machine learning in stock price trend forecasting by pradeep kumar reddy musku
TRANSCRIPT
Machine Learning in Stock Price Trend Forecasting
BY
PRADEEP KUMAR REDDY MUSKU
• Stock prices are dynamic and depends on known and unknown factors
• Stock Prediction Methodologies• Fundamental
• Technical
• Efficient Market Hypothesis(EMH)• Strong
• Semi Strong
• Weak
Implementation
• Data Collection
• Model Selection• Next-Day model
• Long-Term model
• Feature Selection
Data Collection
• Training data is collected from Bloomberg database
• 3M stock was picked and it contains 1471 data points (1/9/2008 to 11/8/2013)
• There are 16 features that can be used for this learning theory. Some of them are
• PE ratio
• 50-day moving average
• Current Enterprise value
Model Selection
• Learning Theories used:• Logistic Regression
• Gaussian Discriminant analysis
• Quadratic Discriminant analysis
• Support Vector Machine(SVM)
• Accuracy = The number of days that the model correctly classified the testing data total no of training days
• Next – Day Model
Model Logistic Regression
GDA QDA SVM
Accuracy 44.5% 46.4% 58.2% 55.2%
• Long – Term Model
Predicting a stock sign of difference between tomorrow’s stock price and that of certain days ego
Feature Selection
Trading Strategy
• Used 990 of the 1470 data points to fit the model.
• Made the investment decision based on the model.
Comparison
• This model has outrun the performance of the stock, with an annualized return of 19.3% vs 12.5%