Real-time news has already made its way into high speed trading strategies and social media is not far behind. But these big data sources need to be handled with care, and focused analytic solutions may be the best way to start.
This was the conclusion from a panel discussion addressing ‘News and Social Media for Trading – Analytics over Latency?’ at the Low Latency Summit in London last week. Moderator Andrew Delaney, editor-in-chief at A-Team Group (host of Low-Latency.com), kicked off with the premise that we will see more pre-trade analytics around news and social media, and the panel appeared to agree, albeit with a number of caveats.
Graeme Burnett, head of infrastrucutre at Azurian Capital Research, was first to voice some dissent, when he said analytics are not finding their feet in high frequency trading as yet and raised concerns about the capabilities of machine learning. On the upside and beyond trading, he pointed to social media potential such as Google-style analytics in banks and the possiblity of emulating Google- and Yahoo-style temporal networks to find skilled professionals.
Defending the role of machine learning in high speed trading, Bram Stalknecht, CEO of SemLab, a specialist in semantic software applications, said: “With machine learning, one system will not fit all, its best use is with a specific ontology. You need to focus, perhaps using an ontology covering how a competitor is doing. This could be a small, fast, low latency ontology.” Burnett endorsed the need for focused ontologies, adding that humans, not algos, should be the masters of data categorisation.
Hugh Taggart, head of sales and business development at RavenPack, a specialist in turning unstructured news into structured data, is equally convinced that there is a place for news and social media analytics in high frequency trading. He explained: “The opportunities in low latency, high frequency trading are getting slimmer and slimmer, so firms are looking for new opportunities. Big data is here, it can be processed faster and that is the way the market will go. The challenge in analytics is choosing the right data source – what is the history of the data, is there enough data to back test and can it be sourced in a structured format? These issues are transient and while this kind of data is very expensive at the moment, the cost will come down. It is also important to have the right people, there are not many data scientists who can join the dots between the market and the data.”
Moving along the spectrum from news to social media, data becomes more difficult to access and its uses are, for the time being, more limited. Cynan Rhodes, co-founder of FSWire, the provider of a social market data feed that filters Twitter to deliver market data by asset class to financial markets, said: “Access to Twitter data is still very difficult and very expensive. In internet circles, Twitter is seen as real time, but it is not low latency as tweets queue to be delivered to followers and this can take up to five seconds. The issue is how to get the data in the fastest possible way. There are three Twitter data resellers, but they add latency, about a one or two second delay. The fastest way is to take the data direct from Twitter’s application programming interfaces, but there are limits on the amount of data you can take. This means low latency services using Twitter data are not possible at the moment, but with an IPO of Twitter coming soon a product for finance can be expected.”
While these developments suggest more news and social media data will be made available to high speed trading, Delaney questioned exactly how the data will help. Professor Gautam Mitra, managing director of OptiRisk Systems, a provider of optimisation and risk analysis applications including financial news analytical services, explained: “Social media is just coming in line with other data. We are working with FSWire to integrate Twitter data with our news feeds. One of the issues with social data is trust. There are thousands of newswires that are edited and their content is trusted, but can you trust Twitter data? There need to be approval models and firms need to consider to what degree they trust the data. The impact of social media micro blogs, such as Twitter, and news flows will be in their integration with market data.”
On the point of trust, Rhodes pointed out, by way of example, that not many people make comments on Twitter, Facebook and LinkedIn on bonds, giving data that is posted on the subject a high degree of trust. Looking at broader potential for Twitter data, he said: “There are many data sources related to US equities, so new information from Twitter doesn’t add much value. Twitter data can be more useful in emerging markets, where there are no big news agencies on the ground, it is hard to reach people and news travels more slowly. Here, Twitter can cut the time lag of data delivery. Its biggest successes to date are in emerging markets, fixed income and foreign exchange.”
Turning to trading strategies in a low latency environment that can benefit from pre-trade analysis of news and social media, Taggart said: “The strategies our clients are finding successful are momentum-based strategies that look at aggregated sentiment data. They are also running volatility strategies and mean reversion strategies. The most difficult trading strategies to run are event-based strategies as they depend on very good data sources and these are hard to find. RavenPack can provide news analysis, but it is hard to source social media date. There is not a lot in the market at the moment, but that is improving. Mostly, our clients use event-based strategies to protect existing positions and for market making.” Burnett added: “There could be a no-news strategy that looks for companies with no news. Companies like this are often strong performers.”
As news and social media become more prevalent in trading, they are bound to pose problems. Delaney asked the panel how disparate data sources, updated at different frequencies, could be pulled together. Mitra responded: “There is no silver bullet. Data coming in at different frequencies needs to be collected and managed depending on the trading strategy. Research into big data is largely about structured data, but a second wave is looking at streaming data. Predictive analysis tools will be developed to mange data that comes in at different frequencies.”
In terms of advice for those adopting news and social media data and analytics into high speed trading, Taggart commented: “You need to be very clear about the data you are working with.” Stalknecht added: “You need to think about the trading strategy and data sources. Essentially, you need a target, a team, a strategy, data sources, and the ability to back test.” Summing up the views and experience of the panel members, Rhodes concluded: “Whatever your opinion of social data, it is important and moving into trading.”