Far from the Manhattan crowd 2

Anthony Pratt |
6 min read
Download Article

Picture this: A thronging crowd slowly making its way down Fifth Avenue towards Central Park. The passing public looks on, confused – somehow the sight doesn’t quite add up. It seems too smart, too polite, for a protest march, and lacks the energy of other marches. The protestors are dressed in a mix of office wear, the casual preppy look of Brooks Brothers, and the occasional padded gilet adorned with a company logo. Good, smart clothes, but maybe not this year’s fashions. The placards read: “Minimum pay for hedgies”, “Hedged off!” Suddenly there’s a loud smash as the window of a high-end car dealership is broken. Still this does not notably enliven the crowd, the mood remaining subdued and disconsolate. Some mumblings are heard – “I nearly bought one of them once” – “My boss has two. Well, my old boss anyway” – as the protestors continue on their way.

In part one of a two part paper, I discussed crowdsourced trading signal platforms that allow funds to access the quantitative modelling capabilities of contributors who may live far from major financial centres.

Part two discusses the strategies firms have used to engage the online crowd, the nuances of how these platforms are built, and the effects this could have in the future. Could the job markets in London and New York begin to lose out to online outsourced workers?

How do you create a new crowd?

Once a web-based coding platform has become active and established, the vibrancy of the community helps attract other people to the community. But how does a fund or web business initially attract people to its platform? How do you start a new online crowd? An obvious route is through a strong online presence. Managers connecting to this space seem to be more aware of the need to be seen as thought leaders and content creators than other hedge fund managers, and to better understand the importance of managing their social media message.

Another way to initiate a crowd is to target specific universities. Online targeting of students is not a new concept for quantitative hedge fund managers, some of which have historically advertised online financial competitions in the careers departments of specific universities that they are intending to target for graduate-level recruitment. These universities are usually relatively local to the London or New York bases of the hiring hedge funds. However, the universities being targeted for a crowdsourced platform encompass a much wider geography; we have even spoken to one group which is considering launching its crowdsourcing initiative by targeting Pakistani universities that it feels have been disregarded by some of the more established crowdsourced quant businesses.

However, not all campaigns focus on university students. Numerai, a platform that focuses specifically on the data science/machine learning community, has used a variety of original ways to build its profile and number of users. These have included press targeting of the tech community through Wired, a magazine/website focusing on emerging technologies, and, a leading blog host; viral marketing campaigns within the data science community; and marketing itself to the top tiers of Kaggle users. Kaggle is an online platform for data scientists that runs a range of prediction competitions. It is no stranger to hedge funds, its competition leader board having been a fertile source for quant fund hiring for some time. Indeed, some successful hedge funds have been formed after leading predictive modellers have been introduced to finance through such predictive modelling competitions, and Two Sigma Investments LP, one of the largest quantitative hedge fund businesses, has partnered with Kaggle since 2016 to run a range of financial data-focused predictive competitions.[1] Numerai, however, took its marketing a step further, by giving away one million dollars of its own cryptocurrency token to Kaggle users who signed up to its platform[2], which led to a number of attempts by hackers to hack Kaggle accounts to try and steal the token[3].

How should a fund structure its online portal and what should it be trying to achieve?

A crowd typically interacts with a firm through the firm’s web portal, so it is extremely important that this is structured such that the firm retains its access to this global talent pool. What may seem like subtle differences in how the portal is structured, and what exactly a coder is allowed to do within the portal, could lead to the crowd leaving and practicing its quantitative finance skills on other platforms. Differences can come in various forms: How often are data sets updated? Which data sets are available? What coding languages can be used? How many parameters can be changed?

As well as trying to keep the crowd engaged, managers need to ensure that the output from the crowd has not been over fit. “Overfitting is a modelling error which occurs when a function is too closely fit to a limited set of data points. Overfitting the model generally takes the form of making an overly complex model to explain idiosyncrasies in the data under study. In reality, the data studied often has some degree of error or random noise within it. Thus attempting to make the model conform too closely to slightly inaccurate data can infect the model with substantial errors and reduce its predictive power.”[4] The platform has to ask itself: How much freedom should a modeller be given? Should there be a tight range of parameters, giving more control around the output received, or more freedom, which may allow the most innovative modellers to better express themselves? How can the platform guard against overfitting of models without making the portal frustrating to use? Should datasets be downloadable, reducing the manager’s ability to control whether sound data handling practices are being followed with regard to overfitting, but potentially reducing the modeller’s ability to use some of the more system intensive statistical techniques.

