Impact of different predictors on stock price forecasting performance

This paper aims to investigate the impact of different predictors on the performance of stock price forecasting. The paper also proposes a novel forecasting pipeline and architecture which gives a holistic approach to stock price forecasting. A comprehensive literature review has been performed to i...

وصف كامل

محفوظ في:
التفاصيل البيبلوغرافية
المؤلف الرئيسي: Chong, Jie Sheng
مؤلفون آخرون: Vidya Sudarshan
التنسيق: Final Year Project
اللغة:English
منشور في: Nanyang Technological University 2023
الموضوعات:
الوصول للمادة أونلاين:https://hdl.handle.net/10356/166007
الوسوم: إضافة وسم
لا توجد وسوم, كن أول من يضع وسما على هذه التسجيلة!
المؤسسة: Nanyang Technological University
اللغة: English
الوصف
الملخص:This paper aims to investigate the impact of different predictors on the performance of stock price forecasting. The paper also proposes a novel forecasting pipeline and architecture which gives a holistic approach to stock price forecasting. A comprehensive literature review has been performed to identify different categories of predictors and predictive models used in stock price forecasting. Research has also been conducted to explore feature selection and feature importance approaches. The project first identifies a suitable stock price forecasting model (Bi-LSTM) to be used as the baseline model. Secondly, 197 indicators were retrieved and feature engineered from 3 different data sources. Lastly, an ensemble of both feature selection (hybrid multi-criterion filter-wrapper) and feature importance (model-centric) methods are then used to select an optimal feature subset of important predictors. Predictors identified include a combination of technical indicators, economic indicators, fundamental indicators, and sentiment analysis. The proposed solution reduced the dimensionality of the dataset from 197 predictors to 29 predictors, leading to an improved forecasting performance. The final forecasting performance is evaluated using RMSE as the key metric. With the optimal feature subset, the model achieves an RMSE of 0.0175, surpassing all of our previous models and four out of five benchmarks. The originality of our proposed solution is that the suggested architecture provides an end-to-end architecture that can analyze all available features related to stock prices to perform optimal feature subset selection and subsequently forecast short-term stock prices. The project also enables the investigation of the impact (ranking) of each predictor in stock price forecasting, therefore providing investors with additional information about such predictors.