예제 #1
0
 def insert_balancesheets(self):
     print( "Insert new balance sheets" )
     mdb_query = Query()
     iex = Iex()
     #Get all symbols in MongoDB
     mdb_symbols = mdb_query.get_active_companies()
     #Get current date
     currDate = datetime.datetime.now().strftime("%Y-%m-%d")
     threeMonthsAgo = (pandas.Timestamp(currDate) + pandas.DateOffset(days=-120)).strftime('%Y-%m-%d')
     #Get latest balancesheets in MongoDB for each symbol
     mdb_balancesheets = mdb_query.get_balancesheets( mdb_symbols.tolist(), currDate, "latest" )
     #Initial call to print 0% progress
     printProgressBar(0, len(mdb_symbols.index), prefix = 'Progress:', suffix = '', length = 50)
     #flag = False
     #Loop through symbols
     for index, mdb_symbol in mdb_symbols.iteritems():
         #if mdb_symbol == 'YETI':
         #    flag = True
         #if not flag:
         #    continue
         #Get matching balancesheet in MongoDB
         if not mdb_balancesheets.empty:
             mdb_balancesheet = mdb_balancesheets[ mdb_balancesheets['symbol'] == mdb_symbol ]
         else:
             mdb_balancesheet = mdb_balancesheets
         #Skip is less than 3 months since most recent
         if not mdb_balancesheet.empty: 
             mask = mdb_balancesheet['reportDate'] > threeMonthsAgo
             mdb_recent_balancesheet = mdb_balancesheet.loc[mask]
             if not mdb_recent_balancesheet.empty:
                 #Update progress bar
                 printProgressBar(index+1, len(mdb_symbols.index), prefix = 'Progress:', suffix = "Data too recent for " + mdb_symbol + "      ", length = 50)
                 continue
         #Get balancesheets from IEX
         iex_balancesheets = iex.get_balancesheets( mdb_symbol )
         #Select balancesheets more recent than MongoDB
         if not mdb_balancesheet.empty and not iex_balancesheets.empty:
             mdb_balancesheet = mdb_balancesheet.sort_values(by='reportDate', ascending=False, axis='index')
             mask = iex_balancesheets['reportDate'] > mdb_balancesheet['reportDate'].iloc[0]
             iex_balancesheets = iex_balancesheets.loc[mask]
         #Insert if balancesheets exist
         if not iex_balancesheets.empty:
             #Update progress bar
             printProgressBar(index+1, len(mdb_symbols.index), prefix = 'Progress:', suffix = "Inserting balancesheets for " + mdb_symbol + "      ", length = 50)
             self.db.iex_balancesheets.insert_many( iex_balancesheets.to_dict('records') )
         else:
             #Update progress bar
             printProgressBar(index+1, len(mdb_symbols.index), prefix = 'Progress:', suffix = "No new data for " + mdb_symbol + "      ", length = 50)
예제 #2
0
    def calculate_top_stocks(self, ref_date):
        """
        Calculate ranked list of stocks
        @params:
            ref_date    - Required  : date YYYY-MM-DD (Str)
        """

        mdb_query = Query()
        #Get ranked stock list for given date
        symbols = mdb_query.get_active_companies().tolist()
        print("Query balance sheets")
        balancesheets = mdb_query.get_balancesheets(symbols, ref_date,
                                                    "latest")
        #earnings = earnings[["EPSReportDate","actualEPS","fiscalEndDate","fiscalPeriod","symbol"]]
        #print( earnings )
        #Get financials within 6 months
        #print( "Query financials" )
        sixMonthsBeforeDate = (
            pandas.Timestamp(ref_date) +
            pandas.DateOffset(months=-6)).strftime('%Y-%m-%d')
        #financials = mdb_query.get_financials(symbols, sixMonthsBeforeDate, "after")
        balancesheets = balancesheets[
            balancesheets['reportDate'] >= sixMonthsBeforeDate]
        #financials = financials[["symbol","reportDate","netIncome","shareholderEquity"]]
        #print( financials )
        #Get prices for inception date
        print("Query prices")
        idx_min = 0
        query_num = 100
        prices = pandas.DataFrame()
        while idx_min < len(symbols):
            idx_max = idx_min + query_num
            if idx_max > len(symbols):
                idx_max = len(symbols)
            symbols_split = symbols[idx_min:idx_max]
            prices_split = mdb_query.get_quotes(symbols_split, ref_date,
                                                "latest")
            prices = prices.append(prices_split, ignore_index=True, sort=False)
            idx_min = idx_min + query_num
        #Get prices within 7 days
        fiveDaysBeforeDate = (pandas.Timestamp(ref_date) +
                              pandas.DateOffset(days=-7)).strftime('%Y-%m-%d')
        prices = prices[prices['date'] >= fiveDaysBeforeDate]
        prices.reset_index(drop=True, inplace=True)
        #print( prices )
        #Get company data
