Computational Intelligence in Economics and Finance

Forecasting is an instrumental tool for strategic decision- making in any business activity. Good forecasts can reduce the uncertainty about the future and, hence, help managers make better decisions. Virtually all statistical forecasting techniques depend on the continuity of historical data and time series and may not predict a discontinuous change in the business environment. Often times, this discontinuity is known to managers who then must rely on their judgment to make forecast adjustments. In this paper, we discuss the role of judgmental forecasting and take the problem of estimating future hotel room demand as a practical business application. Next, we propose IS-JFK: an intelligent system to support judgmental forecasting and knowledge of managers. To account for vagueness in the knowledge elicited from managers and the approximate nature of their reasoning, the system is built around fuzzy IF-THEN rules and uses fuzzy logic for decision inference. ISJFK supports two methods for forecast adjustments: 1) a direct approach and 2) an approach based on fuzzy intervention analysis. Actual data from a hotel property are used in some case-scenario simulations to illustrate the merits of the intelligent support system.

In an uncertain economic decision environment, an expert’s knowledge about discounting cash flows consists of a lot of vagueness instead of randomness. Cash amounts and interest rates are usually estimated by using educated guesses based on expected values or other statistical techniques to obtain them. Fuzzy numbers can capture the difficulties in estimating these parameters. Ill this chapter, the formulas for the analysis of fuzzy present value, fuzzy equivalent uniform annual value, fuzzy future value, fuzzy benefit-cost ratio, and fuzzy payback period are developed and given sonic numeric examples. Then the examined cash flows are expanded to geometric and trigonometric cash flows and using these cash flows fuzzy present value, fuzzy future value, and fuzzy annual value formulas are developed for both discrete compounding and continuous compounding. The fuzzy dynamic programming is applied to the situation where each investment in the set has the following characteristics: the amount to be invested has several possible values, and the rate of return varies with the amount invested. Each sum that may be invested represents a distinct level of investment, and the investment therefore has multiple levels. A fuzzy present worth based dynamic programming approach is used. A numeric example for a multilevel investment with fuzzy geometric cash flows is given. A computer software named FUZDYN is developed for various problems such as alternatives having different lives, different uniform cash flows, and different ranking methods.

The classification of a set of objects into predefined homogenous groups is a problem with major practical interest in many fields. Over the past two decades several non-parametric approaches have been developed to address the classification problem, originating from several scientific fields. This paper is focused on a rule induction approach based on the rough sets theory and the investigation of its performance as opposed to traditional multivariate statistical classification procedures, namely the linear discriminant analysis, the quadratic discriminant analysis and the logit analysis. For this purpose an extensive Monte Carlo simulation is conducted to examine the performance of these methods under different data conditions.

This chapter investigates the information content of SGX-DT (Singapore Exchange-Derivatives Trading Limited) MSCI (Morgan Stanley Capital International) Taiwan futures contracts and its underlying cash market during the non-cash-trading (NCT) period. Previous day’s cash market closing index and the grey forecasts by using the futures during the NCT period are used to forecast the 09:00 AM opening cash price index by the neural networks model. To demonstrate the effectiveness of our proposed method, the five-minute intraday data of spot and futures index from October 1, 1998 to December 31, 1999 was evaluated using the special neural network modeling. Analytic results demonstrate that the proposed model of integrating grey forecasts and neural networks outperforms the neural network model with previous day’s closing index as the input variable, the random walk model and AR,IMA forecasting. It therefore indicates that there is valuable information involved in the futures prices during the NCT period in forecasting the opening cash price index. Besides, grey forecasts provide a better initial solution that speeds up the learning procedure for the neural networks which in turn give better forecasting results.

This paper discusses the feasibility of using the support vector machine (SVM) to build empirical models of currency crises. The main idea is to develop an estimation algorithm, by training a model on a data set, which provide reasonably accurate model outputs. The proposed approach is illustrated to model currency crises in Venezuela.

This chapter deals with the application of saliency analysis to Support. Vector Machines (SVMs) for feature selection. The importance of feature is ranked by evaluating the sensitivity of the network output to the feature input in terms of the partial derivative. A systematic approach to remove irrelevant features based on the sensitivity is developed. Two simulated non-linear time series and five real financial time series are examined in the experiment. The simulation results show that that saliency analysis is effective in SVMs for identifying important features.

