The quantitative finance community has long relied on news signals and is relooking sentiment analysis given the attractions of alternative data and the power of machine learning. The aim is alpha: salient and timely data that can drive trading, fund management, and risk strategies.
In our data economy, we derive alpha from combinations of traditional transaction-derived data and data from online, social, and media sources, made accessible via machine learning and, when the source is text or speech, via natural language processing (NLP).
Sentiment signals are particularly compelling: We recognize that emotion drives markets. Yet sentiment inferred only from market activity and price movement – from traditional data – has limited utility, and news-derived sentiment, from another traditional source, provides a very incomplete picture. We need to go broader and deeper, hence the look to alternative sources – to social media in particular – and investments in machine learning that aim to discover fine-grained sentiment – linked to entities, topics, and events – that is more clearly predictive of market and risk outcomes.
Let’s break out three interlinked “alternative” elements of concern to finance professionals (and equally to those involved in consumer and business-to-business markets): Data, Models, Outcomes.
On the data front, salience – suitability for a given task – is key. We don’t want big data, we want the right data.
Consider Pierce Crosby’s definition: “Alternative data in asset management is anything out there that doesn’t qualify as traditional, examples being satellite, social, and ledger.” According to Pierce, who is Business Director and Data Evangelist at StockTwits, “big data doesn’t usually apply to alpha generation or risk metrics, whereas the specific use case of most alternative data is to generate an actionable signal to be used throughout the investment process or business process. Alternative data is often big, but big data is not often alternative.”
Pierce will moderate a panel, Alternative Data: Signal and Noise, at the up-coming Sentiment Analysis Symposium, June 27-28 in New York. The panel, part of a June 28 symposium workshop on NLP and Sentiment in Finance, will focus on completing two tasks, according to Pierce: “1. Laying out the State of the Market for asset management and alternative data and 2. Explaining some of the most common use cases for alternative data.” Participating are Noam Tasch from iSentium, Jamie Wise from BUZZ Indexes, and “Wall Street’s Top Psychiatrist,” Richard Peterson from MarketPsych, who will also present, on Trading on Media Emotion. “We will also discuss successes and failures we have endured or have seen companies endure, when developing alternative data businesses,” reports Pierce.
Element #2, models, will be covered by Finance workshop speakers Evan Schnidman from Prattle, speaking alongside Basis Technology President Steve Cohen, Brendan Herger from Capital One, and Marcus Hassler from econob, a Vienna, Austria technology company.
Outcomes, from a fund manager’s perspective, will be subject matter for the workshop’s Trading Reality panel, organized and moderated by Bartt Kellermann, Battle of the Quants with participation by Irene Aldridge from AbleMarkets.com and Kevin Shea from Disciplined Alpha.
A morning symposium workshop on Data Science and Technology – that is, Practical AI – will also interest quants and investors.
Other parts of of the symposium program feature consumer-markets focused content. Speakers and panelists represent agencies Kantar TNS, Ipsos, Mediabrands, Nielsen, and FleishmanHillard; YouTube, Verizon, Uber, U.S. Bank, IBM, and Airbus; and established and startup tech providers. Sentiment Analysis Symposium registration is online.