Skip to content

R
robo-dividend
Commit e141cb82 authored by wenwen.tang's avatar wenwen.tang 😕
Browse files
Options

加入arc和prr

parent f3817b3b
Hide whitespace changes
Inline Side-by-side
Showing with 288 additions and 29 deletions
api.py
View file @ e141cb82
......@@ -59,8 +59,10 @@ class LoggerType(Enum):
class SignalType(Enum):
NORMAL = 1
SignalType.NORMAL.p_type = PortfoliosType.NORMAL
class DataSync(ABC):
'''
数据同步服务，需要同步数据的服务，可以实现该接口
......@@ -653,4 +655,4 @@ class RebalanceSignal(ABC):
:param min_date: 指定的起始时间
:param risk: 指定的风险等级
'''
pass
\ No newline at end of file
pass
asset_pool/asset_optimize.py
View file @ e141cb82
......@@ -146,9 +146,13 @@ class FundDividendSortinoAssetOptimize(SortinoAssetOptimize):
def get_groups(self):
funds = pd.DataFrame(self.get_filtered_funds())
result = []
include = list(self.asset_include.keys())[0]
for key, fund_group in funds.groupby(by=include):
if key in self.asset_include[include]:
if self.asset_include:
include = list(self.asset_include.keys())[0]
for key, fund_group in funds.groupby(by=include):
if key in self.asset_include[include]:
result.append(tuple(fund_group['id']))
else:
for (category, asset_type), fund_group in funds.groupby(by=['category', 'assetType']):
result.append(tuple(fund_group['id']))
return result
......
config-svrobo6.yml
View file @ e141cb82
......@@ -40,7 +40,7 @@ py-jftech:
backtest: robo_executor.BacktestExecutor
datum: basic.datum.DefaultDatum
hold-report: portfolios.holder.DivHoldReportor
mpt: portfolios.builder.PoemPortfoliosBuilder
mpt: portfolios.builder.PoemARCPortfoliosBuilder
dividend-holder: portfolios.holder.InvTrustPortfoliosHolder
email:
server: smtphz.qiye.163.com
......@@ -50,14 +50,14 @@ py-jftech:
max-workers: ${MAX_PROCESS:4}
basic: # 基础信息模块
sync:
start-date: 2007-01-01 # 同步数据开始日期
start-date: 2020-01-01 # 同步数据开始日期
datum: # 资料模块
change:
date: ${DATUM_CHANGE_DATE}
file: ${DATUM_CHANGE_FILE}
excludes: # 排除的资料彭博ticker
backtest:
- 'TEMUSGI LX Equity'
# backtest:
# - 'TEMUSGI LX Equity'
real:
- 'FGFSACU LX Equity'
- 'TEMUSGI LX Equity'
......@@ -74,7 +74,7 @@ asset-pool: # 资产池模块
weight: 0.3
- years: 1
weight: 0.2
asset-include: {'category':['US_STOCK','US_IG_BOND','US_HY_BOND']}
asset-include: {'customType':[1,2,3,4]}
optimize-count: 3 #基金优选个数
