Prediction of trend of Stock Price (of coming 7..10 days)
Approach : Regression (predict change of status (Rise/Drop), with estimated range)
- Regression, based on parameter of previous 30 days, 60 days, or 90 days
- Historical records of f particular stock
- Interface with “Stocks API” to automate the process
Steps of analysis
1. Collect data, records of below fields of previous 30 days
[Features]
- Max, Min, open, close
- Status (Drop/Rise compared with previous date)
- Number of successive days of drop or rise
- Total # of lots of transaction (purchase, sold), total amount of transaction (purchase, sold). total # of transaction (purchase, sold, eg: 1 transaction may involve 20 logs of transaction)
[Output, compared with previous day]
- Status (rise / drop)
- % of change of stock price (rise / drop)
2. Conduct correlation analysis of above fields
3.
Based on result of correlation
analysis, select fields (features) for analysis
4.
Divide records into “Training
Set” and “Test Set”
5.
By using Polynormal
Regression, change degree, check accuracy à stop
if overfit
6.
Test prediction upon data of
tomorrow (status, and % of rise/drop)
7.
Repeat above, based on records
of previous 60 days and 90 days
[Further estimation of next 7..10 days]
8.
Assume there are 4 features and 1 result, use result of prediction and 3 features to estimated value of the 4th
feature. Use this approach to estimate values of all the 4 features of tomorrow.
9.
Based on estimated features of
tomorrow, predict status (and % of drop/rise) of the day after tomorrow. Use
this recursive approach to predict stock price of coming 7..10 days.
**Will be
interesting to compare the prediction with actual trend of stock price for the
coming 7..10 days.
Comments
Post a Comment