3.2 分析师推荐 • 分析师的金手指？
在我们的观点中，分析师对股票的评级以及EPS的估计，更多的是对该之股票过去一段时间表现的总结，并没有明确的预测未来的能力。鉴于分析师估计的延迟特点，在我们的策略中我们将分析师估计作为反向指标使用。粗略的说，在固定的期限内，我们买入分析师调低预期的股票，卖出分析师调高预期的股票。
本策略的参数如下：
-
起始日期： 2011年1月1日
-
结束日期： 2015年3月19日
-
股票池： 沪深300
-
业绩基准： 沪深300
-
起始资金： 100000元
-
调仓周期： 3个月
本策略使用的主要数据API有：
这里我们使用了来自于第三方朝阳永续的数据API（需要在数据商城中购买）
-
CGRDReportGGGet获取朝阳永续分析师一致评级 -
CESTReportGGGet获取朝阳永续分析师一致预期
import pandas as pd
start = datetime(2011,1, 1) # 回测起始时间
end = datetime(2015, 3, 19) # 回测结束时间
benchmark = 'HS300' # 策略参考标准
universe = set_universe('HS300') # 股票池
#universe = ['600000.XSHG', '000001.XSHE']
capital_base = 100000 # 起始资金
commission = Commission(0.0,0.0)
longest_history = 1
def CGRDwithBatch(universe, batch, startDate, endDate):
res = pd.DataFrame()
totalLength = len(universe)
count = 0
while totalLength > batch:
tmp = DataAPI.GG.CGRDReportGGGet(secID = universe[count * batch : (count + 1) * batch], BeginPubDate = startDate, EndPubDate = endDate)
count += 1
totalLength -= batch
res = res.append(tmp)
tmp = DataAPI.GG.CGRDReportGGGet(secID = universe[(count * batch):], BeginPubDate = startDate, EndPubDate = endDate)
res = res.append(tmp)
return res
def CESTwithBatch(universe, batch, startDate, endDate):
res = pd.DataFrame()
totalLength = len(universe)
count = 0
while totalLength > batch:
tmp = DataAPI.GG.CESTReportGGGet(secID = universe[count * batch : (count + 1) * batch], BeginPubDate = startDate, EndPubDate = endDate)
count += 1
totalLength -= batch
res = res.append(tmp)
tmp = DataAPI.GG.CGRDReportGGGet(secID = universe[(count * batch):], BeginPubDate = startDate, EndPubDate = endDate)
res = res.append(tmp)
return res
def MktEqudwithBatch(universe, batch, startDate, endDate):
res = pd.DataFrame()
totalLength = len(universe)
count = 0
while totalLength > batch:
tmp = DataAPI.MktEqudGet(secID = universe[count * batch : (count + 1) * batch], beginDate = startDate, endDate = endDate)
count += 1
totalLength -= batch
res = res.append(tmp)
tmp = DataAPI.MktEqudGet(secID = universe[count * batch : (count + 1) * batch], beginDate = startDate, endDate = endDate)
res = res.append(tmp)
return res
def regressionTesting(universe, startDate, endDate):
import statsmodels.api as sm
res1 = CGRDwithBatch(universe, 50, startDate, endDate).sort('publishDate')
res2 = CESTwithBatch(universe, 50, startDate, endDate).sort('publishDate')
res1 = res1[res1.RatingType == 1]
res2 = res2[res2.PnetprofitType == 1]
# got expRating change
lastRating = res1.groupby('secID').last()
firstRating = res1.groupby('secID').first()
lastRating['previousRating'] = firstRating.Rating
lastRating['chg_exp'] = lastRating.Rating / firstRating.Rating - 1.0
lowerP = lastRating['chg_exp'].quantile(0.05)
highP = lastRating['chg_exp'].quantile(0.95)
lastRating = lastRating[(lastRating['chg_exp']>lowerP) & (lastRating['chg_exp']<highP)]
lastRating['chg_exp'] = (lastRating.chg_exp - lastRating.chg_exp.mean())/lastRating.chg_exp.std()
expRating = lastRating[['secShortName', 'publishDate', 'Rating', 'previousRating', 'chg_exp']]
# got expEps change
lastEps = res2.groupby('secID').last()
firstEps = res2.groupby('secID').first()
lastEps['previousEps'] = firstEps.EPS_con
lastEps['chg_eps'] = lastEps.EPS_con / firstEps.EPS_con - 1.0
lowerP = lastEps['chg_eps'].quantile(0.05)
highP = lastEps['chg_eps'].quantile(0.95)
lastEps = lastEps[(lastEps['chg_eps']>lowerP) & (lastEps['chg_eps']<highP)]
lastEps['chg_eps'] = (lastEps.chg_eps - lastEps.chg_eps.mean())/lastEps.chg_eps.std()
expEps = lastEps[['secShortName', 'publishDate', 'EPS_con', 'previousEps', 'chg_eps']]
# Weighted Average Ranking
rankRes = expEps.copy()
rankRes['chg_exp'] = expRating.chg_exp
rankRes['ranking'] = expEps.chg_eps + expRating.chg_exp
# Current period return
mktDate = MktEqudwithBatch(universe, 50, startDate, endDate)
group = mktDate.groupby('secID')
returnRes = group.last().closePrice / group.first().closePrice - 1.0
rankRes['currentReturn'] = (returnRes - returnRes.mean()) / returnRes.std()
rankRes.dropna(inplace=True)