portfolios: # 投组模块
holder: # 持仓投组相关
......@@ -86,6 +86,9 @@ portfolios: # 投组模块
warehouse-frequency: 1 #每隔1个月调一次仓
redeem-list: [ 'TEUSAAU LX Equity', 'LIGTRAA ID Equity', 'TEMFHAC LX Equity', 'LUSHUAA ID Equity' ] #从持仓中的低风险资产“直接”按序赎回
solver: # 解算器相关
model: prr # 结算模型 ARC ,PRR, ~ 标准解算器
arc: on #是否开启ARC
bRR: 0.01
tol: 1E-10 # 误差满足条件
navs: # 净值要求
range: # 需要净值数据的区间, days: 90 表示90自然日，months: 3 表示3个自然月
......@@ -93,21 +96,16 @@ portfolios: # 投组模块
max-nan: # 最大缺失净值条件
asset: 8 # 单一资产最多缺少多少交易日数据，则踢出资产池
day: 0.5 # 单一交易日最多缺少百分之多少净值，则删除该交易日
normal-ratio: #US_STOCK：US_HY_BOND：US_IG_BOND三者分别对应低中高风险所占比率
US_STOCK: [ 0.5, 0.5, 0.7 ]
US_HY_BOND: [ 0.4, 0.4, 0.2 ]
US_IG_BOND: [ 0.1, 0.1, 0.1 ]
riskctl-ratio:
US_STOCK: [ 0.2, 0.4, 0.6 ]
US_HY_BOND: [ 0.5, 0.3, 0.1 ]
US_IG_BOND: [ 0.3, 0.3, 0.3 ]
risk: [] # 资产风险等级要求，可分开写也可以合并写，e.g. risk：[ 2, 3 ] 则表示 所有投组资产风险等级都是 2 或 3
LARC: [0, 0, 0, 0] #低阈值
UARC: [1, 1, 1, 1] #高阈值
matrix-rtn-days: 20 # 计算回报率矩阵时，回报率滚动天数
asset-count: [3,3] # 投组资产个数。e.g. count 或 [min, max] 分别表示 最大最小都为count 或 最小为min 最大为max，另外这里也可以类似上面给不同风险等级分别配置
asset-count: [5,5] # 投组资产个数。e.g. count 或 [min, max] 分别表示 最大最小都为count 或 最小为min 最大为max，另外这里也可以类似上面给不同风险等级分别配置
mpt: # mpt计算相关
cvar-beta: 0.2 # 计算Kbeta 需要用到
quantile: 0.9 # 分位点，也可以给不同风险等级分别配置
low-weight: 0.05 # 最低权重
# high-weight: [ 1 ] # 最高权重比例，可给一个值，也可以给多个值，当多个值时，第一个表示只有一个资产时权重，第二个表示只有两个资产时权重，以此类推，最后一个表示其他资产个数时的权重
high-weight: [ 0.35 ] # 最高权重比例，可给一个值，也可以给多个值，当多个值时，第一个表示只有一个资产时权重，第二个表示只有两个资产时权重，以此类推，最后一个表示其他资产个数时的权重
poem: # poem相关
cvar-scale-factor: 0.1 # 计算时用到的系数
reports: # 报告模块相关
......@@ -234,12 +232,12 @@ robo-executor: # 执行器相关
use: ${ROBO_EXECUTOR:backtest} # 执行哪个执行器，优先取系统环境变量ROBO_EXECUTOR的值，默认backtest
sync-data: ${SYNC_DATA:off} # 是否开启同步资料数据
backtest: # 回测执行器相关
start-date: 2022-09-22 # 回测起始日期
end-date: 2023-07-03 # 回测截止日期
start-date: 2022-02-16 # 回测起始日期
end-date: 2023-01-03 # 回测截止日期
sealing-period: 10 #调仓封闭期
start-step: ${BACKTEST_START_STEP:3} # 回测从哪一步开始执行 1:计算资产池；2:计算最优投组：3:计算再平衡信号以及持仓投组
start-step: ${BACKTEST_START_STEP:2} # 回测从哪一步开始执行 1:计算资产池；2:计算最优投组：3:计算再平衡信号以及持仓投组
end-step: ${BACKTEST_END_STEP:3} # 回测从哪一步执行完成后结束执行 1:计算资产池；2:计算最优投组：3:计算再平衡信号以及持仓投组
clean-up: off
clean-up: on
real: # 实盘执行器
export: ${EXPORT_ENABLE:off} # 是否开启报告
start-date: 2023-05-08 # 实盘开始时间
......
