Doctor of Philosophy / This thesis is based on research on financial time series analysis using pattern recognition methods. The first part of this research focuses on univariate time series analysis using different pattern recognition methods. First, probabilities of basic patterns are used to represent the features of a section of time series. This feature can remove noise from the time series by statistical probability. It is experimentally proven that this feature is successful for pattern repeated time series. Second, a multiscale Gaussian gravity as a pattern relationship measurement which can describe the direction of the pattern relationship is introduced to pattern clustering. By searching for the Gaussian-gravity-guided nearest neighbour of each pattern, this clustering method can easily determine the boundaries of the clusters. Third, a method that unsupervised pattern classification can be transformed into multiscale supervised pattern classification by multiscale supervisory time series or multiscale filtered time series is presented. The second part of this research focuses on multivariate time series analysis using pattern recognition. A systematic method is proposed to find the independent variables of a group of share prices by time series clustering, principal component analysis, independent component analysis, and object recognition. The number of dependent variables is reduced and the multivariate time series analysis is simplified by time series clustering and principal component analysis. Independent component analysis aims to find the ideal independent variables of the group of shares. Object recognition is expected to recognize those independent variables which are similar to the independent components. This method provides a new clue to understanding the stock market and to modelling a large time series database.

Submitted by Ronaldo Cesar Dametto (rdametto@uol.com.br) on 2018-09-18T19:17:34Z No. of bitstreams: 1 Dissertação_Completa_Final.pdf: 2885777 bytes, checksum: 05b2d5417efbec72f927cf8a62eef3fb (MD5) / Approved for entry into archive by Lucilene Cordeiro da Silva Messias null (lubiblio@bauru.unesp.br) on 2018-09-20T12:19:07Z (GMT) No. of bitstreams: 1 dametto_rc_me_bauru.pdf: 2877027 bytes, checksum: cee33d724090a01372e1292109af2ce9 (MD5) / Made available in DSpace on 2018-09-20T12:19:07Z (GMT). No. of bitstreams: 1 dametto_rc_me_bauru.pdf: 2877027 bytes, checksum: cee33d724090a01372e1292109af2ce9 (MD5) Previous issue date: 2018-08-06 / O aprendizado de máquina vem sendo utilizado em diferentes segmentos da área financeira, como na previsão de preços de ações, mercado de câmbio, índices de mercado e composição de carteira de investimento. Este trabalho busca comparar e combinar três tipos de algoritmos de aprendizagem de máquina, mais especificamente, o método Ensemble de Redes Neurais Artificias com as redes Multilayer Perceptrons (MLP), auto-regressiva com entradas exógenas (NARX) e Long Short-Term Memory (LSTM) para predição do Índice Bovespa. A amostra da série do Ibovespa foi obtida pelo Yahoo!Finance no período de 04 de janeiro de 2010 a 28 de dezembro de 2017, de periodicidade diária. Foram utilizadas as séries temporais referentes a cotação do Dólar, além de indicadores numéricos da Análise Técnica como variáveis independentes para compor a predição. Os algoritmos foram desenvolvidos através da linguagem Python usando framework Keras. Para avaliação dos algoritmos foram utilizadas as métricas de desempenho MSE, RMSE e MAPE, além da comparação entre as previsões obtidas e os valores reais. Os resultados das métricas indicam bom desempenho de predição pelo modelo Ensemble proposto, obtendo 70% de acerto no movimento do índice, porém, não conseguiu atingir melhores resultados que as redes MLP e NARX, ambas com 80% de acerto. / Different segments of the financial area, such as the forecast of stock prices, the foreign exchange market, the market indices and the composition of investment portfolio, use machine learning. This work aims to compare and combine two types of machine learning algorithms, the Artificial Neural Network Ensemble method with Multilayer Perceptrons (MLP), auto-regressive with exogenous inputs (NARX) and Long Short-Term Memory (LSTM) for prediction of the Bovespa Index. The Bovespa time series samples were obtained daily, using Yahoo! Finance, from January 4th, 2010 to December 28th, 2017. Dollar quotation, Google trends and numerical indicators of the Technical Analysis were used as independent variables to compose the prediction. The algorithms were developed using Python and Keras framework. Finally, in order to evaluate the algorithms, the MSE, RMSE and MAPE performance metrics, as well as the comparison between the obtained predictions and the actual values, were used. The results of the metrics indicate good prediction performance by the proposed Ensemble model, obtaining a 70% accuracy in the index movement, but failed to achieve better results than the MLP and NARX networks, both with 80% accuracy.

