AbstractStock market activity cycles through periods of trends and fluctuations due to external economic factors and the psychology of participants. Many stock prediction models exist for predicting prices, trends and volatility. However, models focusing on individual or few prediction methods suffer from a lack of adaptability, meaning they perform well at specific stages of the market cycle rather than over the whole range. A prediction model developed and tested during a period of strong growth may perform well under these conditions, but fail during market downturns.
It is desirable that a prediction model adapts to new circumstances so investors can profit across the entire market cycle. For this reason and because of the absence of adaptive models in the literature, this research has developed a dynamic stock investment system that combines the intelligence of multiple predictors using a scoring system to give more weight to predictions that have performed best under recent market conditions and a filtering system to identify the most potentially profitable trades, thereby effectively adapting to market behaviour.
Differently to other research in this area, the performance of the new system is not evaluated based on the accuracy of predictions, but primarily by investment metrics such as profit, drawdown and the Sharpe Ratio, the latter two of which also account for the risk of the system.
The experimental results show that our model works effectively on more than 100 stocks from the UK, US, Chinese and Singaporean markets. These stocks come from more than 10 different market sectors covering a wide-range of market conditions during the testing periods. We concluded that our model shows an excellent capability of handling predictions in fluctuated situations and is effective regardless of the characteristics of
the stock data.
|Date of Award||8 Sep 2021|
|Supervisor||Marina De Vos (Supervisor), Julian Padget (Supervisor) & Tom Fincham Haines (Supervisor)|
- machine design