multi-factor.py

#在在线平台上的代码
import pandas as pd
import numpy as np
import math
from sklearn.model_selection import GridSearchCV  
from sklearn.model_selection import learning_curve
from sklearn.linear_model import LinearRegression
from sklearn.linear_model import LinearRegression
from sklearn import preprocessing
import jqdata

#该函数能够获得指定时间，指定股票成分股的截面数据
def getdata(stock,datetime,gap):
    #调取截面数据、计算月收益率数据
    sample = get_index_stocks(stock, date=datetime)#股票除选择沪深300银行指数成分股
    q = query(valuation.code, 
              valuation.market_cap,#市值
              valuation.circulating_market_cap,
              balance.total_assets - balance.total_liability,#净资产
              balance.total_assets / balance.total_liability, 
              indicator.net_profit_to_total_revenue, #净利润/营业总收入
              indicator.inc_revenue_year_on_year,  #营业收入增长率（同比）
              balance.development_expenditure, #RD
              valuation.pe_ratio, #市盈率（TTM）
              valuation.pb_ratio, #市净率（TTM）
              indicator.inc_net_profit_year_on_year,#净利润增长率（同比）
              balance.dividend_payable,
              indicator.roe,
              indicator.roa,
              income.operating_profit / income.total_profit, #OPTP
              indicator.gross_profit_margin, #销售毛利率GPM
              balance.fixed_assets / balance.total_assets, #FACR
              valuation.pcf_ratio, #CFP
              valuation.ps_ratio #PS
            ).filter(
                valuation.code.in_(sample)
            )
    df = get_fundamentals(q, date =datetime)
    #df.index = df.code.values
    del df['code']
    df = df.fillna(0)
    
    #解决数据日期问题 选取未来一个月的日期
    year = int(datetime[:4])
    month = int(datetime[5:7])
    day=int(datetime[8:10])
    if month == 12:       
        nexttime = str(year+1)+'-'+str(gap)+'-'+str(day)
    else:  
        nexttime = str(year)+'-'+str(month+gap)+'-'+str(day)
        
    #构建特征和标签
    sample2 = get_index_stocks(stock, date=nexttime)#股票除选择沪深300银行指数成分股
    q2 = query(valuation.code, 
              valuation.market_cap,#市值
            ).filter(
                valuation.code.in_(sample2)
            )
    df2=get_fundamentals(q2, date = nexttime)
    del df2['code']
    monthreturn=pd.DataFrame(df2['market_cap'].values/df['market_cap'].values, columns=['monthreturn'])
    df=pd.concat([df,monthreturn],axis=1)
    #df即为我们的训练集 用来训练回归模型 筛选变量和权重
    return df

#构建一个日期表，从这些日期中选取截面数据
def return_datelist(start_date,end_date):
    datelist=[]
    #start_date:str, 初始日期
    #end_date:str,终止日期
    start_year = int(start_date[:4])
    end_year = int(end_date[:4])
    start_month = int(start_date[5:7])
    end_month = int(end_date[5:7])
    start_day=int(start_date[8:10])
    end_day=int(end_date[8:10])
    for i in range(end_month):
        if(end_month-i<10):
            date=str(end_year)+'-'+'0'+str(end_month-i)+'-'+str(end_day)
            datelist=datelist+[date]
        else:
            date=str(end_year)+'-'+str(end_month-i)+'-'+str(end_day)
            datelist=datelist+[date]
    for i in range(13-start_month):
        if(start_month+i<10):
            date=str(start_year)+'-'+'0'+str(start_month+i)+'-'+str(start_day)
            datelist=datelist+[date]
        else:
            date=str(start_year)+'-'+str(start_month+i)+'-'+str(start_day)
            datelist=datelist+[date]          
        current_year=start_year+1
    while(current_year!=end_year):
        for i in range(11):
            if(i+1<10):
                date=str(current_year)+'-'+'0'+str(i+1)+'-'+str(start_day)
                datelist=datelist+[date]
            else:
                date=str(current_year)+'-'+str(i+1)+'-'+str(start_day)
                datelist=datelist+[date]
        current_year=current_year+1
    return datelist    

