准则集合

LogWealth

描述:

LogWealth 准则计算投资组合每日收益的负均值对数。该准则旨在最大化投资组合财富的期望值。

函数调用:

compute_log_wealth_loss()

代码:

>>> def compute_log_wealth_loss(self, portfolios: torch.Tensor, labels: torch.Tensor) -> torch.Tensor:
>>>     """
>>>     Compute the ``LogWealth`` loss, which calculates the negative mean logarithm of the portfolio's daily returns.

>>>     :param portfolios: Portfolio weights tensor of shape (batch_size, num_assets).
>>>     :param labels: Label tensor representing asset returns of shape (batch_size, num_assets).
>>>     :return: ``LogWealth`` loss tensor, representing the negative mean logarithm of the portfolio's daily returns.
>>>     """
>>>     daily_returns = torch.sum(portfolios * labels, dim=-1)
>>>     log_returns = torch.log(torch.clamp(daily_returns, min=1e-6))
>>>     loss = - torch.mean(log_returns)
>>>     return loss

LogWealthL2Diversification

描述:

LogWealthL2Diversification 准则通过引入基于连续投资组合权重向量之间差异的 L2 范数的多样化惩罚，扩展了 LogWealth 准则。该惩罚鼓励投资组合实现多样化，降低集中风险。

函数调用:

compute_log_wealth_l2_diversification_loss()

代码:

>>> def compute_log_wealth_l2_diversification_loss(self, portfolios: torch.Tensor, labels: torch.Tensor) -> torch.Tensor:
>>>     """
>>>     Compute the ``LogWealthL2Diversification`` loss, which extends the ``LogWealth`` loss by incorporating diversification measures.
>>>
>>>     :param portfolios: Portfolio weights tensor of shape (batch_size, num_assets).
>>>     :param labels: Label tensor representing asset returns of shape (batch_size, num_assets).
>>>     :return: ``LogWealthL2Diversification`` loss tensor, a modified version of ``LogWealth`` loss with diversification metrics included.
>>>     """
>>>     daily_returns = torch.sum(portfolios * labels, dim=-1)
>>>     log_returns = torch.log(torch.clamp(daily_returns, min=1e-6))
>>>
>>>     # Calculate the similarity loss
>>>     diff = portfolios[1:, :] - portfolios[:-1, :]
>>>     diff_loss = torch.norm(diff, p=2, dim=-1)
>>>
>>>     # The final loss combines the log wealth and the L2 norm of the differences
>>>     loss = - torch.mean(log_returns) + self.config["LAMBDA_L2"] * torch.mean(diff_loss)
>>>     return loss

LogWealthL2Concentration

描述:

LogWealthL2Concentration 准则通过引入基于连续投资组合权重向量之间差异的负 L2 范数的集中惩罚，扩展了 LogWealth 准则。该惩罚鼓励投资组合集中，从而增加更高收益的潜力。

函数调用:

compute_log_wealth_l2_concentration_loss()

代码:

>>> def compute_log_wealth_l2_concentration_loss(self, portfolios: torch.Tensor, labels: torch.Tensor) -> torch.Tensor:
>>>     """
>>>     Compute the ``LogWealthL2Concentration`` loss, which extends the ``LogWealth`` loss by incorporating concentration measures.
>>>
>>>     :param portfolios: Portfolio weights tensor of shape (batch_size, num_assets).
>>>     :param labels: Label tensor representing asset returns of shape (batch_size, num_assets).
>>>     :return: ``LogWealthL2Concentration`` loss tensor, a modified version of ``LogWealth`` loss with concentration metrics included.
>>>     """
>>>     daily_returns = torch.sum(portfolios * labels, dim=-1)
>>>     log_returns = torch.log(torch.clamp(daily_returns, min=1e-6))
>>>
>>>     # Calculate the similarity loss
>>>     diff = portfolios[1:, :] - portfolios[:-1, :]
>>>     diff_loss = torch.norm(diff, p=2, dim=-1)
>>>
>>>     # The final loss combines the log wealth and the L2 norm of the differences
>>>     loss = - torch.mean(log_returns) - self.config["LAMBDA_L2"] * torch.mean(diff_loss)
>>>     return loss

L2Diversification

描述:

L2Diversification 准则基于连续投资组合权重向量之间差异的 L2 范数来衡量投资组合的多样化。该准则鼓励投资组合实现多样化，降低集中风险。

函数调用:

compute_l2_diversification_loss()

代码:

>>> def compute_l2_diversification_loss(self, portfolios: torch.Tensor, labels: torch.Tensor) -> torch.Tensor:
>>>     """
>>>     Compute the ``L2Diversification`` loss, which measures the diversification of a portfolio based on the L2 norm of consecutive portfolio weight differences.

>>>     This loss function calculates the mean L2 norm of the differences between consecutive portfolio weight vectors.