portfolios/builder.py
View file @ e141cb82
......@@ -83,7 +83,7 @@ class MptPortfoliosBuilder(PortfoliosBuilder):
rmp.delete(min_date=day, risk=risk)
@component(bean_name='poem')
@component(bean_name='mpt')
class PoemPortfoliosBuilder(MptPortfoliosBuilder):
def build_portfolio(self, day, type: PortfoliosType):
......@@ -110,3 +110,88 @@ class PoemPortfoliosBuilder(MptPortfoliosBuilder):
return result
@component(bean_name='mpt')
class MptARCPortfoliosBuilder(MptPortfoliosBuilder):
def get_portfolios(self, day, risk: PortfoliosRisk, type: PortfoliosType = PortfoliosType.NORMAL):
try:
portfolio = rmp.get_one(day, type, risk)
if not portfolio:
result, detail = self.build_portfolio(day, type)
for build_risk, datas in result.items():
try:
rmp.insert({
**datas,
'risk': build_risk,
'type': type,
'date': day
})
except IntegrityError as e:
code, msg = e.args
if code != constants.ER.DUP_ENTRY:
raise e
portfolio = rmp.get_one(day, type, risk)
if SolveType(portfolio['solve']) is not SolveType.INFEASIBLE:
result = json.loads(portfolio['portfolio'])
return {int(x[0]): x[1] for x in result.items()}
return None
except Exception as e:
logger.exception(f"build protfolio of type[{type.name}] and risk[{risk.name}] with date[{format_date(day)}] failure.", exc_info=e)
raise e
def build_portfolio(self, day, type: PortfoliosType):
result = {}
detail = {}
risk = PortfoliosRisk.FT3
logger.info(
f"start to build protfolio of type[{type.name}] and risk[{risk.name}] with date[{format_date(day)}]")
solver = self._factory.create_solver(risk, type)
solver.reset_navs(day)
logger.debug({
'Khist': len(solver.rtn_history),
'beta': solver.get_config('mpt.cvar-beta'),
'Kbeta': solver.k_beta,
})
max_rtn, max_var, minCVaR_whenMaxR = solver.solve_max_rtn()
min_rtn, min_var, maxCVaR_whenMinV = solver.solve_min_rtn()
portfolio, cvar = solver.solve_mpt(min_rtn, max_rtn)
result[risk] = {
'solve': SolveType.MPT,
'portfolio': json.dumps(portfolio),
'cvar': cvar
} if portfolio else {
'solve': SolveType.INFEASIBLE
}
detail[risk] = {
'max_rtn': max_rtn,
'max_var': max_var,
'minCVaR_whenMaxR': minCVaR_whenMaxR,
'min_rtn': min_rtn,
'min_var': min_var,
'maxCVaR_whenMinV': maxCVaR_whenMinV,
}
return result, detail
@component(bean_name='mpt')
class PoemARCPortfoliosBuilder(MptARCPortfoliosBuilder):
def build_portfolio(self, day, type: PortfoliosType):
result, detail = super(PoemARCPortfoliosBuilder, self).build_portfolio(day, type)
risk = PortfoliosRisk.FT3
# if result[risk]['solve'] is SolveType.INFEASIBLE:
# continue
solver = self._factory.create_solver(risk, type)
solver.reset_navs(day)
min_rtn = detail[risk]['min_rtn']
max_rtn = detail[risk]['max_rtn']
mpt_cvar = result[risk]['cvar']
maxCVaR_whenMinV = detail[risk]['maxCVaR_whenMinV']
portfolio, cvar = solver.solve_poem(min_rtn, max_rtn, mpt_cvar, maxCVaR_whenMinV)
if portfolio:
result[risk] = {
'solve': SolveType.POEM,
'portfolio': json.dumps(portfolio),
'cvar': cvar
}
detail[risk]['mpt_cvar'] = mpt_cvar
return result, detail
portfolios/holder.py
View file @ e141cb82
......@@ -317,7 +317,7 @@ class InvTrustPortfoliosHolder(DividendPortfoliosHolder):
day_div = sum(
map(lambda k: share[k] * fund_dividend[k], filter(lambda k: k in fund_dividend, share.keys())))
fund_dividend = last_nav['fund_div'] + day_div
# 基金净值+基金配息+产品配息
# todo 基金净值+基金配息+产品配息
asset_nav = fund_av + fund_dividend + portfolio_div
nav = last_nav['nav'] * asset_nav / last_nav['asset_nav']
div_forecast = last_nav['div_forecast']
......
