Bridging tech and quant

As CIO and Chief Strategist at Tradeworx, Mahjouri led a team that fostered synergies between tech and quant: “Whereas other firms siloed technology teams from trading teams, we put some of the most talented techies and quants in the world together to solve problems. This was a powerful source of edge, which led to efficient systems trading on microstructure phenomena such as HFT. Our team is united in our vision of implementing and evolving alpha-generating strategies and systems amid rapid alpha decay”.

Designing datasets and systems

Tradeworx’s novel approach to alpha generation created a hybrid between statistical arbitrage (mid-frequency trading, or “MFT”) and HFT, intended to transform liquidity provision and institutional inefficiencies. This entailed a gargantuan appetite for data.

In 2017, a Rice University study ranked Tradeworx as one of the largest users of SEC data, on a list where most other firms ran up to tens of billions in assets under management. In 2013, Tradeworx had been retained to build the SEC MIDAS (Market Information and Data Analysis System). The Tradeworx dataset and systems are part of the intellectual property that Blueshift acquired when it spun out in 2018. “Our 20 years of data is a source of advantage, not only in itself but also because we built all of the HFT systems and wrote nearly every line of code. We have architected the system, so we know how long the data takes to affect stock prices and we have continuous snapshots of exchange latency, which can precisely measure all variables affecting the system and how alpha drives prices. The system has been battle tested through all the market turbulence of the past decade and has seamlessly transitioned from Tradeworx to Blueshift. We did not have to unplug a single cable and we transitioned with a trading hiatus of less than two weeks,” recalls Mahjouri.

Quantamental investing 

The data itself has evolved rapidly: “The Rice University study was really surprising at the time, but it is not actually a huge amount of data by today’s standards. We have grown our ability to quantitatively understand the data and drive alpha through models,” explains Mahjouri. Blueshift defines “quantamental” as using both systematic, quant-driven approaches and fundamental data such as news and events. “The juxtaposition of these is the modern advent of quant, which marks the start of a massive paradigm shift. We are viewing markets with atomic resolution and codifying how discretionary analysts think, with big and new data. The three key trends are an explosion in computer power, data and computational complexity. There are more phenomena to study and more questions to ask, and we can ask more nuanced questions in a non-linear world. All of this plays to our advantages and lets us systematize inputs,” says Mahjouri.

For example, alternative data has further expanded the opportunity set in areas including higher frequency forecasting or “nowcasting”. “Alternative data has become bigger and more granular than ever with billions of records being captured in the cloud, where we were one of the first financial firms to use Amazon’s cloud offering. We gather traditional data as well as more opaque and subtle information. We can use shipping, credit card receipts or fill rates in shopping mall parking lots to estimate company sales and consumer spending before the trends become understood more broadly,” says Mahjouri. Mahjouri has authored a book on alternative data, to be released shortly.

A pure play asset manager

Blueshift synthesizes years of development at Tradeworx – and also moved from a trading and technology business model to a pure play asset manager. (The technology consulting arm of Tradeworx was renamed Thesys Technologies and is independent of Blueshift.) “We wanted to spin out the trading unit because the team wanted to focus purely on institutional asset management. Though there were synergies between technology consultancy and trading, we wanted to function as an independent trading business. We also wanted to ensure that we could recycle revenue into the business and acquire talent, rather than sharing profits with a parent company. Most traders are entrepreneurial and Blueshift is majority owned by the staff. An independent company with clear corporate governance is also more attractive to investors,” explains Mahjouri.

Anchor investors included pension funds and endowments, and no seed economics were given up. “We spoke to seeding platforms but determined that White Oak was the best fit for attracting stable institutional capital. Seasoned seeder White Oak did not actually provide seed capital but did provide working capital that allowed Blueshift to acquire intellectual property from Tradeworx. White Oak has also provided invaluable know-how and consulting, to help us build a world class business that meets investor requirements on cash controls, collateral and so on,” says Mahjouri. Blueshift has a substantial team of 23, including 11 PhDs, and has been profitable from the start based only on management fee income. 

