互联网+量化投资 大数据指数手把手

来源：https://uqer.io/community/share/55263359f9f06c8f3390457b

策略简介

从公司基本面、市场驱动指标、市场情绪等多维度验证拥有“天时、地利、人和”的大牛股，让每个人都能生产符合自己投资理念的大数据指数。实现中参考了水星社区中的牛人@吴宇笛的因子计分卡策略。

本策略的参数如下：

• 起始日期： 2014年1月1日
• 结束日期： 2016年5月18日
• 股票池： 上证50
• 业绩基准： 上证50
• 起始资金： 100000元
• 调仓周期： 1个月

策略参数获取：

• 十日移动均线(MA10) 60日移动均线(MA60) 资产回报率(ROA) 市盈率(PE) 对数市值(LCAP) 波幅中位数(DHILO) 净利润/营业总收入(NPToTOR) 产权比率(DebtEquityRatio) 营业利润同比增长(OperatingProfitGrowRate) 总资产同比增长(TotalAssetGrowRate) 均可以通过DataAPI.MktStockFactorsDateRangeGet获得
• 市场新闻热度指标可以通过DataAPI.NewsHeatIndexGet获得
• 市场情绪指标可以通过DataAPI.NewsSentimentIndexGet获得；与新闻热度指标一样，都是DataYes利用大数据分析从海量关联新闻中提取出来的

调仓策略

(1) 对每只股票获取之前的120个交易日的收盘价，计算20日累计收益，共得到100个收益率数据

(2) 获取该股票同期的100个交易日的基本面、市场驱动指标和市场热度、情绪指标，分别计算均值、标准差，并进行中心化

(3) 以该股票20日累计收益率为因变量，基本面、市场驱动指标和市场热度、情绪指标为自变量进行弹性网 ( ElasticNet ) 回归

(4) 获取该股票前一日的基本面、市场驱动指标和市场热度、情绪指标

(5) 对该股票前一日的基本面、市场驱动指标和市场热度、情绪指标，依据前100个交易日的均值和标准差，置相对大小为 （前一日值 - 均值）/ 标准差 并四舍五入，作为在该项因子上的得分

(6) 根据之前计算出的权重对这些得分进行加总，得到该股票的得分，并以此为指数进行股票筛选

(7) 根据指数得分排序，选取总分最高的前五支股票作为买入列表

(8) 根据买入列表调仓

import pandas as pd
import numpy  as np
import statsmodels.api as sm
import statsmodels.regression.linear_model as lm
from sklearn.linear_model import ElasticNet
from CAL.PyCAL import *

used_factors = ['MA10', 'MA60', 'ROA', 'PE', 'LCAP', 'DHILO', 'DebtEquityRatio', 'OperatingProfitGrowRate', 'TotalAssetGrowRate', 'NPToTOR']

#used_factors = ['ASSI', 'EBITToTOR', 'ETP5', 'MA60', 'HSIGMA', 'PE', 'VOL60', 'SUE', 'DAVOL20', 'TotalAssetGrowRate']

def StockFactorsGet(universe, trading_days):
    data_all = {}
    for i,stock in enumerate(universe):
        try:
            data = DataAPI.MktStockFactorsDateRangeGet(secID = stock, beginDate = trading_days[0], endDate = trading_days[-1], field = ['tradeDate'] + used_factors)
            # data['tradeDate'] = pd.to_datetime(data['tradeDate'])
        except Exception, e:
            print e

        try:
            news_data = DataAPI.NewsHeatIndexGet(secID = stock, beginDate = trading_days[0], endDate = trading_days[-1])
            heatIndex = news_data.set_index('newsPublishDate').sort_index().reset_index()[['heatIndex','newsPublishDate']]
            heatIndex['flag'] = heatIndex['newsPublishDate'].apply(lambda x: True if x in data.tradeDate.values else False)
            heatIndex = heatIndex[heatIndex.flag].reset_index()
            data = pd.merge(data, heatIndex, how = 'inner', left_index = 'tradeDate', right_index = 'newsPublishDate').drop(['index','newsPublishDate','flag'], 1)
        except Exception, e:
            data['heatIndex'] = 0

        try:
            emotion_data = DataAPI.NewsSentimentIndexGet(secID = stock, beginDate = trading_days[0], endDate = trading_days[-1])
            emotionIndex = emotion_data.set_index('newsPublishDate').sort_index().reset_index()[['sentimentIndex','newsPublishDate']]
            emotionIndex['flag'] = emotionIndex['newsPublishDate'].apply(lambda x: True if x in data.tradeDate.values else False)
            emotionIndex = emotionIndex[emotionIndex.flag].reset_index()
            data = pd.merge(data, emotionIndex, how = 'inner', left_index = 'tradeDate', right_index = 'newsPublishDate').drop(['index','newsPublishDate','flag'], 1)
        except Exception, e:
            # print 'emotion', stock, e
            data['sentimentIndex'] = 0

        data['news_emotion'] = data['heatIndex'] * data['sentimentIndex']

