Hello,
I am currently testing out Open Office spread sheet calculator with a data sample of about 10,000 data points. This spreadsheet has the historical data of the Foreign Exchange rates of the Major currencies (EUR/USD, GBP/USD, USD/JPY, USD/CHF) with columns: date, time (four hour charts), Open, High, Low, Close, Volume. This is seven years worth of data in the Spot market. I also have this data available for daily, weekly and hourly charts as well. Besides this historical data exported as CVS files, I may get tick by tick data as a DDE link. The data I have put in order is seven years of four hour chart data of the major global currencies. I am interested in nonlinear diagnostics and other statistic based tests but I am not formaly educated in this field.
As a guide, I am following the paper "Nonlinear Diagnostics and Simple Trading Rules for High-Frequency Foreign Exchange Rates" from the book "Time Series Prediction" edited by Weigend Gershenfel. I am also following methods mentioned in another paper by the same author, Blake LeBaron, titled "An Evolutionary Bootsrap Approach to Neural Network Pruning and Generalization".
I am currently learning about Bootsraping and wanted to find out about this Java based application for Open Office Calculator. Because the bootstraping method is similar to a monte carlo, in many random drawings aspect, I am unfamiliar with the amount of computation time required to crunch all the numbers. They use Matlab at the universities and I am familiar with open source versions such as Freemat for these type of calculations. I am testing the limits of Excel and Open Office Calculator in a sense with resampling and time series analysis of historical data with the Foriegn Exchange. Direction is appreciated.
Besides Resampling methods as Bootstraping for historical data of the FX (Foreign Exchange) I would like to conduct a BDS (Brock, Dechert, and Scheinkman) test statistic for independence applied to the standardised residuals of an estimated GARCH. I appreciate help to comprehend this test and how to implement it. Again I have not been formally educated in this field so explaining what the test is, what it is designed for and how to implement it is greatly appreciated. I would like a better understanding of what a Lyapunov exponent is and the concept of incremental mutual information or redundancy applied to time series prediction.
Other topics in the papers I am using as guides include: Bootstrap Cross-Validation, Network Evolution, Henon Map Time Series Forecasting, Mean Square Error Prediction, Network Pruning.
Ideally this work from Blake LeBaron is the ground work for more recent data testing in other software program applications (such as Java or open source Freemat, or Octave from Orange Hat Networks or even in the Metaqoutes language of the MT4 platrorm) as well as testing the post 2000 Foreign Exchange (Spot market) data with a Feigenbaum constant based model to represent chaotic price fluctuations (testing new models) and taking a new look at the historical data of the US discount rate, UK base rate or LIBOR rate from each of the central banks of each of the countries currency's along with historical price data from the FX exchange within a model that is governed by the properties of Mode Locking, also known as Synchronization of two Oscillators. The testing of these two new models upon the groundwork of LeBaron with more recent data and different software applications to conduct the tests would be my dissertation.
Thank you for your time and consideration
David