News Analytics and Sentiment Analysis

The field of natural language processing (NLP) has evolved significantly in recent years. In this chapter we consider two leading and well-established methodologies, namely, those due to Loughran McDonald, and FinBERT. We then contrast our approach to these two approaches and compare our performance against these methods which are considered to be benchmarks. We use S&P 500 market data for our investigations and describe the results obtained following our strategies. Our main consideration is the Earnings Calls for the S&P 500 stocks. We vindicate our findings and present the performance of our trading and fund management strategy which shows better results.

Keywords— Natural Language Processing, Sentiment Analysis, Earnings Calls

The explosive development of electronic media has brought to the market participants thousands of pieces of financial news which are released on different platforms every day. Many news wires published online are editorially controlled and can be relied as factual summary as opposed to fake news or disinformation. These news
items provide a rich source of textual information which in a summative way represents the sentiment of the market. The sentiments influence or impact the asset price as well as the volatility of individual assets. In this study we have tested sentiment enhanced daily trading strategies. Alexandria Technology has provided us news sentiment metadata, which is used in this study. We have also resorted to ‘asset filters’ which we use to restrict the universe of assets chosen for daily trades. We have considered quantified news sentiment and its impact on the movement of asset prices as a second time series data, which is used together with the asset price/return time series data. Our asset allocation strategy uses Second Order Stochastic Dominance (SSD); see Roman et al. (2006, 2013). Following this modelling paradigm we compute daily trade schedules using a time series of historical equity price data. In contrast to classical mean-variance method this approach improves the tail risk as well as the upside of the return. In our recent research we have introduced news sentiment indicators such as News RSI (NRSI) and Derived RSI (DRSI) filters. These filters restrict the choice of asset universe for trading. Consistent performance improvement achieved in back-testing vindicates our approach.

Keywords— Trading Strategy, Sentiment Analysis, News Meta Data, Asset Filter

We investigate how “news sentiment” in general and the “impact of news” in particular can be utilised in designing equity trading strategies. News is an event that moves the market in a small way or a big way. We have introduced a derived measure of news impact score which takes into consideration news flow and decay of sentiment. Since asset behaviour is characterised by return, volatility and liquidity we first consider a predictive analytic model in which market data and impact scores are the inputs and also the independent variables of the model. We finally describe the trading strategies which take into consideration the three important characteristics of an asset, namely, return, volatility and liquidity. The minute-bar market data as well as intraday news sentiment metadata have been provided by Thomson Reuters.

The explosive development of electronic media has brought to the market participants thousands of pieces of financial news which are released on different platforms every day. Many news wires published online are editorially controlled and can be relied as factual summary as opposed to fake news or disinformation. These news items provide a rich source of textual information which in a summative way represents the sentiment of the market. The sentiments influence or impact the asset price as well as the volatility of individual assets. In this study we have tested sentiment enhanced daily trading strategies. Alexandria Technology has provided us news sentiment metadata, which is used in this study. We have also resorted to ‘asset filters’ which we use to restrict the universe of assets chosen for daily trades. We have considered quantified news sentiment and its impact on the movement of asset prices as a second time series data, which is used together with the asset price/return time series data. Our asset allocation strategy uses Second Order Stochastic Dominance (SSD); see Roman et al. (2006, 2013). Following this modelling paradigm we compute daily trade schedules using a time series of historical equity price data. In contrast to classical mean-variance method this approach improves the tail risk as well as the upside of the return. In our recent research we have introduced news sentiment indicators such as News RSI (NRSI) and Derived RSI (DRSI) filters. These filters restrict the choice of asset universe for trading. Consistent performance improvement achieved in back-testing vindicates our approach.

Key Words: Trading Strategy, Sentiment Analysis, News Meta Data, Asset Filter

We propose a method of incorporating macroeconomic news into a predictive model for forecasting prices of crude oil futures contracts. Since these futures contracts are more liquid than the underlying commodity itself, accurate forecasting of their prices is of great value to multiple categories of market participants. We utilize the Kalman filtering framework for forecasting arbitrage-free (futures) prices, and assume that the volatility of oil (futures) price is influenced by macroeconomic news. The impact of quantified news sentiment on the price volatility is modelled through a parametrized, non-linear functional map. This approach is motivated by the successful use of a similar model structure in our earlier work, for predicting individual stock volatility using stock-specific news. We claim the proposed model structure for incorporating macroeconomic news together with historical (market) data is novel and improves the accuracy of price prediction quite significantly. We report results of extensive numerical experiments which justify our claim.

