Mémoire numérisé par la Direction des bibliothèques de l'Université de Montréal.

This work concentrates on exploring the influence of social networks to financial markets. We have introduced a novel approach to Twitter sentiment analysis, in which we collect continuous stream of data and analyze it. Our original data set contains over 200 million English written Tweets from the period between July 1, 2014 and October 9, 2014. Twitter sentiment is used as a good representative of investors' mood. On hourly data we investigate how investors are influenced by basic emotions, moods and sentiment in their decision making processes as well as the influence of keywords related to specific securities and FOREX symbols. Particularly, we examine the relationships between Twitter-based variables and returns as well as volatility of several financial instruments on a wide range of data including commodities, currencies and S&P 500 Cash Index. We show that Twitter sentiment influences volatility of securities' returns, tested and shown on both conditional and realized volatility models. We also describe the effect of Twitter sentiment on securities' returns. Moreover, we reveal the influence of basic emotions on investors' decision making processes. Our results suggest that investors are influenced by emotions and moods, especially at longer investment horizons. The impact of emotions at shorter...

This thesis examines the applicability of Google Econometrics - the use of search volume data of particular queries as explanatory variables in time se- ries modeling - in the case of the Czech Republic. We analyze the contribu- tion of Google data by comparing out-of-sample nowcasting performance and in-sample fit with control variables in three related areas: using an auto- regressive model for unemployment, vector autoregression and logit models for GDP and household consumption, and Granger causality test for consum- er confidence. The improvement in quality of unemployment nowcasting is modest but statistically significant; sentiment index based on Google queries shows reciprocal relationship with the official Consumer Confidence Indicator, and it also provides superior nowcasts for household consumption as well as in- sample fit in logit models; its performance in GDP nowcasting is average among control variables. In overall, the results suggest that Google Econometrics is applicable also to the Czech Republic, despite the fact that the internet penetration rate and Google popularity was lower over the analyzed period compared with developed economies where these methods were usually tested. In the future, Google data may be used together with other leading and coincident indica- tors to...

Notre recherche se situe dans le contexte de l'après-cancer et a pour objectif l'étude de l'hormonothérapie dans le cadre du traitement adjuvant du cancer du sein chez la femme ménopausée. Nous avons cherché à évaluer l'impact du traitement anti-hormonal sur le sentiment de guérison que la femme s'approprie pour accéder à une guérison sur le plan psychique, après le cancer du sein. Cette étude qualitative a été menée auprès de 17 femmes ayant eu un cancer du sein et étant sous hormonothérapie adjuvante actuellement. Des entretiens semi-directifs ont été conduits auprès de ces sujets et complétés par un questionnaire d'auto-évaluation de la qualité de vie. Nous avons repéré trois degrés de sentiment de guérison (faible, moyen et fort) qui semblent corrélés à la qualité de vie actuelle du sujet et à son ressenti d'effets secondaires à l'hormonothérapie. Ainsi les patientes détentrices d'un faible sentiment de guérison du cancer du sein sont celles qui ressentent le plus d'effets secondaires au traitement et endurent la qualité de vie la plus détériorée. Elles semblent se sentir encore proches d'un statut de malade et le moins engagées dans un processus de guérison psychique. A l'opposé les patientes au fort sentiment de guérison ne manifestent pas d'effets secondaires au traitement ; une élaboration psychique est repérable dans leurs discours, témoin du travail de la guérison à l'œuvre chez ces sujets. Enfin, dans cette population de femmes déjà ménopausées lors du diagnostic, nous avons par ailleurs remarqué que l'hormonothérapie pouvait réactiver la période ménopausique, particulièrement chez les sujets dont l'expérience de la ménopause a été vécue difficilement. / Our research deals with the "after cancer" period and aims to study hormonotherapy within the framework of the adjuvating treatment of breast cancer on menopaused women. We tried to estimate the impact of the anti-hormonal treatment on the feeling of cure that women have to make their own, after a breast cancer, to reach a cure on the psychic plan. This qualitative study was led with 17 women who suffered from a breast cancer and are now under adjuvante hormonotherapy. The subject went through semi-directive conversations wich were complemented with a questionnaire of self-assessment of the quality of life. We defined three degrees of feeling of cure (weak, medium and strong) which seem correlated to the actual quality of life of the subjects and to their suffering regarding side effects in the hormonotherapy. So, on one hand, the patients experiencing a weak feeling of cure of a breast cancer are the ones who suffer the most from side effects during the treatment and bear the most damaged quality of life. They seem to feel like they are still patient's and don't seem to be engaged in a process of psychic cure. On the other hand, the patients who are experiencing strong feeling of cure do not show side effects of the treatment ; a psychic elaboration is noticeable in their speeches, and witness the "work of cure" these subjects are going through. Finally, among this population of women already menopaused when the diagnosis was made, we besides noticed that the hormonotherapy could reactivate the menopausic period, especially for subjects who suffered from the experience of menopause.