#获得训练数据集
datelist=return_datelist(start_date='2013-10-15',end_date='2015-10-15')
df=getdata(stock='399951.XSHE',datetime='2013-10-15',gap=1)
for i in datelist:
    dfnew=getdata(stock='399951.XSHE',datetime=i,gap=1)
    df=pd.concat([df,dfnew],axis=0)
    
min_max_scaler = preprocessing.MinMaxScaler()
df1=min_max_scaler.fit_transform(df)
df2 = pd.DataFrame(df1, columns=df.columns)

y=df2['monthreturn']
x=df2.drop(columns = ['monthreturn'])

#利用线性模型得到权重
from sklearn.linear_model import LinearRegression
lr=LinearRegression()
lr.fit(x, y)

#利用线性模型得到权重
from sklearn.linear_model import Lasso
lasso = Lasso(fit_intercept=False)
lasso.set_params(alpha=0.004)
lasso.fit(x, y)

#利用线性模型得到权重
from sklearn.linear_model import ElasticNet
enet=ElasticNet(alpha=.01,l1_ratio=0.06)
enet.fit(x, y)


def initialize(context):
    set_params()
    set_backtest()
    run_daily(trade, 'every_bar')
    
def set_params():
    g.days = 0
    g.refresh_rate = 10
    g.stocknum = 10
    
def set_backtest():
    set_benchmark('399951.XSHE')
    set_option('use_real_price', True)
    log.set_level('order', 'error')
    
def trade(context):
    if g.days % 10 == 0:
        #开始模拟投资 读取当天的相关截面数据 预测收益率 并对股票按收益率排序
        sample3 = get_index_stocks('399951.XSHE', date=None)#股票除选择沪深300银行指数成分股
        q3 = query(valuation.code, 
        valuation.market_cap,#市值
        valuation.circulating_market_cap,
        balance.total_assets - balance.total_liability,#净资产
        balance.total_assets / balance.total_liability, 
        indicator.net_profit_to_total_revenue, #净利润/营业总收入
        indicator.inc_revenue_year_on_year,  #营业收入增长率（同比）
        balance.development_expenditure, #RD
        valuation.pe_ratio, #市盈率（TTM）
        valuation.pb_ratio, #市净率（TTM）
        indicator.inc_net_profit_year_on_year,#净利润增长率（同比）
        balance.dividend_payable,
        indicator.roe,
        indicator.roa,
        income.operating_profit / income.total_profit, #OPTP
        indicator.gross_profit_margin, #销售毛利率GPM
        balance.fixed_assets / balance.total_assets, #FACR
        valuation.pcf_ratio, #CFP
        valuation.ps_ratio #PS
        ).filter(
            valuation.code.in_(sample3)
            )
        df = get_fundamentals(q3,date=None)
        df.index = df.code.values
        del df['code']
        df = df.fillna(0)

        #实现预测+排序
        min_max_scaler = preprocessing.MinMaxScaler()
        df1=min_max_scaler.fit_transform(df)
        df2 = pd.DataFrame(df1, columns=df.columns,index=df.index)
        x=df2
        factor = pd.DataFrame(lr.predict(x), index = df.index, columns = ['monthreturn'])
        factor = factor.sort_index(by = 'monthreturn')#做了个排序
        stockset = list(factor.index[:10])#选排名前十的股票

        sell_list = list(context.portfolio.positions.keys())
        for stock in sell_list:
            if stock not in stockset[:g.stocknum]:
                stock_sell = stock
                order_target_value(stock_sell, 0)
            
        if len(context.portfolio.positions) < g.stocknum:
            num = g.stocknum - len(context.portfolio.positions)
            cash = context.portfolio.cash/num
        else:
            cash = 0
            num = 0
        for stock in stockset[:g.stocknum]:
            if stock in sell_list:
                pass
            else:
                stock_buy = stock
                order_target_value(stock_buy, cash)
                num = num - 1
                if num == 0:
                    break
        g.days += 1
    else:
        g.days = g.days + 1