Genetic Programming of Minimal Neural Nets Using Occam's Razor describes a  genetic program\ming method for constructing minimal neural networks. Using an Occam's razor in the fitness function, the method prefers a simple network architecture to a complex one.

Designing a neural network for forecasting economic time series. Excellent paper with many general hints and tips. An essential introduction.

Essential reading is Financial Prediction, Some Pointers, Pitfalls, and Common Errors by Kevin Swingler. Focusing specifically on financial applications, this paper investigates the common methods for forecasting time series in the face of the efficient market hypothesis and outlines the major hazards.

The paper Incorporating Prior Knowledge about Financial Markets through Neural Multitask Learning discusses the incorporation of prior knowledge (hints) into the inductive learning system of neural networks. An empirical example illustrates the improvement in the accuracy of a neural network used to forecast the price movements of five major German stocks.

Forecasting the 30-year US Treasury Bond with a System of Neural Networks. Thirty-two feed-forward neural networks were used to predict market movements a week in the future.

Forecasting financial markets using neural networks: an analysis of methods and accuracy. Not terribly insightful market-wise, but an example of a typical approach.

A nonparametric approach to pricing and hedging derivative via learning networks. How a network learned and then applied the Black and Scholes model - impressive! Quite mathematical.

A voluminous list of NN references (Word doc) compiled by Athanasios Episcopos, with some links on neural networks, finance and economics. These references are intended for the researcher who wants to use artificial neural networks (NN) in finance and economics.

Neural Computing Publications Worldwide, from the IEEE. Provides journal links and a useful book bibliography, as well as fuzzy systems and evolutionary computation.

The following two products are less comprehensive but are Excel add-ins, making them easier for input/ output:

Neuralyst by Cheshire Engineering. We have used this program and it works well. A disadvantage is the lack of sensitivity analysis while a time-saving feature is the use of genetic algorithms to optimise the network architecture.

Braincel by Promised Land Technologies. Not tested extensively. Has the sensitivity analysis but not the GAs!

For a larger, more sophisticated systems, Neuroshell and Genehunter by Ward Systems provides trading system software.

NeuroDimension provides NeuroSolutions and TradingSolutions, products that combine traditional technical analysis with AI techniques. One can use combinations of financial indicators in conjunction with NNs and genetic algorithms to create trading models.

There is also excellent neural net shareware like EasyNN.

Don Tveter's page on neural networking software, The Basis of AI NN provides the code for various algorithms; programmers and quants are likely to get the most benefit out of these.

MJ Futures is a market prediction company that specializes in neural software and services for market forecasting. The comment on their home page that "At first it seemed that neural nets were the answer to all of our prediction problems, but we soon learned that they were not the Holy Grail they were advertised to be..."  suggests they have evolved beyond the run-of-the-mill, naive price-predictors.

Timestat - a reputable freeware application for dealing with fast Fourier transforms and wavelets in data pre-processing (although the documentation is rather terse). Download Timestat1.2.zip here. For other statistical applications go to the Simstat site.

Sigma Research Associates. Books and trial software on statistics and NNs and a bit of chaos too.

Talking of books, Amazon's Professional and Technical section has an excellent selection of AI titles.

Yahoo links. A short but eclectic mix of links.