Spectral Analysis of Symmetric and Anti-Symmetric Pairwise Kernels

This work investigates theoretical properties of symmetric and anti-symmetric kernels. First chapters give an overview of the theory of kernels used in supervised machine learning. Central focus is on the regularized least squares algorithm, which is motivated as a problem of function reconstruction through an abstract inverse problem. Brief review of reproducing kernel Hilbert spaces shows how kernels define an implicit hypothesis space with multiple equivalent characterizations and how this space may be modified by incorporating prior knowledge. Mathematical results of the abstract inverse problem, in particular spectral properties, pseudoinverse and regularization are recollected and then specialized to kernels. Symmetric and anti-symmetric kernels are applied in relation learning problems which incorporate prior knowledge that the relation is symmetric or anti-symmetric, respectively. Theoretical properties of these kernels are proved in a draft this thesis is based on and comprehensively referenced here. These proofs show that these kernels can be guaranteed to learn only symmetric or anti-symmetric relations, and they can learn any relations relative to the original kernel modified to learn only symmetric or anti-symmetric parts. Further results prove spectral properties of these kernels, central result being a simple inequality for the the trace of the estimator, also called the effective dimension. This quantity is used in learning bounds to guarantee smaller variance.

Modelling the simultaneous effects of intellectual capital and knowledge management on the organisational performance of Finnish companies

The aim of this study was to contribute to the current knowledge-based theory by focusing on a research gap that exists in the empirically proven determination of the simultaneous but differentiable effects of intellectual capital (IC) assets and knowledge management (KM) practices on organisational performance (OP). The analysis was built on the past research and theoreticised interactions between the latent constructs specified using the survey-based items that were measured from a sample of Finnish companies for IC and KM and the dependent construct for OP determined using information available from financial databases. Two widely used and commonly recommended measures in the literature on management science, i.e. the return on total assets (ROA) and the return on equity (ROE), were calculated for OP. Thus the investigation of the relationship between IC and KM impacting OP in relation to the hypotheses founded was possible to conduct using objectively derived performance indicators. Using financial OP measures also strengthened the dynamic features of data needed in analysing simultaneous and causal dependences between the modelled constructs specified using structural path models. The estimates were obtained for the parameters of structural path models using a partial least squares-based regression estimator. Results showed that the path dependencies between IC and OP or KM and OP were always insignificant when analysed separate to any other interactions or indirect effects caused by simultaneous modelling and regardless of the OP measure used that was either ROA or ROE. The dependency between the constructs for KM and IC appeared to be very strong and was always significant when modelled simultaneously with other possible interactions between the constructs and using either ROA or ROE to define OP. This study, however, did not find statistically unambiguous evidence for proving the hypothesised causal mediation effects suggesting, for instance, that the effects of KM practices on OP are mediated by the IC assets. Due to the fact that some indication about the fluctuations of causal effects was assessed, it was concluded that further studies are needed for verifying the fundamental and likely hidden causal effects between the constructs of interest. Therefore, it was also recommended that complementary modelling and data processing measures be conducted for elucidating whether the mediation effects occur between IC, KM and OP, the verification of which requires further investigations of measured items and can be build on the findings of this study.

Capital structure of Finnish firms – Estimating the effects of firm characteristics on capital structure

One of the most disputable matters in the theory of finance has been the theory of capital structure. The seminal contributions of Modigliani and Miller (1958, 1963) gave rise to a multitude of studies and debates. Since the initial spark, the financial literature has offered two competing theories of financing decision: the trade-off theory and the pecking order theory. The trade-off theory suggests that firms have an optimal capital structure balancing the benefits and costs of debt. The pecking order theory approaches the firm capital structure from information asymmetry perspective and assumes a hierarchy of financing, with firms using first internal funds, followed by debt and as a last resort equity. This thesis analyses the trade-off and pecking order theories and their predictions on a panel data consisting 78 Finnish firms listed on the OMX Helsinki stock exchange. Estimations are performed for the period 2003–2012. The data is collected from Datastream system and consists of financial statement data. A number of capital structure characteristics are identified: firm size, profitability, firm growth opportunities, risk, asset tangibility and taxes, speed of adjustment and financial deficit. A regression analysis is used to examine the effects of the firm characteristics on capitals structure. The regression models were formed based on the relevant theories. The general capital structure model is estimated with fixed effects estimator. Additionally, dynamic models play an important role in several areas of corporate finance, but with the combination of fixed effects and lagged dependent variables the model estimation is more complicated. A dynamic partial adjustment model is estimated using Arellano and Bond (1991) first-differencing generalized method of moments, the ordinary least squares and fixed effects estimators. The results for Finnish listed firms show support for the predictions of profitability, firm size and non-debt tax shields. However, no conclusive support for the pecking-order theory is found. However, the effect of pecking order cannot be fully ignored and it is concluded that instead of being substitutes the trade-off and pecking order theory appear to complement each other. For the partial adjustment model the results show that Finnish listed firms adjust towards their target capital structure with a speed of 29% a year using book debt ratio.