# Do linear regression for current return
x = rankRes[['chg_eps','chg_exp']].values
y = rankRes.currentReturn.values
x = sm.add_constant(x)
model = sm.OLS(y, x)
results = model.fit()
rankRes['resid'] = results.resid
return rankRes
def initialize(account): # 初始化虚拟账户状态
account.traded = False
account.universe = universe
account.tradingMonth = set([1,4,7,10])
account.currentTradedMonth = 0
account.previousRatingExp = None
account.previousEpsExp = None
account.holdings = set()
account.first = True
account.chosen = 0.05
def handle_data(account): # 每个交易日的买入卖出指令
today = Date(account.current_date.year, account.current_date.month, account.current_date.day)
if today.month() in account.tradingMonth and not account.traded:
hist = account.get_history(1)
account.traded = True
account.currentTradedMonth = today.month()
endDate = today
startDate = endDate - '3m'
endStr = ''.join(endDate.toISO().split('-'))
startStr = ''.join(startDate.toISO().split('-'))
res = regressionTesting(account.universe, startStr, endStr)
chosenNumber = int(account.chosen * len(res))
secids = res.sort('resid')[:chosenNumber].index.values
print today.toISO() + ' ' + str(chosenNumber) + u' 股票被选择：' + str(secids)
# clean current position
c = account.cash
for s in account.holdings:
c += hist[s]['closePrice'][-1] * account.secpos.get(s, 0)
order_to(s, 0)
equalAmount = c / chosenNumber
# order equal amount
for s in secids:
approximationAmount = int(equalAmount / hist[s]['closePrice'][-1])
order(s, approximationAmount)
account.holdings = secids
if today.month() != account.currentTradedMonth:
account.traded = False
!{}(img/20160730104832.jpg)
2011-01-05 8 股票被选择：['002252.XSHE' '000338.XSHE' '600031.XSHG' '600741.XSHG' '002024.XSHE'
'000869.XSHE' '600027.XSHG' '600588.XSHG']
2011-04-01 9 股票被选择：['600406.XSHG' '300024.XSHE' '002081.XSHE' '000776.XSHE' '002310.XSHE'
'002375.XSHE' '601933.XSHG' '600570.XSHG' '002065.XSHE']
2011-07-01 9 股票被选择：['600873.XSHG' '600415.XSHG' '002344.XSHE' '002400.XSHE' '300133.XSHE'
'002415.XSHE' '601166.XSHG' '002422.XSHE' '600887.XSHG']
2011-10-10 8 股票被选择：['600085.XSHG' '000598.XSHE' '002594.XSHE' '000157.XSHE' '600999.XSHG'
'600208.XSHG' '600252.XSHG' '600585.XSHG']
2012-01-04 9 股票被选择：['600516.XSHG' '601901.XSHG' '600348.XSHG' '600395.XSHG' '601928.XSHG'
'600352.XSHG' '600827.XSHG' '000629.XSHE' '600547.XSHG']
2012-04-05 9 股票被选择：['601929.XSHG' '300146.XSHE' '002450.XSHE' '300133.XSHE' '002603.XSHE'
'600050.XSHG' '600252.XSHG' '601800.XSHG' '600267.XSHG']
2012-07-02 9 股票被选择：['002230.XSHE' '600143.XSHG' '002310.XSHE' '000729.XSHE' '600157.XSHG'
'601258.XSHG' '600170.XSHG' '300133.XSHE' '002385.XSHE']
2012-10-08 9 股票被选择：['000869.XSHE' '002146.XSHE' '000338.XSHE' '601169.XSHG' '601336.XSHG'
'000729.XSHE' '600031.XSHG' '002594.XSHE' '600115.XSHG']
2013-01-04 9 股票被选择：['002007.XSHE' '002065.XSHE' '601928.XSHG' '000858.XSHE' '600633.XSHG'
'600519.XSHG' '600406.XSHG' '002603.XSHE' '603000.XSHG']
2013-04-01 9 股票被选择：['600809.XSHG' '000568.XSHE' '000060.XSHE' '000069.XSHE' '600549.XSHG'
'000858.XSHE' '601377.XSHG' '002653.XSHE' '000338.XSHE']
2013-07-01 9 股票被选择：['600157.XSHG' '002475.XSHE' '000001.XSHE' '600886.XSHG' '002344.XSHE'
'600028.XSHG' '600535.XSHG' '002429.XSHE' '600188.XSHG']
2013-10-08 9 股票被选择：['600372.XSHG' '600010.XSHG' '002146.XSHE' '002051.XSHE' '000999.XSHE'
'600519.XSHG' '600518.XSHG' '000024.XSHE' '601117.XSHG']
2014-01-02 8 股票被选择：['300251.XSHE' '600880.XSHG' '600633.XSHG' '601928.XSHG' '002416.XSHE'
'600637.XSHG' '600332.XSHG' '300058.XSHE']
2014-04-01 8 股票被选择：['002344.XSHE' '600880.XSHG' '002385.XSHE' '002310.XSHE' '600597.XSHG'
'600315.XSHG' '600188.XSHG' '002415.XSHE']
2014-07-01 8 股票被选择：['300146.XSHE' '000413.XSHE' '002065.XSHE' '002456.XSHE' '300058.XSHE'
'600633.XSHG' '000024.XSHE' '000400.XSHE']
2014-10-08 7 股票被选择：['600887.XSHG' '600863.XSHG' '300017.XSHE' '002292.XSHE' '002594.XSHE'
'601169.XSHG' '000400.XSHE']
2015-01-05 8 股票被选择：['600880.XSHG' '002653.XSHE' '300017.XSHE' '603000.XSHG' '002456.XSHE'
'002292.XSHE' '000963.XSHE' '300133.XSHE']