The rapid rise of social media communication has touched upon all aspects of our social and commercial life. In particular, the rise of social media as the most preferred way of connecting people on-line has led to new models of information communication amongst the peers. Of these media Twitter has emerged as a particularly strong platform and in the financial domain tweets by market participants are of great interest and value. News in general, and commercial and financial news wires, in particular provide the market sentiment and in turn influence the asset price behaviour in the financial markets. In a comparable way micro-blogs of tweets generate sentiment and has an impact on market behaviour, that is , the price as well as the volatility of stock prices.

In one of our recent research we have introduced news sentiment based filters such as News RSI (NRSI) and Derived RSI (DRSI), which restrict the choice of asset universe for trading. In this present study, we have extended the same approach to StockTwit’s data. We use the filter approach of asset selection and restrict the available asset universe. We then apply our daily trading strategy using the Second Order Stochastic Dominance (SSD) as an asset allocation model. Our trading model is instantiated by two time series data, namely, (i) historical market price data and (ii) StockTwits sentiment (scores) data. Instead of NRSI we compute the Micro-blog RSI (MRSI) and using this a DRSI is computed. The resulting combined filter (DRSI) leads to an enhancement of the SSD based trading and asset allocation strategy. Empirical experimental results of constructing portfolios are reported for S&P 500 Index constituents.

Key Words: Trading Strategy, Sentiment Analysis, Micro blog data, Asset Filter

Volatility prediction plays an important role in the financial domain. The GARCH family of prediction models is very popular and efficient in using past returns to forecast volatility. It has also been observed that news, scheduled and unscheduled, have an impact on return volatility of assets. An enhanced GARCH model, called News Augmented GARCH (NAGARCH) includes an additional component for news sentiment. With the rise in popularity of the world wide web and social media, it has become a rich source for opinions and sentiments. Twitter is one such platform. It is a micro-blogging site and a popular source for public view on different topics. StockTwits is a social media platform that started as an application built using Twitter’s API. It has since grown into an independent financial social media platform for news and sentiment. StockTwits is a rich source of opinions from subject experts and analysts. This data provides first systematic exploration of social media. It reflects raw sentiments of traders, investors, media, public companies, and investment professionals as opposed to sentiments from curated news wires. This research attempts to determine if the sentiment on stocks from StockTwits micro-blogs can improve volatility prediction. The experiment is performed on 9 NASDAQ100 stocks. The GARCH model with stock returns, and the NA-GARCH model with stock returns and micro-blog sentiment are tuned and their prediction results are evaluated. NA-GARCH, with the sentiment data from StockTwits performed better than the GARCH model in 7 out of the 9 cases.

Key Words: Volatility prediction, GARCH, NA-GARCH, Sentiment, Micro-blog, StockTwits

We enhance the modelling and forecasting of sovereign bond spreads by taking into account quantitative information gained from macro-economic news sentiment. We investigate sovereign bonds spreads of five European countries and improve the prediction of spread changes by incorporating news sentiment from relevant entities and macro-economic topics. In particular, we create daily news sentiment series from sentiment scores as well as positive and negative news volume and investigate their effects on yield spreads and spread volatility. We conduct a correlation and rolling correlation analysis between sovereign bond spreads and accumulated sentiment series and analyse changing correlation patterns over time. Market regimes are detected through correlation series and the impact of news sentiment on sovereign bonds in different market circumstances is investigated. We find best-suited external variables for forecasts in an ARIMAX model set-up. Error measures for forecasts of spread changes and volatility proxies are improved when sentiment is considered. These findings are then utilised to monitor sovereign bonds from European countries and detect changing risks through time.