Sabe-se que, na literatura, existem muitos modelos para se fazer previsão para séries temporais financeiras. Sabe-se também que não há um modelo perfeito e que os mais utilizados atualmente são os modelos de redes neurais recorrentes e os da família GARCH. Referências internacionais apontam que existe uma técnica de medição de uma janela temporal para se identificar o tipo de comportamento existente em uma série temporal; tal técnica é conhecida como Expoente de Hurst. É uma medida que qualifica a série como persistente ou anti-persistente. Este trabalho analisou se o Expoente de Hurst, interfere na qualidade das previsões feitas com o modelo de redes neurais recorrentes com e sem o uso do filtro de ondaletas, utilizando os preços diários das principais commodities, ações negociadas no mercado e a taxa de câmbio. no período de janeiro de 1998 a dezembro de 2010. Com a pesquisa observa-se, na maioria dos casos, há uma possível melhora na qualidade das previsões para as séries antipersistentes. / It is known that there are a lot of models to forecast financial time series. It is known, also, that there is not a perfect model and the most used nowadays are the Recurrent Neural Network models and those from the GARCH family. International references point to a technique of measurement using windowing in order to identify the kind of behavior that is present in time series. This technique is known as Hurst Exponent. It is a measure that qualifies the time series as persistent or anti-persistent. This work analyzed if the Hurst Exponent interferes in the quality of the forecasts made with the Neural Network models with and without the wavelet filter, using the main commodities, stock prices, Ibovespa index and the Dollar/Real exchange rate in the period ranging from January 1998 to December 2010. The initial conclusions concerning the models worked out are positives.

Neste trabalho estudamos diferentes variantes dos modelos GARCH quando consideramos a chegada da informação heterogênea sob a forma de dados de alta freqüência. Este tipo de modelos, conhecidos como HARCH(n), foram introduzidos por Muller et al. (1997). Para entender a necessidade de incorporar esta característica da heterogeneidade da informação, estudamos o problema da agregação temporal para processos GARCH e a modelagem destes em dados de alta freqüência e veremos quais são as desvantagens destes modelos e o porquê da necessidade de corrigi-lo. Propusemos um novo modelo que leva em conta a heterogeneidade da informação do mercado financeiro e a memória longa da volatilidade, generalizando assim o modelo proposto por Müller et al.(1997), e estudamos algumas das propriedades teóricas do modelo proposto. Utilizamos estimação via máxima verossimilhança e amostrador de Griddy-Gibbs, e para avaliar o desempenho destes métodos realizamos diversas simulações. Também fizemos aplicações a duas séries de alta freqüência, a taxa de câmbio Euro- Dólar e o índice Ibovespa. Uma modificação ao algoritmo de Griddy-Gibbs foi proposta, para ter uma janela móvel de pontos, para a estimação das distribuições condicionais, a cada iteração. Este procedimento foi validado pela proximidade das estimações com a técnica de máxima verossimilhança. Disponibilizaremos algumas bibliotecas para o pacote S-Plus em que as análises descritas neste trabalho poderão ser reproduzidas. Informações relativas a tais bibliotecas estarão disponíveis na página Web http://www.ime.usp.br/~ruilova. / In this work we study different variants of GARCH models to analyze the arrival of heterogeneous information in high frequency data. These models, known as HARCH(*n*) models, were introduced by Müller et al.(1997). To understand the necessity to incorporate this characteristic, heterogeneous information, we study temporal aggregation on GARCH processes for high frequency data, and show some problems in the application of these models and the reason why it is necessary to develop new models. We propose a new model, that incorporates the heterogeneous information present in the financial market and the long memory of the volatility, generalizing the model considered by Müller et al.(1997). We propose to estimate the model via maximum likelihood and Griddy-Gibbs sampler. To assess the performance of the suggested estimation procedures we perform some simulations and apply the methodology to two time series, namely the foreign exchange rate Euro-Dollar and the series of the Ibovespa index. A modification of the algorithm of Griddy-Gibbs sampler was proposed to have a grid of points in a mobile window, to estimate the condicional distributions, in each iteration. This was validated by the similar results between maximum likelihood and Griddy-Gibbs sampler estimates obtained. We implemented the methods described in this work creating some libraries for the SPlus package. Information concerning these libraries is available in the Web page http://www.ime.usp.br/~ruilova.

