NN IP Gains From Sentiment Analysis
Mark Robertson, senior portfolio manager, multi-asset portfolios at Dutch fund manager NN IP, pointed to outperformance in financial and utilities last year as evidence of the benefits of integrating digital news and sentiment analysis into its investment process.
The Dutch fund manager bases its top-down multi-asset strategy on a combination of fundamental and behavioural analysis, through incorporating investor emotion and behaviour from digital news channels such as financially oriented news platforms, blogs and social media. A proprietary toolkit provides data for assessing markets which strategists and portfolio managers can use with their own research to form views.
Robertson told Markets Media that the firm began thinking about whether social media, artificial intelligence and big data had a place in its investment process in 2009. NN Investment Partners, formerly known as ING Investment Management, is part of NN Group a publicly traded company listed on Euronext Amsterdam. NN IP managed nearly €200bn ($224bn) in assets at the end of September last year for institutional and retail investors.
NN IP partnered with MarketPsych, an independent consultancy that develops financial applications from behavioral economics, and began testing the use of sentiment analysis in its investment process in 2013. Robertson said: ‘We found it had predictive value.”
He continued that NN IP selected MarketPsych because of the richness of the data it provides. “It does more than just give positive or negative signals. Rather than just saying whether an earnings report met analyst expectations for example, the data analyses the proximity of the word “higher” to “earnings” to indicate sentiment,” said Robertson.
In 2011 the Thomson Reuters MarketPsych Indices were launched for countries, currencies, commodities and industries through analyzing news and social media in real-time. MarketPsych built advanced statistical language processing software to quantify sentiments – such as fear, joy, uncertainty, trust, and urgency – which are known to predict risk taking. The system’s logic takes into account all the specific mentions in various news and social media posts and aggregates that data into the specific indices.
In 2015 Thomson Reuters extended the analytics offered through its MarketPsych Indices to include individual companies. Thomson Reuters said the indices draw from a pool of 40,000 primary global news sources and 7,000 social media sites from 1998 to the present.
In a blog last year Robertson described a meeting Dr Richard Peterson, the founder of MarketPsych, during a visit to San Francisco. “The innovative use of algorithms that can read news articles alongside impressive computing power enables market sentiment to be analysed, and quantified, at a much deeper level than had previously been available,” wrote Robertson. “Indices are constructed from thousands of web sources, many within milliseconds of publication, across both professional and social media domains. The self-learning algorithms developed by Dr Peterson and his team can classify this vast amount of data into sentiment indices such as optimism, fear, joy or conflict.”
NN IP uses the sentiment data in different ways in its investment process. For example, analysis of emotions including conflict, violence and government trust are grouped together to give a measure of political risk.
“The use of human judgement, which is still better able to weigh the intricacies of geopolitical events, shifting political alliances, changes in corporate governance and central bank policy, alongside the use of data-derived models that protect against known human biases, creates a system that is well placed to deal with the complex, ever-changing world in which we invest,” wrote Robertson.
He told Markets Media that in equities, for example, NN IP uses a scorecard of about 30 independent signals in the investment process and five or six will come from sentiment analysis.”We have found that sentiment data makes a meaningful contribution of between 10% to 15% in our investment scorecards,” he added.
He gave the example of the example of gains in financials that were partly driven by sentiment data last year.
“Last June and July financials were unloved,” Robertson said. “Sentiment picked up and by September, when incorporated with financial analysis, there was a strong signal to buy. By the end of 2016 we had 10% outperformance over the benchmark [financials sector relative to the World BM].”
Sentiment analysis also led to NN IP downgrading utilities in September, and by the end of 2016 the sector had underperformed the market by 10%. Robertson said: “We had a 20% outperformance driven by sentiment data.”
In equities NN IP uses sentiment analysis across countries, regions and individual sectors and Robertson said NN IP would like to extend the process to individual stocks but has no current plans. In credit sentiment analysis is being used in emerging markets, high-yield and investment grade.
“We want to develop the commodities scorecard,” said Robertson. “There is a long way to go with the process and there is a willingness to explore new unique strategies using the data.”
Technology has enhanced capabilities of surveilling larger and more disparate data sets.
With Eugene Kanevsky, James Redbourn, and Joanna Wong, CLSA
AI/ML on the buy-side trading desk is a long-term program with short- and mid-term deliverables.
Aiden uses the computational power of deep reinforcement learning to improve trading results and insights.
BondDroid’s AI-generated prices are integrated directly into LTX’s pre-trade analytical tools.