Test RandomForest On Stocks

1. Precision are more important.

   I need to find only a few stocks with positive perspectives. More stocks are meaningless because I cannot buy too many. However, negative stocks (result in loss) should be strongly avoided.

   The worst situation, is I didn't buy any stock, which means I have no gain no loss on that day.

2. use Random Forest as the predicting model

   Why random Forest?

• it is easy to implement and use in R, which is current language used in this project.
• do not need feature selection.
• its principal is quite suitable to stock modeling:

1. decision based algorithm is easy and intuitive to understand; no black box.
2. each day the stock may have different traits, the model can be built based on each day's data, respectively.
3. consensus modeling is a good way to find the strongest candidates.

3. How to define the dataset?

1. Raw data need to be updated every time the model was constructed; -- data_retrieve()
   The raw data is overwriten in data/data_newest.RData.
   Historical file does not need to be kept.

2. However, only missing days are needed to be updated; -- dataset_generating()

Three types of data need to be generated:
(1) Raw data (the closing price of each stock in previous n Days) (2) Basic statistics of Raw data -- get_rawdata_index(rawdata) (3) Stock Indexing (measured by quantmod functions)