Skip to content

A
alpha-mind
Commit 8541e9e2 authored by Dr.李's avatar Dr.李
Browse files
Options

greatly enhance the optimizer performance by using factor model

parent 949d03a9
Hide whitespace changes
Inline Side-by-side
Showing with 264 additions and 77 deletions
alphamind/cython/optimizers.pyx
View file @ 8541e9e2
......@@ -10,6 +10,7 @@ cimport numpy as cnp
from libcpp.string cimport string
from libcpp.vector cimport vector
import numpy as np
from PyFin.api import pyFinAssert
cdef extern from "lpoptimizer.hpp" namespace "pfopt":
......@@ -71,7 +72,11 @@ cdef extern from "tvoptimizer.hpp" namespace "pfopt":
double*,
double*,
double,
double) except +
double,
int,
double*,
double*,
double*) except +
vector[double] xValue()
double feval()
int status()
......@@ -81,6 +86,7 @@ cdef class CVOptimizer:
cdef TVOptimizer* cobj
cdef int n
cdef int m
cdef int f
def __cinit__(self,
double[:] expected_return,
......@@ -91,20 +97,26 @@ cdef class CVOptimizer:
double[:] clbound=None,
double[:] cubound=None,
double target_low=0.0,
double target_high=1.0):
double target_high=1.0,
cnp.ndarray[double, ndim=2] factor_cov_matrix=None,
cnp.ndarray[double, ndim=2] factor_loading_matrix=None,
double[:] idsync_risk=None):
self.n = lbound.shape[0]
self.m = 0
cdef double[:] cov = cov_matrix.flatten(order='C')
self.f = factor_cov_matrix.shape[0] if factor_cov_matrix is not None else 0
cdef double[:] cov = cov_matrix.flatten(order='C') if cov_matrix is not None else None
cdef double[:] cons
cdef double[:] factor_cov = factor_cov_matrix.flatten(order='C') if factor_cov_matrix is not None else None
cdef double[:] factor_loading = factor_loading_matrix.flatten(order='C') if factor_loading_matrix is not None else None
if cons_matrix is not None:
self.m = cons_matrix.shape[0]
cons = cons_matrix.flatten(order='C');
cons = cons_matrix.flatten(order='C')
self.cobj = new TVOptimizer(self.n,
&expected_return[0],
&cov[0],
&cov[0] if cov is not None else NULL,
&lbound[0],
&ubound[0],
self.m,
......@@ -112,11 +124,15 @@ cdef class CVOptimizer:
&clbound[0],
&cubound[0],
target_low,
target_high)
target_high,
self.f,
&factor_cov[0] if factor_cov is not None else NULL,
&factor_loading[0] if factor_loading is not None else NULL,
&idsync_risk[0] if idsync_risk is not None else NULL)
else:
self.cobj = new TVOptimizer(self.n,
&expected_return[0],
&cov[0],
&cov[0] if cov is not None else NULL,
&lbound[0],
&ubound[0],
0,
......@@ -124,7 +140,11 @@ cdef class CVOptimizer:
NULL,
NULL,
target_low,
target_high)
target_high,
self.f,
&factor_cov[0] if factor_cov is not None else NULL,
&factor_loading[0] if factor_loading is not None else NULL,
&idsync_risk[0] if idsync_risk is not None else NULL)
def __dealloc__(self):
del self.cobj
......@@ -150,7 +170,11 @@ cdef extern from "mvoptimizer.hpp" namespace "pfopt":
double*,
double*,
double*,
double) except +
double,
int,
double*,
double*,
double*) except +
vector[double] xValue()
double feval()
int status()
......@@ -171,30 +195,36 @@ cdef extern from "qpalglib.hpp" namespace "pfopt":
cdef class QPOptimizer:
cdef MVOptimizer* cobj
cdef QPAlglib* cobj2
cdef cnp.ndarray er
cdef cnp.ndarray cov
cdef double risk_aversion
cdef int n
cdef int m
cdef int f
def __cinit__(self,
double[:] expected_return,
cnp.ndarray[double, ndim=2] cov_matrix,
double[:] lbound,
double[:] ubound,
cnp.ndarray[double, ndim=2] cons_matrix=None,
double[:] clbound=None,
double[:] cubound=None,
double risk_aversion=1.0):
double[:] expected_return,
cnp.ndarray[double, ndim=2] cov_matrix,
double[:] lbound,
double[:] ubound,
cnp.ndarray[double, ndim=2] cons_matrix=None,
double[:] clbound=None,
double[:] cubound=None,
double risk_aversion=1.0,
cnp.ndarray[double, ndim=2] factor_cov_matrix=None,
cnp.ndarray[double, ndim=2] factor_loading_matrix=None,
double[:] idsync_risk=None):
self.n = lbound.shape[0]
self.m = 0
self.f = factor_cov_matrix.shape[0] if factor_cov_matrix is not None else 0
self.er = np.array(expected_return)
self.cov = np.array(cov_matrix)