Consumer sentiment has the ability to provide researchers with many avenues to test existing Finance and Economic theories. Chapter 1 introduces the issues that I seek to explore within the area of Behavioral Finance. Chapter 2 utilizes thirty years of consumer sentiment data to explore extant economic theories and hypotheses. In particular, I study the Prospect Theory and the Life Cycle Investment Hypothesis. In addition, I also study how changes in consumer sentiment can foretell future stock returns for firms in different industries and of different sizes. By studying how individuals of different ages display optimism and pessimism through consumer sentiment surveys, I am able to contribute to the literature by shedding additional light on just how the important age is with respect to a person's economic outlook. One particular phenomenon that I discuss in this chapter is downside risk. I will provide further support to the existing literature which shows that gains and losses are not viewed equally by individuals. To account for this discrepancy, this paper models the time series relationship between consumer sentiment and stock returns using asymmetric response models. Chapter 3 builds upon the previous chapter's findings by using consumer sentiment to explore if this index can forecast housing market variables such as changes in home sales and home prices. Given the recent financial market turmoil that stemmed from the U.S. housing market debacle, this chapter is timely. Using widely cited housing indices, I explore regional differences in the U.S. housing market and how the sentiment of local consumers can possibly affect their housing markets. I also include analyses in which the age of the consumer is accounted for to see if evidence of the Life Cycle Investment Hypothesis emerges. This theory postulates that younger individuals are more likely to demand housing as a financial asset and if this were true, I hypothesize that changes in younger individuals' sentiment would have more forecasting power with respect to future housing sales and price changes. Lastly, I conclude this dissertation with Chapter 4 which includes additional discussions of the issues studied.

This dissertation addresses issues concerning liquidity and its volatility. It consists of two essays. The first essay, "Liquidity, Macro Factors and the U.S. Equity Flows to Emerging Markets", examines the role of liquidity on equity flows from the U.S. to fifteen emerging markets around the world. Since liquidity has many dimensions, an emphasis is placed on utilizing various measures of liquidity. Moreover, both static and dynamic analyses, as well as short and long-horizon regressions, are performed to investigate the research questions. The results suggest that a liquid market attracts flows, after controlling for market size, political openness, exchange rate and other macro factors. Additionally, evidence indicates that the importance of liquidity varies across regions. For instance in the Asian region, the relation between equity flows and volume-related liquidity is weak while that between flows and price impacts of trading is strong. Evidence also supports the relevance of macro factors such as a country's economic freedom. The second essay, "Liquidity Risk Premium Puzzle and Possible Explanations", attempts to resolve the liquidity risk puzzle: a negative relation between returns and liquidity risk, documented by Chordia, Subrahmanyam, and Anshuman (2001b), by employing alternative liquidity measures and by incorporating factors that might potentially affect the relation. The main findings are as follows. The relation between stock returns and volatility of liquidity depends on the measure of liquidity. When liquidity measures are based on trading volume, the results are largely mixed, but when liquidity is measured based on price impact of trading, the relation between returns and volatility of price impacts is positive, as expected. The results are sensitive to time periods examined. Moreover, during extreme down markets, the aversion to liquidity volatility is lower, suggesting behavioral bias might potentially address the puzzle. Empirical findings also suggest that liquidity risk premium tends to be greater for small stocks. Finally, when the VIX index is included as a proxy for investor sentiment, the results indicate that the relation between returns and liquidity risk is significantly positive in four out of five liquidity measures. In sum, the empirical analysis partially but not completely addresses the puzzle.