Using Self-organizing maps (SOM), tins paper formalizes chartists’ behavior in searching of patterns (charts). By applying a 6 by 6 two-dimensional SOM to a time series data of TAIEX (Taiwan Stock Index), 36 patterns are established. To see whether these 36 patterns transmit profitable signals, a “normalized” equity curve is drawn for each pattern up to 20 days after observing the pattern. Many of these equity curves are either monotonically increasing or decreasing, and none of them exhibits random fluctuation. Therefore, it is concluded that the patterns established by SOM can help us foresee the movement of stock index in the near future. We further test profitability performance by trading on these SOM-induced financial patterns. The equity-curve results show that SOM-induced trading strategy is able to outperform the buy-and-hold strategy in quite a significant period of time.

The Economic and Monetary Union (EMU) is the culmination of the European integration process from a financial perspective, whose main aim is the implantation of a one and only one currency for all the member states included in this integration project taking place in early 2002. To decide the relation of countries included in this phase it was established a group of economical regulations (macroeconomics rules) known as Convergence Criteria or Maastricht Criteria which must be fulfilled to guarantee the economic convergence among countries sharing the same currency. Nevertheless, these criteria are not enough to assure the effective convergence among states as far as enterprises is concerned, due to internal national differences in microeconomic structures, which affect competitiveness among firms and could distort a lot the effect of the union in favour of some countries and against others. The use of unsupervised artificial neural networks, specifically the employment of self-adaptive models based on Kohonen’s proposal, makes easier to analyze the microeconomic differences by getting a visual image of a concrete dispersion of the differences through two-dimensional topological maps.

A methodology on how to combine wavelets with Self-organizing Maps (SOM) for financial time-series visualisation and interpretation is presented. For the denoising of the stock time-series wavelet packets are used because of their optimal signal compression and denoising capabilities. The visualisation of transient shocks like crashes, in higher order wavelet coefficients is presented. The Self-organising Map Neural Network is introduced to aid the visualisation of the behaviour of the daily closing value of S&P 500 and the daily closing value of two example stocks. The features that are used for the visualisation are based on the wavelet coefficients of 32-day trading periods with daily sampling of the closing value. The trajectory formed on the U-matrix of SOM shows the evolution of the individual stock and indicator data and aids the detection of abrupt changes in the behaviour of the time-series.

Based on a algorithm for pattern matching in character strings, a pattern description language (PDL) is developed. The compilation of a regular expression, that conforms to the PDL, creates a non deterministic pattern matching machine (PMM) that can be used as a searching device for detecting sequential patterns or functional (statistical) relationships in multidimensional data. As an example, a chart pattern of ex ante unknown length is encoded and its occurrences are searched for in financial data.

This study proposes a temporal data mining method to discover qualitative and quantitative patterns in time series databases. The method performs discrete-valued time series (DIS) analysis on time series databases to search for any similarity and periodicity of patterns that are used for knowledge discovery. In our method there are three levels for mining patterns. At the first level, a structural search based on distance measure models is employed to find pattern structures; the second level performs a value-based search on the discovered patterns using a local polynomial analysis; the third level, based on hidden Markov models (HMMs), finds global patterns from a DTS set. As a result, similar and periodic patterns are successfully extracted. We demonstrate our method on the analysis of “Exchange Rate Patterns” between the U.S. dollar and Australian dollar.

Using Quinlan’s Cubist, this paper examines whether there is a consistent, interpretation of the efficient market hypothesis between financial econometrics and machine learning. In particular, we ask whether machine learning can be useful only in the case when the market is not efficient. Based on the forecasting performance of Cubist in our artificial returns, some evidences seems to support this consistent interpretation. However, there are a few cases whereby Cubist can beat the random walk even though the series is independent. As a result, we do not consider that the evidence is strong enough to convince one to give up his reliance on machine learning even though the efficient market hypothesis is sustained.

The purpose of this paper is to contribute to the debate on the relevance of non-linear predictors of high-frequency data in foreign exchange markets. To that end, we apply nearest-neighbour (NN) predictors, inspired by the literature on forecasting in non-linear dynamical systems, to exchange-rate series. The forecasting performance of univariate and multivariate versions of such NN predictors is first evaluated from the statistical point of view, using a battery of statistical tests. Secondly, we assess if NN predictors are capable of producing valuable economic signals in foreign exchange markets. The results show the potential usefulness of NN predictors not only as a helpful tool when forecast daily exchange data but also as a technical trading rules.

In this chapter, we shall review some early applications of genetic programming to financial data mining and knowledge discovery, including some analyses of its statistical behavior. These early applications are known as symbolic regression in GP. In this type of application, genetic programming is formally demonstrated as an engine searching for the hidden relationships among observations. Here, we find evidence of the closest step ever made toward the original motivation of John Holland’s invention of genetic algorithms: Instead of trying to write your programs to perform a task you don’t quite know how to do, evolve them.