portfolios/solver.py
View file @ e141cb82
import math
import os
import sys
from logging import DEBUG, getLogger
import numpy as np
import pandas as pd
from dateutil.relativedelta import relativedelta
from numpy import NAN
from py_jftech import component, autowired, get_config
from py_jftech import component, autowired, get_config, filter_weekend
from pyomo.environ import *
from api import SolverFactory as Factory, PortfoliosRisk, PortfoliosType, AssetPool, Navs, Solver, Datum, DatumType
......@@ -27,14 +29,26 @@ def create_solver():
@component
class DefaultFactory(Factory):
def create_solver(self, risk: PortfoliosRisk, type: PortfoliosType = PortfoliosType.NORMAL) -> Solver:
return DefaultSolver(risk, type)
def __init__(self):
self._config = get_config(__name__)
@property
def solver_model(self):
return self._config['model'].upper() if 'model' in self._config and self._config['model'] is not None else None
def create_solver(self, risk: PortfoliosRisk = None, type: PortfoliosType = PortfoliosType.NORMAL) -> Solver:
if self.solver_model == 'ARC':
return ARCSolver(type=type, risk=risk)
if self.solver_model == 'PRR':
if risk == PortfoliosRisk.FT3:
return PRRSolver(type=type, risk=risk)
return DefaultSolver(type=type, risk=risk)
class DefaultSolver(Solver):
@autowired
def __init__(self, risk: PortfoliosRisk, type: PortfoliosType, assets: AssetPool = None, navs: Navs = None,
def __init__(self, type: PortfoliosType, risk: PortfoliosRisk, assets: AssetPool = None, navs: Navs = None,
datum: Datum = None):
self._category = None
self._transfer_type = None
......@@ -284,3 +298,159 @@ class DefaultSolver(Solver):
'port_CVaR': self.calc_port_cvar(model)
})
logger.debug('-------------------------------')
class ARCSolver(DefaultSolver):
def __init__(self, type: PortfoliosType, risk: PortfoliosRisk, assets: AssetPool = None, navs: Navs = None,
datum: Datum = None):
super().__init__(type, risk)
self.__date = None
@property
def date(self):
return self.__date
def calc_port_weight(self, model):
id_list = self.navs.columns
weight_list = [model.w[i]._value * model.z[i]._value for i in model.indices]
df_w = pd.DataFrame(data=weight_list, index=id_list, columns=['weight'])
df_w.replace(0, math.nan, inplace=True)
df_w.dropna(axis=0, inplace=True)
df_w['weight'] = pd.Series(format_weight(dict(df_w['weight'])))
dict_w = df_w.to_dict()['weight']
return dict_w
@property
def max_count(self):
count = self.get_config('asset-count')
return count[1] if isinstance(count, list) else count
@property
def min_count(self):
count = self.get_config('asset-count')
return min(count[0] if isinstance(count, list) else count, len(self.rtn_annualized))
def create_model(self):
low_weight = self.get_config('mpt.low-weight')
high_weight = self.get_config('mpt.high-weight')
if isinstance(high_weight, list):
high_weight = high_weight[min(len(self.navs.columns), self.min_count, len(high_weight)) - 1]
model = ConcreteModel()
model.indices = range(0, len(self.navs.columns))
model.w = Var(model.indices, domain=NonNegativeReals)
model.z = Var(model.indices, domain=Binary)
model.cons_sum_weight = Constraint(expr=sum([model.w[i] for i in model.indices]) == 1)
model.cons_num_asset = Constraint(
expr=inequality(self.min_count, sum([model.z[i] for i in model.indices]), self.max_count, strict=False))
model.cons_bounds_low = Constraint(model.indices, rule=lambda m, i: m.z[i] * low_weight <= m.w[i])
model.cons_bounds_up = Constraint(model.indices, rule=lambda m, i: m.z[i] * high_weight >= m.w[i])
if self._config['arc']:
LARC = self._config['LARC']
UARC = self._config['UARC']
numARC = len(LARC) # this is the M in the doc
numAsset = len(self.navs.columns)