With the rapid growth and maturity of Question-Answering (QA) domain, non-factoid Question-Answering tasks are in high demand. However, existing Question-Answering systems are either fact-based, or highly keyword related and hard-coded. Moreover, if QA is to become more personable, sentiment of the question and answer should be taken into account. However, there is not much research done in the field of non-factoid Question-Answering systems based on sentiment analysis, that would enable a system to retrieve answers in a more emotionally intelligent way. This study investigates to what extent could prediction of the best answer be improved by adding an extended representation of sentiment information into non-factoid Question-Answering.

L'analyse de sentiments est un des nouveaux défis apparus en traitement automatique des langues avec l'avènement des réseaux sociaux sur le WEB. Profitant de la quantité d'information maintenant disponible, la recherche et l'industrie se sont mises en quête de moyens pour analyser automatiquement les opinions exprimées dans les textes. Pour nos travaux, nous nous plaçons dans un contexte multilingue et multi-domaine afin d'explorer la classification automatique et adaptative de polarité.Nous proposons dans un premier temps de répondre au manque de ressources lexicales par une méthode de construction automatique de lexiques affectifs multilingues à partir de microblogs. Pour valider notre approche, nous avons collecté plus de 2 millions de messages de Twitter, la plus grande plate-forme de microblogging et avons construit à partir de ces données des lexiques affectifs pour l'anglais, le français, l'espagnol et le chinois.Pour une meilleure analyse des textes, nous proposons aussi de remplacer le traditionnel modèle n-gramme par une représentation à base d'arbres de dépendances syntaxiques. Dans notre modèles, les n-grammes ne sont plus construits à partir des mots mais des triplets constitutifs des dépendances syntaxiques. Cette manière de procéder permet d'éviter la perte d'information que l'on obtient avec les approches classiques à base de sacs de mots qui supposent que les mots sont indépendants.Finalement, nous étudions l'impact que les traits spécifiques aux entités nommées ont sur la classification des opinions minoritaires et proposons une méthode de normalisation des décomptes d'observables, qui améliore la classification de ce type d'opinion en renforçant le poids des termes affectifs.Nos propositions ont fait l'objet d'évaluations quantitatives pour différents domaines d'applications (les films, les revues de produits commerciaux, les nouvelles et les blogs) et pour plusieurs langues (anglais, français, russe, espagnol et chinois), avec en particulier une participation officielle à plusieurs campagnes d'évaluation internationales (SemEval 2010, ROMIP 2011, I2B2 2011). / Sentiment analysis is a challenging task today for computational linguistics. Because of the rise of the social Web, both the research and the industry are interested in automatic processing of opinions in text. In this work, we assume a multilingual and multidomain environment and aim at automatic and adaptive polarity classification.We propose a method for automatic construction of multilingual affective lexicons from microblogging to cover the lack of lexical resources. To test our method, we have collected over 2 million messages from Twitter, the largest microblogging platform, and have constructed affective resources in English, French, Spanish, and Chinese.We propose a text representation model based on dependency parse trees to replace a traditional n-grams model. In our model, we use dependency triples to form n-gram like features. We believe this representation covers the loss of information when assuming independence of words in the bag-of-words approach.Finally, we investigate the impact of entity-specific features on classification of minor opinions and propose normalization schemes for improving polarity classification. The proposed normalization schemes gives more weight to terms expressing sentiments and lower the importance of noisy features.The effectiveness of our approach has been proved in experimental evaluations that we have performed across multiple domains (movies, product reviews, news, blog posts) and multiple languages (English, French, Russian, Spanish, Chinese) including official participation in several international evaluation campaigns (SemEval'10, ROMIP'11, I2B2'11).