self.risk_aversion = risk_aversion
cdef double[:] cov = cov_matrix.flatten(order='C')
cdef double[:] cov = cov_matrix.flatten(order='C') if cov_matrix is not None else None
cdef double[:] cons
cdef double[:] factor_cov = factor_cov_matrix.flatten(order='C') if factor_cov_matrix is not None else None
cdef double[:] factor_loading = factor_loading_matrix.flatten(order='C') if factor_loading_matrix is not None else None
if cons_matrix is not None:
self.m = cons_matrix.shape[0]
......@@ -202,48 +232,49 @@ cdef class QPOptimizer:
self.cobj = new MVOptimizer(self.n,
&expected_return[0],
&cov[0],
&cov[0] if cov is not None else NULL,
&lbound[0],
&ubound[0],
self.m,
&cons[0],
&clbound[0],
&cubound[0],
risk_aversion)
risk_aversion,
self.f,
&factor_cov[0] if factor_cov is not None else NULL,
&factor_loading[0] if factor_loading is not None else NULL,
&idsync_risk[0] if idsync_risk is not None else NULL)
else:
self.cobj2 = new QPAlglib(self.n,
&expected_return[0],
&cov[0],
&lbound[0],
&ubound[0],
risk_aversion)
# self.cobj2 = new QPAlglib(self.n,
# &expected_return[0],
# &cov[0] if cov is not None else NULL,
# &lbound[0],
# &ubound[0],
# risk_aversion)
self.cobj = new MVOptimizer(self.n,
&expected_return[0],
&cov[0] if cov is not None else NULL,
&lbound[0],
&ubound[0],
self.m,
NULL,
NULL,
NULL,
risk_aversion,
self.f,
&factor_cov[0] if factor_cov is not None else NULL,
&factor_loading[0] if factor_loading is not None else NULL,
&idsync_risk[0] if idsync_risk is not None else NULL)
def __dealloc__(self):
if self.cobj:
del self.cobj
else:
del self.cobj2
del self.cobj
def feval(self):
if self.cobj:
return self.cobj.feval()
else:
x = np.array(self.cobj2.xValue())
return 0.5 * self.risk_aversion * x @ self.cov @ x - self.er @ x
return self.cobj.feval()
def x_value(self):
if self.cobj:
return np.array(self.cobj.xValue())
else:
return np.array(self.cobj2.xValue())
return np.array(self.cobj.xValue())
def status(self):
if self.cobj:
return self.cobj.status()
else:
status = self.cobj2.status()
if 1 <= status <= 4:
return 0
else:
return status
return self.cobj.status()
alphamind/data/engines/sqlengine.py
View file @ 8541e9e2
......@@ -371,7 +371,7 @@ class SqlEngine(object):
res['chgPct'] = df.chgPct
res = res.loc[ref_date]
res.index = list(range(len(res)))
return res.drop_duplicates(['trade_date', 'code'])
return res
def fetch_factor_range(self,
universe: Universe,
......
pfopt @ 40976afd
Subproject commit 1dcecd88728512ff50730f418d9d58195bf33851
Subproject commit 40976afd3ea03b921177cef364fa8a6b37bf4dda
alphamind/portfolio/meanvariancebuilder.py
View file @ 8541e9e2
......@@ -51,7 +51,10 @@ def mean_variance_builder(er: np.ndarray,
ubound: Union[np.ndarray, float],
risk_exposure: Optional[np.ndarray],
risk_target: Optional[Tuple[np.ndarray, np.ndarray]],
lam: float=1.) -> Tuple[str, float, np.ndarray]:
lam: float=1.,
factor_cov: np.ndarray=None,
factor_loading: np.ndarray=None,
idsync: np.ndarray=None) -> Tuple[str, float, np.ndarray]:
lbound, ubound, cons_mat, clbound, cubound = _create_bounds(lbound, ubound, bm, risk_exposure, risk_target)
optimizer = QPOptimizer(er,
......@@ -61,7 +64,10 @@ def mean_variance_builder(er: np.ndarray,
cons_mat,
clbound,
cubound,
lam)
lam,
factor_cov,
factor_loading,
idsync)
return _create_result(optimizer, bm)
......@@ -74,7 +80,10 @@ def target_vol_builder(er: np.ndarray,
risk_exposure: Optional[np.ndarray],
risk_target: Optional[Tuple[np.ndarray, np.ndarray]],
vol_low: float = 0.,
vol_high: float = 1.)-> Tuple[str, float, np.ndarray]:
vol_high: float = 1.,
factor_cov: np.ndarray = None,
factor_loading: np.ndarray = None,
idsync: np.ndarray = None)-> Tuple[str, float, np.ndarray]:
lbound, ubound, cons_mat, clbound, cubound = _create_bounds(lbound, ubound, bm, risk_exposure, risk_target)
optimizer = CVOptimizer(er,
......@@ -85,7 +94,10 @@ def target_vol_builder(er: np.ndarray,
clbound,
cubound,
vol_low,
vol_high)
vol_high,
factor_cov,
factor_loading,
idsync)
return _create_result(optimizer, bm)
......
