The project aims at predicting the direction of volatility index which is the measure of expected volatility of the stock market. The team has used rich data set comprising of main financial market indices,option contracts, macro-economic data and sentiment data. I feel the team has done a good job in data collection utilizing different resources available.
What I like about the report:
The initial data exploration is done really well. They have considered different methods to visualize the data like violin plot, histograms. They have also calculated correlation to eliminate the redundant features having high correlation with other features in the data. As many features in the data set were in perfect correlation, it makes sense to eliminate such features at the outset for better prediction. They have further used random forest method to eliminate other redundant features from the data-set.
Lastly they have made the prediction using the extra trees model which is kind of appropriate for the given data set.
Areas of Improvement
During the data-preprocessing , to bring parity in the frequency they have filled the lower frequency data value. It would be nicer if they could explain the logic behind doing this. Also, it would have been better if they could have explained the extra trees model in detail and provided a diagnostic of the model. Lastly, since it is a classification problem, I am not sure if using linear and lasso regression will give better prediction. May be they can explore other classification models like logistic regression or multinomial. regression.
Overall, I really like their mid-term report and I am looking forward to read their final report.
The project aims at predicting the direction of volatility index which is the measure of expected volatility of the stock market. The team has used rich data set comprising of main financial market indices,option contracts, macro-economic data and sentiment data. I feel the team has done a good job in data collection utilizing different resources available. What I like about the report: The initial data exploration is done really well. They have considered different methods to visualize the data like violin plot, histograms. They have also calculated correlation to eliminate the redundant features having high correlation with other features in the data. As many features in the data set were in perfect correlation, it makes sense to eliminate such features at the outset for better prediction. They have further used random forest method to eliminate other redundant features from the data-set. Lastly they have made the prediction using the extra trees model which is kind of appropriate for the given data set. Areas of Improvement During the data-preprocessing , to bring parity in the frequency they have filled the lower frequency data value. It would be nicer if they could explain the logic behind doing this. Also, it would have been better if they could have explained the extra trees model in detail and provided a diagnostic of the model. Lastly, since it is a classification problem, I am not sure if using linear and lasso regression will give better prediction. May be they can explore other classification models like logistic regression or multinomial. regression.
Overall, I really like their mid-term report and I am looking forward to read their final report.