Forecasting Stock Index Returns: A Hybrid ARIMA and Neural Network Approach

Coblenz, Maximilian (2012) Forecasting Stock Index Returns: A Hybrid ARIMA and Neural Network Approach. [Dissertation (University of Nottingham only)] (Unpublished)

[img] PDF - Registered users only - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader
Download (502kB)


Forecasting of stock indices has become a crucial task for investors and other market participants due to the increasing use of stock index related instruments for hedging and trading. A forecasting model proposed by Zhang (2003) is used on return data from the ATX, AEX, CAC 40, DAX, FTSE 100, Hang Seng, KOSPI 200, NIKKEI 225, RTS and S&P 500 indices. The model combines linear predictions from an ARIMA model and nonlinear predictions from a neural network to yield higher accuracy. At first glance the model seems to outperform an ARIMA model and a simple random walk. However, in the majority of the cases the results are statistically not significant. Hence, the accuracy improvement is poor in relation to the increased estimation effort. Also, a trading simulation is conducted. Yet, statistical tests do not support the superior profits generated from using the forecasts. Additionally, in a realistic setting the trading simulation results are unlikely to be reproducible. Therefore, the results do not challenge the Efficient Market Hypothesis.

Nonetheless, there are some positive results. A newly introduced handling of transaction costs turns out to be quite useful for future analyses of forecasting in financial markets. In addition to that, trading is also simulated for index futures and index-tracking exchange traded funds; securities that are actually traded in real markets. Hence, a trading simulation does not have to rely on the unrealistic assumption anymore that the index is traded directly.

Item Type: Dissertation (University of Nottingham only)
Depositing User: EP, Services
Date Deposited: 08 Apr 2013 11:13
Last Modified: 07 Jan 2018 10:25

Actions (Archive Staff Only)

Edit View Edit View