Investment using evolutionary learning methods and technical rules

In this paper, I propose a genetic learning approach to generate technical trading systems for stock timing. The most informative technical indicators are selected from a set of almost 5000 signals by a multi-objective genetic algorithm with variable string length. Successively, these signals are combined into a unique trading signal by a learning method. I test the expert weighting solution obtained by the plurality voting committee, the Bayesian model averaging and Boosting procedures with data from the S&P 500 Composite Index, in three market phases, up-trend, down-trend and sideways-movements, covering the period 2000–2006. Computational results indicate that the near-optimal set of rules varies among market phases but presents stable results and is able to reduce or eliminate losses in down-trend periods.     

A strategy based on mean reverting property of markets and applications to foreign exchange trading with trailing stops

We propose a strategy for automated trading, outline theoretical justification of the profitability of this strategy, and overview the backtesting results in application to foreign currencies trading. The proposed methodology relies on the assumption that processes reflecting the dynamics of currency exchange rates are in a certain sense similar to the class of Ornstein–Uhlenbeck processes and exhibit the mean reverting property. In order to describe the quantitative characteristics of the projected return of the strategy, we derive the explicit expression for the running maximum of the Ornstein–Uhlenbeck process stopped at maximum drawdown and look at the correspondence between derived characteristics and the observed ones. Copyright © 2017 John Wiley & Sons, Ltd.     

The application of neural networks to predict abnormal stock returns using insider trading data

Until now, data mining statistical techniques have not been used to improve the prediction of abnormal stock returns using insider trading data. Consequently, an investigation using neural network analysis was initiated. The research covered 343 companies for a period of 4½ years. Study findings revealed that the prediction of abnormal returns could be enhanced in the following ways: (1) extending the time of the future forecast up to 1 year; (2) increasing the period of back aggregated data; (3) narrowing the assessment to certain industries such as electronic equipment and business services and (4) focusing on small and midsize rather than large companies. Copyright © 2002 John Wiley & Sons, Ltd.     

Modeling stock index returns by means of partial least‐squares methods: An out‐of‐sample analysis for three stock markets

We analyze the underlying economic forces of the stock markets in Germany, the U.K. and the U.S. Identifying a number of variables evincing return predictability, we follow a partial least‐squares (PLS) approach to combine these observables into a few latent factors. Conditional on European markets, our findings indicate (i) superior prediction performance of PLS‐based schemes in comparison with both, a random walk and a first‐order autoregressive benchmark model, (ii) consistent profitable trading on the German and British market, (iii) profitable linear forecast combinations, (iv) the U.S. stock market is diagnosed as informationally efficient. Copyright © 2010 John Wiley & Sons, Ltd.     

Deep learning for spatio‐temporal modeling: Dynamic traffic flows and high frequency trading

Deep learning applies hierarchical layers of hidden variables to construct nonlinear high dimensional predictors. Our goal is to develop and train deep learning architectures for spatio‐temporal modeling. Training a deep architecture is achieved by stochastic gradient descent and dropout for parameter regularization with a goal of minimizing out‐of‐sample predictive mean squared error. To illustrate our methodology, we first predict the sharp discontinuities in traffic flow data, and secondly, we develop a classification rule to predict short‐term futures market prices using order book depth. Finally, we conclude with directions for future research.     

Subset-conjunctive rules for breast cancer diagnosis

The objective of this study was to distinguish within a population of patients with and without breast cancer. The study was based on the University of Wisconsin's dataset of 569 patients, of whom 212 were subsequently found to have breast cancer. A subset-conjunctive model, which is related to Logical Analysis of Data, is described to distinguish between the two groups of patients based on the results of a non-invasive procedure called Fine Needle Aspiration, which is often used by physicians before deciding on the need for a biopsy. We formulate the problem of inferring subset-conjunctive rules as a 0-1 integer program, show that it is NP-Hard, and prove that it admits no polynomial-time constant-ratio approximation algorithm. We examine the performance of a randomized algorithm, and of randomization using LP rounding. In both cases, the expected performance ratio is arbitrarily bad. We use a deterministic greedy algorithm to identify a Pareto-efficient set of subset-conjunctive rules; describe how the rules change with a re-weighting of the type-I and type-II errors; how the best rule changes with the subset size; and how much of a tradeoff is required between the two types of error as one selects a more stringent or more lax classification rule. An important aspect of the analysis is that we find a sequence of closely related efficient rules, which can be readily used in a clinical setting because they are simple and have the same structure as the rules currently used in clinical diagnosis.