def calcDayDescriptor(self, stock_code: str, timepoint: str):
try:
period = self.params['N']
except Exception as e:
raise ("计算平均持股成本缺少回溯期数N")
timepoint = datetime.strptime(timepoint, '%Y%m%d')
priceInfo = fundamentalApi.quotes(
asset_code=stock_code,
starttime=None,
endtime=timepoint - timedelta(days=1),
period=period,
fields=['Open', 'High', 'Low', 'Close', 'Tor_f'],
freq='d',
adj='hfq')
if ((priceInfo['Tor_f'] > 0).sum() < period * 2 / 3
or priceInfo.__len__() == 0):
cgoValue = np.nan
else:
yesterdayClose = priceInfo['Close'].values[0]
priceInfo['Tor_f'] = priceInfo['Tor_f'].clip(
upper=1) #换手率超过100%的视为100%
priceInfo.sort_index(ascending=False, inplace=True)
priceInfo['remianPart'] = (
1 - priceInfo['Tor_f']).shift(1).fillna(1).cumprod()
priceInfo['relativeWeight'] = priceInfo['remianPart'] * priceInfo[
'Tor_f']
priceInfo['priceWeight'] = priceInfo['relativeWeight'] / priceInfo[
'relativeWeight'].sum()
priceInfo['avgPrice'] = (priceInfo['High'] + priceInfo['Low'] +
priceInfo['Open'] +
priceInfo['Close']) / 4
referencePrice = (priceInfo['priceWeight'] *
priceInfo['avgPrice']).sum()
cgoValue = yesterdayClose / referencePrice - 1
return cgoValue
def calcDayDescriptor(self, stock_code: str, timepoint: str):
timepoint_dt = datetime.strptime(timepoint, "%Y%m%d")
try:
t = self.params['t'] #计算量价背离所用回溯期数
if (type(t) is not int or t < 1):
raise Exception("回溯期t必须为大于1的整数!")
except Exception as e:
print("参数异常！")
raise (e)
priceInfo = fundamentalApi.quotes(
asset_code=stock_code,
starttime=None,
endtime=timepoint_dt - timedelta(days=1),
period=t,
fields=['Open', 'High', 'Low', 'Close', 'Volume'],
freq='d',
adj='hfq')
try:
depart = 0.25 * (
np.corrcoef(priceInfo['Open'], priceInfo['Volume']) +
np.corrcoef(priceInfo['High'], priceInfo['Volume']) +
np.corrcoef(priceInfo['Low'], priceInfo['Volume']) +
np.corrcoef(priceInfo['Close'], priceInfo['Volume']))[0, 1]
except:
depart = np.nan
return depart
def calcDayDescriptor(self, stock_code: str, timepoint: str):
timepoint_dt = datetime.strptime(timepoint, "%Y%m%d")
try:
n1, n2 = self.params['wait'], self.params['cum'] #间隔期,总长度
if (type(n1) is not int or type(n2) is not int):
raise Exception("间隔期wait、区间长度cum必须为非负整数!")
if (n2 < 1):
raise Exception("区间长度cum至少为1!")
except Exception as e:
print("参数异常！")
raise (e)
priceInfo = fundamentalApi.quotes(asset_code=stock_code,
starttime=None,
endtime=timepoint_dt -
timedelta(days=1),
period=n1 + n2,
fields=['Open', 'Close'],
freq='d',
adj='hfq')
try:
momentum = priceInfo['Close'].values[n1] / priceInfo[
'Open'].values[n1 + n2 - 1] - 1
except:
momentum = np.nan
return momentum
def __settleOrder(self,stockcode:str,dealprice:float,holdAmt:int,topModelPeriod:int):
#被fire调用时执行
#以股票交易价格作为已结模型的关闭仓位价格、新开模型的开仓价格，每只股票关闭时，尝试对OrderBook表、OrderBookDetail表登记实现损益
#(这玩意不影响交易流程，但运行的时机和实际交易时机一致)
#1.标记平仓行实际收益率real_ret
#1.1 select出OrderBookDetail中所有待平仓行(所有已执行(即标记了开仓价及开仓后仓位)，但未平仓，且指令源模型已到平仓时间的指令)
#1.2 逐行循环，找出复权因子
#1.3 计算实现平仓复权价与实现开仓复权价之间的收益率
#1.4 记录收益率
CalendarCount = fApi.trade_cal('SSE', self.MODEL_BEGIN_DATE, datetime.now().strftime('%Y%m%d'), '1').__len__()
markCloseSql = 'select a.Id,a.AdjDt,a.stockcode,a.open_price from OrderBookDetail a ' \
'left join MultiFactorOptimizor b on a.Id=b.Id ' \
'where a.post_position is not null and a.real_ret is null' \
'and {CalendarCount}%b.ForcastPeriod=0 '.format(CalendarCount=CalendarCount)
closeList = pd.read_sql_query(markCloseSql,fApi.conn)
