[.uk] Neural Networks in Finance: Gaining Predictive Edge in ... (ISBN 0124859674)

More Mathematical than Technical:
Defiantly more of a math book than a programming guide, but that was what I was expecting. This book explains how to use neural networks in the field of finance. It does so very logically and mathematically. You are shown how to apply neural networks to many different financial problems. But you are mostly left to yourself to actually implement the neural networks on a computer system. Some example source code is provided for MathCad, which is an expensive software package you can buy separately. If you are already comfortable with neural network programming, and are looking to learn to apply neural networks to finance, this book is great. Being a Java programmer I used the open source JOONE package to implement some of the book's examples in Java. Though JOONE is not suited to all examples in the book, it is a good start for a Java programmer. The book shows how neural networks can be applied to many real world financial problems. The book pays particular interest to international finance. The book examines Hong Kong and Japan, examining inflation, deflation, currency volatility, and other issues. I found the book to be very useful in giving me an introduction to neural networks in finance. The table of contents follows: Chapter 1: Introduction Part 1: Econometric Foundations Chapter 2: What Are Neural Networks? Chapter 3: Estimation of a Network with Evolutionary Computation Chapter 4: Evaluation of Network Estimation Part 2: Applications and Examples Chapter 5: Estimating and Forecasting with Artificial Data Chapter 6: Time Series: Examples from Industry and Finance Chapter 7: Inflation and Deflation: Hong Kong and Japan Chapter 8: Classification: Credit Card Default and Bank Failures Chapter 9: Dimensionality Reduction and Implied Volatility Forecasting

Great Book but Horrible Matlab Code:
I've only been through the first 4 chapters so far. I found the way the material was presented to be very good and the authors did a very good job presenting and explaining the mater. Having understood the material which I would credit to the author's great clarity and presentation, I decided to run the Matlab code the author provides. This is were everything started going wrong. The functions are full of error and would not run. I had to make changes to the m-file for the proram to run. This was also very hard since the code is very poorly documented (input variables are not even explained). Even after fixing the erros, the programs did not give the results the author claims. In the example on page 78, the author claims that the genetic algorithm gives a result very close to 4 which is not true (some results were less than 2). I then tried to work the example on page 81. Again I got errors trying to run the program. In the file ffnet9.m, the author has an if statement if the number of arguments is 8 instead of the 12 expected by the function while in the example, the number of arguments is 9 and therefore you get an error trying to run the function ffnet9. second, it seems the author had modified a previous function which took 8 arguments since the function is actually called ffnet8 in the file while the file is called ffnet9.m (very bad programming). After fixing the problem, the linear model gave an R-squared in the 0.55 range and the second degree polynomials gave a result in the range of 0.91 however, the neural network R-squared was in the range of 0.73 and not 0.99 as claimed by the author! the line search in the function fminunc is exiting due to the line search. By the way, don't run the program on page 81 1000 times as done in the for loop as this will take forever and I'm not sure way the author did it.