Modelling and Trading the English and German Stock Markets with Novelty Optimization Techniques

• Author: Chia Chun Lo,Sovan Mitra,Andreas Karathanasopoulos,Konstantinos Skindilias
• Published date: 01 December 2017
• Date: 01 December 2017
• DOI: http://doi.org/10.1002/for.2445

Modelling and Trading the English and German Stock Markets with

Novelty Optimization Techniques

ANDREAS KARATHANASOPOULOS,

1

*SOVAN MITRA,

2

KONSTANTINOS SKINDILIAS

3

AND CHIA CHUN LO

4

1

Department of Finance, Accounting and Managerial Economics, American University of Beirut,

Lebanon

2

Department of Mathematical Sciences, University of Liverpool, UK

3

School of Computing and Mathematical Sciences, University of Greenwich, London, UK

4

Faculty of Business Administration, University of Macau, Taipa, Macau China

ABSTRACT

The motivation for this paper was the introduction of novel short-term models to trade the FTSE 100 and DAX 30

exchange-traded funds (ETF) indices. There are major contributions in this paper which include the introduction of

an input selection criterion when utilizing an expansive universe of inputs, a hybrid combination of partial swarm op-

timizer (PSO) with radial basis function (RBF) neural networks, the application of a PSO algorithm to a traditional

autoregressive moving model (ARMA), the application of a PSO algorithm to a higher-order neural network and, fi-

nally, the introduction of a multi-objective algorithm to optimize statistical and trading performance when trading an

index. All the machine learning-based methodologies and the conventional models are adapted and optimized to model

the index. A PSO algorithm is used to optimize the weights in a traditional RBF neural network, in a higher-order neu-

ral network (HONN) and the AR and MA terms of an ARMA model. In terms of checking the statistical and empirical

accuracy of the novel models, we benchmark them with a traditional HONN, with an ARMA, with a moving average

convergence/divergence model (MACD) and with a naïve strategy. More specifically, the trading and statistical perfor-

mance of all models is investigated in a forecast simulation of the FTSE 100 and DAX 30 ETF time series over the

period January 2004 to December 2015 using the last 3 years for out-of-sample testing. Finally, the empirical and sta-

tistical results indicate that the PSO-RBF model outperforms all other examined models in terms of trading accuracy

and profitability, even with mixed inputs and with only autoregressive inputs. Copyright © 2016 John Wiley & Sons,

Ltd.

key words particle swarm optimization; radial basis function; confirmation filters; FTSE 100; DAX

30 day trading

INTRODUCTION

Modelling and trading financial indices remains a challenging and demanding task for market participants. Forecast-

ing financial time series can be extremely difficult because they are influenced by a large number of variables. Much

of the analysed data displays periods of erratic behaviour and, as a result, drastic declines and spikes in the data series

are experienced. Existing linear methods are limited as they only focus on one time series. Some of the older

machine-learning models also have trouble producing accurate and profitable forecasts owing to their rigid architec-

tures. In this paper the proposed models improve on these inefficiencies to make the models more dynamic, similar to

the time series they are tasked with forecasting. This is particularly important in times of crises as the correlations

between different asset classes and time series increase. These inadequacies have been studied in great depth by

the scientific community and many methodologies have been proposed to overcome the disadvantages of previous

models (Li and Ma, 2010). The main disadvantage of existing nonlinear financial forecasting and trading methodol-

ogies is that most of them search for global optimal estimators. The problem with this approach is that most of the

time global estimators do not exist owing to the dynamic nature of financial time series. Moreover, the algorithms

that are used for modelling financial time series have many algorithms that need to be tuned, and if this procedure

is performed without careful consideration the accuracy of extracted prediction models will suffer and in some cases

result in a data-snooping effect. Finally, the training of a prediction model is usually performed separately from the

construction viable trading signals and thus the overall performance is reduced. Specifically, most machine learning

algorithms that are designed for forecasting financial time series deploy only statistical metrics for the optimization

steps of their training phase and do not apply any optimization step for improving their trading performances. Here,

a multi-objective algorithm is employed to optimize both statistical properties and trading performance.

The motivation for this paper is to introduce in a hybrid neural network architecture of particle swarm optimization

(PSO) combined with radial basis function (RBF), which tries to overcome some of these limitations. More

*Correspondence to: Andreas Karathanasopoulos, Department of Finance, Accounting and Managerial Economics, American University of

Beirut, Lebanon.

E-mail: andreas.kara@hotmail.com

Journal of Forecasting,J. Forecast. 36, 974–988 (2017)

Published online 6 November 2016 in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/for.2445

Copyright © 2016 John Wiley & Sons, Ltd.