def test_simulate_covariance():
"""
Test the deriving an empirical vector of means and an empirical covariance matrix.
"""
nco = NCO()
# Real values used for deriving empirical ones
mu_vec = np.array([0, 0.1, 0.2, 0.3])
cov_mat = np.array([[1, 0.1, 0.2, 0.3], [0.1, 1, 0.1, 0.2],
[0.2, 0.1, 1, 0.1], [0.3, 0.2, 0.1, 1]])
num_obs = 100000
# Using the function
mu_empir, cov_empir = nco._simulate_covariance(mu_vec, cov_mat,
num_obs, False)
# Testing that with a high number of observations empirical values are close to real ones
np.testing.assert_almost_equal(mu_empir.flatten(),
mu_vec.flatten(),
decimal=2)
np.testing.assert_almost_equal(cov_mat, cov_empir, decimal=2)
# Also testing the Ledoit-Wolf shrinkage
mu_empir_shr, cov_empir_shr = nco._simulate_covariance(
mu_vec, cov_mat, num_obs, True)
np.testing.assert_almost_equal(mu_empir_shr.flatten(),
mu_vec.flatten(),
decimal=2)
np.testing.assert_almost_equal(cov_mat, cov_empir_shr, decimal=2)
def test_allocate_cvo():
"""
Test the estimates of the Convex Optimization Solution (CVO).
"""
nco = NCO()
# Covariance matrix and the desired mu vector
cov_matrix = np.array([[0.01, 0.002, -0.001], [0.002, 0.04, -0.006],
[-0.001, -0.006, 0.01]])
mu_vec = np.array([1, 1, 1]).reshape(-1, 1)
# Expected weights for minimum variance allocation
w_expected = np.array([[0.37686939], [0.14257228], [0.48055833]])
# Finding the optimal weights
w_cvo = nco.allocate_cvo(cov_matrix, mu_vec=None)
# Also when manually inputting the vector mu
w_cvo_mu = nco.allocate_cvo(cov_matrix, mu_vec=mu_vec)
# Testing if the optimal allocation is right and if the custom mu works
np.testing.assert_almost_equal(w_cvo, w_expected, decimal=4)
np.testing.assert_almost_equal(w_cvo, w_cvo_mu, decimal=4)
def test_estim_errors_mcos():
"""
Test the computation the true optimal allocation w, and compares that result with the estimated ones by MCOS.
"""
nco = NCO()
# Weights from CVO and NCO, and data to estimate the true optimal weights
np.random.seed(0)
w_cvo = pd.DataFrame([[0.249287, 0.256002, 0.242593, 0.252118],
[0.257547, 0.265450, 0.242453, 0.234551]])
w_nco = pd.DataFrame([[0.248396, 0.243172, 0.250751, 0.257680],
[0.257547, 0.265450, 0.242453, 0.234551]])
mu_vec = np.array([0, 0.1, 0.2, 0.3])
cov_mat = np.array([[1, 0.1, 0.2, 0.3], [0.1, 1, 0.1, 0.2],
[0.2, 0.1, 1, 0.1], [0.3, 0.2, 0.1, 1]])
min_var_portf = True
# Expected errors
err_cvo_expected = 0.0062604
err_nco_expected = 0.0111115
# Finding the errors in estimations
err_cvo, err_nco = nco.estim_errors_mcos(w_cvo, w_nco, mu_vec, cov_mat,
min_var_portf)
# Testing if the error computation is right
np.testing.assert_almost_equal(err_cvo, err_cvo_expected, decimal=4)
np.testing.assert_almost_equal(err_nco, err_nco_expected, decimal=4)
def test_allocate_nco():
"""
Test the estimates the optimal allocation using the (NCO) algorithm
"""
nco = NCO()
# Covariance matrix and the custom mu vector
np.random.seed(0)
cov_matrix = np.array([[0.01, 0.002, -0.001], [0.002, 0.04, -0.006],
[-0.001, -0.006, 0.01]])
mu_vec = np.array([1, 1, 1]).reshape(-1, 1)
# Expected weights for minimum variance allocation
w_expected = np.array([[0.43875825], [0.09237016], [0.4688716]])
max_num_clusters = 2
# Finding the optimal weights
w_nco = nco.allocate_nco(cov_matrix, max_num_clusters=max_num_clusters)
# Finding the optimal weights using the custom mu vector
w_nco_mu = nco.allocate_nco(cov_matrix,
mu_vec=mu_vec,
max_num_clusters=max_num_clusters)
# Testing if the optimal allocation is right and if the custom mu works
np.testing.assert_almost_equal(w_nco, w_expected, decimal=4)
np.testing.assert_almost_equal(w_nco, w_nco_mu, decimal=4)
def test_cluster_kmeans_base(self):
"""
Test the finding of the optimal partition of clusters using K-Means algorithm.
