Stock Market is always being the most important role in modern capital market. And Stock Market is becoming one the most popular investment tools these days. Because of the Globalization of capital markets, the spreading of capital becomes faster and easier. The development of capital markets evoke the interesting of scholars and the field of stock market prediction attract scholars and researchers from different background. There are two approaches of predicting stock market - fundamental analysis and technical analysis. The purpose of my work was to predict three stock markets in the world, namely Taiwan Weighted Index (IDXWT), Hong Kong Hang Seng Index (HSI) and Dow Jones Industrial Average (DJI) using technical analysis and Dynamic Bayesian Network (DBN).This thesis is based on Wang¡¦s thesis [Wan05] ¡§Investment Decision Support with Dynamic Bayesian Networks¡¨. According to different characteristic of 3 stock markets, we divide 3 different markets into 3 experiments. For each market, we expect we can find the best indicators and trading signals. The first experiment involves Taiwan Weighted Index as our prediction target; the second one uses Hong Kong Hang Seng Index and the third experiment employs Dow Jones Industrial Average. As a result, Taiwan Stock market (both 15-day and 20-day Moving Average)can make higher returns than buy-and-hold, RSI_6 and KD. And we also have the same conclusion of Hang Seng Index and Dow Jones Industrial Average. The best return from 15-day MA and 20-day MA of Taiwan Stock market is 47.95% and 60.21%, respectively. Moreover, the best result of Hang Seng Index is 60.06% for 4 years and 25.83% for Dow Jones Industrial Average. All of the best results can make higher returns than each of their buy-and-hold, RSI_6 and KD. In the conclusion, we may say that this paper can provide a direction to investors while they are using these technical indicators to predict these particular stock markets.

In this work we made an attempt to predict the upwards/downwards movement of the S&P 500 index from the news articles published by Bloomberg and Reuters. We employed the SVM classifier and conducted multiple experiments aiming at understanding the shape of the data and the specifics of the task better. As a result, we established the common evaluation settings for all our subsequent experiments. After that we tried incorporating various features into the model and also replicated several approaches previously suggested in the literature. We were able to identify some non-trivial dependencies in the data which helped us achieve a high accuracy on the development set. However, none of the models that we built showed comparable performance on the test set. We have come to the conclusion that whereas some trends or patterns can be identified in a particular dataset, such findings are usually barely transferable to other data. The experiments that we conducted support the idea that the stock market is changing at random and a high quality of prediction may only be achieved on particular sets of data and under very special settings, but not for the task of stock market prediction in general. 1

Ingeniero Civil Eléctrico / En el mundo financiero, la decisión de compraventa de activos se suele asentar en el análisis fundamental a largo plazo, combinado con análisis técnico a corto plazo; con el objetivo de establecer un momento adecuado para la adquisición y enajenación de activos. En la última década, se ha verificado un crecimiento exponencial en la capacidad de procesamiento y de manejo de bases de datos; siendo la minería de estos vastamente estudiada y aplicada exitosamente en distintos campos, entre los cuales se encuentran las finanzas. En el presente trabajo, se estudia la existencia de estructura con capacidad predictiva en activos financieros, con el fin de anticipar cambios de tendencia y así obtener retornos por sobre el mercado. Para esto, se desarrolla a cabalidad el proceso de extracción de conocimiento de bases de datos, el que considera desde la generación de variables, hasta la obtención de información, a partir de los datos transaccionales de las acciones que componen el Índice de Precios Selectivo de Acciones (IPSA) 2013. En este sentido, es importante precisar que la metodología clásica en la predicción de series de tiempo, se basa en la utilización de precios anteriores para así predecir el precio futuro, utilizando ventanas de tiempo estáticas. En este trabajo se estudia un método nuevo, donde la variable objetivo, en vez de ser retornos en ventanas temporales, son tanto retornos como ventanas dinámicas, extraídas a partir de extensiones no causales de retracciones porcentuales del precio (indicador ZigZag) de las acciones, las que representan mínimos y máximos locales de la serie de tiempo; evitando así sobreajuste temporal y acomodándose a los cambios de ciclo del activo en estudio. Se generan variables independientes a partir de datos de transacciones realizadas por parte de miembros de las compañías (Insiders) e indicadores técnicos tales como cruces, divergencias y zonas de agotamiento a partir de Medias Móviles Convergentes/Divergentes, Índice de fuerza Relativa y Oscilador Estocástico. Se realiza selección de características mediante Forward Selection y Backward Elimination, para encontrar un subconjunto de atributos adecuado y analizar su impacto predictivo. Se aplican algoritmos de aprendizaje supervisado con capacidad de extraer patrones altamente no lineales, destacando Redes Neuronales de Retropropagación, Máquinas de Soporte Vectorial y Métodos Basados en Similitud. Con el fin de determinar el ciclo del mercado al que mejor se ajustan los atributos extraídos y el mejor modelo predictor sobre la base de datos no balanceada, se evalúa la combinación de predicciones de compraventa (anticipaciones de cambio de tendencia) utilizando clasificador Bayesiano ingenuo y operadores lógicos. Finalmente, se realiza una evaluación tanto cualitativa (visual) como cuantitativa (mediante un simulador de inversiones) del comportamiento de las recomendaciones de compraventa; analizando la distribución de retorno, drawdown y tiempo de apertura de las operaciones. De lo anterior puede concluirse que dentro de lo caótico del mercado bursátil, subyace estructura altamente no lineal con poder anticipativo de cambios de tendencia de los activos; la cual se puede atribuir a que, en Chile, el mercado es poco profundo, ilíquido o ineficiente.