Momentum strategy is one of the most popular strategies that market participants use to make investment decisions. In the past two decades, many researchers have shown that momentum strategy beats the market, and provides attractive portfolio returns. In this study we investigate Dow Jones Industry Average (DJIA) index and include news data and social media sentiment data to improve the performance of momentum strategy.Particularly, we select StockTwits as the social media source. Four weekly momentum strategies are built and compared over a five-year back-testing period. This research starts with using market data to calculate 5-day Relative Strength Indicator (RSI) that captures the momentum of price. A momentum strategy is constructed based on the overbought/oversold (70/30) signals of RSI proposed by Wilder (1978). Furthermore,the news and social media sentiment data are applied separately to enhance the RSI selections of momentum strategy. News impact scores are used to give more precise evaluations toward news sentiment. Finally, news and social media sentiment data are applied as a double filter to enhance the momentum strategy. The results show that news sentiment and social media data improves the performance of the momentum strategy.

In this study, we introduce a new method of assessing the credit risk of corporate bonds;where in addition to the historical market data news sentiment data is used. Typically,a higher yield spread is usually associated with higher credit risk. By predicting the upward/downward movement of yield and yield spread accurately, the credit risk associated to the bonds can be detected precisely. The corporate bonds studied are issued after 1 January 2007 by seven chosen companies listed in Euro Stoxx 50 index.The time series of bond yields and news sentiment cover the period from 1 January 2007 to 15 May 2017. The modelling of the dynamics of corporate bond yields and credit spreads are based on ARIMA and ARIMAX models. In the ARIMAX model, macroeconomic and firm-specific news sentiment are used as the external explanatory variable. We examine the effect of several categories of macroeconomics news sentiment and firm-specific news sentiment on corporate bond yield spreads.Furthermore, we separate the positive and negative sentiment and investigate their impact on the forecast of corporate bond yields. It is found that negative country news sentiment and central bank news sentiment are effective during a recession period and positive country news sentiment is effective in the recovery period. Negative government and firm-specific news sentiment, in general, affect corporate bond yield spreads more than positive government and company news sentiment.

Forecasting of stock return volatility plays an important role in the financial markets. GARCH model is one of the most common models used for predicting asset price volatility from the return time series. In this study, we have considered quantified news sentiment as a second source of information, which is used together with the GARCH model to predict the volatility of asset price returns. We call this NA-GARCH (news augmented GARCH) model. Our empirical investigation compares volatility prediction of returns of 12 different stocks (from two different stock markets), with 9 data sets for each stock. Our results clearly demonstrate that NA-GARCH provides a superior prediction of volatility than the “plain vanilla” GARCH model. These results vindicate some recent findings regarding the utility of news sentiment as a predictor of volatility, and also vindicate the utility of our novel model structure combining the proxies for past news sentiments and the past asset price returns.

We describe a method for generating daily trading signals to construct trade portfolios of exchange traded securities. Our model uses Second Order Stochastic Dominance (SSD) as the choice criterion for both long and short positions. We control dynamic risk of ‘draw down’ by applying money management. The asset choice for long and short positions are influenced by market sentiment; the market sentiments are in turn acquired from news wires and microblogs. The solution method is challenging as it requires processing stochastic integer programming (SIP) models as well as computing the impact of market sentiment. The computation of SSD portfolios are well known to be computationally hard as this involves processing of large discrete MIP problems. The solution approach is based on our well-established solver system FortSP which uses CPLEX as its embedded solver engine to process SIP models.

In this study we investigate how the prediction of future volatility is improved by using news (meta)data. We use three input time series, namely: (i) market data, (ii) news sentiment impact scores, as explained by Yu (2014), and (iii) the news volume. We compare the results of predicting volatility by using a “vanilla” GARCH model, which uses market data only, and the news enhanced GARCH, as described above. Finally, the forecasted volatility is compared with the realized volatility, allowing an assessment of the robustness and precision of the model. RavenPack and Thomson Reuters provided news data and market data, respectively. The main findings are that the inclusion of scheduled news and the inclusion of news volume characterized by negative sentiment improve the forecasted volatility. The added value of scheduled news to volatility predictions is in line with Li and Engle (1998).

