Problem with annualization data and annualized / non-annualized metrics

[mariobajo]

I am having problems with the annualization of returns and the annualized/non annualized metrics. Here is an example for monthly data: freq=12.

1. with this code, I get this result:

Efficient Frontier

-----------------------------------------------------------------------

st.markdown("### 🟥 Efficient Frontier")

Crear BasePortfolios

portfolio_base_train = BasePortfolio(
returns=X_train,
observations=X_train.index,
annualized_factor=freq,
risk_free_rate=0,
value_at_risk_beta=0.99,
cvar_beta=0.99
)

portfolio_base_test = BasePortfolio(
returns=X_test,
observations=X_test.index,
annualized_factor=freq,
risk_free_rate=0,
value_at_risk_beta=0.99,
cvar_beta=0.99
)

Entrenamiento del modelo MVO_FE

model_MVO = MeanRisk(
risk_measure=RiskMeasure.STANDARD_DEVIATION,
efficient_frontier_size=100,
portfolio_params=dict(name="MVO")
)

model_MVO.fit(portfolio_base_train)

Predicciones

portfolio_MVO_train = model_MVO.predict(portfolio_base_train)
portfolio_MVO_test = model_MVO.predict(portfolio_base_test)

portfolio_MVO_train.set_portfolio_params(tag="Train")
portfolio_MVO_test.set_portfolio_params(tag="Test")

Concatenar resultados

portfolio_MVO = portfolio_MVO_train + portfolio_MVO_test

Chart Frontera eficiente

fig = portfolio_MVO.plot_measures(
x=RiskMeasure.STANDARD_DEVIATION,
y=PerfMeasure.MEAN,
color_scale=RatioMeasure.SHARPE_RATIO,
hover_measures=[RiskMeasure.MAX_DRAWDOWN, RiskMeasure.CVAR],
)

fig.update_layout(
title='Efficient frontier (Train & Test)',
xaxis_title='Annualized volatility (%)',
yaxis_title='Annualized return (%)',
font=dict(family="Arial", size=12),
)

Calcular media y desviación estándar anualizadas por activo

assets_returns = X_train.mean()*freq
assets_vol = X_train.std()*np.sqrt(freq)

Añadir cada activo como un scatter point en la frontera eficiente

fig.add_trace(go.Scatter(
x=assets_vol,
y=assets_returns,
mode='markers+text',
marker=dict(color='red', size=10, symbol='circle'),
text=assets_vol.index,
textposition='top center',
name='Assets'
))

st.plotly_chart(fig, use_container_width=True)

2. with this code, I get this result:

Efficient Frontier

-----------------------------------------------------------------------

st.markdown("### 🟥 Efficient Frontier")

Crear BasePortfolios

portfolio_base_train = BasePortfolio(
returns=X_train,
observations=X_train.index,
annualized_factor=freq,
risk_free_rate=0,
value_at_risk_beta=0.99,
cvar_beta=0.99
)

portfolio_base_test = BasePortfolio(
returns=X_test,
observations=X_test.index,
annualized_factor=freq,
risk_free_rate=0,
value_at_risk_beta=0.99,
cvar_beta=0.99
)

Entrenamiento del modelo MVO_FE

model_MVO = MeanRisk(
risk_measure=RiskMeasure.ANNUALIZED_STANDARD_DEVIATION,
efficient_frontier_size=100,
portfolio_params=dict(name="MVO")
)

model_MVO.fit(portfolio_base_train)

Predicciones

portfolio_MVO_train = model_MVO.predict(portfolio_base_train)
portfolio_MVO_test = model_MVO.predict(portfolio_base_test)

portfolio_MVO_train.set_portfolio_params(tag="Train")
portfolio_MVO_test.set_portfolio_params(tag="Test")

Concatenar resultados

portfolio_MVO = portfolio_MVO_train + portfolio_MVO_test

Chart Frontera eficiente

fig = portfolio_MVO.plot_measures(
x=RiskMeasure.ANNUALIZED_STANDARD_DEVIATION,
y=PerfMeasure.ANNUALIZED_MEAN,
color_scale=RatioMeasure.ANNUALIZED_SHARPE_RATIO,
hover_measures=[RiskMeasure.MAX_DRAWDOWN, RiskMeasure.CVAR],
)

fig.update_layout(
title='Efficient frontier (Train & Test)',
xaxis_title='Annualized volatility (%)',
yaxis_title='Annualized return (%)',
font=dict(family="Arial", size=12),
)

Calcular media y desviación estándar anualizadas por activo

assets_returns = X_train.mean()*freq
assets_vol = X_train.std()*np.sqrt(freq)

Añadir cada activo como un scatter point en la frontera eficiente

fig.add_trace(go.Scatter(
x=assets_vol,
y=assets_returns,
mode='markers+text',
marker=dict(color='red', size=10, symbol='circle'),
text=assets_vol.index,
textposition='top center',
name='Assets'
))

st.plotly_chart(fig, use_container_width=True)

The mean-risk of the individual assets is correct.

[HugoDelatte]

In the plot you have to use the annualized measures so you compare the same metrics as your individual assets:

fig = portfolio_MVO.plot_measures(
    x=RiskMeasure.ANNUALIZED_STANDARD_DEVIATION,
    y=PerfMeasure.ANNUALIZED_MEAN,
    color_scale=RatioMeasure.SHARPE_RATIO,
    hover_measures=[RiskMeasure.MAX_DRAWDOWN, RiskMeasure.CVAR],
)

On the fit and predict, it's interesting to see that you use a BasePortfolio as input. We usually use a DataFrame.
In that case it's working because inside the fit/predict we call np.asarray which transform the BasePortfolio into a numpy array.

In any case, you need to pass the annualized_factor as a portfolio_params in your optimization:

model_MVO = MeanRisk(
    risk_measure=RiskMeasure.STANDARD_DEVIATION,
    efficient_frontier_size=100,
    portfolio_params=dict(name="MVO", annualized_factor=freq)
)

model_MVO.fit(X_train)

portfolio_MVO_train = model_MVO.predict(X_train)
portfolio_MVO_test = model_MVO.predict(X_test)
)

This will fix your issue.

On a side note, I never thought of passing a BasePortfolio as input instead of a DataFrame so its params are not passed down, but it could be an interesting idea indeed. I'll explore further.

[mariobajo]

Thank you , Hugo. Working.

I could not find how set the annualization factor for different periodic data in the library. I tried with the concept of BasePortfolio since I understood in the doc that is the way to parametrize the basic portfolio characteristics before running a optimization.

In fact, when running other types of portoflio optimization, it works perfectly adding the annualized_factor in the portfolio after the optimization is done in order to get the proper ANNUALIZED_stataistic:

for cartera in selected_portfolios:
if cartera == "Inverse Volatility":
try:
model_iv = InverseVolatility()
model_iv.fit(X_train)
portfolios["Inverse Volatility"] = model_iv.predict(X_test)
portfolios["Inverse Volatility"].annualized_factor = freq <------------- not in model_iv = InverseVolatility()
except Exception as e:
st.error(f"Error in Inverse Volatility model: {e}")