for k,v in closeList.iterrows():
id,adjdt,stkcode,open_price = v
openAdjFactor = fApi.quotes(stkcode,None,datetime.strptime(adjdt,'%Y%m%d'),1,'Adjfactor','d',None)['Adjfactor'].values[0]
nowAdjFactor = fApi.quotes(stkcode,None,datetime.now()-timedelta(days=1),1,'Adjfactor','d',None)['Adjfactor'].values[0]
openAdjPrice = open_price * openAdjFactor
closeAdjPrice = dealprice * nowAdjFactor
realRet = closeAdjPrice/openAdjPrice -1
sql = 'replace into OrderBookDetail' \
'(id, adjdt, stockcode, fore_position, adj_amount_plan, post_position_plan, post_position, open_price, exp_ret, real_ret)' \
'select a.id, a.adjdt, a.stockcode, a.fore_position, a.adj_amount_plan, a.post_position_plan' \
', a.post_position, {dealprice}, a.exp_ret, {realRet}' \
'from OrdeBookDetail a' \
'where a.stockcode=\'{stockcode}\' and a.Id={Id} and a.AdjDt=\'{adjdt}\''.format(dealprice=dealprice,realRet=realRet,stockcode=stockcode,Id=id,adjdt=adjdt)
fApi.conn.execute(sql)
#2.标记开仓行的实际开仓价格和仓位(当前阶段，将现实仓位与计划仓位差异全归于顶层模型)
#找出已下达但未执行(即未标记开仓价格和开仓后仓位）的指令，写入开仓价及开仓后仓位
markOpenSql = 'replace into OrderBookDetail' \
'(id, adjdt, stockcode, fore_position, adj_amount_plan, post_position_plan, post_position, open_price, exp_ret, real_ret)' \
'select a.id, a.adjdt, a.stockcode, a.fore_position, a.adj_amount_plan, a.post_position_plan' \
', case when c.ForcastPeriod={topModelPeriod} then {holdAmt} else a.post_position_plan end' \
', {dealprice}, a.exp_ret, a.real_ret' \
'from OrdeBookDetail a' \
'left join OrderBook b on a.Id=b.Id and a.AdjDt=b.AdjDt' \
'left join MultiFactorOptimozor c on a.Id=c.Id'\
'where a.stockcode=\'{stockcode}\' and b.IsClosed=0 and a.post_position is null'.format(topModelPeriod=topModelPeriod,holdAmt=holdAmt,dealprice=dealprice,stockcode=stockcode)
fApi.conn.execute(markOpenSql)
fApi.conn.commit()
def calcWeekDescriptor(self, stock_code: str, timepoint: str):
try:
period = self.params['period']
indexCode = self.params['index']
except Exception as e:
raise ("计算股票beta缺少回溯期数period或参照指数index")
timepoint = datetime.strptime(timepoint, '%Y%m%d')
stockReturn = fundamentalApi.quotes(asset_code=stock_code,
starttime=None,
endtime=timepoint -
timedelta(days=1),
period=period,
fields=['Close'],
freq='d',
adj='hfq')['Close'].pct_change()
indexReturn = fundamentalApi.quotes(
asset_code=indexCode,
starttime=None,
endtime=timepoint - timedelta(days=1),
period=period,
fields=['Close'],
freq='d',
adj=None,
asset_type='index')['Close'].pct_change()
mergeReturn = pd.DataFrame({
'stock': stockReturn,
'index': indexReturn
})
try:
model = sm.OLS(mergeReturn['stock'].values.reshape(-1, 1),
sm.add_constant(mergeReturn['index'].values.reshape(
-1, 1)),
missing='drop')
rst = model.fit()
Beta = rst.params[1]
except Exception as e:
Beta = np.nan
return Beta
def calcWeekDescriptor(self, stock_code: str, timepoint: str):
timepoint_dt = datetime.strptime(timepoint, "%Y%m%d")
size_df = fundamentalApi.quotes(asset_code=stock_code,
starttime=None,
endtime=timepoint_dt -
timedelta(days=1),
period=1,
fields='TotalMv',
freq='d',
adj=None)
if (size_df.__len__() == 0):
return np.nan
else:
return size_df.values[0][1]
def makeOptimizeOrder(self):
#从表MultiFactorOptimizor中获取今日激活运行的优化器，从基层到
#高层(短周期)逐个执行；
#每执行完一层改变一层当前对象base，并将新target插入OrderBook表；
#执行前检查base周期是否可整除当前模型周期；若不可整除则跳过该层模型。
#每日第二个执行
rst = self.getActiveModelStackFromDB()
num_of_models = rst.__len__()
if(num_of_models==0):
#本日非调仓日，无需操作
print('本日无模型调仓!')
