Sammlung Bachelor- und Masterarbeiten 2017

Page 31

Bachelorarbeit – Banking and Finance

Predicting Financial Times Series in the Short Run With Echo State Networks (ESNs) GRADUATE Simon Blum SUPERVISOR Dr. Simon Rentzmann

Financial time series forecasting in stock and foreign ex­

The MATLAB simulations trading results of MT with the

change markets has attracted the attention of investors,

broker FXCM were used to deliver commonly used perfor­

speculators, and researchers over recent decades due to

mance metrics trading profits, including all transaction

its enormous potential returns. Various advances in ana­

costs and spreads. The timeframe resolution for the ESN

lytical and computational procedures have led to a number

time series is usually one day with a one-step ahead pre­

of promising new approaches to financial time series min­

diction. This paper investigates, in contrast, a higher time­

ing, based on nonlinear and nonstationary models. Among

frame resolution – one hour with a one-step ahead fore­

an expanding variety of methods, artificial neural networks

cast.

are the most widely used models to analyze and handle stochastic and nonlinear systems. Financial time series are

The findings confirm that simulations tend to overfit and

widely known as chaotic, nonstationary, and difficult to

lead to poor outcomes in terms of profits and winning

predict, which in fact lies within the scope of neural net­

trade ratios. It was possible, however, to tune the entry

works.

parameters and adjust the settings manually to avoid over­ fitting, which gives better results.

This Bachelor’s thesis presents a study of a novel recurrent neural network – the Echo State Network (ESN) – to pre­

It could, therefore, be shown that ESN can be applied to

dict the next closing price in foreign exchange markets.

foreign exchange markets successfully and that it can pro­

The main purpose of this paper was to verify whether ESN

vide important forecasts in the short run. In this context,

could be used as a prediction model for the foreign ex­

ESN is a state-of-the-art prediction model in forecasting

change market in the short run.

financial time series with a higher resolution in certain time­ frames, as well as having still great potential in terms of

A review of the current state of the literature revealed

developing the network structure.

agreement among researchers concerning the predictive power of Echo State Networks in the financial domain and yielded a selection of the most relevant applications to stock and foreign exchange markets. The thesis suggests various improvements to the base model provided by Her­ bert Jaeger. New in this field of research is the direct linking of the ESN to a trading account of a large foreign exchange (FOREX) and contract for difference (CFD) broker through METATRADER (MT) and MATLAB. This relationship is de­ scribed in detail and used to evaluate the prediction accu­ racy and performance of the ESN.

31


Turn static files into dynamic content formats.

Create a flipbook
Issuu converts static files into: digital portfolios, online yearbooks, online catalogs, digital photo albums and more. Sign up and create your flipbook.
Sammlung Bachelor- und Masterarbeiten 2017 by ZHAW School of Management and Law - Issuu