In this research we investigate the relationship between multifactor investing and Artificial Neural Network (ANN) and contribute to modern stock market prediction. We present the components for multifactor investing i.e. value, quality, size, low volatility & momentum as well as a methodology for ANN which provides the theory for the results. The return for the multifactor funds tested in this research is recorded below the benchmark used. However, the factors do have a dynamic relationship when testing for correlation and the multifactor regression analysis showed a high explanatory power (R2) for the funds. Based on the methodology of an ANN we establish that it is possible to use the knowledge from multifactor investing to train the technology with. When summarizing peer reviewed journals, we find that momentum have already been recurrently used in previous stock market prediction systems based on ANN, but the remaining factors have not. We conclude that there is an opportunity to use several factors to train an ANN due to their dynamic relationship and unique characteristics.

Prediction of financial time series is described as one of the most challenging tasks of time series prediction, due to its characteristics and dynamic nature. In any investment activity, having an accurate prediction system will significantly benefit investors by guiding decision making, especially in trading, asset management and risk management. Thus, the attempts to build such systems have attracted the attention of practitioners in the market and also researchers for many decades. Furthermore, the purpose of this thesis is to investigate and develop a new approach to predicting financial time series with consideration given to their dynamic nature. In this thesis, the prediction procedures will be carried out in three phases. The first phase proposes a new hybrid dynamic model based on Ensemble Empirical Mode Decomposition (EEMD), Back Propagation Neural Network (BPNN), Recurrent Neural Network (RNN), Support Vector Regression (SVR) and EEMD-Genetic Algorithm (GA)-Weighted Average (WA) to predict stock index closing price. EEMD in this phase is introduced as a preprocessing step to historical observation for the first time in the literature. The experimental results show that the EEMDD-GA-WA model performance is a notch above the other methods utilised in this phase. The second phase proposes a new hybrid static model based on Wavelet Transform (WT), RNN, Support Vector Machine (SVM), Nave Bayes and WT-GA-WA to predict the exact change of the stock index closing price. In this phase, the experimental results showed that the proposed WT-GA-WA model outperformed the rest of the models utilised in this phase. Moreover, the input data that are fed into the hybrid model in this phase are technical indicators. The third phase in this research introduces a new Hybrid Heuristic-Rules-based System (HHRS) for stock price prediction. This phase intends to combine the output of the hybrid models in phase one and two in order to enhance the final prediction results. Thus,to the best of our knowledge, this study is the only one to have carried out and tested this approach with a real data set. The results show that the HHRS outperformed all suggested models over all the data sets. Thus, this indicates that combining di↵erent techniques with diverse types of information could enhance prediction accuracy.