Weak Form Tests of Efficient Market Hypothesis: Evidence from CEE-Countries and Russian Stock Markets

Tämän tutkielman tavoitteena on selvittää Venäjän, Slovakian, Tsekin, Romanian, Bulgarian, Unkarin ja Puolan osakemarkkinoiden heikkojen ehtojen tehokkuutta. Tämä tutkielma on kvantitatiivinen tutkimus ja päiväkohtaiset indeksin sulkemisarvot kerättiin Datastreamin tietokannasta. Data kerättiin pörssien ensimmäisestä kaupankäyntipäivästä aina vuoden 2006 elokuun loppuun saakka. Analysoinnin tehostamiseksi dataa tutkittiin koko aineistolla, sekä kahdella aliperiodilla. Osakemarkkinoiden tehokkuutta on testattu neljällä tilastollisella metodilla, mukaan lukien autokorrelaatiotesti ja epäparametrinen runs-testi. Tavoitteena on myös selvittääesiintyykö kyseisillä markkinoilla viikonpäiväanomalia. Viikonpäiväanomalian esiintymistä tutkitaan käyttämällä pienimmän neliösumman menetelmää (OLS). Viikonpäiväanomalia on löydettävissä kaikilta edellä mainituilta osakemarkkinoilta paitsi Tsekin markkinoilta. Merkittävää, positiivista tai negatiivista autokorrelaatiota, on löydettävissä kaikilta osakemarkkinoilta, myös Ljung-Box testi osoittaa kaikkien markkinoiden tehottomuutta täydellä periodilla. Osakemarkkinoiden satunnaiskulku hylätään runs-testin perusteella kaikilta muilta paitsi Slovakian osakemarkkinoilla, ainakin tarkastellessa koko aineistoa tai ensimmäistä aliperiodia. Aineisto ei myöskään ole normaalijakautunut minkään indeksin tai aikajakson kohdalla. Nämä havainnot osoittavat, että kyseessä olevat markkinat eivät ole heikkojen ehtojen mukaan tehokkaita

Market efficiency and calendar anomalies in the Russian stock market

Tämän tutkimuksen tarkoituksena on tarkastella esiintyykö Venäjän osakemarkkinoilla kalenterianomalioita. Tutkimus keskittyy Halloween-, kuukausi-, kuunvaihde-, viikonpäivä- ja juhlapäiväanomalioiden tarkasteluun. Tutkimusaineistona käytetään RTS (Russian Trading System) indeksiä. Tarkasteluaika alkaa 1. syyskuuta 1995 ja loppuu 31. joulukuuta 2005. Havaintojen kokonaismäärä on 2584. Tutkimusmenetelmänä käytetään pienimmän neliösumman menetelmää (OLS). Tutkimustulokset osoittavat, että Venäjän osakemarkkinoilla esiintyy Halloween-, kuunvaihde- ja viikonpäiväanomalioita. Sen sijaan kuukausi- ja juhlapäiväanomalioita ei tulosten mukaanesiinny Venäjän osakemarkkinoilla. Tulokset osoittavat lisäksi, että suurin osaanomalioista on merkittävämpiä nykyään kuin Venäjän osakemarkkinoiden ensimmäisinä vuosina. Näiden tulosten perusteella voidaan todeta, että Venäjän osakemarkkinat eivät ole vielä tehokkaat.