        company = mdb_query.get_company(symbols)
        company = company[['symbol', 'companyName']]
        #Merge dataframes together
        print("Merge dataframes")
        #merged = pandas.merge(earnings,financials,how='inner',left_on=["symbol","fiscalEndDate"],right_on=["symbol","reportDate"],sort=False)
        merged = pandas.merge(balancesheets,
                              prices,
                              how='inner',
                              on="symbol",
                              sort=False)
        merged = pandas.merge(merged,
                              company,
                              how='inner',
                              on='symbol',
                              sort=False)
        #Remove any rows with missing values
        merged = merged.dropna(
            axis=0,
            subset=['shareholderEquity', 'close', 'marketCap', 'peRatio'])
        #Calculate ROE
        #close / peRatio = EPS
        #marketCap / close = sharesOutstanding
        #sharesOutstanding * EPS = netIncome
        #netIncome / shareholderEquity = returnOnEquity
        merged = merged[merged.peRatio != 0]
        merged["EPS"] = merged.close / merged.peRatio
        merged = merged[merged.close != 0]
        merged["sharesOutstanding"] = merged.marketCap / merged.close
        merged["netIncome"] = merged.sharesOutstanding * merged.EPS
        merged = merged[merged.shareholderEquity != 0]
        merged["returnOnEquity"] = merged.netIncome / merged.shareholderEquity
        merged = merged[merged.returnOnEquity != 0]
        merged["peROERatio"] = merged.peRatio / merged.returnOnEquity
        #Count number of stocks above mcap value
        # A useful indicator of how universe compares to S&P500
        print("Universe before cuts...")
        print("mcap > 50M: " +
              str(merged[merged["marketCap"] > 50000000].count()["marketCap"]))
        print(
            "mcap > 100M: " +
            str(merged[merged["marketCap"] > 100000000].count()["marketCap"]))
        print(
            "mcap > 500M: " +
            str(merged[merged["marketCap"] > 500000000].count()["marketCap"]))
        print(
            "mcap > 1B: " +
            str(merged[merged["marketCap"] > 1000000000].count()["marketCap"]))
        print(
            "mcap > 5B: " +
            str(merged[merged["marketCap"] > 5000000000].count()["marketCap"]))
        print("mcap > 10B: " + str(merged[
            merged["marketCap"] > 10000000000].count()["marketCap"]))
        print("mcap > 50B: " + str(merged[
            merged["marketCap"] > 50000000000].count()["marketCap"]))
        print("mcap > 100B: " + str(merged[
            merged["marketCap"] > 100000000000].count()["marketCap"]))
        #Rank stocks
        #Cut negative PE and ROE
        merged = merged[(merged.peRatio > 0) & (merged.returnOnEquity > 0)]
        #Remove invalid stock symbols, and different voting options
        # Do the different voting options affect marketCap?
        #forbidden = [ "#", ".", "-" ]
        #merged = merged[ merged.apply( lambda x: not any( s in x['symbol'] for s in forbidden ), axis=1 ) ]
        #Remove American Depositary Shares
        #ads_str = 'American Depositary Shares'
        #merged = merged[ merged.apply( lambda x: ads_str not in x['companyName'], axis=1 ) ]
        #Remove industries that do not compare well
        # e.g. Companies that have investments as assets
        #forbidden_industry = ['Brokers & Exchanges','REITs','Asset Management','Banks']
        #merged = merged[ ~merged.industry.isin( forbidden_industry ) ]
        #Count number of stocks after cuts
        print("Universe after cuts...")
        print("mcap > 50M: " +
              str(merged[merged["marketCap"] > 50000000].count()["marketCap"]))
        print(
            "mcap > 100M: " +
            str(merged[merged["marketCap"] > 100000000].count()["marketCap"]))
        print(
            "mcap > 500M: " +
            str(merged[merged["marketCap"] > 500000000].count()["marketCap"]))
        print(
            "mcap > 1B: " +
            str(merged[merged["marketCap"] > 1000000000].count()["marketCap"]))
        print(
            "mcap > 5B: " +
            str(merged[merged["marketCap"] > 5000000000].count()["marketCap"]))
        print("mcap > 10B: " + str(merged[
            merged["marketCap"] > 10000000000].count()["marketCap"]))
        print("mcap > 50B: " + str(merged[
            merged["marketCap"] > 50000000000].count()["marketCap"]))
        print("mcap > 100B: " + str(merged[
            merged["marketCap"] > 100000000000].count()["marketCap"]))
        #Order by peROERatio
        merged = merged.sort_values(by="peROERatio",
                                    ascending=True,
                                    axis="index")

        return merged