In this work we compare the prediction performance of three optimized technical indicators with a Support Vector Machine Neural Network. For the indicator part we picked the common used indicators: Relative Strength Index, Moving Average Convergence Divergence and Stochastic Oscillator. For the Support Vector Machine we used a radial-basis kernel function and regression mode. The techniques were applied on financial time series brought from the Swedish stock market. The comparison and the promising results should be of interest for both finance people using the techniques in practice, as well as software companies and similar considering to implement the techniques in their products.

This degree project aims to investigate different evaluation strategies for clustering methodsused to cluster multivariate financial time series. Clustering is a type of data mining techniquewith the purpose of partitioning a data set based on similarity to data points in the same cluster,and dissimilarity to data points in other clusters. By clustering the time series of mutual fundreturns, it is possible to help individuals select funds matching their current goals and portfolio. Itis also possible to identify outliers. These outliers could be mutual funds that have not beenclassified accurately by the fund manager, or potentially fraudulent practices. To determine which clustering method is the most appropriate for the current data set it isimportant to be able to evaluate different techniques. Using robust evaluation methods canassist in choosing the parameters to ensure optimal performance. The evaluation techniquesinvestigated are conventional internal validation measures, stability measures, visualizationmethods, and evaluation using domain knowledge about the data. The conventional internalvalidation methods and stability measures were used to perform model selection to find viableclustering method candidates. These results were then evaluated using visualization techniquesas well as qualitative analysis of the result. Conventional internal validation measures testedmight not be appropriate for model selection of the clustering methods, distance metrics, or datasets tested. The results often contradicted one another or suggested trivial clustering solutions,where the number of clusters is either 1 or equal to the number of data points in the data sets.Similarly, a stability validation metric called the stability index typically favored clustering resultscontaining as few clusters as possible. The only method used for model selection thatconsistently suggested clustering algorithms producing nontrivial solutions was the CLOSEscore. The CLOSE score was specifically developed to evaluate clusters of time series bytaking both stability in time and the quality of the clusters into account. We use cluster visualizations to show the clusters. Scatter plots were produced by applyingdifferent methods of dimension reduction to the data, Principal Component Analysis (PCA) andt-Distributed Stochastic Neighbor Embedding (t-SNE). Additionally, we use cluster evolutionplots to display how the clusters evolve as different parts of the time series are used to performthe clustering thus emphasizing the temporal aspect of time series clustering. Finally, the resultsindicate that a manual qualitative analysis of the clustering results is necessary to finely tune thecandidate clustering methods. Performing this analysis highlights flaws of the other validationmethods, as well as allows the user to select the best method out of a few candidates based onthe use case and the reason for performing the clustering.