>>>     :param portfolios: Portfolio weights tensor of shape (batch_size, num_assets).
>>>     :param labels: Label tensor representing asset returns of shape (batch_size, num_assets).
>>>     :return: ``L2Diversification`` loss tensor, indicating the degree of diversification based on the L2 norm of portfolio weight differences.
>>>     """
>>>     # Calculate the similarity loss
>>>     # We use the L2 norm to measure the difference between consecutive predictions
>>>     diff = portfolios[1:, :] - portfolios[:-1, :]
>>>     diff_loss = torch.norm(diff, p=2, dim=-1)

>>>     # By minimizing the L2 norm of the difference vectors, you will tend to get two vectors that are similar, since the
>>>     # L2 norm measures the Euclidean distance between vectors. When the L2 norm is small, it means that the difference
>>>     # between two vectors is small.
>>>     loss = torch.mean(diff_loss)
>>>     return loss

L2Concentration

描述:

L2Concentration 准则基于连续投资组合权重向量之间差异的负 L2 范数来衡量投资组合的集中程度。该准则鼓励投资组合集中，从而增加更高收益的潜力。

函数调用:

compute_l2_concentration_loss()

代码:

>>> def compute_l2_concentration_loss(self, portfolios: torch.Tensor, labels: torch.Tensor) -> torch.Tensor:
>>>     """
>>>     Compute the ``L2Concentration`` loss, which measures the concentration of a portfolio based on the L2 norm of consecutive portfolio weight differences.

>>>     This loss function calculates the negative mean L2 norm of the differences between consecutive portfolio weight vectors.

>>>     :param portfolios: Portfolio weights tensor of shape (batch_size, num_assets).
>>>     :param labels: Label tensor representing asset returns of shape (batch_size, num_assets).
>>>     :return: ``L2Concentration`` loss tensor, indicating the degree of concentration based on the negative mean L2 norm of portfolio weight differences.
>>>     """
>>>     # Calculate the similarity loss
>>>     # We use the L2 norm to measure the difference between consecutive predictions
>>>     diff = portfolios[1:, :] - portfolios[:-1, :]
>>>     diff_loss = torch.norm(diff, p=2, dim=-1)

>>>     # By minimizing the L2 norm of the difference vectors, you will tend to get two vectors that are similar, since the
>>>     # L2 norm measures the Euclidean distance between vectors. When the L2 norm is small, it means that the difference
>>>     # between two vectors is small.
>>>     loss = - torch.mean(diff_loss)
>>>     return loss

SharpeRatio

描述:

SharpeRatio 准则使用夏普比率评估投资组合的风险调整收益。该准则旨在最大化夏普比率，即均值超额收益与超额收益标准差的比率。

函数调用:

compute_sharpe_ratio_loss()

代码:

>>> def compute_sharpe_ratio_loss(self, portfolios: torch.Tensor, labels: torch.Tensor) -> torch.Tensor:
>>>     """
>>>     Compute the ``SharpeRatio`` loss, which evaluates the risk-adjusted return of the portfolios using the Sharpe Ratio.

>>>     This loss function calculates the negative Sharpe Ratio, which is the ratio of the mean excess return to the standard deviation of the excess return.

>>>     :param portfolios: Portfolio weights tensor of shape (batch_size, num_assets).
>>>     :param labels: Label tensor representing asset returns of shape (batch_size, num_assets).
>>>     :return: ``SharpeRatio`` loss tensor, indicating the negative Sharpe Ratio value for the portfolios.
>>>     """
>>>     daily_returns = torch.sum(portfolios * labels, dim=-1)
>>>     # log_returns = torch.log(torch.clamp(daily_returns, min=1e-6))
>>>     excess_returns = daily_returns  # risk_free_rate = 0

>>>     mean_excess_return = torch.mean(excess_returns)
>>>     std_excess_return = torch.std(excess_returns)

>>>     sharpe_ratio = mean_excess_return / (std_excess_return + 1e-6)
>>>     loss = - sharpe_ratio
>>>     return loss

Volatility

描述:

Volatility 准则基于每日收益的标准差来衡量投资组合的波动性。该准则旨在最小化模型的波动性，从而降低大额损失的风险。

函数调用:

compute_volatility_loss()

代码:

>>> def compute_volatility_loss(self, portfolios: torch.Tensor, labels: torch.Tensor) -> torch.Tensor:
>>>     """
>>>     Compute the ``Volatility`` loss, which measures the volatility of the portfolios based on the standard deviation of daily returns.

>>>     This loss function calculates the standard deviation of the daily portfolio returns.

>>>     :param portfolios: Portfolio weights tensor of shape (batch_size, num_assets).
>>>     :param labels: Label tensor representing asset returns of shape (batch_size, num_assets).
>>>     :return: ``Volatility`` loss tensor, representing the volatility of the portfolios based on the standard deviation of daily returns.
>>>     """
>>>     daily_returns = torch.sum(portfolios * labels, dim=-1)
>>>     volatility = torch.std(daily_returns)
>>>     loss = volatility
>>>     return loss

CustomCriterion

描述:

CustomCriterion 准则是用户定义自定义损失准则的占位符。该准则允许用户实现任何他们所需的自定义损失函数，以满足具体需求。

函数调用:

compute_custom_criterion_loss()

代码:

>>> def compute_custom_criterion_loss(self, portfolios: torch.Tensor, labels: torch.Tensor) -> torch.Tensor:
>>>     """
>>>     Compute the ``CustomCriterion`` loss,  which is left for the user to define.

>>>     This loss function is a placeholder for the user to implement their own custom loss criterion.

>>>     :param portfolios: Portfolio weights tensor of shape (batch_size, num_assets).
>>>     :param labels: Label tensor representing asset returns of shape (batch_size, num_assets).
>>>     :return: ``CustomCriteria`` loss tensor, representing the user-defined loss criterion.
>>>     """
>>>     # This is a placeholder for the user to implement their own custom loss function.
>>>     # The implementation of the custom loss function is left to the user.
>>>     loss = torch.tensor(0.0, requires_grad=True)
>>>     return loss