def get_weights(self):
code_list = list_jq2wind(self.code_list)
code_weights = {}
w.start()
weight_data = np.array(
w.wss(code_list, "mkt_cap_float",
"unit=1;tradeDate=" + self.date + ";currencyType=").Data[0])
weight_data = weight_data / np.sum(weight_data)
for i in range(len(code_list)):
code = code_list[i]
code_weights[list_wind2jq([code])[0]] = weight_data[i]
return code_weights
def algo(context):
date_now = context.now.strftime('%Y-%m-%d')
date_previous = get_trading_date_from_now(date_now, -1,
ql.Days) # 前一个交易日，用于获取因子数据的日期
if date_now not in trading_date_list: # 非调仓日
pass # 预留非调仓日的微调空间
else: # 调仓日执行算法
print(date_now + '日回测程序执行中...')
try:
code_list = list_gm2wind(
list(
get_history_constituents(
INDEX,
start_date=date_previous,
end_date=date_previous)[0]['constituents'].keys()))
except IndexError:
code_list = w.wset("sectorconstituent", "date=" + date_previous +
";windcode=" + INDEX).Data[1]
I = trading_date_list.index(date_now)
trading_dates = trading_date_list[I - HISTORY_LENGTH:I + 1]
data_dfs = []
for i in range(len(trading_dates) - 1):
date_start = get_trading_date_from_now(
trading_dates[i], -1, ql.Days) # 计算因子值的日子，买入前一日的因子值
date_end = get_trading_date_from_now(trading_dates[i + 1], -1,
ql.Days) # 计算收益率到期的日子-收盘
factors_df = get_factor_from_wind(code_list, FACTOR_LIST,
date_start) # 获取因子
return_df = get_return_from_wind(code_list, date_start, date_end)
factors_df_and_return_df = pd.concat(
[factors_df, return_df], axis=1).dropna() # 去掉因子或者回报有空缺值的样本
factors_df_and_return_df = sort_data(
factors_df_and_return_df) # 使用排序数据作为输入
data_dfs.append(factors_df_and_return_df)
factors_return_df = pd.concat(data_dfs,
axis=0) # 获取的最终训练数据拼接，return为目标
# 根据data_df训练模型
model = OrdinaryLinearRegression()
model.fit(factors_return_df)
# 根据factor_date_previous选取股票
factor_date_previous_df = get_factor_from_wind(code_list, FACTOR_LIST,
date_previous).dropna()
sorted_codes = model.predict(
factor_date_previous_df) # 获取预测收益率从小到大排序的股票列表
sorted_codes = list_wind2jq(sorted_codes)
# 根据股票列表下单
stock_codes = sorted_codes[-STOCK_NUMBER:]
stock_now = {}
for stock_code in stock_codes: # 平均持仓持股
stock_now[stock_code] = 1. / STOCK_NUMBER
stock_dict[date_now] = stock_now
def algo(context):
date_now = context.now.strftime('%Y-%m-%d')
date_previous = get_trading_date_from_now(date_now, -1, ql.Days) # 前一个交易日，用于获取因子数据的日期
if date_now not in trading_date_list: # 非调仓日
pass # 预留非调仓日的微调空间
else: # 调仓日执行算法
print(date_now+'日回测程序执行中...')
# 根据指数获取股票候选池的代码
select_code_list = list_wind2jq(INCLUDED_INDEX)
if len(select_code_list) > 0: # 有可选股票时选取合适的股票
stock_now = WEIGHTS(select_code_list, date_previous).get_weights()
stock_dict[date_now] = stock_now
else:
stock_dict[date_now] = {}
def get_weights(self):
code_list = list_jq2wind(self.code_list)
w.start()
return_value = np.array(
w.wsd(code_list, "pct_chg", "ED-" + str(self.N - 1) + "TD",
self.date, "").Data)
return_value = return_value.transpose()
return_value = pd.DataFrame(return_value, columns=code_list)
optimizer = HRPOpt(return_value)
hrp_portfolio = optimizer.hrp_portfolio()
hrp_portfolio = dict(
zip(list_wind2jq(list(hrp_portfolio.keys())),
hrp_portfolio.values()))
return hrp_portfolio
def algo(context):
date_now = context.now.strftime('%Y-%m-%d')
date_previous = get_trading_date_from_now(date_now, -1, ql.Days) # 前一个交易日，用于获取因子数据的日期
if date_now not in trading_date_list: # 非调仓日
pass # 预留非调仓日的微调空间
else: # 调仓日执行算法
print(date_now+'日回测程序执行中...')
# 根据指数获取股票候选池的代码
code_list = SelectedStockPoolFromListV1(INCLUDED_INDEX, EXCLUDED_INDEX, date_previous).get_stock_pool()
# 按照选股策略选股，与选股策略接口保持一致
strategy = STRATEGY(code_list, date_previous)
select_code_list = list_wind2jq(strategy.select_code())
if len(select_code_list) > 0: # 有可选股票时选取合适的股票
stock_now = WEIGHTS(select_code_list, date_previous).get_weights()
stock_dict[date_now] = stock_now
else:
stock_dict[date_now] = {}
def get_weights(self):
code_list = list_jq2wind(self.code_list)
SW1_code_list = [t[0] for t in SW1_INDEX]
weight_value = self._calc_weights(SW1_code_list) # 提取行业权重
industry_list = w.wss(code_list, "indexcode_sw", "tradeDate="+self.date+";industryType=1").Data[0]
weight_value_temp = []
for i in range(len(code_list)):
industry_temp = industry_list[i]
if industry_temp is None: # 个股无行业分类数据的处理
weight_value_temp.append(0.0)
else:
industry_index = SW1_code_list.index(industry_temp)
weight_value_temp.append(weight_value[industry_index])
weight_value_temp = np.array(weight_value_temp)
weight_value_temp = weight_value_temp / np.sum(weight_value_temp) # 权重归一化
code_weights = dict([[list_wind2jq([code_list[i]])[0], weight_value_temp[i]] for i in range(len(code_list))])
return code_weights
def algo(context):
date_now = context.now.strftime('%Y-%m-%d')
date_previous = get_trading_date_from_now(date_now, -1,
ql.Days) # 前一个交易日，用于获取因子数据的日期
select_time_value = select_time_model[date_now] # 择时信号计算
select_time_dict[date_now] = select_time_value
print(date_now + ('日回测程序执行中...，择时值：%.2f' % select_time_value))
if date_now not in trading_date_list: # 非调仓日
pass
else: # 调仓日执行算法
# 根据指数获取股票候选池的代码
code_list = SelectedStockPoolFromListV1(
INCLUDED_INDEX, EXCLUDED_INDEX, date_previous).get_stock_pool()
strategy = STRATEGY(code_list, date_previous, 0.9)
select_code_list = list_wind2jq(strategy.select_code())
if len(select_code_list) > 0: # 有可选股票时选取合适的股票
stock_now = WEIGHTS(select_code_list, date_previous).get_weights()
else:
stock_dict[date_now] = {}
def algo(context):
date_now = context.now.strftime('%Y-%m-%d')
print(date_now + '日回测程序执行中...')
if date_now not in context.trading_date_list: # 非调仓日
pass # 预留非调仓日的微调空间
else: # 调仓日执行算法
date_previous = get_trading_date_from_now(
date_now, -1, ql.Days) # 前一个交易日，用于获取因子数据的日期
code_list = SelectedStockPoolFromListExcluded(
context.INCLUDED_INDEX, context.EXCLUDED_INDEX,
date_previous).get_stock_pool()
factor = context.FACTOR(date_previous, code_list) # 读取单因子的代码
df = factor.get_factor().dropna()
quantiles = df.quantile(context.QUANTILE).values # 取对应的分位数值
stock_codes = list(
df[(df[factor.factor_name] >= quantiles[0][0])
& (df[factor.factor_name] < quantiles[1][0])].index.values)
stock_codes = list_wind2jq(stock_codes) # 仓位配置前转换为聚宽代码
stock_now = context.WEIGHTS_FUNCTION(stock_codes,
date_previous).get_weights()
stock_dict[date_now] = stock_now
def algo(context):
global position_now, position_target
date_now = context.now.strftime('%Y-%m-%d')
date_previous = get_trading_date_from_now(date_now, -1,
ql.Days) # 前一个交易日，用于获取因子数据的日期
select_time_value = select_time_model[date_now] # 择时信号计算
if date_now not in trading_date_list: # 非调仓日
pass
else: # 调仓日执行算法，更新position_target
position_now = False # 虚拟上，调仓日需要提前清仓
stock_dict[date_now] = {}
# 根据指数获取股票候选池的代码
all_code_set = set()
for index in INDEX:
code_set = set(
w.wset("sectorconstituent",
"date=" + date_previous + ";windcode=" + index).Data[1])
all_code_set = all_code_set | code_set
code_list = list(all_code_set) # 股票候选池
strategy = STRATEGY(code_list, date_previous, 0.9)
select_code_list = list_wind2jq(strategy.select_code())
if len(select_code_list) > 0: # 有可选股票时记录下可选股票
stock_now = WEIGHTS(select_code_list, date_previous).get_weights()
position_target = stock_now
else:
position_target = {}
# 择时判定
if select_time_value >= 0 and not position_now and position_target != {}: # LLT择时信号为正,空仓且有目标持仓状态
stock_dict[date_now] = position_target
position_now = True
elif select_time_value < 0 and position_now and position_target != {}: # LLT择时信号为负且持仓状态:
stock_dict[date_now] = {}
position_now = False
# 打印择时信号和仓位配置信息
print(date_now + ('日回测程序执行中...，择时值：%.2f' % select_time_value))
print(date_now + '日回测程序执行中...，选股：')
print(position_target)
def get_weights(self):
code_list = list_jq2wind(self.code_list)
code_weights = {}
return_value = self._get_return(code_list)
def optimization_target(weight_temp):
weight_temp = weight_temp[:, np.newaxis]
portfolio_return_value = np.matmul(weight_temp.transpose(),
return_value)[0, :] # 组合的投资回报序列
mean = np.mean(portfolio_return_value)
var = np.var(portfolio_return_value)
skew = scipy.stats.skew(portfolio_return_value)
kurtosis = scipy.stats.kurtosis(portfolio_return_value)
loss = self.w2 * var + self.w3 * skew + self.w4 * kurtosis # 对方差、偏度、丰度进行优化，期望值最小越好
return loss
def constraint(weight_temp):
return np.sum(weight_temp) - 1.0
n = len(self.code_list) # 资产个数
x_0 = np.ones((n, )) * 1.0 / n
bounds = n * ((0.0, None), ) # 权重非负约束条件
res = scipy.optimize.minimize(optimization_target,
x_0,
method='SLSQP',
bounds=bounds,
constraints={
'type': 'eq',
'fun': constraint
})
# print(res)
if res.success: # 优化成功
x = res.x
else: # 优化失败，用等权
x = x_0
for i in range(len(code_list)):
code = code_list[i]
code_weights[list_wind2jq([code])[0]] = x[i]
return code_weights
def get_weights(self):
code_list = list_jq2wind(self.code_list)
stock_ZX1_industry_code_list = w.wss(
code_list, "indexcode_citic",
"industryType=1;tradeDate=" + self.date).Data[0]
ZX1_industry_code_list = list(set(stock_ZX1_industry_code_list))
weight_value = self._calc_weights(list(
set(ZX1_industry_code_list))) # 计算股票涉及到的行业权重
# industry_list = w.wss(code_list, "indexcode_sw", "tradeDate="+self.date+";industryType=1").Data[0]
weight_value_temp = []
for i in range(len(code_list)):
industry_temp = stock_ZX1_industry_code_list[i]
if industry_temp is None: # 个股无行业分类数据的处理
weight_value_temp.append(0.0)
else:
industry_index = ZX1_industry_code_list.index(industry_temp)
weight_value_temp.append(weight_value[industry_index])
weight_value_temp = np.array(weight_value_temp)
weight_value_temp = weight_value_temp / np.sum(
weight_value_temp) # 权重归一化
code_weights = dict(
[[list_wind2jq([code_list[i]])[0], weight_value_temp[i]]
for i in range(len(code_list))])
return code_weights
risk_budget.append(value_ratio)
risk_bounds.append([0.4, 0.85]) # 权重约束设置
elif value_code in [
'513050.SH', '513100.SH', '513500.SH', '518880.SH'
]: # 外盘黄金类
stock_pool.append(value_code)
risk_budget.append(value_ratio)
risk_bounds.append([0.0, 0.2]) # 权重约束设置
else: # 国内股票类
stock_pool.append(value_code)
risk_budget.append(value_ratio)
risk_bounds.append([0.0, 1.0]) # 权重约束设置
S_result.append([stock_pool, risk_budget, np.array(risk_bounds)])
# 策略0部分
stock_pool_0 = list_wind2jq(S_result[0][0])
risk_budget_0 = S_result[0][1]
risk_bounds_0 = S_result[0][2]
stock_weights_0 = 风险预算组合_模块求解基本版_带约束(stock_pool_0,
DATE,
risk_budget=risk_budget_0,
bounds=risk_bounds_0).get_weights()
# 加入信用债品种，对国债配置比例进行修正
stock_weights_0 = adjust_weights(stock_weights_0, BOND_ADJUST, STOCK_ADJUST)
# 输出文字信息
string_output_0 = '策略0的风险预算权重为：\n'
string_output_0 += str(dict(zip(stock_pool_0, risk_budget_0))) + '\n'
string_output_0 += '策略0的选股结果（选股日为 ' + DATE + ' 收盘）：\n'
string_output_0 += str(stock_weights_0) + '\n'
print(string_output_0)
stock_pool_bond = ['511010.SH']
risk_budget_bond = [1.0] # 候选目标[0.2, 0.8, 0.8]
risk_bounds_bond = np.array([[0.0, 1.0]])
# 国内股票部分配置方案
stock_pool_stock = ['159949.SZ', '159928.SZ', '510050.SH', '510500.SH']
risk_budget_stock = [1.0, 1.0, 1.0, 1.0] # 总权重固定为4个单位
risk_bounds_stock = np.array([[0.0, 1.0]] * len(stock_pool_stock))
# 国际部分配置方案
stock_pool_global = ['513100.SH', '513500.SH', '518880.SH']
risk_budget_global = [0.75, 0.75, 0.8] # 在区间[0.75, 0.75, 0.8]-[1.0, 1.0, 1.0]之间灵活调整
risk_bounds_global = np.array([[0.0, 1.0]] * len(stock_pool_global))
# 合并为整体证券池
stock_pool = stock_pool_bond + stock_pool_stock + stock_pool_global
risk_budget = risk_budget_bond + risk_budget_stock + risk_budget_global
risk_bounds = np.concatenate([risk_bounds_bond, risk_bounds_stock, risk_bounds_global])
stock_pool = list_wind2jq(stock_pool)
EXCLUDED_INDEX = [] # 剔除的股票代码
TRADING_DATES_LIST = ['15'] # 每月的调仓日期，非交易日寻找下一个最近的交易日
# 用于记录调仓信息的字典
stock_dict = {}
# 根据回测阶段选取好调仓日期
trading_date_list = [] # 记录调仓日期的列表
def init(context):
# 调仓日期获取
global trading_date_list
trading_date_list = get_trading_date_list_by_day_monthly(BACKTEST_START_DATE, BACKTEST_END_DATE, TRADING_DATES_LIST)
# 每天time_rule定时执行algo任务，time_rule处于09:00:00和15:00:00之间
views_vector.append(ws.cell(ws.nrows - 2, j).value)
views_cov.append(ws.cell(ws.nrows - 1, j).value)
views_matrix = np.array(views_matrix).transpose()
views_vector = np.array(views_vector).reshape(-1, 1)
views_cov = np.diag(np.array(views_cov)**2)
return views_matrix, views_vector, views_cov
if __name__ == '__main__':
# 输入数据
date = '2019-11-04'
file_path = '投资示例.xlsx'
# 数据整理
code_list = 品种获取(file_path)
views_matrix, views_vector, views_cov = 观点信息获取(file_path)
code_weights = WEIGHTS(list_wind2jq(code_list), date).get_weights()
# 风险平价收益和历史数据估计协方差数据
rp_weights = [code_weights[code] for code in code_weights]
return_cov = 方差风险_历史数据(code_list, date, 60).return_cov * 240
# 主观观点标准差与历史估计标准差的比较
views_variance = get_variance_of_views_from_weights_covariances(
return_cov, views_matrix)
views_std = np.sqrt(views_variance)
print('观点的历史数据估计标准差为：')
print(views_std)
print('观点的主观判断标准差为：')
print(np.sqrt(np.diag(views_cov)))
# BL模型的参数设定
delta_value = 50.0
def equal_transfer(code_list, date):
# 按照date日的参数等权调仓
stock_now = {}
for code in code_list:
stock_now[list_wind2jq([code])[0]] = 1. / len(code_list)
return stock_now
# 国内债券部分配置方案
stock_pool_1_bond = ['511010.SH']
risk_budget_1_bond = [1.0]
risk_bounds_1_bond = np.array([[0.4, 1.0]])
# 国内股票部分配置方案
stock_pool_1_stock = [
'159928.SZ', '159949.SZ', '510050.SH', '512660.SH', '512720.SH',
'515000.SH'
]
risk_budget_1_stock = [0.7, 0.7, 0.7, 0.5, 0.8, 0.6] # 总权重固定为4个单位
risk_bounds_1_stock = np.array([[0.0, 1.0]] * len(stock_pool_1_stock))
# 合并为整体证券池
stock_pool_1 = stock_pool_1_bond + stock_pool_1_stock
risk_budget_1 = risk_budget_1_bond + risk_budget_1_stock
risk_bounds_1 = np.concatenate([risk_bounds_1_bond, risk_bounds_1_stock])
stock_pool_1 = list_wind2jq(stock_pool_1)
stock_weights_1 = 风险预算组合_模块求解基本版_带约束(stock_pool_1,
date_1,
risk_budget=risk_budget_1,
bounds=risk_bounds_1).get_weights()
# 加入信用债品种，对国债配置比例进行修正
stock_weights_1['160618.XSHE'] = 0.05
stock_weights_1['161713.XSHE'] = 0.03
stock_weights_1['161716.XSHE'] = 0.09
stock_weights_1['167501.XSHE'] = 0.05
stock_weights_1['511010.XSHG'] = stock_weights_1['511010.XSHG'] - 0.22
# 输出文字信息
string_output_1 = '策略一的风险预算权重为：\n'
string_output_1 += str(dict(zip(stock_pool_1, risk_budget_1))) + '\n'
string_output_1 += '策略一的选股结果（选股日为 ' + date_1 + ' 收盘）：\n'
string_output_1 += str(stock_weights_1) + '\n'
def algo(context):
global position_now, position_target
date_now = context.now.strftime('%Y-%m-%d')
date_previous = get_trading_date_from_now(date_now, -1,
ql.Days) # 前一个交易日，用于获取因子数据的日期
select_time_value = select_time_model[date_now] # 择时信号计算
print(date_now + ('日回测程序执行中...，择时值：%.2f' % select_time_value))
if date_now not in trading_date_list: # 非调仓日
pass # 预留非调仓日的微调空间
else: # 调仓日执行算法
position_now = False # 虚拟上，调仓日需要提前清仓
# 根据指数获取股票候选池的代码
code_list = SelectedStockPoolFromListV1(
INCLUDED_INDEX, EXCLUDED_INDEX, date_previous).get_stock_pool()
I = trading_date_list.index(date_now)
trading_dates = trading_date_list[I - HISTORY_LENGTH:I + 1]
# 提取训练数据并训练模型
data_dfs = []
for i in range(len(trading_dates) - 1):
date_start = get_trading_date_from_now(
trading_dates[i], -1, ql.Days) # 计算因子值的日子，买入前一日的因子值
date_end = get_trading_date_from_now(trading_dates[i + 1], -1,
ql.Days) # 计算收益率到期的日子-收盘
# 提取因子和收益数据
factors_df = get_factor_from_wind(code_list, FACTOR_LIST,
date_start) # 获取因子
return_df = get_return_from_wind(code_list, date_start, date_end)
factors_df_and_return_df = pd.concat(
[factors_df, return_df], axis=1).dropna() # 去掉因子或者回报有空缺值的样本
factors_df_and_return_df = factor_and_return_process(
factors_df_and_return_df) # 对因子和回报进行预处理# 使用排序数据作为输入
data_dfs.append(factors_df_and_return_df)
factors_return_df = pd.concat(data_dfs,
axis=0) # 获取的最终训练数据拼接，return为目标
# 根据data_df训练模型
model = OrdinaryLinearRegression(select_number=SELECT_NUMBER)
model.fit(factors_return_df)
# 根据factor_date_previous选取股票，使用模型
factor_date_previous_df = get_factor_from_wind(code_list, FACTOR_LIST,
date_previous).dropna()
factor_date_previous_df = factor_process(
factor_date_previous_df) # 对因子进行预处理
select_code_list = model.predict(
factor_date_previous_df) # 返回选取的股票代码，Wind格式
# 行业轮动部分
sw1_industry = get_SW1_industry(date_now, select_code_list)
industry_wm_result = industry_wheel_movement[date_now]
select_code_list = [
stock for stock in select_code_list
if sw1_industry[stock] is not None
and industry_wm_result[sw1_industry[stock]] == 1
] # 忽略无行业信息的股票并根据行业择时信号选择候选股票
# 转化为聚宽代码格式
select_code_list = list_wind2jq(select_code_list)
# 根据股票列表下单
if len(select_code_list) > 0: # 有可选股票时记录下可选股票
stock_now = WEIGHTS(select_code_list, date_previous).get_weights()
position_target = stock_now
else:
position_target = {}
# 择时判定
if select_time_value >= 0 and not position_now and position_target != {}: # LLT择时信号为正,空仓且有目标持仓状态
stock_dict[date_now] = position_target
position_now = True
elif select_time_value < 0 and position_now and position_target != {}: # LLT择时信号为负且持仓状态:
stock_dict[date_now] = {}
position_now = False
