Nelson Areal

On the Predictability of Stock Market Behavior Using StockTwits Sentiment and Posting Volume

In this study, we explored data from StockTwits, a microblogging platform exclusively dedicated to the stock market. We produced several indicators and analyzed their value when predicting three market variables: returns, volatility and trading volume. For six major stocks, we measured posting volume and sentiment indicators. We advance on the previous studies on this subject by considering a large time period, using a robust forecasting exercise and performing a statistical test of forecasting ability. In contrast with previous studies, we find no evidence of return predictability using sentiment indicators, and of information content of posting volume for forecasting volatility. However, there is evidence that posting volume can improve the forecasts of trading volume, which is useful for measuring stock liquidity (e.g. assets easily sold).

Socially Responsible Investing in the Global Market: The Performance of US and European Funds

This paper investigates the style and performance of US and European global socially responsible funds. Several specifications of the return generating process are applied as well as their corresponding conditional versions. Most European global socially responsible funds do not show significant performance differences in relation to both conventional benchmarks and socially responsible benchmarks. US funds and Austrian funds show evidence of underperformance. By applying conditional models, we find evidence of time-varying betas, but not of time-varying alphas. With respect to investment style, we have found evidence that socially responsible funds are strongly exposed to small cap and growth stocks. While these results are consistent with previous studies, they uncover some misclassification issues in these funds. Finally, we have also documented a significant home bias for global socially responsible funds.

The impact of microblogging data for stock market prediction: Using Twitter to predict returns, volatility, trading volume and survey sentiment indices

Abstract In this paper, we propose a robust methodology to assess the value of microblogging data to forecast stock market variables: returns, volatility and trading volume of diverse indices and portfolios. The methodology uses sentiment and attention indicators extracted from microblogs (a large Twitter dataset is adopted) and survey indices (AAII and II, USMC and Sentix), diverse forms to daily aggregate these indicators, usage of a Kalman Filter to merge microblog and survey sources, a realistic rolling windows evaluation, several Machine Learning methods and the Diebold-Mariano test to validate if the sentiment and attention based predictions are valuable when compared with an autoregressive baseline. We found that Twitter sentiment and posting volume were relevant for the forecasting of returns of S&P 500 index, portfolios of lower market capitalization and some industries. Additionally, KF sentiment was informative for the forecasting of returns. Moreover, Twitter and KF sentiment indicators were useful for the prediction of some survey sentiment indicators. These results confirm the usefulness of microblogging data for financial expert systems, allowing to predict stock market behavior and providing a valuable alternative for existing survey measures with advantages (e.g., fast and cheap creation, daily frequency).

Stock market sentiment lexicon acquisition using microblogging data and statistical measures

Lexicon acquisition is a key issue for sentiment analysis. This paper presents a novel and fast approach for creating stock market lexicons. The approach is based on statistical measures applied over a vast set of labeled messages from StockTwits, which is a specialized stock market microblog. We compare three adaptations of statistical measures, such as Pointwise Mutual Information (PMI), two new complementary statistics and the use of sentiment scores for affirmative and negated contexts. Using StockTwits, we show that the new lexicons are competitive for measuring investor sentiment when compared with six popular lexicons. We also applied a lexicon to easily produce Twitter investor sentiment indicators and analyzed their correlation with survey sentiment indexes. The new microblogging indicators have a moderate correlation with popular Investors Intelligence (II) and American Association of Individual Investors (AAII) indicators. Thus, the new microblogging approach can be used alternatively to traditional survey indicators with advantages (e.g., cheaper creation, higher frequencies). Proposal of an automatic procedure for the creation of stock market lexicons.The procedure uses diverse statistical measures on StockTwits labeled messages.The new lexicons obtain better investor sentiment indicators than general lexicons.The new Twitter sentiment indicators correlate with survey sentiment indicators.

Some experiments on modeling stock market behavior using investor sentiment analysis and posting volume from Twitter

The analysis of microblogging data related with stock markets can reveal relevant new signals of investor sentiment and attention. It may also provide sentiment and attention indicators in a more rapid and cost-effective manner than other sources. In this study, we created several indicators using Twitter data and investigated their value when modeling relevant stock market variables, namely returns, trading volume and volatility. We collected recent data from nine major technological companies. Several sentiment analysis methods were explored, by comparing 5 popular lexical resources and two novel lexicons (emoticon based and the merge of all 6 lexicons) and sentiment indicators produced using two strategies (based on daily words and individual tweet classifications). Also, we measured posting volume associated with tweets related to the analyzed companies. While a short time period is considered (32 days), we found scarce evidence that sentiment indicators can explain these stock returns. However, interesting results were obtained when measuring the value of using posting volume for fitting trading volume and, in particular, volatility.