"""
nco = NCO()
# Correlation matrix and parameters used in the K-Means algorithm.
np.random.seed(0)
corr_matrix = pd.DataFrame([[1, 0.1, -0.1],
[0.1, 1, -0.3],
[-0.1, -0.3, 1]])
max_num_clusters = 2
n_init = 10
# Expected correlation matrix of clustered elements, clusters and, Silhouette Coefficient series
expected_corr = pd.DataFrame([[1, 0.1, -0.1],
[0.1, 1, -0.3],
[-0.1, - 0.3, 1]])
expected_clust = {0: [0, 1], 1: [2]}
expected_silh_coef = pd.Series([0.100834, 0.167626, 0], index=[0, 1, 2])
# Finding the clustered corresponding values
corr, clusters, silh_coef = nco._cluster_kmeans_base(corr_matrix, max_num_clusters, n_init)
# When maximum number of clusters is not predefined
corr_no_max, _, _ = nco._cluster_kmeans_base(corr_matrix, None, n_init)
# Testing if the values are right
self.assertTrue(clusters == expected_clust)
np.testing.assert_almost_equal(np.array(corr), np.array(expected_corr), decimal=4)
np.testing.assert_almost_equal(np.array(silh_coef), np.array(expected_silh_coef), decimal=4)
np.testing.assert_almost_equal(np.array(corr), np.array(corr_no_max), decimal=4)
def test_form_true_matrix():
"""
Test the creation of a random vector of means and a random covariance matrix.
"""
nco = NCO()
# Parameters to build the vector of means and a covariance matrix
np.random.seed(0)
num_blocks = 2
block_size = 2
block_corr = 0.3
std = 0.3
# Alternative std parameter
std_alt = None
# Expected vector of means and covariance matrix
mu_expected = np.array([[0.5936214],
[0.97226796],
[0.8602674],
[0.00681664]])
cov_expected = pd.DataFrame([[0.09, 0.027, 0, 0],
[0.027, 0.09, 0, 0],
[0, 0, 0.09, 0.027],
[0, 0, 0.027, 0.09]])
mu_alt_expected = np.array([[0.34260886],
[0.12059827],
[0.24366456],
[0.17248975]])
# Finding the random vector of means and covariance matrix
mu_vec, cov_matrix = nco.form_true_matrix(num_blocks, block_size, block_corr, std)
# Also when the std is default
mu_vec_alt, _ = nco.form_true_matrix(num_blocks, block_size, block_corr, std_alt)
# Testing if the results are right
np.testing.assert_almost_equal(mu_vec, mu_expected, decimal=4)
np.testing.assert_almost_equal(np.array(cov_matrix), np.array(cov_expected), decimal=4)
np.testing.assert_almost_equal(mu_vec_alt, mu_alt_expected, decimal=4)
def test_form_block_matrix():
"""
Test the creation of a block correlation matrix with given parameters.
"""
nco = NCO()
# Parameters to build the block correlation matrix with
num_blocks = 2
block_size = 2
block_corr = 0.3
# Expected corr matrix
corr_expected = np.array([[1, 0.3, 0, 0], [0.3, 1, 0, 0],
[0, 0, 1, 0.3], [0, 0, 0.3, 1]])
# Finding the errors in estimations
corr_matrix = nco._form_block_matrix(num_blocks, block_size,
block_corr)
# Testing if the error computation is right
np.testing.assert_almost_equal(corr_matrix, corr_expected, decimal=4)
def test_allocate_mcos():
"""
Test the estimates of the optimal allocation using the Monte Carlo optimization selection
"""
nco = NCO()
# Covariance matrix, mean vector and other variables for the method
np.random.seed(0)
mu_vec = np.array([0, 0.1, 0.2, 0.3])
cov_mat = np.array([[1, 0.1, 0.2, 0.3], [0.1, 1, 0.1, 0.2],
[0.2, 0.1, 1, 0.1], [0.3, 0.2, 0.1, 1]])
num_obs = 100
num_sims = 2
kde_bwidth = 0.25
min_var_portf = True
lw_shrinkage = False
# Alternative set of values
min_var_portf_alt = False
kde_bwidth_alt = 0
# Expected weights for minimum variance allocation
w_cvo_expected = pd.DataFrame(
[[0.249287, 0.256002, 0.242593, 0.252118],
[0.257547, 0.265450, 0.242453, 0.234551]])
w_nco_expected = pd.DataFrame(
[[0.248396, 0.243172, 0.250751, 0.257680],
[0.257547, 0.265450, 0.242453, 0.234551]])
# Expected weights for maximum Sharpe ratio allocation
w_cvo_sr_expected = pd.DataFrame(
[[-1.081719, 1.810936, 1.218067, 3.978880],