Dans ce travail de recherche, nous allons étudier les facteurs déterminants inhérents aux salariés, qui interviennent dans leur prise de décision de se former. Nous nous demandons pourquoi certains salariés sont capables d’affronter tous les obstacles et fournissent d’énormes efforts pour se former, alors que d’autres ne parviennent pas à surmonter les difficultés et abandonnent plus facilement. Les déterminants de la capacité des adultes à prendre la décision de suivre une formation professionnelle continue ont préoccupé les chercheurs depuis plusieurs décennies (Peretti, 1994 ; Dubar, 2004 ; Galand et Vanlède, 2004 ; Carré et Fenouillet, 2009). Ces auteurs sont convaincus que la FPC est un outil d’évolution dans l’entreprise et pour ce faire, le salarié doit avoir la capacité de s’engager dans ce processus astreignant nécessitant beaucoup d’efforts et de concentration. Or, dans son livre « Auto-efficacité : « le sentiment d’efficacité personnelle » Bandura et ses collaborateurs (1997, 2003) ont effectué des centaines de recherches, menées à partir des années soixante-dix. Ils expliquent à travers leurs travaux, les raisons qui font que des individus avec le même niveau d’étude, les mêmes capacités intellectuelles et les mêmes statuts socioprofessionnels agissent différemment face à l’apprentissage. Bandura et ses collaborateurs (1997, 2003) soulignent de fait, l’importance du sentiment d’efficacité personnelle dans la décision de s’engager ou pas dans un processus de FPC. / In this research work, we will study the determining factors inherent to employees, who intervene in their decision-making to train. We wonder why some employees are able to face all obstacles and make enormous efforts to train, while others fail to overcome difficulties and give up more easily.The determinants of adults' ability to make the decision to follow continuing professional education have been of concern to researchers for several decades (Peretti 1994, Dubar 2004, Galand and Vanlède 2004, Carré and Fenouillet 2009). These authors are convinced that the CVT is a tool of evolution in the company and to do this, the employee must have the capacity to engage in this demanding process requiring a lot of effort and concentration. However, in his book "Self-efficacy:" the feeling of self-efficacy "Bandura and his collaborators (1997, 2003) carried out hundreds of research, carried out from the seventies. Through their work, they explain the reasons why individuals with the same level of study, the same intellectual abilities and the same socio-professional status act differently in the face of learning. Bandura et al. (1997, 2003) emphasize the importance of self-efficacy in deciding whether or not to engage in a CPD process.

This thesis explores the possibility of applying sentiment analysis to extract tonality of user reviews on the Nordic languages. Data processing is performed in the form of preprocessing through tokenization and padding. A model is built in a framework called Keras. Models for classification and regression were built using LSTM and GRU architectures. The results showed how the dataset influences the end result and the correlation between observed and predicted values for classification and regression. The project shows that it is possible to implement NLP in the Nordic languages and how limitations in input and performance in hardware affected the result. Some questions that arose during the project consist of methods for improving the dataset and alternative solutions for managing information related to big data and GDPR. / Denna avhandling undersöker möjligheten att tillämpa sentiment analys för att extrahera tonalitet av användarrecensioner på nordiska språk. Databehandling utförs i form av förprocessering genom tokenisering och padding. En modell är uppbyggd i en ramverkad Keras. Modeller för klassificering och regression byggdes med LSTM och GRU-arkitekturer. Resultaten visade hur datasetet påverkar slutresultatet och korrelationen mellan observerade och förutspådda värden för klassificering och regression. Projektet visar att det är möjligt att implementera NLP på de nordiska språken och hur begränsningar i input och prestanda i hårdvara påverkat resultatet. Några frågor som uppstod under projektet består av metoder för att förbättra datasetet och alternativa lösningar för hantering av information relaterad till stora data och GDPR.