Michael Holden Faculty Sponsor: Professor Gordon H. Dash
ANN is structured after a biological neural network A mathematical model that attempts to mine, predict, and forecast data Provides Artificial Intelligence (AI)
A process of pattern recognition and manipulation is based on: ◦ Massive Parallelism ◦ Connectionism ◦ Associative Distributed Memory
Brain contains an interconnected net of approximately 10 billion neurons (cortical cells) Biological Neuron The simple “arithmetic computing” element
Mathematical Model of human- brain principles of computations Consists of elements called the biological neuron prototype ◦ Interconnected by direct links (connections) ◦ Cooperate to perform PDP to solve a computational task
New paradigms of computing mathematics consists of the combination of artificial neurons into artificial neural net ? Brain-Like Computer
Data Acquisition Data Analysis Interpretation and Decision Making Signals & parameters Characteristics & Estimations Rules & Knowledge Productions Data Acquisition Data Analysis Decision Making Knowledge Base Adaptive Machine Learning via Neural Network
Independent VariablesDependent Variables 30-Day Treasury Bill 20-Year Treasury Bond Volatility Index (VIX) -Equity Market Neutral -Event Driven -Global Macro -Long/Short Equity
WinORS e-AI Windows Operating Research System with e-data and artificial intelligence capabilities Developed by NKD-Group, Inc.
Neural Network is not programmed – it learns Training = Learning Validating = Testing 33.3%
Kajiji-4 is the algorithm GCV is Generalized Cross Validation Gaussian transfers information between nodes
RBF – Parameters RBF – Weights RBF - Predicted
Equity Market Neutral Event Driven Global Macro Long/Short Equity Computed Measures Actual Error 1.33E E E E+00 Training Error 1.66E E E E-03 Validation Error 1.73E E E E-03 Fitness Error 1.71E E E E-03
Performance Measures Equity Market Neutral Event Driven Global Macro Long/Short Equity Direction Modified Direction TDPM R-Square 99.99%99.45%99.89%99.98% AIC Schwarz MAPE
Gives relativity of independent variables Absolute numbers > signs *Global Macro and Event Driven
Actual Return -Predicted Return Residual How well did it learn?
Small Residuals ◦ Most < 1bp Very Fit Model
2 Factors ◦ Global vs. Domestic Principal Component Analysis Explains Majority of Variance ◦ Some variance not captured by residuals
Fit Model ◦ Learned very well Small Residuals ◦ Trained very well Factors explained 90.4% of variance ◦ Include global and domestic independent variable next time Excellent Predictive Ability