Show simple item record

dc.contributor.authorAhmed, Abdillahi M
dc.date.accessioned2023-02-08T06:31:08Z
dc.date.available2023-02-08T06:31:08Z
dc.date.issued2022
dc.identifier.urihttp://erepository.uonbi.ac.ke/handle/11295/162318
dc.description.abstractThe relevance and success of financial time-series modelling is predicated upon the inefficiencies of the securities markets. Each forecasting attempt first denies that prices of securities mimic a random-walk process and are hence unpredictable. As traders and investors exploit predictive models successfully, the market efficiency increases, which renders previous strategies ineffective. This prompts investors and researchers to deploy newer and better approaches. While Machine Learning algorithms have already received sparse attention in the Kenyan financial literature, their use has been limited to the rudimentary, vanilla Feed-forward Networks (FNN). In this study, we explored the predictive ability of Recurrent Neural Networks (RNNs) using a sliding window approach on the Kenyan bourse. The population consisted of all the stocks listed on the Nairobi Securities Exchange and the sample was the twenty stocks listed as part of the NSE-20 index through the non-probabilistic purposive sampling technique. We collected ten years' price and volume data from the NSE. Common technical indicators were used to transform the raw features into nine independent inputs. We used the Python coding language and its libraries to perform the predictive modelling in this study. 70% of the data was used for training while 30% of the data tested the models. We then developed two Recurrent Networks, the Gated Recurrent Unit (GRU) and the Long-Short-term-Memory (LSTM) on each stock and benchmarked them against two traditional approaches; The Box- Jenkins ARIMA and the vanilla Artificial Neural Networks (FNN). The R-Squared and Root Mean-Squared Error metrics formed a basis of comparison between the Recurrent models and the traditional models. Through a paired T-test for the difference between means of the R-Squared (95% level of confidence), we found that one of the Recurrent Network models, the GRU, performed significantly better than all other models in the study. The other Recurrent Network (LSTM) was the second-best model by the means of both the R-Squared and the RMSE, followed by the FNN and last, the ARIMA. Further, we explored different parameters and found that a shorter predictive window of {20:10}, and a smaller batch size (40) resulted in better fits for the neural network models on the stocks sample selected. The study has broadened the variety of market research tools available to the Kenyan market participants and has added Recurrent Neural Networks to the Kenyan empirical literature.en_US
dc.language.isoenen_US
dc.publisherUniversity of Nairobien_US
dc.rightsAttribution-NonCommercial-NoDerivs 3.0 United States*
dc.rights.urihttp://creativecommons.org/licenses/by-nc-nd/3.0/us/*
dc.titleAssessing Recurrent Neural Networks as a Prediction Tool for Quoted Stock Prices on the Nairobi Securities Exchangeen_US
dc.typeThesisen_US


Files in this item

Thumbnail
Thumbnail

This item appears in the following Collection(s)

Show simple item record

Attribution-NonCommercial-NoDerivs 3.0 United States
Except where otherwise noted, this item's license is described as Attribution-NonCommercial-NoDerivs 3.0 United States