        data_all[stock] = data
    return data_all

def StockRegDataGet(stock, trading_days, factors, shift = 20):
    start = trading_days[0]
    end   = trading_days[-1]
    data  = factors[(factors.tradeDate >= start.strftime('%Y-%m-%d')) & (factors.tradeDate <= end.strftime('%Y-%m-%d'))][:-shift]

    ret = DataAPI.MktEqudGet(secID = stock, beginDate = start.strftime('%Y%m%d'), endDate = end.strftime('%Y%m%d'), field = ['tradeDate', 'closePrice'])
    ret['fwdPrice'] = ret['closePrice'].shift(-shift)
    ret['return'] = ret['fwdPrice'] / ret['closePrice'] - 1.
    ret = ret[:-shift]

    data = data.merge(ret, how = 'inner', left_on = ['tradeDate'], right_on = ['tradeDate'])
    data = data.loc[:, ['return', 'heatIndex', 'sentimentIndex', 'news_emotion'] + used_factors]
    return data

def GetRegressionResult(data):
    data = data.dropna()

    all_factors = ['heatIndex', 'sentimentIndex', 'news_emotion'] + used_factors
    for f in all_factors:
        if data[f].std() == 0:
            continue
        data[f] = (data[f] - data[f].mean()) / data[f].std()

    y = np.array(data['return'].tolist())
    x = []
    for f in all_factors:
        x.append(data[f].tolist())
    x = np.column_stack(tuple(x))
    x = np.array( [ np.append(v,1) for v in x] ) 

    en = ElasticNet(fit_intercept=True, alpha=0)
    en.fit(x, y)
    res = en.coef_[:-1]
    w = dict(zip(all_factors, res))
    return w

def preparing(universe, date, factors_all):
    date = Date(date.year, date.month, date.day)

    cal = Calendar('China.SSE')
    start = cal.advanceDate(date, '-120B', BizDayConvention.Following)
    end   = cal.advanceDate(date, '-1B',   BizDayConvention.Following)

    start = datetime(start.year(), start.month(), start.dayOfMonth())
    end   = datetime(  end.year(),   end.month(),   end.dayOfMonth())

    trading_days = quartz.utils.tradingcalendar.get_trading_days(start, end)
    datas, means, vols, weights = {}, {}, {}, {}
    for i,stock in enumerate(universe):
        try:
            datas[stock]   = StockRegDataGet(stock, trading_days, factors_all[stock])
            means[stock]   = dict(datas[stock].mean())
            vols[stock]    = dict(datas[stock].std())
            weights[stock] = GetRegressionResult(datas[stock])
        except Exception, e:
            pass
    return means, vols, weights

from datetime import datetime
end   = datetime(2016, 5, 18)
f_start = datetime(2014, 1, 1)
universe = set_universe('SH50')
f_days = quartz.utils.tradingcalendar.get_trading_days(f_start, end)
factors_all = StockFactorsGet(universe, f_days)

from datetime import datetime
start = datetime(2014, 6, 1)
end   = datetime(2016, 5, 18)
benchmark = 'SH50'
universe = set_universe('SH50')
capital_base = 100000
refresh_rate = 20

# f_start = datetime(2012, 6, 1)
# f_days = quartz.utils.tradingcalendar.get_trading_days(f_start, end)
# factors_all = StockFactorsGet(universe, f_days)

def initialize(account):
    pass

def handle_data(account):
    print account.current_date
    means, vols, weights = preparing(account.universe, account.current_date, factors_all)

    cal  = Calendar('China.SSE')
    date = Date(account.current_date.year, account.current_date.month, account.current_date.day)
    date = cal.advanceDate(date, '-1B', BizDayConvention.Following)
    date = datetime(date.year(), date.month(), date.dayOfMonth())

    factors_cur = StockFactorsGet(account.universe, [date])

    score = {}
    all_factors = ['heatIndex', 'sentimentIndex', 'news_emotion'] + used_factors
    for stock in account.universe:
        if stock not in weights:
            continue

        fac = factors_cur[stock]
        s = 0
        for f in all_factors:
            try:
                x = fac[f].iloc[-1]
                x = (x - means[stock][f])/vols[stock][f]
                s += weights[stock][f] * int(round(x))
            except:
                pass
        score[stock] = s

    buylist = sorted(score.keys(), key = lambda x: score[x])[-5:]
    rebalance(account, buylist)

def rebalance(account, buylist):
    for stock in account.valid_secpos:
        if stock not in buylist:
            order_to(stock, 0)

    for stock in buylist:
        order(stock, account.referencePortfolioValue / len(buylist) / account.referencePrice[stock])

2014-06-03 00:00:00
2014-07-01 00:00:00
2014-07-29 00:00:00
2014-08-26 00:00:00
2014-09-24 00:00:00
2014-10-29 00:00:00
2014-11-26 00:00:00
2014-12-24 00:00:00
2015-01-23 00:00:00
2015-02-27 00:00:00
2015-03-27 00:00:00
2015-04-27 00:00:00
2015-05-26 00:00:00
2015-06-24 00:00:00
2015-07-22 00:00:00
2015-08-19 00:00:00
2015-09-18 00:00:00
2015-10-23 00:00:00