This chapter introduces a new heuristics for solving the optimal consumption path in one-sector growth model, a typical stochastic dynamical optimization problem in economics. The proposed method avoids the ex-ante specification of a functional form for the policy function that solves the optimization problem. This novel approach has an advantage over other approaches like the Linear Quadratic Approximation (LQA) and the Parameterized Expectation (PE) methods. Instead, the functional form arises endogenously according to the characteristic of the problem the method is seeking to deal with. The heuristics combines Radial Basis Network (RBN) as a. representation of the potential solutions and an Evolutionary Strategy (ES) as a mechanism to prune the search space. Experiments were performed on different versions of a stochastic growth model and some satisfactory results were consequently obtained. In most cases the approximation obtained with the proposed method indeed outperforms the approximation reached by both the LQA and PE methods, based on not only one criterion but several different quality criteria.

Previous work has used neural networks to predict the rate of inflation in Taiwan using four measures of ‘money’ (simple sum and three divisia measures). In this work we develop a new approach that uses an evolutionary strategy as a predictive tool. This approach is simple to implement yet produces results that are favourable with the neural network predictions. Computational results are given.

This chapter successfully introduces the ants algorithm into the business failure prediction problem domain. The original ants algorithm is also modified and improved in both transition probability and pheromone trail update mechanism. The distinct advantages of this modified ants algorithm (MAA) consist of no special demand on the problem’s form, lower computer storage, and less CPU time for computation. The empirical results based on the real-world data demonstrate the effectiveness of its application to the business failure prediction problem domain and also show its advantages compared with RPA (recursive partition algorithm), DA (Discriminate Analysis) and GP (Genetic Programming).

We propose a knowledge discovery and knowledge management process for equity management institutions. We realize the process with a financial knowledge management system, FKMS, that is a system platform being able to convert various sources of data into the data warehouse, to retrieve data cubes based on different power users’ commands for subsequent valuation modeling or data mining applications. We then introduced a data mining solution for equity portfolio construction using the simulated annealing algorithm. Two data sets consist of small stocks ranging from 11/86 to 10/91 and from 6/93 to 5/96 are used. The corresponding rates of return of Russell 2000 index are collected as benchmarks for evaluation based on the Sharpe ratios and the turnover ratios. The result of the simulated annealing algorithm has shown to outperform the market index as well as the gradient maximization method.

When the Temporal Factor Analysis (TFA) model is used for classical Arbitrage Pricing Theory (APT) analysis in finance, it is necessary to perform white noise tests on the residual in order to validate the model adequacy. We carry out white noise tests and obtain results that provide assurance for further statistical analysis using the TFA model. We also explore empirically the relationship between macroeconomic time series and Gaussian statistically uncorrelated hidden factors recovered by TFA. Based on the statistical hypothesis test results, we conclude that each of the four economic time series is linearly related to the uncorrelated factors. Consequently, APT economic factors can be synthesized from those statistically uncorrelated factors.

In this paper we expand the well-known methodology of deriving cross spectra from autoregressive lag models to the class of time-varying parameter models. That enables us to analyse all frequency properties at all points in time (in contrast to the wavelet methodology, where one can only analyse certain frequencies at all points in time). This allows us to separate the evolution of an equilibrium from dynamic adjustments to it and the process of learning about it. Using these results, we analyse the behaviour of short terni interest rates in response to monetary policy changes in Britain during and following the European Exchange Rate Mechanism (ERM) crisis of 1992/3. We hod that the British monetary transmission mechanisms is very stable even in an event like the ERM crisis. This is possible due to the adjustment of the risk premium.

In this paper, we incorporate a Self-organizing state-space model into the Markov-switching model, and propose the Self-organizing Markov-switching state-space (SOMS) model. The approximate Monte Carlo filter is applied for state estimation, including the latent Markov chain, of this model. The SOMS model allows us to evaluate complex systems. We apply it to an analysis of international transmission of business cycles between the U.S. and Germany.

The purpose of tins paper is to discuss how IT, information technology, creates additional business value. Firstly, in order to articulate the IT added value creation mechanism, we introduce a new framework, called the “3C-DRIVE,” composed of two axes: (1) 3C: Company, Competitor, and Customer, and (2) Direction, Readiness, Information Technology, and Value Evaluation. The 3C-DRIVE framework has bird’s-eye-view, dynamic, and emerging characteristics. Secondly, we apply tins framework to the existing research and perspectives regarding the relationship between business and IT. Thirdly, we design a mathematical model for computer simulation task and apply it to several conceivable cases, taking into consideration of the factors such as organization, readiness and business environment in addition to the IT investment. As a result we may conclude that complementary factors other than the IT investment greatly enhance the added business value front IT especially in the light of the perspective of the current EC era.