Testing Weak-Form Efficiency of the Chinese Stock Market

Tämän tutkielman tavoitteena on tarkastella Kiinan osakemarkkinoiden tehokkuutta ja random walk -hypoteesin voimassaoloa. Tavoitteena on myös selvittää esiintyykö viikonpäiväanomalia Kiinan osakemarkkinoilla. Tutkimusaineistona käytetään Shanghain osakepörssin A-sarjan,B-sarjan ja yhdistelmä-sarjan ja Shenzhenin yhdistelmä-sarjan indeksien päivittäisiä logaritmisoituja tuottoja ajalta 21.2.1992-30.12.2005 sekä Shenzhenin osakepörssin A-sarjan ja B-sarjan indeksien päivittäisiä logaritmisoituja tuottoja ajalta 5.10.1992-30.12.2005. Tutkimusmenetelminä käytetään neljä tilastollista menetelmää, mukaan lukien autokorrelaatiotestiä, epäparametrista runs-testiä, varianssisuhdetestiä sekä Augmented Dickey-Fullerin yksikköjuuritestiä. Viikonpäiväanomalian esiintymistä tutkitaan käyttämällä pienimmän neliösumman menetelmää (OLS). Testejä tehdään sekä koko aineistolla että kolmella erillisellä ajanjaksolla. Tämän tutkielman empiiriset tulokset tukevat aikaisempia tutkimuksia Kiinan osakemarkkinoiden tehottomuudesta. Lukuun ottamatta yksikköjuuritestien saatuja tuloksia, autokorrelaatio-, runs- ja varianssisuhdetestien perusteella random walk-hypoteesi hylättiin molempien Kiinan osakemarkkinoiden kohdalla. Tutkimustulokset osoittavat, että molemmilla osakepörssillä B-sarjan indeksien käyttäytyminenon ollut huomattavasti enemmän random walk -hypoteesin vastainen kuin A-sarjan indeksit. Paitsi B-sarjan markkinat, molempien Kiinan osakemarkkinoiden tehokkuus näytti myös paranevan vuoden 2001 markkinabuumin jälkeen. Tutkimustulokset osoittavat myös viikonpäiväanomalian esiintyvän Shanghain osakepörssillä, muttei kuitenkaan Shenzhenin osakepörssillä koko tarkasteluajanjaksolla.

Kalenterianomaliat pohjoismaisilla sähkömarkkinoilla

Osakemarkkinoilta on jo useiden vuosien ajan julkaistu lukuisia tutkimuksia, joissa on esitetty havaintoja ajallisesta säännönmukaisuudesta osakkeiden hinnoissa, joita ei pystytä selittämään markkinakohtaisilla fundamenteilla. Nämä niin kutsutut kalenterianomaliat esiintyvät tyypillisesti ajallisissa käännepisteissä, kuten vuoden, kuukauden tai viikon vaihtuessa seuraavaksi. Myös erilaisten katkosten, kuten juhlapyhien, kaupankäyntirutiineissa on havaittu aiheuttavan anomalioita. Tutkimuksen tavoitteena oli tutkia osakemarkkinoilla havaittujen kalenterianomalioiden esiintymistä pohjoismaisilla sähkömarkkinoilla. Tutkitut anomaliat olivat viikonpäivä- kuukausi-, kuunvaihde- ja juhlapyhäanomalia. Näiden lisäksi tutkittiin tuottojen käyttäytymistä optioiden erääntymispäivien läheisyydessä. Yksittäisten tuotteiden sijasta tarkastelut suoritettiin sesonki- ja kvartaalituotteista muodostetuilla vuosituotteilla. Testauksessa käytettiin pienimmän neliösumman menetelmää, huomioidenheteroskedastisuuden, autokorrelaation ja multikollineaarisuuden vaikutukset. Pelkkien kalenterimuuttujien lisäksi testit suoritettiin regressiomalleilla, joissa lisäselittäjinä käytettiin spot-hintaa, päästöoikeuden hintaa ja/tai sade-ennusteita. Tarkastelujakso koostui vuosista 1998-2006.

Direct torque control of permanent magnet synchronous machines - analysis and implementation

Thedirect torque control (DTC) has become an accepted vector control method besidethe current vector control. The DTC was first applied to asynchronous machines,and has later been applied also to synchronous machines. This thesis analyses the application of the DTC to permanent magnet synchronous machines (PMSM). In order to take the full advantage of the DTC, the PMSM has to be properly dimensioned. Therefore the effect of the motor parameters is analysed taking the control principle into account. Based on the analysis, a parameter selection procedure is presented. The analysis and the selection procedure utilize nonlinear optimization methods. The key element of a direct torque controlled drive is the estimation of the stator flux linkage. Different estimation methods - a combination of current and voltage models and improved integration methods - are analysed. The effect of an incorrect measured rotor angle in the current model is analysed andan error detection and compensation method is presented. The dynamic performance of an earlier presented sensorless flux estimation method is made better by improving the dynamic performance of the low-pass filter used and by adapting the correction of the flux linkage to torque changes. A method for the estimation ofthe initial angle of the rotor is presented. The method is based on measuring the inductance of the machine in several directions and fitting the measurements into a model. The model is nonlinear with respect to the rotor angle and therefore a nonlinear least squares optimization method is needed in the procedure. A commonly used current vector control scheme is the minimum current control. In the DTC the stator flux linkage reference is usually kept constant. Achieving the minimum current requires the control of the reference. An on-line method to perform the minimization of the current by controlling the stator flux linkage reference is presented. Also, the control of the reference above the base speed is considered. A new estimation flux linkage is introduced for the estimation of the parameters of the machine model. In order to utilize the flux linkage estimates in off-line parameter estimation, the integration methods are improved. An adaptive correction is used in the same way as in the estimation of the controller stator flux linkage. The presented parameter estimation methods are then used in aself-commissioning scheme. The proposed methods are tested with a laboratory drive, which consists of a commercial inverter hardware with a modified software and several prototype PMSMs.