Financial time-series are not uncommon to research in an academic context. This is possibly not only due to its challenging nature with high levels of noise and non-stationary data, but because of the endless possibilities of features and problem formulations it creates. Consequently, problem formulations range from classification and categorical tasks determining directional movements in the market to regression problems forecasting their actual values. These tasks are investigated with features consisting of data extracted from Twitter feeds to movements from external markets and technical indicators developed by investors. Cryptocurrencies are known for being evermore so volatile and unpredictable, resulting in institutional investors avoiding the market. In contrast, research in academia often applies state-of-the-art machine learning models without the industry’s knowledge of pre-processing. This thesis aims to lessen the gap between industry and academia by presenting a process from feature extraction and selection to forecasting through machine learning. The task involves how well the market movements can be forecasted and the individual features’ role in the predictions for a six-hours ahead regression task. To investigate the problem statement, a set of technical indicators and a feature selection algorithm were implemented. The data was collected from the exchange FTX and consisted of hourly data from Solana, Bitcoin, and Ethereum. Then, the features selected from the feature selection were used to train and evaluate an Autoregressive Integrated Moving Average (ARIMA) model, Prophet, a Long Short-Term Memory (LSTM) and a Transformer on the spread between the spot price and three months futures market for Solana. The features’ relevance was evaluated by calculating their permutation importance. It was found that there are indications of short-term predictability of the market through several forecasting models. Furthermore, the LSTM and ARIMA-GARCH performed best in a scenario of low volatility, while the LSTM outperformed the other models in times of higher volatility. Moreover, the investigations show indications of non-stationary. This phenomenon was not only found in the data as sequence but also in the relations between the features. These results show the importance of feature selection for a time frame relevant to the prediction window. Finally, the data displays a strong mean-reverting behaviour and is therefore relatively well-approximated by a naive walk. / Finansiella tidsserier är inte ovanliga att utforska i ett akademiskt sammanhang. Det beror troligen inte bara på dess utmanande karaktär med höga ljudnivåer och icke-stationära data, utan även till följd av de oändliga möjligheter till inmatning och problemformuleringar som det skapar. Följaktligen sträcker sig problemformuleringarna från klassificering och kategoriska uppgifter som bestämmer riktningsrörelser på marknaden till regressionsproblem som förutsäger deras faktiska värden. Dessa uppgifter undersöks med data extraherad från twitterflöden till rörelser från externa marknader och tekniska indikatorer utvecklade av investerare. Kryptovalutor är kända för att vara volatila och oförutsägbara till sin natur, vilket resulterar i att institutionella investerare undviker marknaden. I kontrast tillämpas forskning inom den akademiska världen ofta med avancerade maskininlärningsmodeller utan branschens typiska förbearbetningsarbete. Detta examensarbete syftar till att minska klyftan mellan industri och akademi genom att presentera en process från dataextraktion och urval till prognoser genom maskininlärning. Arbetet undersöker hur väl marknadsrörelserna kan prognostiseras och de enskilda variablernas roll i förutsägelserna för ett regressionsproblem som prognotiserar en sex timmar fram i tiden. Därmed implementerades en uppsättning tekniska indikatorer tillsammans med en algoritm för variabelanvändning. Datan samlades in från börsen FTX och bestod av timdata från Solana, Bitcoin och Ethereum. Sedan användes variablerna som valts för att träna och utvärdera en Autoregressive Integrated Moving Average (ARIMA)-modell, Prophet, en Long Short-Term Memory (LSTM) och en Transformer på skillnaden mellan spotpriset och tre månaders framtidsmarknad för Solana. Variablernas relevans utvärderades genom att beräkna deras vikt vid permutation. Slutsatsen är att det finns indikationer på kortsiktig förutsägbarhet av marknaden genom flera prognosmodeller. Vidare noterades det att LSTM och ARIMA-GARCH presterade bäst i ett scenario med låg volatilitet, medan LSTM överträffade de andra modellerna i vid högre volatilitet. Utöver detta visar undersökningarna indikationer på icke-stationäritet inte bara för datan i sig, utan också för relationerna mellan variablerna. Detta visar vikten av att välja variabler för en tidsram som är relevant för prediktionsfönstret. Slutligen visar tidsserien ett starkt medelåtergående beteende och är därför relativt väl approximerad av en naiv prediktionsmodell.