The advance in technologies such as massive storage devices and high speed internet has led to an enormous increase in the volume of available documents in electronic form. These documents represent information in a complex and rich manner that cannot be analysed using conventional statistical data mining methods. Consequently, text mining is developed as a growing new technology for discovering knowledge from textual data and managing textual information. Processing and analysing textual information can potentially obtain valuable and important information, yet these tasks also requires enormous amount of computational resources due to the sheer size of the data available. Therefore, it is important to enhance the existing methodologies to achieve better scalability, efficiency and accuracy. / The emerging Grid technology shows promising results in solving the problem of scalability by splitting the works from text clustering algorithms into a number of jobs, each to be executed separately and simultaneously on different computing resources. That allows for a substantial decrease in the processing time and maintaining the similar level of quality at the same time. / To improve the quality of the text clustering results, a new document encoding method is introduced that takes into consideration of the semantic similarities of the words. In this way, documents that are similar in content will be more likely to be group together. / One of the ultimate goals of text mining is to help us to gain insights to the problem and to assist in the decision making process together with other source of information. Hence we tested the effectiveness of incorporating text mining method in the context of stock market prediction. This is achieved by integrating the outcomes obtained from text mining with the ones from data mining, which results in a more accurate forecast than using any single method.

Background. Stock market prediction is an active yet challenging research area. A lot of effort has been put in by both academia and practitioners to produce accurate stock market predictions models, in the attempt to maximize investment objectives. Tree-based ensemble machine learning methods such as XGBoost have proven successful in practice. At the same time, there is a growing trend to incorporate multiple data sources in prediction models, such as historical prices and text, in order to achieve superior forecasting performance. However, most applications and research have so far focused on the American or Asian stock markets, while the Swedish stock market has not been studied extensively from the perspective of hybrid models using both price and text derived features.  Objectives. The purpose of this thesis is to investigate whether augmenting a numerical dataset based on historical prices with sentiment features extracted from financial news improves classification performance when predicting the daily price trend of the Swedish stock market index, OMXS30. Methods. A dataset of 3,517 samples between 2006 - 2020 was collected from two sources, historical prices and financial news. XGBoost was used as classifier and four different metrics were employed for model performance comparison given three complementary datasets: the dataset which contains only the sentiment feature, the dataset with only price-derived features and finally, the dataset augmented with sentiment feature extracted from financial news.  Results. Results show that XGBoost has a good performance in classifying the daily trend of OMXS30 given historical price features, achieving an accuracy of 73% on the test set. A small improvement across all metrics is recorded on the test set when augmenting the numerical dataset with sentiment features extracted from financial news.  Conclusions. XGBoost is a powerful ensemble method for stock market prediction, reflected in a satisfactory classification performance of the daily movement direction of OMXS30. However, augmenting the numerical input set with sentiment features extracted from text did not have a powerful impact on classification performance in this case, as the improvements across all employed metrics were small.

This degree project investigates if a company’s stock price development can be predicted using the general opinion expressed in tweets about the company. The project starts off with the model from a previous project and then tries to improve the results using state-of-the-art neural network sentiment analysis and more tweet data. This project also attempts to perform hourly predictions along with daily predictions in order to investigate the method further. The results show a decrease in accuracy compared to the previous project. The results also indicate that the neural network sentiment analysis improves the accuracy of the stock price development when compared to the baseline model under comparable conditions. / Detta examensarbete undersöker om ett företags aktievärdesutveckling kan förutspås genom att använda sig av den generella opinionen hos tweets skrivna om företaget. Examensarbetet utgår ifrån en model från ett tidigare projekt och försöker förbättra resultaten från denna genom att använda sig av dels state-of-the-art sentimentanalys med neurala nätverk, dels mer tweet data. Examensarbetet undersöker både prognoser timvis samt dygnsvis för att undersöka metoden djupare. Resultaten tyder på en minskad träffsäkerhet jämfört med det tidigare projektet. Resultaten indikerar också att sentimentanalys med neurala nätverk förbättrar träffsäkerheten hos aktievärdesprognosen jämfört med tidigare sentimentanalysmetod givet jämförbara förutsättningar.

