Mark Jurik – Neural Networks & Financial Forecasting

Description

 

Learn technique for building neural nets

 

Apply new types of indicators

 

 

This is a bound collection of published and unpublished technical reports by Jurik Research & Consulting. The author explains with amazing clarity both neural networks and fuzzy logic, as well as techniques and tricks on forecasting that Jurik Research reveals to its biggest clients.

 

For example, one of the best ways to improve forecast reliability is to reduce the number of inputs to a model. But popular approaches (elimination by pairwise correlation analysis or sensitivity analysis) are fraught with pitfalls. Discover the correct procedure and how easy it is to perform.

 

 

 

 

You will learn how to …

 

 

avoid months of trial and error experiments

make advanced technical indicators

compare commercial neural network packages

prepare your data for time series forecasting

preprocess data four important ways

grade the performance of a trading system, … and more.

 

 

Contents

 

 

Wall Street Forecast with a Neural Network – highly acclaimed introduction to how neural networks work and how to apply them on financial spreadsheets to forecast market activity.

 

Consumer Guide to Software for Smart Forecasting – Reviews numerous software modeling packages for both their user friendly features and ability to forecast 30 year Treasury Bond prices.

 

Indicator Development issues in the Space Domain – Reveals many advanced techniques for making financial data talk. This report has been hailed as a model of lucidity, published in AI in Finance and the book Virtual Trading.

 

Methodology Report #1: Train performance versus Test Performance – Is a model’s performance on data set A during training less or more important than its performance on data set B after training has finished? You’ll be surprised!

 

Methodology Report #2: Partitioning the Data Set – Partitioning financial time-series data for training and testing models is not a trivial matter due to the complex behavior of the markets.

 

Methodology Report #3: Scoring a Trading System – A way to evaluate the quality of a trading system by running its profits and losses though a unique function that produces a single number. Ideal for genetic algorithm development of financial trading systems.

 

Application Note #1: Optimal Phase in Crossover Moving Averages – Shows how noisy oscillators can rob you of good trades and how to get optimal lag in a moving average by relating lag to phase angles of financial cycles.

 

Application Note #2: Efficient Trading with Adaptive Moving Averages – Shows how to increase the smoothness of an exponential moving average without inducing significant lag, and then compares the result with Jurik’s moving average.

 

Estimating Optimal Forecast Distance using Chaos Analysis – Shows how to estimate the optimal distance in the future to forecast. Report applies mathematics to T-Bond data, but can be applied to all markets.

 

Neural Networks and Regression – Addresses commonly asked questions on the difference between neural networks and standard linear regression.

 

Introduction to a Neural Network Algorithm – Simple mathematical explanation of the most popular neural net algorithm in the world today, Back-Propagation.

 

Executive Overview of Neural Nets and Fuzzy Logic – A intuitive approach to describing these technologies with little math and lots of pictures.

 

Technical Discussion of Neural Nets and Fuzzy Logic – A unique way of unifying almost all neural net architectures. The fuzzy logic report is also very intuitive and not at all confusing. Just the right amount of math, pictures and analogies.

 

Some Problems with the Back-Error Propagation Method – Although popular for its simplicity, this algorithm has some problems. This report give a brief summary of these findings.

 

Back-Error Propagation: A Critique – This detailed report works from ground truth up to expose the fundamental problems with Back-propagation. For mathematically inclined readers only.

 

Backpercolation – A non-proprietary report introducing a better algorithm for training neural nets. This algorithm is available only in Braincel, an add-in module for Microsoft’s Excel for Windows spreadsheet program. Braincel is available from Jurik Research.