alphamind/tests/cython/test_optimizers.py
View file @ 8541e9e2
......@@ -54,6 +54,36 @@ class TestOptimizers(unittest.TestCase):
[0.1996, 0.3004, 0.5000],
4)
def test_qpoptimizer_with_factor_model(self):
objective = np.array([0.1, 0.2, 0.3])
lbound = np.array([0.0, 0.0, 0.0])
ubound = np.array([1.0, 1.0, 1.0])
factor_var = np.array([[0.5, -0.3], [-0.3, 0.7]])
factor_load = np.array([[0.8, 0.2], [0.5, 0.5], [0.2, 0.8]])
idsync = np.array([0.1, 0.3, 0.2])
cons = np.array([[1., 1., 1.]])
clbound = np.array([1.])
cubound = np.array([1.])
optimizer = QPOptimizer(objective,
None,
lbound,
ubound,
cons,
clbound,
cubound,
1.,
factor_var,
factor_load,
idsync)
# check against cvxpy result
np.testing.assert_array_almost_equal(optimizer.x_value(),
[0.2866857, 0.21416417, 0.49915014],
4)
def test_qpoptimizer_with_identity_matrix(self):
objective = np.array([-0.02, 0.01, 0.03])
cov = np.diag([1., 1., 1.])
......@@ -122,6 +152,38 @@ class TestOptimizers(unittest.TestCase):
[-0.3, -0.10919033, 0.40919033],
4)
def test_cvoptimizer_with_factor_model(self):
objective = np.array([0.1, 0.2, 0.3])
lbound = np.array([0.0, 0.0, 0.0])
ubound = np.array([1.0, 1.0, 1.0])
factor_var = np.array([[0.5, -0.3], [-0.3, 0.7]])
factor_load = np.array([[0.8, 0.2], [0.5, 0.5], [0.2, 0.8]])
idsync = np.array([0.1, 0.3, 0.2])
cons = np.array([[1., 1., 1.]])
clbound = np.array([1.])
cubound = np.array([1.])
target_vol = 0.5
optimizer = CVOptimizer(objective,
None,
lbound,
ubound,
cons,
clbound,
cubound,
0.,
target_vol,
factor_var,
factor_load,
idsync)
# check against cvxpy result
np.testing.assert_array_almost_equal(optimizer.x_value(),
[0.26595552, 0.21675092, 0.51729356],
4)
def test_cvoptimizer_with_cons_and_ieq(self):
objective = np.array([0.1, 0.2, 0.3])
cov = np.array([[0.05, 0.01, 0.02],
......
