Kernel-Based Ranking. Methods for Learning and Performance Estimation

Machine learning provides tools for automated construction of predictive models in data intensive areas of engineering and science. The family of regularized kernel methods have in the recent years become one of the mainstream approaches to machine learning, due to a number of advantages the methods share. The approach provides theoretically well-founded solutions to the problems of under- and overfitting, allows learning from structured data, and has been empirically demonstrated to yield high predictive performance on a wide range of application domains. Historically, the problems of classification and regression have gained the majority of attention in the field. In this thesis we focus on another type of learning problem, that of learning to rank. In learning to rank, the aim is from a set of past observations to learn a ranking function that can order new objects according to how well they match some underlying criterion of goodness. As an important special case of the setting, we can recover the bipartite ranking problem, corresponding to maximizing the area under the ROC curve (AUC) in binary classification. Ranking applications appear in a large variety of settings, examples encountered in this thesis include document retrieval in web search, recommender systems, information extraction and automated parsing of natural language. We consider the pairwise approach to learning to rank, where ranking models are learned by minimizing the expected probability of ranking any two randomly drawn test examples incorrectly. The development of computationally efficient kernel methods, based on this approach, has in the past proven to be challenging. Moreover, it is not clear what techniques for estimating the predictive performance of learned models are the most reliable in the ranking setting, and how the techniques can be implemented efficiently. The contributions of this thesis are as follows. First, we develop RankRLS, a computationally efficient kernel method for learning to rank, that is based on minimizing a regularized pairwise least-squares loss. In addition to training methods, we introduce a variety of algorithms for tasks such as model selection, multi-output learning, and cross-validation, based on computational shortcuts from matrix algebra. Second, we improve the fastest known training method for the linear version of the RankSVM algorithm, which is one of the most well established methods for learning to rank. Third, we study the combination of the empirical kernel map and reduced set approximation, which allows the large-scale training of kernel machines using linear solvers, and propose computationally efficient solutions to cross-validation when using the approach. Next, we explore the problem of reliable cross-validation when using AUC as a performance criterion, through an extensive simulation study. We demonstrate that the proposed leave-pair-out cross-validation approach leads to more reliable performance estimation than commonly used alternative approaches. Finally, we present a case study on applying machine learning to information extraction from biomedical literature, which combines several of the approaches considered in the thesis. The thesis is divided into two parts. Part I provides the background for the research work and summarizes the most central results, Part II consists of the five original research articles that are the main contribution of this thesis.

Performance characteristic of moving average trading rules in the Finnish stock market

The thesis examines the profitability of DMAC trading rules in the Finnish stock market over the 1996-2012 period. It contributes to the existing technical analysis literature by comparing for the first time the performance of DMAC strategies based on individual stock trading portfolios to the performance of index trading strategies based on the trading on the index (OMX Helsinki 25) that consists of the same stocks. Besides, the market frictions including transaction costs and taxes are taken into account, and the results are reported from both institutional and individual investor’s perspective. Performance characteristic of DMAC rules are evaluated by simulating 19,900 different trading strategies in total for two non- overlapping 8-year sub-periods, and decomposing the full-sample-period performance of DMAC trading strategies into distinct bullish- and bearish-period performances. The results show that the best DMAC rules have predictive power on future price trends, and these rules are able to outperform buy-and-hold strategy. Although the performance of the DMAC strategies is highly dependent on the combination of moving average lengths, the best DMAC rules of the first sub-period have also performed well during the latter sub-period in the case of individual stock trading strategies. According to the results, the outperformance of DMAC trading rules over buy-and-hold strategy is mostly attributed to their superiority during the bearish periods, and particularly, during stock market crashes.

Insiders´ views as a predictor of stock performance : can insider trading be used to predict stock performance on the United States stock market?

This thesis studies the possibility of using information on insiders’ transactions to forecast future stock returns after the implementation of Sarbanes Oxley Act in July 2003. Insider transactions between July 2003 and August 2009 are analysed with regression tests to identify the relationships between insiders’ transactions and future stock returns. This analysis is complemented with rudimentary bootstrapping procedures to verify the robustness of the findings. The underlying assumption of the thesis is that insiders constantly receive pieces of information that indicate future performance of the company. They may not be allowed to trade on large and tangible pieces of information but they can trade on accumulation of smaller, intangible pieces of information. Based on the analysis in the thesis insiders’ profits were found not to differ from the returns from broad stock index. However, their individual transactions were found to be linked to future stock returns. The initial model was found to be unstable but some of the predictive power could be sacrificed to achieve greater stability. Even after sacrificing some predictive power the relationship was significant enough to allow external investors to achieve abnormal profits after transaction costs and taxes. The thesis does not go into great detail about timing of transactions. Delay in publishing insiders’ transactions is not taken into account in the calculations and the closed windows are not studied in detail. The potential effects of these phenomena are looked into and they do not cause great changes in the findings. Additionally the remuneration policy of an insider or a company is not taken into account even though it most likely affects the trading patterns of insiders. Even with the limitations the findings offer promising opportunities for investors to improve their investment processes by incorporating additional information from insiders’ transaction into their decisions. The findings also raise questions on how insider trading should be regulated. Insiders achieve greater returns than other investors based on superior information. On the other hand, more efficient information transfer could warrant more lenient regulation. The fact that insiders’ returns are dominated by the large investment stake they maintain all the time in their own companies also speaks for more leniency. As Sarbanes Oxley Act considerably modified the insider trading landscape, this analysis provides information that has not been available before. The thesis also constitutes a thorough analysis of insider trading phenomenon which has previously been somewhat separated into several studies.

Stability and predictive ability of beta

The purpose of this paper is to examine the stability and predictive abilities of the beta coefficients of individual equities in the Finnish stock market. As beta is widely used in several areas of finance, including risk management, asset pricing and performance evaluation among others, it is important to understand its characteristics and find out whether its estimates can be trusted and utilized.