New kernel functions and learning methods for text and data mining

Recent advances in machine learning methods enable increasingly the automatic construction of various types of computer assisted methods that have been difficult or laborious to program by human experts. The tasks for which this kind of tools are needed arise in many areas, here especially in the fields of bioinformatics and natural language processing. The machine learning methods may not work satisfactorily if they are not appropriately tailored to the task in question. However, their learning performance can often be improved by taking advantage of deeper insight of the application domain or the learning problem at hand. This thesis considers developing kernel-based learning algorithms incorporating this kind of prior knowledge of the task in question in an advantageous way. Moreover, computationally efficient algorithms for training the learning machines for specific tasks are presented. In the context of kernel-based learning methods, the incorporation of prior knowledge is often done by designing appropriate kernel functions. Another well-known way is to develop cost functions that fit to the task under consideration. For disambiguation tasks in natural language, we develop kernel functions that take account of the positional information and the mutual similarities of words. It is shown that the use of this information significantly improves the disambiguation performance of the learning machine. Further, we design a new cost function that is better suitable for the task of information retrieval and for more general ranking problems than the cost functions designed for regression and classification. We also consider other applications of the kernel-based learning algorithms such as text categorization, and pattern recognition in differential display. We develop computationally efficient algorithms for training the considered learning machines with the proposed kernel functions. We also design a fast cross-validation algorithm for regularized least-squares type of learning algorithm. Further, an efficient version of the regularized least-squares algorithm that can be used together with the new cost function for preference learning and ranking tasks is proposed. In summary, we demonstrate that the incorporation of prior knowledge is possible and beneficial, and novel advanced kernels and cost functions can be used in algorithms efficiently.

Learning Preferences with Kernel-Based Methods

Learning of preference relations has recently received significant attention in machine learning community. It is closely related to the classification and regression analysis and can be reduced to these tasks. However, preference learning involves prediction of ordering of the data points rather than prediction of a single numerical value as in case of regression or a class label as in case of classification. Therefore, studying preference relations within a separate framework facilitates not only better theoretical understanding of the problem, but also motivates development of the efficient algorithms for the task. Preference learning has many applications in domains such as information retrieval, bioinformatics, natural language processing, etc. For example, algorithms that learn to rank are frequently used in search engines for ordering documents retrieved by the query. Preference learning methods have been also applied to collaborative filtering problems for predicting individual customer choices from the vast amount of user generated feedback. In this thesis we propose several algorithms for learning preference relations. These algorithms stem from well founded and robust class of regularized least-squares methods and have many attractive computational properties. In order to improve the performance of our methods, we introduce several non-linear kernel functions. Thus, contribution of this thesis is twofold: kernel functions for structured data that are used to take advantage of various non-vectorial data representations and the preference learning algorithms that are suitable for different tasks, namely efficient learning of preference relations, learning with large amount of training data, and semi-supervised preference learning. Proposed kernel-based algorithms and kernels are applied to the parse ranking task in natural language processing, document ranking in information retrieval, and remote homology detection in bioinformatics domain. Training of kernel-based ranking algorithms can be infeasible when the size of the training set is large. This problem is addressed by proposing a preference learning algorithm whose computation complexity scales linearly with the number of training data points. We also introduce sparse approximation of the algorithm that can be efficiently trained with large amount of data. For situations when small amount of labeled data but a large amount of unlabeled data is available, we propose a co-regularized preference learning algorithm. To conclude, the methods presented in this thesis address not only the problem of the efficient training of the algorithms but also fast regularization parameter selection, multiple output prediction, and cross-validation. Furthermore, proposed algorithms lead to notably better performance in many preference learning tasks considered.

Advanced chemometric methods: Applicability on industrial data

Recent years have produced great advances in the instrumentation technology. The amount of available data has been increasing due to the simplicity, speed and accuracy of current spectroscopic instruments. Most of these data are, however, meaningless without a proper analysis. This has been one of the reasons for the overgrowing success of multivariate handling of such data. Industrial data is commonly not designed data; in other words, there is no exact experimental design, but rather the data have been collected as a routine procedure during an industrial process. This makes certain demands on the multivariate modeling, as the selection of samples and variables can have an enormous effect. Common approaches in the modeling of industrial data are PCA (principal component analysis) and PLS (projection to latent structures or partial least squares) but there are also other methods that should be considered. The more advanced methods include multi block modeling and nonlinear modeling. In this thesis it is shown that the results of data analysis vary according to the modeling approach used, thus making the selection of the modeling approach dependent on the purpose of the model. If the model is intended to provide accurate predictions, the approach should be different than in the case where the purpose of modeling is mostly to obtain information about the variables and the process. For industrial applicability it is essential that the methods are robust and sufficiently simple to apply. In this way the methods and the results can be compared and an approach selected that is suitable for the intended purpose. Differences in data analysis methods are compared with data from different fields of industry in this thesis. In the first two papers, the multi block method is considered for data originating from the oil and fertilizer industries. The results are compared to those from PLS and priority PLS. The third paper considers applicability of multivariate models to process control for a reactive crystallization process. In the fourth paper, nonlinear modeling is examined with a data set from the oil industry. The response has a nonlinear relation to the descriptor matrix, and the results are compared between linear modeling, polynomial PLS and nonlinear modeling using nonlinear score vectors.

Testing the Random Walk Hypothesis on the Baltic Stock Markets

The purpose of the thesis is to analyze whether the returns of general stock market indices of Estonia, Latvia and Lithuania follow the random walk hypothesis (RWH), and in addition, whether they are consistent with the weak-form efficiency criterion. Also the existence of the day-of-the-week anomaly is examined in the same regional markets. The data consists of daily closing quotes of the OMX Tallinn, Riga and Vilnius total return indices for the sample period from January 3, 2000 to August 28, 2009. Moreover, the full sample period is also divided into two sub-periods. The RWH is tested by applying three quantitative methods (i.e. the Augmented Dickey-Fuller unit root test, serial correlation test and non-parametric runs test). Ordinary Least Squares (OLS) regression with dummy variables is employed to detect the day-of-the-week anomalies. The random walk hypothesis (RWH) is rejected in the Estonian and Lithuanian stock markets. The Latvian stock market exhibits more efficient behaviour, although some evidence of inefficiency is also found, mostly during the first sub-period from 2000 to 2004. Day-of-the-week anomalies are detected on every stock market examined, though no longer during the later sub-period.

The influence of crystallization conditions on the filtration characteristics of sulphathiazole suspensions

Cooling crystallization is one of the most important purification and separation techniques in the chemical and pharmaceutical industry. The product of the cooling crystallization process is always a suspension that contains both the mother liquor and the product crystals, and therefore the first process step following crystallization is usually solid-liquid separation. The properties of the produced crystals, such as their size and shape, can be affected by modifying the conditions during the crystallization process. The filtration characteristics of solid/liquid suspensions, on the other hand, are strongly influenced by the particle properties, as well as the properties of the liquid phase. It is thus obvious that the effect of the changes made to the crystallization parameters can also be seen in the course of the filtration process. Although the relationship between crystallization and filtration is widely recognized, the number of publications where these unit operations have been considered in the same context seems to be surprisingly small. This thesis explores the influence of different crystallization parameters in an unseeded batch cooling crystallization process on the external appearance of the product crystals and on the pressure filtration characteristics of the obtained product suspensions. Crystallization experiments are performed by crystallizing sulphathiazole (C9H9N3O2S2), which is a wellknown antibiotic agent, from different mixtures of water and n-propanol in an unseeded batch crystallizer. The different crystallization parameters that are studied are the composition of the solvent, the cooling rate during the crystallization experiments carried out by using a constant cooling rate throughout the whole batch, the cooling profile, as well as the mixing intensity during the batch. The obtained crystals are characterized by using an automated image analyzer and the crystals are separated from the solvent through constant pressure batch filtration experiments. Separation characteristics of the suspensions are described by means of average specific cake resistance and average filter cake porosity, and the compressibilities of the cakes are also determined. The results show that fairly large differences can be observed between the size and shape of the crystals, and it is also shown experimentally that the changes in the crystal size and shape have a direct impact on the pressure filtration characteristics of the crystal suspensions. The experimental results are utilized to create a procedure that can be used for estimating the filtration characteristics of solid-liquid suspensions according to the particle size and shape data obtained by image analysis. Multilinear partial least squares regression (N-PLS) models are created between the filtration parameters and the particle size and shape data, and the results presented in this thesis show that relatively obvious correlations can be detected with the obtained models.