Due to its significance, forecasting asset volatility has been an active area of research in recent decades. In this whitepaper we aim to take into account the stylised facts of volatility to improve predictive power of a simple GARCH model. We investigate the power of three GARCH models (GARCH, EGARCH, GJRGARCH) using implied volatility and news sentiment data as external regressors in order to enhance forecasts of stock return volatility. We also explore the impact of the use of fat-tailed and skewed distributions. Analysis is conducted on 5 constituents of the S&P500. In terms of in-sample performance, the findings suggest that a GJR-GARCH(1,1) model incorporating a student-t distribution, implied volatility and news sentiment data consistently out-performs a simple GARCH(1,1) with a normal distribution. When comparing out-of-sample forecast performance, the enhanced models were able to improve volatility predictions for four out of five stocks.

We report an empirical study of a predictive analysis model for equities; the model uses high frequency (minute-bar) market data and quantified news sentiment data. The purpose of the study is to identify a predictive model which can be used in designing automated trading strategies. Given that trading strategies take into consideration three important characteristics of an asset, namely, return, volatility and liquidity, our model is designed to predict these three parameters for a collection of assets. The minute-bar market data as well as intraday news sentiment metadata have been provided by Thomson Reuters.

Computer trading in financial markets is a rapidly developing field with a growing number of applications. Automated analysis of news and computation of market sentiment is a related applied research topic which impinges on the methods and models deployed in the former. In this review we have first explored the asset classes which are best suited for computer trading. We present in a summary form the essential aspects of market microstructure and the process of price formation as this takes place in trading. We critically analyse the role of different classes of traders and categorise alternative types of automated trading. We introduce alternative measures of liquidity which have been developed in the context of bid-ask of price quotation and explore its connection to market microstructure and trading. We review the technology and the prevalent methods for news sentiment analysis whereby qualitative textual news data is turned into market sentiment. The impact of news on liquidity and automated trading is critically examined. Finally we explore the interaction between manual and automated trading.

A review of news analytics and its applications in finance is given in this chapter. In particular we review the multiple facets of current research and some of the major applications. It is widely recognised news plays a key role in financial markets. The sources and volumes of news continue to grow. New technologies that enable automatic or semi-automatic news collection, extraction, aggregation and categorisation are emerging. Further machine learning techniques can be used to process the textual input of news stories to determine quantitative sentiment scores. We consider the various types of news available and how these can be processed to form inputs to financial models. We report applications of news, for prediction of abnormal returns, for trading strategies, for diagnostic applications as well as the use of news for risk control.

Equity portfolio management problems require fund managers to make decisions about what portfolio to hold (ex-ante) without knowing what future equity returns will be. Though these returns are uncertain, market participants try to understand the nature of the uncertainty and make decisions based on their beliefs about the market environment. Traditionally, portfolio managers have used variants of Markowitz mean-variance analysis to determine the optimal portfolio to hold and this is still fairly standard practice in industry. Mean-variance portfolio decision models fall into the more general group of mean-risk models, where portfolio risk and expected return are traded-off when making asset choices. Variance and standard deviation both measure the spread of a distribution about its mean. Since the variance of a portfolio can be easily calculated from the co-variances of the pairs of asset returns and the asset weights used in the portfolio, variance is predominantly used in portfolio formation. In contrast to computing the asset variances and covariances directly using historical data, multifactor models provide an accurate and efficient way to provide these estimates. They decompose an assets return into returns derived from exposure to common factors and an asset-specific component. The common factors can be understood as representing different risk (uncertainty) aspects, which all the assets are exposed to in varying degrees (factor sensitivities). By describing a group of asset returns through a set of key common factors, the size of the estimation problem is significantly reduced. The new problem faced is to estimate the covariance matrix of common sources of risk, the variances of the specific returns and estimates of each security factor exposures. These models capture the natural intuition that firms with similar characteristics will behave similarly. Active portfolio managers seek to incorporate their investment insight to beat the market. An accurate description of asset price uncertainty is key to the ability to outperform the market. Tetlock et al. (2008) develop a fundamental factor model that incorporates news as a factor. Investors perceptions of the riskiness of an asset are determined by their knowledge about the company and its prospects, that is, by their information sets. They note that these are determined from three main sources: analysts forecasts, quantifiable publicly disclosed accounting variables and linguistic descriptions of the firms current and future profit generating activities. If the first two sources of information are incomplete or biased, the third may give us relevant information for equity prices. We seek to extract an improved understanding of equity price uncertainty using a quantified measure of market sentiment to update a traditional factor model. This may give us the tools to make improved portfolio (management) decisions.