notebooks/Example 7 - Portfolio Optimizer Performance.ipynb
View file @ 8541e9e2
......@@ -243,7 +243,7 @@
"metadata": {},
"outputs": [],
"source": [
"from cvxpy import quad_form\n",
"from cvxpy import *\n",
"\n",
"df = pd.DataFrame(columns=u_names, index=['cvxpy', 'alphamind'])\n",
"number = 1\n",
......@@ -262,16 +262,36 @@
" lbound = -np.ones(n) * np.inf\n",
" ubound = np.ones(n) * np.inf\n",
"\n",
" status, y, x1 = mean_variance_builder(er, sec_cov, bm, lbound, ubound, None, None, lam=1)\n",
" elasped_time1 = timeit.timeit(\"mean_variance_builder(er, sec_cov, bm, lbound, ubound, None, None, lam=1)\",\n",
" status, y, x1 = mean_variance_builder(er,\n",
" None,\n",
" bm,\n",
" lbound,\n",
" ubound,\n",
" None,\n",
" None,\n",
" lam=1,\n",
" factor_cov=risk_cov / 10000.,\n",
" factor_loading=risk_exposure,\n",
" idsync=(special_risk ** 2.) / 10000)\n",
" elasped_time1 = timeit.timeit(\"\"\"mean_variance_builder(er,\n",
" None,\n",
" bm,\n",
" lbound,\n",
" ubound,\n",
" None,\n",
" None,\n",
" lam=1,\n",
" factor_cov=risk_cov / 10000.,\n",
" factor_loading=risk_exposure,\n",
" idsync=(special_risk ** 2.) / 10000)\"\"\",\n",
" number=number, globals=globals()) / number * 1000\n",
" \n",
" w = cvxpy.Variable(n)\n",
" risk = quad_form(w, sec_cov)\n",
" risk = sum_squares(mul_elemwise(special_risk / 100., w)) + quad_form((w.T * risk_exposure).T, risk_cov / 10000.)\n",
" objective = cvxpy.Minimize(-w.T * er + 0.5 * risk)\n",
" prob = cvxpy.Problem(objective)\n",
" prob.solve(solver='CVXOPT')\n",
" elasped_time2 = timeit.timeit(\"prob.solve(solver='CVXOPT')\",\n",
" prob.solve(solver='ECOS')\n",
" elasped_time2 = timeit.timeit(\"prob.solve(solver='ECOS')\",\n",
" number=number, globals=globals()) / number * 1000\n",
" \n",
" u1 = -x1 @ er + 0.5 * x1 @ sec_cov @ x1\n",
......@@ -325,18 +345,38 @@
" lbound = np.zeros(n)\n",
" ubound = np.ones(n) * 0.1\n",
"\n",
" status, y, x1 = mean_variance_builder(er, sec_cov, bm, lbound, ubound, None, None, lam=1)\n",
" elasped_time1 = timeit.timeit(\"mean_variance_builder(er, sec_cov, bm, lbound, ubound, None, None, lam=1)\",\n",
" status, y, x1 = mean_variance_builder(er,\n",
" None,\n",
" bm,\n",
" lbound,\n",
" ubound,\n",
" None,\n",
" None,\n",
" lam=1,\n",
" factor_cov=risk_cov / 10000.,\n",
" factor_loading=risk_exposure,\n",
" idsync=(special_risk ** 2.) / 10000)\n",
" elasped_time1 = timeit.timeit(\"\"\"mean_variance_builder(er,\n",
" None,\n",
" bm,\n",
" lbound,\n",
" ubound,\n",
" None,\n",
" None,\n",
" lam=1,\n",
" factor_cov=risk_cov / 10000.,\n",
" factor_loading=risk_exposure,\n",
" idsync=(special_risk ** 2.) / 10000)\"\"\",\n",
" number=number, globals=globals()) / number * 1000\n",
" \n",
" w = cvxpy.Variable(n)\n",
" risk = quad_form(w, sec_cov)\n",
" risk = sum_squares(mul_elemwise(special_risk / 100., w)) + quad_form((w.T * risk_exposure).T, risk_cov / 10000.)\n",
" objective = cvxpy.Minimize(-w.T * er + 0.5 * risk)\n",
" constraints = [w >= lbound,\n",
" w <= ubound]\n",
" prob = cvxpy.Problem(objective, constraints)\n",
" prob.solve(solver='CVXOPT')\n",
" elasped_time2 = timeit.timeit(\"prob.solve(solver='CVXOPT')\",\n",
" prob.solve(solver='ECOS')\n",
" elasped_time2 = timeit.timeit(\"prob.solve(solver='ECOS')\",\n",
" number=number, globals=globals()) / number * 1000\n",
" \n",
" u1 = -x1 @ er + 0.5 * x1 @ sec_cov @ x1\n",
......@@ -393,20 +433,40 @@
" risk_constraints = np.ones((len(er), 1))\n",
" risk_target = (np.array([1.]), np.array([1.]))\n",
"\n",
" status, y, x1 = mean_variance_builder(er, sec_cov, bm, lbound, ubound, risk_constraints, risk_target, lam=1)\n",