A manager must also ensure that the output from the crowdsourced platform fits into the process already established at the fund. Integral to this is what the crowd is being asked to do: Is it being asked a specific or a general question? Is it being asked to find signals that explain the performance of a stock, a sector, or the market in some way, or to build a whole trading model that wouldn’t necessarily have to be combined with other signals? We have heard of a situation where a major quant hedge fund manager set up a crowdsourced platform, but the question the crowd was asked was different to that which the researchers at the manager were hoping to answer. The output was never used and eventually the platform was scrapped. This is clearly an area where subtle differences in how something is presented or posed may lead to notable differences between success and failure.

How should a fund structure its online portal and what should it be trying to achieve?

Working out how to generate online buzz and build an effective simulation platform and managing money effectively are clearly very different skills. So far this space appears to have seen more success from established hedge fund businesses that have changed their branding strategy and begun to engage the online crowd, than from firms that have launched as crowd-focused businesses and moved into money management. However, those businesses that have successfully engaged a crowd have managed to capture a potential research source that could prove to be genuinely alpha generative if used correctly.

A quantitative hedge fund manager typically has several different functions, including data handling, signal research, portfolio optimisation, risk management, and execution. Ultimately the crowdsourced platform is just one part of the signal research process. It is likely that failures in the other functions have held back some of the new crowdsourced hedge funds thus far, though that is not to say that, with time and experience, they won’t improve and become more competitive.

There is potential for the development of these platforms to lead to commoditisation, or outsourcing, of quantitative research. However, it is clear that, even if an entire alpha research group were to be sourced online, there would still be a range of more sensitive tasks, in particular portfolio optimisation, risk management and execution, that would remain in the firm’s main offices, probably in major financial centres.

While there does not, as yet, appear to be any need for quantitative researchers to be worried about their jobs being outsourced to online workers, it is interesting to note how quickly an idea that may seem odd initially can become the norm in time. For example, alpha capture, “a computer system that enables investment banks and other organisations to submit ‘trading ideas’ or ‘trade ideas’ to clients in a written electronic format”[5], which was seen as unique when Marshall Wace LLP began to do it in 2001[5], is now an established part of a sell-side equity salesperson’s day-to-day process. Could the same thing happen with regard to crowdsourced quant research signals?

Overall, the success of the online crowd is still up for debate but it is clear that this is a way of hiring and building a research function that could become more prevalent. It is a research route that is changing the way that some hedge funds look at PR, branding, and their global reach, but, for the moment at least, it isn’t notably threatening the control of the major global financial centres.







You may also like

Monthly hedge fund performance review – April 2023


Hedge fund performance was generally positive in April; the average hedge fund net return across all strategies was 0.51%. April was a month of lower volatility…

Hedge fund industry performance deep dive – Q1 2023


In summary… Global growth surprised markets positively in Q1 2023. Global equities*** and bonds** returned 6.35% and 3.19% respectively. Strategies…

Aurum’s quarterly review – Q1 2023


Performance for Aurum’s commingled fund of hedge funds’ $US classes ranged from marginally negative to -2.4% for Q1. While Aurum’s portfolios delivered…

Monthly hedge fund performance review – March 2023


Hedge fund performance was mixed in March, the average hedge fund net return across all strategies was -0.39%. Strategies exhibiting a higher beta to equities…

A guide to hedge fund fees and redemption terms


In summary Investors determine whether hedge fund fees and redemption terms are appropriate through the manager selection process. Consideration is given…

Monthly hedge fund performance review – February 2023


Hedge fund performance was mixed in February, the average hedge fund net return across all strategies was -0.19%. Strategies with a higher beta to equities…

Monthly hedge fund performance review – January 2023


Hedge fund performance was generally positive in January, the average hedge fund net return across all strategies was 1.83%. Strategies with a higher beta…

Hedge fund industry performance deep dive – Full year 2022


The hedge fund industry was down 2.4% in 2022. Global equities** and global bonds*** have fallen 20.0% and 16.7% respectively. It has been an extremely…

Aurum’s quarterly review – Q4 2022


All of Aurum’s managed funds and bespoke accounts delivered positive returns in the fourth quarter of 2022. Performance for Aurum’s commingled fund…

Monthly hedge fund performance review – December 2022


Hedge fund performance was generally positive in December, the average hedge fund net return across all strategies was 0.50%. Strategies with a lower beta…

Monthly hedge fund performance review – November 2022


Hedge fund performance was generally positive in November, although underlying strategies had mixed performance. Long biased was the top performing master…

Arbitrage deep dive – Oct 22


The period under review includes ‘the worst cross asset selloff since 1981’ with simultaneous declines across equities, commodities, corporate and…