Nested Clusters Optimization

This tutorial introduces the NestedClustersOptimization optimization.

Nested Clusters Optimization (NCO) is a portfolio optimization method developed by Marcos Lopez de Prado.

It uses a distance matrix to compute clusters using a clustering algorithm ( Hierarchical Tree Clustering, KMeans, etc..). For each cluster, the inner-cluster weights are computed by fitting the inner-estimator on each cluster using the whole training data. Then the outer-cluster weights are computed by training the outer-estimator using out-of-sample estimates of the inner-estimators with cross-validation. Finally, the final assets weights are the dot-product of the inner-weights and outer-weights.

Note

The original paper uses KMeans as the clustering algorithm, minimum Variance for the inner-estimator and equal-weight for the outer-estimator. Here we generalize it to all sklearn and skfolio clustering algorithm (Hierarchical Tree Clustering, KMeans, etc.), all portfolio optimizations (Mean-Variance, HRP, etc.) and risk measures (variance, CVaR, etc.). To avoid data leakage at the outer-estimator, we use out-of-sample estimates to fit the outer estimator.

Data

We load the S&P 500 dataset composed of the daily prices of 20 assets from the S&P 500 Index composition starting from 1990-01-02 up to 2022-12-28:

from plotly.io import show
from sklearn.cluster import KMeans
from sklearn.model_selection import train_test_split

from skfolio import Population, RiskMeasure
from skfolio.cluster import HierarchicalClustering, LinkageMethod
from skfolio.datasets import load_sp500_dataset
from skfolio.distance import KendallDistance
from skfolio.optimization import (
    EqualWeighted,
    MeanRisk,
    NestedClustersOptimization,
    ObjectiveFunction,
    RiskBudgeting,
)
from skfolio.preprocessing import prices_to_returns

prices = load_sp500_dataset()

X = prices_to_returns(prices)
X_train, X_test = train_test_split(X, test_size=0.33, shuffle=False)

Model

We create a NCO model that maximizes the Sharpe Ratio intra-cluster and uses a CVaR Risk Parity inter-cluster. By default, the inter-cluster optimization uses KFolds out-of-sample estimates of the inner-estimator to avoid data leakage. and the HierarchicalClustering estimator to form the clusters:

inner_estimator = MeanRisk(
    objective_function=ObjectiveFunction.MAXIMIZE_RATIO,
    risk_measure=RiskMeasure.VARIANCE,
)
outer_estimator = RiskBudgeting(risk_measure=RiskMeasure.CVAR)

model1 = NestedClustersOptimization(
    inner_estimator=inner_estimator,
    outer_estimator=outer_estimator,
    n_jobs=-1,
    portfolio_params=dict(name="NCO-1"),
)
model1.fit(X_train)
model1.weights_

array([4.34583888e-02, 3.11063721e-03, 1.22117587e-12, 4.55249501e-02,
       7.75790832e-02, 6.00652106e-02, 4.72968062e-02, 1.25073031e-01,
       4.62236661e-02, 3.16514560e-02, 2.74284346e-12, 2.48339328e-03,
       8.63901744e-02, 3.19328243e-02, 6.00333111e-02, 4.79313343e-02,
       9.30700985e-02, 5.98357211e-02, 4.48492469e-02, 9.34906672e-02])

Dendrogram

To analyze the clusters structure, we can plot the dendrogram. The blue lines represent distinct clusters composed of a single asset. The remaining colors represent clusters of more than one asset:

model1.clustering_estimator_.plot_dendrogram(heatmap=False)

The horizontal axis represent the assets. The links between clusters are represented as upside-down U-shaped lines. The height of the U indicates the distance between the clusters. For example, the link representing the cluster containing Assets HD and WMT has a distance of 0.5 (called cophenetic distance).

When heatmap is set to True, the heatmap of the reordered distance matrix is displayed below the dendrogram and clusters are outlined with yellow squares:

model1.clustering_estimator_.plot_dendrogram()

Linkage Methods

The hierarchical clustering can be greatly affected by the choice of the linkage method. In the HierarchicalClustering estimator, the default linkage method is set to the Ward variance minimization algorithm, which is more stable and has better properties than the single-linkage method which suffers from the chaining effect.

To show this effect, let’s create a second model with the single-linkage method:

model2 = NestedClustersOptimization(
    inner_estimator=inner_estimator,
    outer_estimator=outer_estimator,
    clustering_estimator=HierarchicalClustering(
        linkage_method=LinkageMethod.SINGLE,
    ),
    n_jobs=-1,
    portfolio_params=dict(name="NCO-2"),
)
model2.fit(X_train)
model2.clustering_estimator_.plot_dendrogram(heatmap=True)

Distance Estimator

The distance metric used has also an important effect on the clustering. The default is to use the distance of the pearson correlation matrix. This can be changed using the distance estimators.

For example, let’s create a third model with a distance computed from the absolute value of the Kendal correlation matrix:

model3 = NestedClustersOptimization(
    inner_estimator=inner_estimator,
    outer_estimator=outer_estimator,
    distance_estimator=KendallDistance(absolute=True),
    n_jobs=-1,
    portfolio_params=dict(name="NCO-3"),
)
model3.fit(X_train)
model3.clustering_estimator_.plot_dendrogram(heatmap=True)

Clustering Estimator

The above models used the default HierarchicalClustering estimator. This can be replaced by any sklearn or skfolio clustering estimators.

For example, let’s create a new model with sklearn.cluster.KMeans:

model4 = NestedClustersOptimization(
    inner_estimator=inner_estimator,
    outer_estimator=outer_estimator,
    clustering_estimator=KMeans(n_init="auto"),
    n_jobs=-1,
    portfolio_params=dict(name="NCO-4"),
)
model4.fit(X_train)
model4.weights_

array([9.37398256e-02, 6.54760361e-02, 2.23827517e-12, 4.19444429e-02,
       7.44202615e-02, 1.21835406e-02, 4.35769437e-02, 1.21100445e-01,
       1.50869871e-02, 3.15598827e-02, 2.65572491e-12, 2.40451542e-03,
       8.04789599e-02, 3.18404370e-02, 5.81265252e-02, 4.77926605e-02,
       8.96000814e-02, 5.96626058e-02, 4.13218833e-02, 8.96839665e-02])

To compare the NCO models, we use an equal weighted benchmark using the EqualWeighted estimator:

bench = EqualWeighted()
bench.fit(X_train)
bench.weights_

array([0.05, 0.05, 0.05, 0.05, 0.05, 0.05, 0.05, 0.05, 0.05, 0.05, 0.05,
       0.05, 0.05, 0.05, 0.05, 0.05, 0.05, 0.05, 0.05, 0.05])

Prediction

We predict the models and the benchmark on the test set:

population_test = Population([])
for model in [model1, model2, model3, model4, bench]:
    population_test.append(model.predict(X_test))

population_test.plot_cumulative_returns()

Composition

Let’s plot each portfolio composition:

fig = population_test.plot_composition()
show(fig)