The long-horizon returns behaviour of the Portuguese stock market1

In the last few years several research studies have challenged the traditional weak-form efficiency tests of the stock market. These studies suggested an alternative to the random walk model, containing temporary and permanent components. If stocks follow such a model then the traditional tests, using returns computed for short intervals would be unable to detect them. To investigate the evidence for such models in the Portuguese stock market ten stock indexes were created. This is a pioneer study of the Portuguese stock market, and uses nominal, real and excess returns, computed for longer horizons. Three methodologies were used: variance ratios, ordinary least squares regressions and weighted least squares regressions. The statistical significance of the results was studied using traditional parametric tests as well as non-parametric tests. The evidence is mixed, as the presence of tendencies towards mean aversion and mean reversion were detected. Results also show that the evidence is very sensitive to the methodology used and the signifcance tests performed. These results, however, do not necessarily reject the weak-form market efficiency hypothesis.

When times get tough, gold is golden

We investigate the dynamic behaviour of conditional correlations between the US market, gold and two gold financial proxies using a multivariate dynamic conditional correlation model over different market regimes. A comprehensive period of time is analysed covering approximately 37 years of daily data, from August 1976 to March 2013, as well as a shorter period, of about 15 years, from September 1998 to March 2013. Both periods include the recent sub-prime financial crisis. Market regimes are defined using bull/bear states and alternatively using volatility regimes from a three-state Markov-switching variance model. An index of US mining companies and a value-weighted portfolio of US gold mutual funds are treated as potential proxies for an investment in gold. Two important conclusions emerge from our study. The first is that, even in the context of a dynamic correlation analysis, gold is always a safe haven; negatively correlated with the stock market under adverse market conditions. The second is that, although the gold proxies considered here exhibit a low correlation with the stock market and therefore offer diversification benefits, they cannot be considered perfect substitutes of gold due to their lack of negative correlations with the market in times of turmoil.

Automatic creation of stock market lexicons for sentiment analysis using StockTwits data

Sentiment analysis has been increasingly applied to the stock market domain. In particular, investor sentiment indicators can be used to model and predict stock market variables. In this context, the quality of the sentiment analysis is highly dependent of the opinion lexicon adopted. However, there is a lack of lexicons adjusted to microblogging stock market data. In this work, we propose an automatic procedure for the creation of such lexicon by exploring a large set of labeled messages from StockTwits, a popular financial microblogging service, and using four statistical measures: adaptations of the known TF-IDF, Information Gain, Class Percentage, and a newly proposed Weighted Class Probability. The obtained lexicons are competitive when compared with a set of six reference lexicons. Moreover, we verified that it is beneficial to use continuous sentiment scores instead of sentiment labels.