Fusing HFT and MFT

Mahjouri has been close to the coalface of HFT since Tradeworx began it in 2008 and has watched the acceleration of timeframes. “The HFT process of dispersing inefficiencies by transferring excess demand, supply, risk and liquidity to other correlated names might have once taken place in pits or trading floors by specialists managing supply or demand, over days or weeks. It now happens much faster on an intraday basis and technology lets you identify dislocations at more precise levels.” The arms race towards zero latency led to huge investments based on the laws and metrics of physics, of which Mahjouri had first-hand experience: “Traders dug through mountains to cut millionths of a second off the time between NYSE and CME and spent half a billion dollars laying submarine cables under the Atlantic Ocean. Because fiber optic cables made of glass are 30pc slower than air, there was a race for microwave tower bandwidth to send signals back and forth,” explains Mahjouri.

Yet Blueshift is not competing in this race – partly because it is a precarious game: “Being the fastest trader is a tough seat and not necessarily a stable one because you cannot control the next innovation. We have colocation and microwave towers, but we never depended on being the absolute fastest. We survived the HFT arms race without spending the most – through innovation and scrappiness”. 

Indeed, Mahjouri sees more compelling opportunities in the mid frequency space: “The ultra-low latency game is about risk transfer at very short time horizons for very high Sharpe ratios. It is very capital intensive on infrastructure, and limits firms to intraday time horizons because they cannot get leverage overnight and have abnormally high risk aversion, especially overnight, which limits the amount of liquidity at that time horizon. That also leaves a lot of alpha on the table to apply the same liquidity provision techniques to longer, multi-day holding periods. This is an underserved area of liquidity that used to be handled by investment banks. If we buy on the bid our goal is to sell on the offer, but we have to await liquidity to do this,” says Mahjouri. 

HFT is used for both alpha signal generation and trade execution, which is monitored closely. “We obsessively keep close track of execution efficiency and nearly always beat the VWAP by capturing bid-offer spread.”

Multiple uncorrelated models

HFT and MFT are fused in the value and liquidity provision strategy, which uses statistical and information arbitrage for factors and behaviours. “We think it is pretty unusual to fuse HFT and MFT. We can axe our bid or ask in ways that do not make sense in a two-sided market, because the MFT motivates that axe, based on our history of over a decade’s worth of the most detailed level of market microstructure data,” explains Mahjouri. 

The other models are information arbitrage, quantamental anticipation (using alternative data), and exotic book alpha based on flows and volumes. 

Blueshift uses machine learning to develop and refine models but follows a supervised hypothesis-based approach rather than a “letting the data speak” approach: “We are prior-driven and use discretion to formulate a hypothesis. Within that context, we do also use machine leaning to formulate how the data impacts stock prices”.

Correlations amongst the strategies are near zero, and all have made positive contributions to performance over time: “Whatever the market regime, there is always some dynamic at work”. Risk budget allocations amongst the five strategies are roughly equivalent ex ante, though there is some flexibility as opportunity sets ebb and flow: “For instance, the value and liquidity provision strategy can automatically scale up to take advantage of higher bid/offer spreads when an event such as Covid or GameStop hits the market and increases trading volumes and demand for liquidity. Conversely, the Covid crash made some fundamental data such as historical 10Ks less relevant, and those signals scaled down,” says Mahjouri.

The models have stayed robust and survived extraordinary market conditions including deleveraging events, and unexpected exogenous events that have wrong-footed some quantitative strategies. Most of the other strategies traded by Blueshift were also traded at Tradeworx, though they have been adapted to multi-day holding periods. Traditional quant factor approaches performed poorly in the past few years, and especially in 2020 partly because momentum became overcrowded and this was amplified by the split between Covid winners and losers, which led to extreme divergence. 9th November 2020 saw a huge momentum crash as Covid names rallied, and this dislocation set up a massive opportunity set for some of Blueshift’s models. Leverage can go up to 6 times to take advantage of opportunities. 

The strategies can also be adapted to long-only mandates, and Blueshift’s NOVA platform can offer joint ventures to asset owners that could involve sharing technology, data, risk and research. “We feel confident about sharing this and offering customization without giving away intellectual property, because we are sharing aspects of expertise that would anyway be difficult to generate in-house. We can work with some of the largest and most sophisticated institutional investors to use our tools, technology, and platform to help them solve their complex investment needs, such as a desire to improve their analytics, execution, or alphas,” says Mahjouri.