By A Paris – Alternative data and data science techniques can help give hedge funds a competitive edge but, it is the symbiotic integration of human and machine which ultimately underpins managers’ success or failure in their use of these techniques.
“We believe we understand these manager datasets better than anybody else and because of this we’re able to come up with factors and signals that nobody else would be able to identify; or rather it would considerably difficult for them to do so,” asserts Michael Perlow, co-founder of Epsilon Asset Management.
Epsilon employs a bottom-up investment process with a systematic approach which includes data science techniques to establish rankings of domestic equity securities, similar to factor investing.
Perlow explains that familiarity and expertise in particular datasets can drive the value a manager can extract from these alternative sources of information: “A quant investor who’s familiar with the data has a massive advantage because they will know how to qualitatively treat that data. So, the data science always needs to be married with that qualitative, fundamental view.”
At Lombard Odier Investment Managers, data science sits at the heart of the first stages of the investment process. Christophe Khaw, Chief Investment Strategist for LOIM’s 1798 Alternatives business says: “We use data science to drive our idea generation. Many managers who claim to use it still maintain the same legacy process but refer to a dataset late in the process to help guide timing and sizing. We are the opposite. We think removing human bias, group think and past experiences from idea generation helps us identify often contrarian ideas and avoid crowding. This has driven our lack of correlation to markets and equity peers.”
Bryan Cross, Head of Quantitative Evidence and Data Science (QED) at UBS Asset Management shares a similar view: “Using data and science to derive insights helps to minimise any bias that can creep into the analysis process and ultimately end up with a more robust thesis around an investment.”
However, although data science is the lynchpin of its idea generation process, Khaw highlights the importance of the balance between man and machine: “We take a different man + machine approach, where broadly speaking we use AI to drive idea generation, while the portfolio manager evaluates those signals relative to the current environment, times and applies those signals based on 20 years of investment experience.
“It is often assumed that quants will dominate here, but this is just not true. Machines and AI are, for now, statistical engines. They can process more volumes of data but they do not have the capability to apply the logic necessary to extract value from data that does not have a statistical trend. Alternative data is very messy and AI cannot evaluate the environment, adapt and respond like a human can. People need perspective. We are 12 to 17 years away from AI replacing a salesperson, 30 years from replacing a surgeon, and 35 years from replacing maths research.”
Overload, noise and commoditisation
Chris Longworth, a Senior Scientist at GAM Systematic Cambridge talks through some of the challenges of using alternative data: “Data analysis can be harder when working with alternative datasets. The more data you have to explore, the more likely it is that you will encounter spurious relationships in the data. Chance patterns can give the appearance of tradable effects that don’t exist in reality.
“Ensuring that you have the statistical analysis framework required to distinguish a genuine signal from the noise is critically important. Many emerging datasets also have a limited amount of history available which makes it harder to establish whether any relationships that your model identifies will persist across different market environments.”
Andrea Leccese President & Portfolio Manager at Bluesky Capital also emphasises the importance of having a robust technology framework to underpin any alternative data or data science efforts: “ Unless you have very good technology and talented quantitative people, especially in computer science and statistics, to support these techniques, you’re not going to get anything from it. Usually these datasets have millions or billions of rows so you need to be very proficient with coding and machine learning to apply techniques that can retrieve alpha from very big datasets.”
Further, as more managers turn to alternative data to carve out performance in challenging markets, the risk of the data becoming commoditised is also very real.
Robert Kosowski, Head of Quantitative Research at Unigestion comments: “We’ll be using a dataset until it becomes commoditised. When that happens, we’ll have to look for a new data set. I think this is now becoming a continuous process within our work.”
Leccese at Bluesky Capital muses: “After a certain number of years these datasets become publicly available and well known. Despite this there are still some opportunities to monetise on them, particularly because some of them are very expensive, and only the biggest hedge funds with high budgets can afford to gain access to them.”
Also, if many investors see the opportunity and everybody starts trading on that view, the opportunity can disappear because the price becomes efficient. Therefore, if managers want to be competitive in the field, they need to have highly skilled data scientists or analysts who can keep researching new data and to find creative ways of retrieving data.
Perlow at Epsilon discusses the long-term impact of data science techniques being used more broadly: “I think it will result in higher efficiency in the marketplace. Ideally, you’ll have less idiosyncratic volatility and capital running to companies that are more deserving of that capital. But it will always be married with the fundamental. Leaning one way, either fundamental or quantitative too heavily will always be more vulnerable than marrying the two in a happy medium.”
Kosowski warns that alternative data is not a salve for an ailing hedge fund: “If you don’t have a person that’s in place to determine whether the traditional data you’re using is adding value or is significant, then there’s no point turning to alternative data.”
Some managers may be attracted by alternative data during a time when traditional signals are not working well. But, to echo Longsworth’s comments, without the right framework and expertise in place, managers can run the risk of spending large amounts of money on data which they are not able to fully utilise.
However, just because many have access to the same data does not necessarily mean that information is no longer valuable. Khaw at LOIM highlights: “We still continue to see massive value in things like credit card data. While there is perception in the market that everyone has access to this and it is not proprietary, I think this is often misunderstood. Execution is very important in this field.
“All fundamental managers have had access to Bloomberg for decades, but there is still huge performance dispersion, managers get different reads on company 10ks, management team tone, etc. Alternative data is no different. Based on how creative you get in massaging the data and applying it on a case by case basis, you can get excellent colour beyond just revenues, margins and promotional intensity. You also need to understand nuances in the data between demographics, regions, biases of each different vendor. You can get much more sophisticated than many understand. This is just a single example which scratches the surface of one type of dataset.”
Capturing relationships
One thing that is certain is the data science and analytics of alternative data are here to stay. A study by UBS and Element 22, in association with Greenwich Associates carried out in 2019 established that more than two thirds of asset management respondents already use advanced analytics and alternative data in their research, portfolio construction and portfolio management functions. Further, 70 percent expect significant growth in the use of alternative data over the next three years.
Longworth at GAM Systematic Cambridge says: “Compared to traditional fundamental datasets, alternative data is an opportunity to provide our investment models with richer, more detailed information about what is actually happening out there in the world. This can allow us to capture relationships that may not be apparent in traditional datasets or might be too weak to identify.”
Cross at UBS AM believes this is a trend that will continue to progress: “More science in investment management is an irreversible trend. Science, data and technology are the cornerstones of how the hedge fund industry is going to scale and will be key to allowing hedge funds to do what they do best, which is generate alpha.” n