Fast Trees for Options with Discrete Dividends

The valuation of options using a binomial non-recombining tree with discrete dividends can be intricate. This paper proposes three different enhancements that can be used alone or combined to value American options with discrete dividends using a nonrecombining binomial tree. These methods are compared in terms of both speed and accuracy with a large sample of options with one to four discrete dividends. This comparison shows that the best results can be achieved by the simultaneous use of the three enhancements. These enhancements when used together result in significant speed/accuracy gains in the order of up to 200 times for call options and 50 times for put options. These techniques allow the use of a non-recombining binomial tree with very good accuracy for valuing options with up to four discrete dividends in a timely manner. ∗The support of the Portuguese Foundation for Science and Technology Project PTDC/GES/78033/2006 is acknowledged. We thank Ana Carvalho for her helpful comments. †School of Economics and Management. University of Minho. 4710-057 Braga (Portugal). E-mail: nareal@eeg.uminho.pt; Phone: +351 253 601 923; Fax:+351 253 601 380. ‡Corresponding author. School of Economics and Management. University of Minho. 4710-057 Braga (Portugal). E-mail: artur.rodrigues@eeg.uminho.pt; Phone: +351 253 601 923; Fax:+351 253 601 380. Fast trees for options with discrete dividends American options valuation using binomial lattices can be cumbersome. Several authors have suggested different approaches to speed the computation or to increase the speed of convergence. Among others, some authors suggested ways to reduce the number of the nodes in the tree, thus speeding up the computation time (e.g.: Baule and Wilkens [2004]). One example of the latter approach is a paper by Curran [1995] that has been largely ignored in the literature. One advantage of Curran’s [1995] method is that it is not an approximation to the value of a binomial tree, since it produces exactly the same result as Cox, Ross, and Rubinstein’s [1979] (CRR) tree with the same number of steps at a fraction of the computation time. The gains of speeding the computation of a binomial tree are more relevant for valuing options with underling assets that pay discrete dividends. In such cases the binomial trees do not recombine and the number of nodes rapidly explode even for a small number of time steps. There are several approximations that allow the use of recombining binomial trees, but all of them can in some occasions produce large valuation errors (usually they occur when a dividend is paid at the very beginning of the option life), or require a very large number of steps to avoid such valuation errors.1 The advantages of using a non-recombining binomial tree are twofold: first it considers the true stochastic process for the underlying asset which excludes arbitrage opportunities; and secondly it eliminates large valuation errors irrespective of when the dividends occur. This, in turn, results in a accurate valuation of such options. The problem of these trees is that the number of nodes in the tree grows exponentially with the number of dividends. This explains why usually a recombining approximation is used instead of the non-recombining tree. Unfortunately the method proposed by Curran [1995] is not directly applied to options on assets which pay discrete dividends. In light of the techniques proposed by Curran [1995], which in turn are based on the work of Kim and Byun [1994], this paper adjusts their acceleration techniques to the valuation of American put options and also proposes a different accelerated binomial method to the valuation of American call options. We also suggest two other improvements with very good results. One, called here Adapted Binomial, consists in making a time step to coincide with the ex-dividend dates. The other is to apply the Black and Scholes [1973] formula to obtain the continuation value in the last steps of the binomial tree. These adaptations along with the improvements here suggested result in significant speed/accuracy gains in the order of up to 200 times for call options and 50 times for put options. These techniques allow the use of a non-recombining binomial tree with very good Examples of such methods are: Schroder [1988], Hull and White [1988], Harvey and Whaley [1992], Wilmott, Dewynne, and Howison [1998], Vellekoop and Nieuwenhuis [2006].

Value-at-Risk Forecasting Ability of Filtered Historical Simulation for Non-Normal GARCH Returns

As a hybrid methodology to estimate VaR, that combines the use of parametric modelling with the use of bootstrapping techniques, filtered historical simulation (FHS) should not be sensitive to the use of alternative distributions assumed in the filtering stage. However, recent studies (Kuester et al. 2006) have found that the distribution used in the filtering stage can influence the VaR estimates obtained in the context of this methodology. Using Extreme Value Theory (EVT) this paper explains that the VaR estimates for lower probabilities should not be sensitive to the distribution assumed in the filtering stage of the FHS method. However, for higher probabilities, the EVT results do not hold and therefore the use of alternative distributions might impact the VaR estimates. These theoretical results are tested using both simulated and real data. Three different realistic data generating processes were considered to generate several series of simulated returns. Additionally, three competing models, differing in the innovations assumption, were tested: a normal-GARCH, a t-GARCH and a skew-t-GARCH. Our backtesting results indicate that FHS can forecast VaR with accuracy for data which exhibits a high incidence of zeros, time-varying skewness, asymmetric effects to return shocks on volatility, as well as other stylized facts. Importantly, our results for the simulated data demonstrate that, for lower probabilities, the choice of the distribution assumed in the filtering stage has no impact on the performance of FHS as an accurate method to forecasting VaR. Additionally, 40 years of daily data on six well known active stock indices are used to empirically evaluate the FHS VaR estimates. Four competing GARCH-type specifications, combined with three different innovation assumptions (normal, Student-t and skew-Student t), are used to capture time series dynamics. Based on a sample of several VaR probabilities, the results of the dynamic quantile (DQ) tests clearly indicate that the use of asymmetric GARCH models (specifically GJR and GJR in Mean) generally improve the VaR forecasting performance of FHS. In addition, the choice of a skew-Student t distribution for the innovation process slightly improves the performance results of the GJR in Mean model. When different VaR probabilities are used, the choice of an appropriate model specification seems to be more important than the choice of a suitable distribution assumption. With respect to the lower VaR probability tested (1%), the results show that, as expected, the VaR estimate is very similar regardless of the GARCH model and distribution assumed.