[-2.431651, 0.594868, -0.210175, 5.117628]])
w_nco_sr_expected = pd.DataFrame(
[[-1.060835, 1.910503, 1.315026, 3.908128],
[-0.937168, 1.886158, -0.389275, 4.884809]])
# Finding the optimal weights for minimum variance
w_cvo, w_nco = nco.allocate_mcos(mu_vec, cov_mat, num_obs, num_sims,
kde_bwidth, min_var_portf,
lw_shrinkage)
# Finding the optimal weights for maximum Sharpe ratio
w_cvo_sr, w_nco_sr = nco.allocate_mcos(mu_vec, cov_mat, num_obs,
num_sims, kde_bwidth_alt,
min_var_portf_alt, lw_shrinkage)
# Testing if the optimal allocation simulations are right
np.testing.assert_almost_equal(np.array(w_cvo),
np.array(w_cvo_expected),
decimal=4)
np.testing.assert_almost_equal(np.array(w_nco),
np.array(w_nco_expected),
decimal=4)
np.testing.assert_almost_equal(np.array(w_cvo_sr),
np.array(w_cvo_sr_expected),
decimal=4)
np.testing.assert_almost_equal(np.array(w_nco_sr),
np.array(w_nco_sr_expected),
decimal=4)
def test_allocate_mcos():
"""
Test the estimates of the optimal allocation using the Monte Carlo optimization selection
"""
nco = NCO()
# Covariance matrix, mean vector and other variables for the method
np.random.seed(0)
mu_vec = np.array([0, 0.1, 0.2, 0.3])
cov_mat = np.array([[1, 0.1, 0.2, 0.3],
[0.1, 1, 0.1, 0.2],
[0.2, 0.1, 1, 0.1],
[0.3, 0.2, 0.1, 1]])
num_obs = 100
num_sims = 2
kde_bwidth = 0.25
min_var_portf = True
lw_shrinkage = False
# Alternative set of values
min_var_portf_alt = False
kde_bwidth_alt = 0
# Expected weights for minimum variance allocation
# Second line in the below test was changed after v.0.11.3 from [0.257547, 0.265450, 0.242453, 0.234551]
# to [0.303161, 0.256327, 0.177548, 0.262963]
# due to scikit-learn changing np.random outputs
w_cvo_expected = pd.DataFrame([[0.249287, 0.256002, 0.242593, 0.252118],
[0.303161, 0.256327, 0.177548, 0.262963]])
# Second line in the below test was changed after v.0.11.3 from [0.257547, 0.265450, 0.242453, 0.234551]
# to [0.301132, 0.247042, 0.198597, 0.25323]
# due to scikit-learn changing np.random outputs
w_nco_expected = pd.DataFrame([[0.248396, 0.243172, 0.250751, 0.257680],
[0.301132, 0.247042, 0.198597, 0.25323]])
# Expected weights for maximum Sharpe ratio allocation
# Values in the below test was changed after v.0.11.3 from [[-1.081719, 1.810936, 1.218067, 3.978880]
# [-2.431651, 0.594868, -0.210175, 5.117628]]
# to [[-0.128849, -0.326671, 0.870183, 2.020053]
# [-2.095454, -0.7403, 2.390951, 1.793756]]
# due to scikit-learn changing np.random outputs
w_cvo_sr_expected = pd.DataFrame([[-0.128849, -0.326671, 0.870183, 2.020053],
[-2.095454, -0.7403, 2.390951, 1.793756]])
# Values in the below test was changed after v.0.11.3 from [[-1.060835, 1.910503, 1.315026, 3.908128]
# [-0.937168, 1.886158, -0.389275, 4.884809]]
# to [[-0.204089, -0.050088, 0.912494, 1.983382]
# [-2.036145, -0.769203, 2.677561, 1.347365]]
# due to scikit-learn changing np.random outputs
w_nco_sr_expected = pd.DataFrame([[-0.204089, -0.050088, 0.912494, 1.983382],
[-2.036145, -0.769203, 2.677561, 1.347365]])
# Finding the optimal weights for minimum variance
w_cvo, w_nco = nco.allocate_mcos(mu_vec, cov_mat, num_obs, num_sims, kde_bwidth, min_var_portf, lw_shrinkage)
# Finding the optimal weights for maximum Sharpe ratio
w_cvo_sr, w_nco_sr = nco.allocate_mcos(mu_vec, cov_mat, num_obs, num_sims, kde_bwidth_alt, min_var_portf_alt, lw_shrinkage)
# Testing if the optimal allocation simulations are right
np.testing.assert_almost_equal(np.array(w_cvo), np.array(w_cvo_expected), decimal=6)
np.testing.assert_almost_equal(np.array(w_nco), np.array(w_nco_expected), decimal=6)
np.testing.assert_almost_equal(np.array(w_cvo_sr), np.array(w_cvo_sr_expected), decimal=6)
np.testing.assert_almost_equal(np.array(w_nco_sr), np.array(w_nco_sr_expected), decimal=6)