to update the contest to include some measure of an algorithm’s originality or uniqueness. This is an early stage research project and we will be relatively conservative with any changes to the contest. We will, as always, be totally transparent and solicit your feedback before we make any rule changes.
Quantopian is first and foremost a community of people with a shared passion for data, modeling, and investing. The Quantopian community, now a quarter of a million strong, has a reputation of being supportive toward both new members and those who remember using BatchTransforms. Thank you to everyone who has put in the effort and creativity to make it into the contest -- you should be proud of that milestone.
If you’ve never submitted an algorithm to the contest, take a shot!
p.s. The Get Funded page will be updated on Monday 5/20/2019 to reflect our new royalty model.