Microblogging hot search stock selection strategy

Preface

This strategy stems from a small talk in the Group ,A Stocks are a magical market , Gu ailing wins the championship ,“ invengo ” Born in the sky ！ While lamenting that investors' imagination is too rich , According to the hot search of Weibo , Grasp the next stock limit ？ Don't ask for accuracy , Let's sort out the related technology implementation first .

Background information

source ： China Securities News

Gu Ailing is really hot ！

2 month 8 Japan , Gu ailing won the gold medal of Freestyle women's ski platform in the Winter Olympics . At that time , Swipe the screen in the circle of friends , Gu Ailing is even more popular in Weibo search . And here's the interesting thing , And Gu ailing A stocks “ Sub Valley ” Appearance —— Yuanwang Valley rare daily limit . Investors rushed to the investor interaction platform , It is suggested that the listed company sign a contract with Gu ailing .

2 month 8 The morning of , The women's Freestyle Ski platform final of the Beijing Winter Olympics was held at Shougang ski platform . People place high hopes on “ Genius girl ” Gu Ailing has lived up to expectations , In the third jump challenge, I never made a super difficult action to succeed , Put on a super reversal , In total 188.25 The score of points won the championship miraculously .

During Gu Ailing's competition , One was named “ invengo ” The stock suddenly changed . The Winter Olympics freestyle skiing women's platform final will take place on 10 when ,14 Minutes later, the stock price of Yuanwang valley began to rise suddenly ; stay 10 when 36 Minute limit , The stock price hit 8 Month high , As of the close , Yuanwang Valley newspaper 6.22 element , The total market value 46 One hundred million yuan .

Regarding this , Some shareholders of the stock bar joked ： In Yuanwang Valley, I watched Gu ailing win the championship from a distance , I didn't expect the limit to rise ！ It's like “ Gu ailing concept stock ”. There are also investors sighing ： It's really difficult to speculate in stocks now , You have to learn to make sentences ！

Get Weibo hot search words

The first is, of course, to get the hot search words of Weibo , For the convenience of back testing , It's better to have a historical record .

Find one API, Let's climb a little

import pandas as pd
import requests
import json

date = "2022-02-09"
url = 'https://google-api.zhaoyizhe.com/google-api/index/mon/sec?date='+date
r = requests.get(url)
message = json.loads(r.content)
data = message['data']
weibo = pd.DataFrame(data)
weibo['pm'] = weibo['pm'].astype('int')
weibo = weibo.sort_values(by="pm").reset_index()

Emotional categories

because A There is no short selling mechanism for stocks , We need to eliminate negative information , So first make an emotional classification

from snownlp import SnowNLP
from snownlp import sentiment

def get_sentiments(text):
    return SnowNLP(text).sentiments

weibo['sentiments'] = weibo['topic'].apply(get_sentiments)
weibo[(weibo['sentiments']>0.5) & (weibo['pm']<=3)]

among sentiments The closer the 1, The more positive ; The closer the 0, The more negative .pm Rank for hot search , Only the top three themes are selected here .

Get all stock names

The hot search topics here are relatively brief , No word bag model ,Tfidf or Word2Vec To calculate the text similarity , Use built-in difflib It will be faster . If you want to match information such as the profile of listed companies , Data cleaning can be further done , Introduce stop words , Optimize the algorithm with a larger model .

import difflib

#  Get all stock information 
df = get_all_securities(types=['stock'], date=None)

def string_similar(s1, s2):
    return difflib.SequenceMatcher(None, s1, s2).quick_ratio()

#  Calculate similarity 
df['similar'] = 0
for word in weibo[(weibo['sentiments']>0.5) & (weibo['pm']<=3)]['topic']:
    df['similar'] += df['display_name'].apply(string_similar, args=(word,))

#  Get the best match 10 Shares 
df[df['similar']>0].sort_values(by="similar", ascending=False)[:10]

Calculate the similarity between the stock name and the hot search words every time , Accumulate all matching values , Then output in reverse order , Sure enough, I can see the legend in the first four " Trading stocks " 了 .

Data back testing

Strategy is very mindless , Get hot search words before opening every day , Then select the best match 10 Shares , Divide the funds into 10 Share purchase ; Sell all stocks that are not in the hot search the next day , Re match new shares .

Although this stock selection is too wonderful and unreasonable , But we still “ rigorous ” some , Do a back test with the data since this year .

Encapsulating utility functions

Encapsulate the obtained hot search data into a letter

List of stocks to be purchased

The main task before opening is to get ready , List of stocks to be purchased

##  Run the function before opening 
def before_market_open(context):
    #  Output run time 
    log.info(' Function run time (before_market_open)：'+str(context.current_dt.time()))

    #  The maximum number of stocks bought per day 
    g.daily_buy_count  = 10
    g.sentiments = 0.5  #  Emotional threshold 
    g.pm = 3  #  Ranking threshold 
    date = context.current_dt
    weibo = get_weibo(date.strftime('%Y-%m-%d'))
    
    if weibo is not None:
        df = get_all_securities(types=['stock'], date=None)
        df['similar'] = 0
        for word in weibo[(weibo['sentiments']>g.sentiments) & (weibo['pm']<=g.pm)]['topic']:
            df['similar'] += df['display_name'].apply(string_similar, args=(word,))
        
        #  Get the purchase list 
        stocks = df[df['similar']>0].sort_values(by="similar", ascending=False)
        stocks = stocks[:g.daily_buy_count].index
        
        g.buy_list = list(stocks)
    else:
        g.buy_list = []
        
    print(g.buy_list)

Buy sell

Sell stocks that are not in the stock pool , Then divide the cash equally , Full position purchase .

##  Run the function at the opening 
def market_open(context):
    log.info(' Function run time (market_open):'+str(context.current_dt.time()))
    
    #  Sell stocks that are not in the stock pool 
    sell_list = set(context.portfolio.positions.keys()) - set(g.buy_list)
    for stock in sell_list:
        order_target_value(stock, 0)  
    
    #  Get current cash , divide equally 
    if len(g.buy_list) > 0:
        cash = context.portfolio.available_cash / len(g.buy_list)
    
    for security in g.buy_list:
        order_value(security, cash)
        #  Record this purchase 
        log.info("Buying %s" % (security))

This strategy is very radical and wild , I didn't expect to beat the index ......

Don't do this in actual combat , If you can provide a more perspective , Let more hot stocks come into view when choosing stocks , It should be all the purposes of this algorithm .

Source download

Relevant documents in this issue , Available in the public number “ Deep awakening ”, The background to reply ：“trade03”, Get download link .