3 Takeaways from Quantopian Shutting Down
Quantopian announced that it is shutting down its community platform. This doesn’t entirely come as a surprise.
Quantopian returned money to investors earlier this year after its investment strategy underperformed. It shut down paper trading in 2019 (having already ended live trading in 2017), then terminated its daily contests in May of this year. Since that time, Quantopian’s website has billed itself simply as a free educational site where anyone can “become an expert in quant finance,” with no suggestion of a business model other than a sidebar link to Quantopian Enterprise, Quantopian’s commercial offering through FactSet. Meanwhile, numerous employees have left or been laid off in the past year, including many long-time and VP-level positions. The writing has been on the wall.
While some of Quantopian’s competitors, in reacting to the news, have scarcely concealed their schadenfreude, I am genuinely sad to see Quantopian go. Not only did the company create and open-source Zipline, one of QuantRocket’s two backtesters, but they also provided a funnel of new QuantRocket customers who honed their research techniques on Quantopian then came to QuantRocket for live trading or for more control of their trading deployment.
Here are my three takeaways from Quantopian’s announcement.
1. You can’t crowdsource alpha
Quantopian’s founding idea was to apply the concept of crowdsourcing to investment management. Crowdsourcing has yielded impressive results in a variety of domains, but investment management turned out not to be one of them. The demise of Quantopian’s investment fund was foreshadowed by Quantopian’s own researchers in a 2016 paper titled “All That Glitters is Not Gold.” The paper examined a large sample of backtests created by Quantopian users and found that backtest performance metrics offered “little value in predicting out of sample performance.” Quantopian spent the next several years trying unsuccessfully to work around this basic problem.
The fatal flaw in crowdsourced investing, I would argue, stems from the issue of intellectual property. If you want the crowd to feel comfortable sharing trading strategies with you, you must protect their IP. That means you can’t look at the source code, only license the algorithm’s outputs. Quantopian understood this and respected it. The problem is that if you can’t open the black box, you can’t understand the strategy’s rationale, or whether it has a rationale at all. You can only look at performance metrics.
This is a recipe for overfitting. With thousands of users running backtests, some backtests will produce attractive results simply by chance. The best way to distinguish strategies with real edges from strategies that got lucky is to understand what market structure or behavioral bias the strategy is exploiting. You can’t do that with a black box strategy.
This problem will beset any marketplace that aims to match algorithm sellers with algorithm buyers: protect the seller’s IP, and the buyer is buying blind. Give the buyer visibility, and the seller has no protection from IP theft.
The lesson for quantitative traders is: make sure you understand why your strategy works. The difference between research and data mining is that a researcher has an idea to go with the data. Without the idea, you’re just data mining. The scientific method does not consist in blindly running experiments to see what works. It consists in the interplay of theory and experiment, forming a hypothesis then seeing if the data support it.
2. If you depend on software, make sure you’re paying for it
Quantopian community members who used the platform not only for learning or casual research but to inform daily trading decisions now find themselves scrambling for an alternative. Some lament that Quantopian gave no advance notice of the shutdown, and several have drawn comparisons to Google, another company known for unceremoniously pulling the rug out from users by suddenly discontinuing popular products. The comparison is apt: what Quantopian and Google have in common is that their users are not their customers. Google’s customers are advertisers; Google needs users of its products only so that it can sell them as target audiences to those advertisers. Similarly, Quantopian’s customers were its investors; the role of Quantopian users was to provide the investment ideas being sold to those investors. As the saying goes, if you aren’t paying for the product, you are the product.
Everyone loves free software, but think of payment as a kind of glue that binds a company to its users. Purchasing a subscription-based product aligns the company’s interests with yours and gives them an incentive to keep you happy so as to secure your future payments. If you depend on a product, it’s prudent to make sure the product’s creators depend on you.
3. The trading software market favors small companies
Quantopian hasn’t announced what’s next for the company, but presumably they will pivot to Quantopian Enterprise, which is essentially the Quantopian platform sold to institutions with an enterprise price tag and no data holdouts. Such a pivot necessitates shuttering the community site: it’s hard to sell an enterprise product when you’re simultaneously offering it for free.
If Quantopian indeed pivots from managing investments to selling software, they will almost certainly need to become a smaller company. Before this year’s staff departures, Quantopian’s head count stood at around 50. Back of the napkin, that suggests an annual expense line of $5–10 million. Had crowdsourcing proved a good way to manage Steve Cohen’s billions, such an expense line would have been pocket change, and Quantopian’s management team could have collected fat incentive fees.
The economics of selling trading software are very different. The trading software market is quite fragmented, consisting of numerous small companies (typically 1–10 employees) rather than a few dominant players. There is a good reason for this. Traders employ many different strategies. There are numerous markets, data sources, trading styles, and execution frequencies, and there is competition among traders to be unique and find the edges others are missing. Trading is fundamentally exploratory, open-ended, and creative. Trading software is not a fungible commodity. It functions more like a planetary rover, the specific design of which affects what experiments the researcher can perform. No single platform can be optimized for all use cases, and therefore no one software company can monopolize the market. This is a boon to new startups, but it sets a ceiling for potential revenue. Large companies need not apply.
If you’re a Quantopian user looking for a new home, no other platform has better feature parity with Quantopian than QuantRocket. QuantRocket integrates numerous open-source Quantopian libraries including Zipline, Pyfolio, and Alphalens, and includes a research environment, 1-minute US stock data with survivorship bias-free history back to 2007 for all listed US stocks, fundamental data from Sharadar, end-of-day international equities data from EDI, live trading through Interactive Brokers and Alpaca, and support for futures as well as equities. Learn more.
Originally published at https://www.quantrocket.com.
