Assessing forecasting performance on gold data using artificial neural network based models
Gold is the most stable commodity when compared to oil, crypto currency and even stock bonds. It is to gold that men turn to in the midst of political and monetary vulnerability, a place of refuge that gives versatile store of value, the common insurance hedge against bankruptcy and chaos. Forecasti...
Saved in:
Main Author: | |
---|---|
Format: | Thesis |
Language: | English |
Published: |
2019
|
Subjects: | |
Online Access: | http://eprints.utm.my/id/eprint/101809/1/MohammadRidhwanReyezMFS2019.pdf |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Summary: | Gold is the most stable commodity when compared to oil, crypto currency and even stock bonds. It is to gold that men turn to in the midst of political and monetary vulnerability, a place of refuge that gives versatile store of value, the common insurance hedge against bankruptcy and chaos. Forecasting gold price and gold demand is an important step to ensure that gold remains as a valueble form of investment or asset instead of being a liablity. Large gold bullions and gold authorities do forecasting with a large set of data but this is not the case for a private investor or consumer since these data are not made easily available to the public. In this research a model was developed to conduct a forecast with a limited set of data and the model is then tested using a large set of data. The quarterly world gold demand from 2010 to 2017 obtained from worldgoldcoundil.org were used as the limited data set and the daily world gold price within 2017 obtained from the London Bullion Market (LBMA) portal were used as the large data set. Artificial Neural Network (ANN) is the most common model for forecasting with a limited data set, but in this study the Multilayer Perceptron (MLP) model is developed with a Wavelet decomposition as well as bootstrapping. Both sets of data were fitted into the ANN model, Bootstrap Artificial Neural Network (BANN) model, Wavelet Artificial Neural Network (WANN) model and finally Wavelet Bootstrap Artificial Neural Network (WBANN) model. It is found that for the limited data set, the best model for the limited data set with the lowest value of Mean Absolute Percentage Error (MAPE) and Root Mean Squared Error (RMSE) is the WBANN model with 20.19 and 232.58, MAPE and RMSE values respectively. Meanwhile the best model for the large data set was also the WBANN model with 0.89, 12.58 MAPE and RMSE values respectively. The accuracy index for the large data set corresponds with that from the limited data set models. Thus, WBANN model is the best model for limited data forecasting. |
---|