return
else:
print('本日以下'+num_of_models.__str__()+'个模型调仓：')
[print('{id}号模型，预测期{p}日'.format(id=str(id),p=str(forcastperiod))) for id,model,optmizor,forcastperiod in rst]
stk_price = fApi.quotes(asset_code=None,starttime=None,endtime=datetime.now(), period=1, fields='Close', freq='day', adj=None)
#todo 这里是以收盘价格近似次日开盘价格，若当日为除权登记日(即这两个价格会有较大差异时)时不可直接近似，需要对stk_price做除权处理
stk_price.set_index('StockCode',inplace=True)
cashrow = pd.DataFrame({'Close':1},index=['cash'])
stk_price = stk_price.append(cashrow)
stk_price = stk_price['Close']
for id,model,optmizor,forcastperiod in rst:
model:MultiFactor
stk_return = model.forcastStkReturn()['e_ret']
#multifactormodel.forcastStkCov需另写
OptimizeOutcome = portfolioOptimizor(stk_return = stk_return
,stk_price = stk_price
,stk_cov=model.forcastStkCov()
,stk_hold=self.base
,feeRate=0.003
,method=optmizor
,gamma=0.1)
target = OptimizeOutcome['after_adj_hold']['hold_quantity'].to_dict() #获取调整后持仓目标
exp_ret = OptimizeOutcome['expRet']
self.base = target
#优化后，把新目标写入self.base、数据库OrderBook表、OrderBookDetail表
sql = 'insert into OrderBook(Id,AdjDt,exp_ret,IsClosed) values(?,?,?,?)'
fApi.conn.execute(sql,(id,datetime.now().strftime('%Y%m%d'),exp_ret,0))
#再将目标插入OrderBookDetail表(内含现金行)
for idx,row in OptimizeOutcome['after_adj_hold'].iterrows():
stockcode = idx
fore_position = row['remain_quantity']
adj_amount_plan = row['trade_shares']
post_position_plan = row['hold_quantity']
stk_exp_ret = row['stk_return']
stk_sql = 'insert into OrderBookDetail(Id,AdjDt,stockcode,fore_position,adj_amount_plan,post_position_plan,exp_ret)' \
'values (?,?,?,?,?,?,?)'
fApi.conn.execute(stk_sql,(id,datetime.now().strftime('%Y%m%d'),stockcode,fore_position,adj_amount_plan,post_position_plan, stk_exp_ret))
fApi.conn.commit()
def calcDayDescriptor(self, stock_code: str, timepoint: str):
timepoint_dt = datetime.strptime(timepoint, "%Y%m%d")
priceInfo = fundamentalApi.quotes(asset_code=stock_code,
starttime=None,
endtime=timepoint_dt -
timedelta(days=1),
period=1,
fields=['PB'],
freq='d',
adj=None)
try:
BTOP = 1 / priceInfo['PB'].values[0]
except Exception as e:
BTOP = np.nan
return BTOP
def calcDayDescriptor(self, stock_code: str, timepoint: str):
timepoint_dt = datetime.strptime(timepoint, "%Y%m%d")
try:
n1 = self.params['n1'] #间隔期
n2 = self.params['n2'] #总长度
if (n2 < 1):
raise ("n2至少为1!")
except Exception as e:
print("参数异常！")
raise (e)
priceInfo = fundamentalApi.quotes(asset_code=stock_code,
starttime=None,
endtime=timepoint_dt -
timedelta(days=1),
period=1,
fields=['Open', 'Close'],
freq='d',
adj=None)
def calcDayDescriptor(self, stock_code: str, timepoint: str):
timepoint_dt = datetime.strptime(timepoint, "%Y%m%d")
try:
N = self.params['N']
except Exception as e:
print("params中必须输入计算非流动性回溯期N!")
raise (e)
priceInfo = fundamentalApi.quotes(asset_code=stock_code,
starttime=None,
endtime=timepoint_dt -
timedelta(days=1),
period=N,
fields=['Open', 'Close', 'Amount'],
freq='d',
adj=None)
priceInfo['dIlliq'] = abs((priceInfo['Close'] / priceInfo['Open'] -
1)) / priceInfo['Amount'] * pow(10, 8)
Illiquidity = priceInfo['dIlliq'].mean()
return Illiquidity
def calcDayDescriptor(self, stock_code: str, timepoint: str):
timepoint_dt = datetime.strptime(timepoint, "%Y%m%d")
try:
t = self.params['t'] #计算量价背离所用回溯期数
if (type(t) is not int or t < 1):
raise Exception("回溯期t必须为大于1的整数!")
except Exception as e:
print("参数异常！")
raise (e)
priceInfo = fundamentalApi.quotes(asset_code=stock_code,
starttime=None,
endtime=timepoint_dt -
timedelta(days=1),
period=t,
fields=['Volume'],
freq='d',
adj='hfq')
try:
abnormalVol = priceInfo['Volume'][0] / np.mean(priceInfo['Volume'])
except:
abnormalVol = np.nan
return abnormalVol