" elasped_time1 = timeit.timeit(\"mean_variance_builder(er, sec_cov, bm, lbound, ubound, None, None, lam=1)\",\n",
" status, y, x1 = mean_variance_builder(er,\n",
" None,\n",
" bm,\n",
" lbound,\n",
" ubound,\n",
" risk_constraints,\n",
" risk_target,\n",
" lam=1,\n",
" factor_cov=risk_cov / 10000.,\n",
" factor_loading=risk_exposure,\n",
" idsync=(special_risk ** 2.) / 10000)\n",
" elasped_time1 = timeit.timeit(\"\"\"mean_variance_builder(er,\n",
" None,\n",
" bm,\n",
" lbound,\n",
" ubound,\n",
" risk_constraints,\n",
" risk_target,\n",
" lam=1,\n",
" factor_cov=risk_cov / 10000.,\n",
" factor_loading=risk_exposure,\n",
" idsync=(special_risk ** 2.) / 10000)\"\"\",\n",
" number=number, globals=globals()) / number * 1000\n",
" \n",
" w = cvxpy.Variable(n)\n",
" risk = quad_form(w, sec_cov)\n",
" risk = sum_squares(mul_elemwise(special_risk / 100., w)) + quad_form((w.T * risk_exposure).T, risk_cov / 10000.)\n",
" objective = cvxpy.Minimize(-w.T * er + 0.5 * risk)\n",
" curr_risk_exposure = risk_constraints.T @ w\n",
" constraints = [w >= lbound,\n",
" w <= ubound,\n",
" curr_risk_exposure == risk_target[0]]\n",
" prob = cvxpy.Problem(objective, constraints)\n",
" prob.solve(solver='CVXOPT')\n",
" elasped_time2 = timeit.timeit(\"prob.solve(solver='CVXOPT')\",\n",
" prob.solve(solver='ECOS')\n",
" elasped_time2 = timeit.timeit(\"prob.solve(solver='ECOS')\",\n",
" number=number, globals=globals()) / number * 1000\n",
"\n",
" u1 = -x1 @ er + 0.5 * x1 @ sec_cov @ x1\n",
......@@ -445,7 +505,7 @@
"source": [
"df = pd.DataFrame(columns=u_names, index=['cvxpy', 'alphamind'])\n",
"number = 1\n",
"target_vol = 0.1\n",
"target_vol = 0.5\n",
"\n",
"\n",
"for u_name, sample_data in zip(u_names, data_set):\n",
......@@ -468,12 +528,34 @@
" risk_constraints = np.ones((n, 1))\n",
" risk_target = (np.array([bm.sum()]), np.array([bm.sum()]))\n",
"\n",
" status, y, x1 = target_vol_builder(er, sec_cov, bm, lbound, ubound, risk_constraints, risk_target, vol_low=0, vol_high=target_vol)\n",
" elasped_time1 = timeit.timeit(\"mean_variance_builder(er, sec_cov, bm, lbound, ubound, None, None, lam=1)\",\n",
" status, y, x1 = target_vol_builder(er,\n",
" None,\n",
" bm,\n",
" lbound,\n",
" ubound,\n",
" risk_constraints,\n",
" risk_target,\n",
" vol_low=0,\n",
" vol_high=target_vol,\n",
" factor_cov=risk_cov / 10000.,\n",
" factor_loading=risk_exposure,\n",
" idsync=(special_risk ** 2.) / 10000)\n",
" elasped_time1 = timeit.timeit(\"\"\"target_vol_builder(er,\n",
" None,\n",
" bm,\n",
" lbound,\n",
" ubound,\n",
" risk_constraints,\n",
" risk_target,\n",
" vol_low=0,\n",
" vol_high=target_vol,\n",
" factor_cov=risk_cov / 10000.,\n",
" factor_loading=risk_exposure,\n",
" idsync=(special_risk ** 2.) / 10000)\"\"\",\n",
" number=number, globals=globals()) / number * 1000\n",
" \n",
" w = cvxpy.Variable(n)\n",
" risk = quad_form(w - bm, sec_cov)\n",
" risk = sum_squares(mul_elemwise(special_risk / 100., w)) + quad_form((w.T * risk_exposure).T, risk_cov / 10000.)\n",
" objective = cvxpy.Minimize(-w.T * er)\n",
" curr_risk_exposure = risk_constraints.T @ w\n",
" constraints = [w >= lbound,\n",
......@@ -481,8 +563,8 @@
" curr_risk_exposure == risk_target[0],\n",
" risk <= target_vol * target_vol]\n",
" prob = cvxpy.Problem(objective, constraints)\n",
" prob.solve(solver='CVXOPT')\n",
" elasped_time2 = timeit.timeit(\"prob.solve(solver='CVXOPT')\",\n",
" prob.solve(solver='ECOS')\n",
" elasped_time2 = timeit.timeit(\"prob.solve(solver='ECOS')\",\n",
" number=number, globals=globals()) / number * 1000\n",
"\n",
" u1 = -x1 @ er\n",
......
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