# This should from DB. We just fake value here for developing the code
datums = self._datum.get_datums(type=DatumType.FUND, datum_ids=list(self.navs.columns))
AssetARC = np.array([x['customType'] for x in datums], dtype=int)
# the above are input data from either config file or DB
# the following are POEM / MPT code
A = np.zeros((numARC, numAsset), dtype=int)
for i in range(numAsset):
A[AssetARC[i] - 1, i] = 1
model.cons_arc_low = Constraint(range(numARC),
rule=lambda m, i: LARC[i] <= sum([A[i, j] * m.w[j] for j in m.indices]))
model.cons_arc_up = Constraint(range(numARC),
rule=lambda m, i: UARC[i] >= sum([A[i, j] * m.w[j] for j in m.indices]))
return model
def reset_navs(self, day):
self.__date = filter_weekend(day)
asset_ids = self._assets.get_pool(self.date)
asset_risk = self.get_config('navs.risk')
datum = self._datum.get_datums(type=DatumType.FUND, datum_ids=asset_ids, risk=asset_risk)
exclude = self.get_config('navs.exclude-asset-type') or []
asset_ids = list(set(asset_ids) & set([x['id'] for x in datum if x['assetType'] not in exclude]))
min_date = self.date - relativedelta(**self.get_config('navs.range'))
navs = pd.DataFrame(self._navs.get_fund_navs(fund_ids=asset_ids, max_date=self.date, min_date=min_date))
navs = navs[navs['nav_date'].dt.day_of_week < 5]
navs['nav_date'] = pd.to_datetime(navs['nav_date'])
navs = navs.pivot_table(index='nav_date', columns='fund_id', values='nav_cal')
navs = navs.sort_index()
navs_nan = navs.isna().sum()
navs.drop(columns=[x for x in navs_nan.index if navs_nan.loc[x] >= self.get_config('navs.max-nan.asset')],
inplace=True)
navs_nan = navs.apply(lambda r: r.isna().sum() / len(r), axis=1)
navs.drop(index=[x for x in navs_nan.index if navs_nan.loc[x] >= self.get_config('navs.max-nan.day')],
inplace=True)
navs.fillna(method='ffill', inplace=True)
if navs.iloc[0].isna().sum() > 0:
navs.fillna(method='bfill', inplace=True)
self.set_navs(navs)
class PRRSolver(ARCSolver):
def __init__(self, type: PortfoliosType, risk: PortfoliosRisk, assets: AssetPool = None, navs: Navs = None,
datum: Datum = None):
super().__init__(type, risk)
self.__risk = None
def create_model(self):
model = super(PRRSolver, self).create_model()
# print(self.risks)
# 创建一个空列表来存储第二列的值
RR = []
# 遍历字典的键值对
for key, value in self.risks.items():
# 将值添加到列表中
RR.append(value)
# 打印第二列的值
# print(RR)
minRRweightWithinTRR = 0.7 + self._config['bRR']
TRR = 3
# RR = np.zeros(len(self.navs.columns), dtype=int)
# # Please note, RR should come from DB with real values. Here, we just assign fake values for coding
# for i in range(len(self.navs.columns)):
# RR[i] = math.ceil((i + 1) / len(self.navs.columns) * 5)
# the following code are real model code ::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
model.cons_TRR = Constraint(expr=sum([model.w[i] * RR[i] for i in model.indices]) <= TRR)
RR_LE_TRR = np.zeros(len(self.navs.columns), dtype=int)
RR_in_1_5 = np.zeros(len(self.navs.columns), dtype=int)
RR_EQ_5 = np.zeros(len(self.navs.columns), dtype=int)
for i in range(len(self.navs.columns)):
if RR[i] <= TRR:
RR_LE_TRR[i] = 1
if RR[i] > 1 and RR[i] < 5:
RR_in_1_5[i] = 1
elif RR[i] == 5:
RR_EQ_5[i] = 1
model.cons_RR_LE_TRR = Constraint(
expr=sum([model.w[i] * RR_LE_TRR[i] for i in model.indices]) >= minRRweightWithinTRR)
if TRR < 5:
model.cons_RR_in_1_5 = Constraint(
expr=sum([model.z[i] * (RR_in_1_5[i] * self.max_count - RR_EQ_5[i]) for i in model.indices]) >= 0)
return model
def reset_navs(self, day):
super(PRRSolver, self).reset_navs(day=day)
datums = self._datum.get_datums(type=DatumType.FUND, datum_ids=list(self.navs.columns))
self.__risk = {x['id']: x['risk'] for x in datums}
# self.__risk = {x['risk'] for x in datums}
@property
def risks(self):
return self.__risk
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment