Stock price forecast

Stock price forecast

This article uses machine learning Methods a practical method to predict the stock price .

But if you want to make money with it , Guess how much you will lose ？

---

• Stock price forecast
• Data sources
• The basic method
• The result of the calculation is

  • Machine learning model “ prejudice ”

Data sources

The data comes from 2021 Years and 2022 The United States since NASDAQ Index data . The data includes

• Opening price （ Open）;
• Closing price （ Close/Last）;
• All day high value （ High）;
• All day low （ Low）.

In form , No matter which value , It is a time series , The sampling interval is 1 God

among , For index , Represents the number of days .

Stock 1

One last little question , there “ Index ” and “ Stock price ” What does it matter ？

The NASDAQ-100 Index includes 100 of the largest domestic and international non-financial companies listed on The NASDAQ Stock Market based on market capitalization.

The Index reflects companies across major industry groups including computer hardware and software, telecommunications, retail/wholesale trade and biotechnology.

Simply speaking , It is also called NASDAQ 100, Is the U.S. 100 Stock price index of major non-financial companies .

In a nutshell , It represents “ US stock market ” The overall trend of . If we can predict its value , Then you can make money in the stock market .

The basic method

What is prediction ？ Is a known past value , Estimate the subsequent value

such as , I want to know tomorrow's index （）, Just use the known today （）、 yesterday （） And the day before yesterday （） To predict it .

This paper adopts three classic machine learning Methods to do this , Namely

• Linear SVR linear regression model ;
• RBF SVR Kernel regression model ;
• Bayesian Ridge Probability model .

# SVR with linear and RBF kernels
from sklearn.svm import SVR
# Bayesian Ridge Regression
from sklearn.linear_model import BayesianRidge

# Some amazing code for stock analysis
# ...

The result of the calculation is

say concretely , I use a long period of data to train the regression model

• The input parameters of the model are as follows 20 The index value of days ;
• The output parameters are “ Tomorrow, ” Index value of .

Then use the trained model , Predict the data of another period , The prediction effect is as follows

• Train in the second half , Use the first half of the time to test

  

  Stock 2

• Use the first half of the training , Test in the second half of the time

  

  Stock 3

so , Linear model （Linear） And probability models （BR） The prediction effect is better , And the nonlinear model of kernel function （RBF） The prediction effect is poor .

On the whole , Such a simple model has been able to achieve “ It looks good ” The prediction effect of .

Machine learning model “ prejudice ”

Next , Let me take a closer look at the details of the prediction , You will find a very interesting phenomenon , This involves

Machine learning model “ prejudice ” problem ,

Judging from the general trend , This index started last year , It has been showing a steady downward trend .

Because the regression model can only learn the data it sees , therefore , It must have “ prejudice ”

• Predict the index in the later period

  

  Stock 4

  Because the previous indexes are higher than the real value , therefore , Regression models include The predicted value should be slightly larger Prejudice ;

• Predict the index of the previous period

  

  Stock 5

  This result is contrary to the previous , Regression models include The predicted value should be slightly smaller Prejudice .

Because this is pure Data driven Result , therefore , From the perspective of big data, there is no problem . But emotionally , Will think that it has a certain “ ethic ” problem .

Because of the bias of previous data , It will affect the later results like a ghost . The stock market is a relatively neutral place , But if the predicted value is human “ Ability ”、“ level ” Or something , It will inevitably infiltrate human “ Moral judgment ”, Thus causing substantial inequality .