Financial markets have long been modelled on one primary assumption and that assumption is a linear relationship between factors involved in financial risk, return and prices. Given that linear models do not have a good track record of providing above average returns or even providing good explanations for market movements, market participants look for other ways of developing models which can provide above average performance and a much higher degree of explanation for market movements. There is now compelling evidence of nonlinearities in the financial markets. We use learning networks which is a non-linear, nonparametric model free modelling technique, to model a variety of international financial and commodity markets such as sixteen international equity markets, fifteen foreign exchange rates with respect to the US$, sterling interbank market including six maturities and Brent crude oil futures market comprising of the first six maturities, and provide above average returns with a significantly higher degree of explanation of market movements. The learning network model performance is evaluated on out - of - sample periods, during periods of high volatility and structural/regime change. The model evaluation basis is primarily percentage returns adjusted for transaction costs although usual error measures such as root mean square error, Thiel's U statistic etc. are also used. We also carry out a detailed analysis of the weight structure of the models and determine the important inputs to the models. This step we believe would remove the common objection that neural networks have a black box which makes it impossible to evaluate neural network applications. We determine clear non-linearities in financial and commodity markets and conclude that the markets are far more complex than previously thought.
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:617791 |
| Date | January 1996 |
| Creators | Dasgupta, Bhaskar |
| Publisher | University of Manchester |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
Page generated in 0.0018 seconds