Séries temporais financeiras, como índices de mercado e preços de ativos, são produzidas por interações complexas dos agentes que participam do mercado. As propriedades fractais e multifractais destas séries fornecem evidências para detectar com antecedência a ocorrência de movimentos bruscos de mercado (crashes). Tais evidências são obtidas ao aplicar o conceito de Calor Específico Análogo C(q), proveniente da equivalência entre a Multifractalidade e Termodinâmica. Na proximidade de um crash, C(q) apresenta um ombro anômalo à direita de sua curva, enquanto que na ausência de um crash, possui o formato parecido com uma distribuição gaussiana. Com base neste comportamento, o presente trabalho propõe um novo indicador temporal IA(i), definido como a taxa de variação da área sob a curva de C(q). O indicador foi construído por intermédio de uma janela temporal de tamanho s que se movimenta ao longo da série, simulando a entrada de dados na série ao longo do tempo. A análise de IA(i) permite detectar com antecedência a ocorrência de grandes movimentos, como os famosos crashes de 1929 e 1987 para os índices Dow Jones, S&P500 e Nasdaq. Além disso, a análise simultânea de medidas como a Energia Livre, a Dimensão Multifractal e o Espectro Multifractal, sugerem que um crash de mercado se assemelha a uma transição de fase. A robustez do método para diferentes ativos e diferentes períodos de tempo, demonstra a importância dos resultados. Além disso, modelos estatísticos não lineares para a volatilidade foram empregados no trabalho para estudar grandes flutuações causadas por crashes e crises financeiras ao longo do tempo. / Financial time series such as market index and asset prices, are produced by complex interactions of agents that trade in the market. The fractal and multifractal properties of these series provides evidence for early detection of the occurrence of sudden market movements (crashes). This evidence is obtained by applying the concept of Analog Specific Heat C(q), from the equivalence between the Multifractal Analysis and Thermodynamics. In the vicinity of a crash, C(q) exhibits a shoulder at the right side of its curve, while in the absence of a crash, C(q) presents a form similar to a Gaussian distribution curve. Based on this behavior, it is proposed in this work a new temporal indicator IA(i) defined here as the area variation rate over the Specific Heat function. We have constructed the mentioned indicator from a window of data with the first points (size s), that moves throughout the series, simulating the actual input of data over time. The indicator IA(i) allows one detecting in advance the occurrence of large financial market movements, such as those occurred in 1929 and 1987 for the marked indexes Dow Jones, Nasdaq and S&P500. Moreover, the simultaneous analysis of measures such as the Free Energy, Multifractal Dimension and Multifractal Spectrum suggest that a market crash resembles a phase transition. The robustness of the method for others assets and different periods of time demonstrates the importance of the results. Moreover, nonlinear statistical models for volatility have been employed in the work to study large fluctuations caused by crashes and financial crises over time.

Séries temporais financeiras, como índices de mercado e preços de ativos, são produzidas por interações complexas dos agentes que participam do mercado. As propriedades fractais e multifractais destas séries fornecem evidências para detectar com antecedência a ocorrência de movimentos bruscos de mercado (crashes). Tais evidências são obtidas ao aplicar o conceito de Calor Específico Análogo C(q), proveniente da equivalência entre a Multifractalidade e Termodinâmica. Na proximidade de um crash, C(q) apresenta um ombro anômalo à direita de sua curva, enquanto que na ausência de um crash, possui o formato parecido com uma distribuição gaussiana. Com base neste comportamento, o presente trabalho propõe um novo indicador temporal IA(i), definido como a taxa de variação da área sob a curva de C(q). O indicador foi construído por intermédio de uma janela temporal de tamanho s que se movimenta ao longo da série, simulando a entrada de dados na série ao longo do tempo. A análise de IA(i) permite detectar com antecedência a ocorrência de grandes movimentos, como os famosos crashes de 1929 e 1987 para os índices Dow Jones, S&P500 e Nasdaq. Além disso, a análise simultânea de medidas como a Energia Livre, a Dimensão Multifractal e o Espectro Multifractal, sugerem que um crash de mercado se assemelha a uma transição de fase. A robustez do método para diferentes ativos e diferentes períodos de tempo, demonstra a importância dos resultados. Além disso, modelos estatísticos não lineares para a volatilidade foram empregados no trabalho para estudar grandes flutuações causadas por crashes e crises financeiras ao longo do tempo. / Financial time series such as market index and asset prices, are produced by complex interactions of agents that trade in the market. The fractal and multifractal properties of these series provides evidence for early detection of the occurrence of sudden market movements (crashes). This evidence is obtained by applying the concept of Analog Specific Heat C(q), from the equivalence between the Multifractal Analysis and Thermodynamics. In the vicinity of a crash, C(q) exhibits a shoulder at the right side of its curve, while in the absence of a crash, C(q) presents a form similar to a Gaussian distribution curve. Based on this behavior, it is proposed in this work a new temporal indicator IA(i) defined here as the area variation rate over the Specific Heat function. We have constructed the mentioned indicator from a window of data with the first points (size s), that moves throughout the series, simulating the actual input of data over time. The indicator IA(i) allows one detecting in advance the occurrence of large financial market movements, such as those occurred in 1929 and 1987 for the marked indexes Dow Jones, Nasdaq and S&P500. Moreover, the simultaneous analysis of measures such as the Free Energy, Multifractal Dimension and Multifractal Spectrum suggest that a market crash resembles a phase transition. The robustness of the method for others assets and different periods of time demonstrates the importance of the results. Moreover, nonlinear statistical models for volatility have been employed in the work to study large fluctuations caused by crashes and financial crises over time.

No description available.