Comparing different SVM Kernels, specifically the linear kernel, polynomial kernel, and the Radial Basis Function Kernel, in their effectiveness when applied to quantitative finance and algorithmic trading.
Featuresets: various technical indicators for a company Labels: - Buy/Sell - Buy - if price is predicted to rise - Sell - of price is predicted to drop
- Linear kernel: 53.3%
- Polynomial Kernel: 57.3%
- Radial Basis Function Kernel: 51.3%
- Effects of feature standardization on SVMs
- Predictability of certain technical indicators, can be studied using Hypothesis Testing
This model is based on the idea that the price trend of certain groups of companies are likely to move together. Each model is generated on a per company basis, while taking into account of all other companies.
Featuresets: Pricing change of all S&P 500 companies on a certain day Labels: Buy/Sell/Hold
- Buy - if price rise more than 2% for the next 7 days
- Sell - if price drop more than 2% for the next 7 days
- Else - Hold
54% mean accuracy for all S&P 500 companies in predicting buy/hold/sell (random arcuracy: 33%)