The Neuroforecasting Club is a technology transfer initiative in the application of neural networks in the capital markets which was established by London Business School in collaboration with University College London. The primary aim of the Club is to produce a coherent set of methodologies for developing and assessing neural forecasting methods with a strong emphasis on their practical use in the capital markets. It is 50% financed by the Department of Trade and Industry through the Neural Computing Technology Transfer programme and 50% by corporate membership. The Club brings together leading research organisations with a proven track record in neural networks development, and financial institutions with methodological needs for efficient asset allocation.
Tactical Asset Allocation
This tasks deals with tactical allocation between asset classes: bonds versus equities versus cash. Initially
for the UK markets and at a later stage at the international and global level. The aim is to estimate
expected differential returns on the basis of (up to seventeen) economic variables which are then used to
construct an optimal portfolio.
Futures Price Sensitivity to volume and open interest
This project tries to identify the relationship between price changes and changes in volume and open
interest in futures contracts. The aim is to confirm (or more difficult refute) the hypothesis that there
indeed exists a relationship and to investigate if this can be used for forecasting or trading purposes.
Tactical intra-day currency trading
The task here is to develop technical systems for tactical intra-day currency trading in selected markets.
Technical indicators such as RSI's and ADX's are fed to neural networks which are trained to generate
optimal buy/sell signals.
Factor Models for Equity Investment
The task here is to model stock returns on the basis of their exposure to fundamental financial factors (e.g.
interest rates) and financial ratios (e.g. PE). After this modelling process is completed the aim is to
construct different portfolio styles: diversified, factor-sensitive, factor-neutral etc.
Modelling and Trading Concurrent Futures Indices
The task here is to estimate differential returns of futures indices on the basis of technical indicators which
are subsequently used for portfolio management. The main activities concentrate on model design and
assessment against ARMA, maximum likelihood processes, and static ordinary least squares neural
networks.
Forecasting Volatilities for Option Pricing
The task here is to produce estimates of implied volatilities to be used in the context of option pricing for
futures contracts. High frequency tick-data from the European markets is being used to develop the
methodology.
Advanced Machine Learning and Methodology
All financial modelling and forecasting experiments are conducted within a special purpose software
simulation environment also developed as part of the project. The Advanced Machine Learning
Simulation Environment consists of Kohonen networks, Backpropagation networks, Genetic Algorithms
for Trading Rule Induction, Bayesian networks, Radial Basis Function networks, recurrent networks, and
mechanisms for combining the decisions obtained from several networks.
We address the following problems of methodology which are common to most applications.
Discounted Least Squares: extending the Ordinary Least Squares (OLS) neural learning procedure to deal with data discontinuity by extended window training in which long term memory effect experience exponential decay.
Non-linear Auto Regression: extending the OLS networks to deal with serial correlation by using error correcting terms as inputs.
Explicit Model representation: We formulate Neural Learning as a nonparametric, multiplicative/additive regression model which allows us to extract explicit representations of the estimated models. This is important in determining (using financial economics theory) whether the estimated models are valid through time or whether they represent temporary and probably unrepeatable market states.
The London Business School and University College London are joint co-ordinators of the project.
Partners from the financial institutions are:
Barclays de Zoede Wedd
CitiCorp
MARS Group
POSTEL Investment Management
SABRE Fund Management
Societe Generale
London Business School
Paul Refenes