The views expressed are those of the individual authors and do not necessarily reflect official positions of the Federal Reserve Bank of St. Louis, the Federal Reserve System, or the Board of Governors. Federal Reserve Bank of St. Louis Working Papers are preliminary materials circulated to stimulate discussion and critical comment. References in publications to Federal Reserve Bank of St. Louis Working Papers (other than an acknowledgment that the writer has had access to unpublished material) should be cleared with the author or authors. Photo courtesy of The Gateway Arch, St. Louis, MO. www.gatewayarch.com

Fourteen linguistically-motivated numeri- cal indicators are evaluated for their abil- ity to categorize verbs as either states or events. The values for each indicator are computed automatically across a corpus of text. To improve classification performance, machine learning techniques are employed to combine multiple indicators. Three machine learning methods are compared for this task: decision tree induction, a genetic algorithm, and log-linear regres- sion.

Technical analysis, also known as “charting, ” has been a part of financial practice for many decades, but this discipline has not received the same level of academic scrutiny and acceptance as more traditional approaches such as fundamental analysis. One of the main obstacles is the highly subjective nature of technical analysis—the presence of geometric shapes in historical price charts is often in the eyes of the beholder. In this paper, we propose a systematic and automatic approach to technical pattern recognition using nonparametric kernel regression, and we apply this method to a large number of U.S. stocks from 1962 to 1996 to evaluate the effectiveness of technical analysis. By comparing the unconditional empirical distribution of daily stock returns to the conditional distribution—conditioned on specific technical indicators such as head-and-shoulders or double-bottoms—we find that over the 31-year sample period, several technical indicators do provide incremental information and may have some practical value. ONE OF THE GREATEST GULFS between academic finance and industry practice

Abstract- Financial investment decision making is extremely difficult due to the complexity of the domain. Many factors could influence the change of share prices. FGP (Financial Genetic Programming) is a genetic programming based forecasting system, which is designed to help users evaluate impact of factors and explore their interactions in relation to future prices. Users channel into FGP factors that they believe are relevant to the prediction. Examples of such factors may include fundamental factors such as "price-earning ratio", "inflation rate " or/and technical factors such as "5-days moving average", "63-days trading range breakout", etc. FGP uses the power of genetic programming to generate decision trees through combination of technical rules with self-adjusted thresholds. In earlier papers, we have reported how FGP used well-known technical analysis rules to make investment decisions. This paper tests the versatility of FGP by testing it on shorter-term investment decisions. To evaluate FGP more thoroughly, we also compare it with C4.5, a well-known machine learning classifier system. We used six and a half years ’ daily closing price of the Dow Jones Industrial Average (DJIA) index for training and over three and half years ’ data for testing and obtained favourable results for FGP. 1

The creation of mathematical, as well as qualitative (or rule-based), models is difficult, time-consuming, and expensive. Recent developments in evolutionary computation hold out the prospect that, for many problems of practical import, machine learning techniques can be used to discover useful models automatically. These prospects are particularly bright, we believe, for such automated discoveries in the context of game theory. This paper reports on a series of successful experiments in which we used a genetic programming regime to discover high-quality negotiation policies. The game-theoretic context in which we conducted these experiments --- a three-player coalitions game with sidepayments --- is considerably more complex and subtle than any reported in the literature on machine learning applied to game theory. 1. Introduction 1 Modeling is difficult, time-consuming, and expensive. Examining a real-world system, collecting data, and summarizing the findings in the form of a valid...

We present a parallel implementation of genetic programming on distributed memory machines. To overcome the time overhead due to uneven load associated with program evaluation, we propose and evaluate a non-preemptive dynamic scheduling algorithm for load balancing. The system is applied to the evolution of trading model strategies which is a compute-intensive application. Our results show that reasonable trading models can be inferred and that the system can produce a nearly linear speedup for that application. 1. Introduction Artificial evolutionary processes, such as genetic algorithms (GA) (Holland 1975), are based on reproduction, recombination and selection of the fittest members in an evolving population of candidate solutions. (Koza 1992) extended this genetic model of learning into the space of programs and thus introduced the concept of genetic programming. Each solution in the search space is represented by a genetic program (GP), traditionally using the Lisp syntax. Gene...

This paper provides a survey of the applications of evolutionary algorithms in macroeconomic models. Discussion is organized around the issues related to stability of equilibria, equilibrium selection, transitional dynamics, and the long-run evolutionary dynamics di erent from rational expectations equilibrium outcomes. The survey also discusses criteria that can be used to evaluate the performance and usefulness of evolutionary algorithms in macroeconomic context.

Please do not quote without permission * Chidambaran is visiting at NYU, on leave from Tulane. Lee holds joint appointments at Tulane and HKUST. Trigueros is at Tulane. We are grateful for the comments from participants at seminars at Tulane

This paper presents a time series forecasting methodology and applies it to generate one--step-- ahead predictions for the daily foreign exchange spot rates. The methodology draws from the disciplines of chaotic time series analysis, clustering, artificial neural networks and evolutionary computation. In brief, clustering is applied to identify neighborhoods in the reconstructed state space of the system; and subsequently neural networks are trained to model the dynamics of each neighborhood separately. The results obtained through this approach are promising.

A genetic programming system for data mining trading rules out of past foreign exchange data is described. The system is tested on real data from the dollar/yen and dollar/DM markets, and shown to produce considerable excess returns in the dollar/yen market. Design issues relating to potential rule complexity and validation regimes are explored empirically. Keeping potential rules as simple as possible is shown to be the most important component of success. Validation issues are more complicated. Inspection of fitness on a validation set is used to cut-off search in hopes of avoiding overfitting. Additional attempts to use the validation set to improve performance are shown to be ineffective in the standard framework. An examination of correlations between performance on the validation set and on the test set leads to an understanding of how such measures can be marginally benificial; unfortunately, this suggests that further attemps to improve performance through...