People have been trying to predict the stock marketsince its inception and financial investors have made it theirprofession. What makes predicting the stock market such ahard task is its seemingly random dependency on everythingfrom Elon Musks tweets to future earnings. Machine learninghandles this apparent randomness with ease and we will try itout by implementing a Hidden Markov Model. We will modeltwo different stocks, Tesla, Inc. and Coca-Cola Company, andtry using the forecasted prices as a template for a simple tradingalgorithm. We used an approach of calculating the log-likelihoodof preceding observations and correlated it with the log-likelihoodof all the preceding subsequences of equivalent size by turningthe time window by one day in the past. The results show thatmodeling two stocks of different volatility is possible, but usingthe result as a template for trading came back inconclusive withless than 50 percent successful trades for both of the modelledstocks. / Människor har försökt förutsäga aktiemarknaden sedan starten och finansiella investerare har gjort det till sitt yrke. Det som gör att förutsäga aktiemarknaden till en så svår uppgift är dess till synes slumpmässiga beroende av allt från Elon Musks tweets till framtida intäkter. Maskininlärning hanterar denna uppenbara slumpmässighet med lätthet och vi kommer att testa det genom att implementera en Hidden Markov-modell. Vi kommer att modellera två olika aktier, Tesla, Inc. och Coca-Cola Company, och försöka använda de prognostiserade priserna som bas för en enkel algoritm att handla på. Vi använde ett tillvägagångssätt för att beräkna log-sannolikheten för föregående observationer och korrelerade den med logsannolikheten för alla föregående följder av motsvarande storlek genom att vrida tidsfönstret med en dag tidigare. Resultaten visar att det är möjligt att modellera två aktier med olika volatilitet, men att använda resultatet som en mall för handel kom tillbaka de med mindre än 50 procent framgångsrika affärer för båda modellerna. / Kandidatexjobb i elektroteknik 2021, KTH, Stockholm

Financial markets, particularly capital and stock markets, play an important role in mobilizing and canalising the idle savings of individuals and institutions to the investment options where they are really required for productive purposes. The prediction of stock prices and returns is carried out in order to enhance the quality of investment decisions in stock markets, but it is considered to be tricky and complicates tasks as these prices behave in a random fashion and vary with time. Owing to the potential of returns and inherent risk factors in stock market returns. Various stock market prediction models and decision support systems such as Capital asset pricing model, the arbitrage pricing theory of Ross, the inter-temporal capital asset pricing model of Merton ,Fama and French five-factor model, and zero beta model to provide investors with an optimal forecast of stock prices and returns. In this research thesis, a stock market prediction model consisting of two parts is presented and discussed. The first is the three factors of the Fama and French model (FF) at the micro level to forecast the return of the portfolios on the Saudi Arabian Stock Exchange (SASE) and the second is a Value Based Management (VBM) model of decision-making. The latter is based on the expectations of shareholders and portfolio investors about taking investment decisions, and on the behaviour of stock prices using an accurate modern nonlinear technique in forecasting, known as Artificial Neural Networks (ANN). This study examined monthly data relating to common stocks from the listed companies of the Saudi Arabian Stock Exchange from January 2007 to December 2011. The stock returns were predicted using the linear form of asset pricing models (capital asset pricing model as well as Fama and French three factor model). In addition, non-linear models were also estimated by using various artificial neural network techniques, and adaptive neural fuzzy inference systems. Six portfolios of stock predictors are combined using: average, weighted average, and genetic algorithm optimized weighted average. Moreover, value-based management models were applied to the investment decision-making process in combination with stock prediction model results for both the shareholders’ perspective and the share prices’ perspective. The results from this study indicate that the ANN technique can be used to predict stock portfolio returns; the investment decisions and the behaviour of stock prices, optimized by the genetic algorithm weighted average, provided better results in terms of error and prediction accuracy compared to the simple linear form of stock price prediction models. The Fama and French model of stock prediction is better suited to Saudi Arabian Stock Exchange investment activities in comparison to the conventional capital assets pricing model. Moreover, the multi-stage type1 model, which is a combination of Fama and French predicted stock returns and a value-based management model, gives more accurate results for the stock market decision-making process for investment or divestment decisions, as well as for observing variation in and the behaviour of stock prices on the Saudi stock market. Furthermore, the study also designed a graphic user interface in order to simplify the decision-making process based upon Fama and French and value-based management, which might help Saudi investors to make investment decisions quickly and with greater precision. Finally, the study also gives some practical implications for investors and regulators, along with proposing future research in this area.