API

entroport.entroport

class entroport.entroport.EntroPort(df, estlength, step=1)

Portfolio allocation with relative entropy minimization

Estimates portfolio weights on a rolling out of sample basis by projecting past returns on an estimated stochastic discount factor.

Parameters:

df : DataFrame

Portfolio time series, net log (excess) returns

estlength : int

Length of the moving estimation window

step : int

The number of observations in the out of sample estimation window (default is 1) - a step size of for ex. 10 would be the same as a 10 period rebalancing.

Notes

In detail: for each estimation window of size T, estimate

\[\boldsymbol{\hat \theta} = \arg\min_{\boldsymbol{\theta}} \sum_{t=1}^{T}\mathrm{kernel}(\boldsymbol{\theta}, \mathbf{R}_t)\]

where

\[\mathrm{kernel}(\boldsymbol{\theta}, \mathbf{R}_t) = e^{\boldsymbol{\theta'} \mathbf{R}_t}\]

The estimated portfolio weights are the coefficients (normalized) obtained from regressing the returns in the estimation window on the kernel evaluated at \(\boldsymbol{\hat \theta}\)

The out of sample information portfolio is these weights multiplied by the out of sample returns.

The out of sample stochastic discount factor is the estimated kernel evaluated at the out of sample returns.

References

[R1]	Ghosh, Julliard, and Taylor A, “One Factor Benchmark Model for Asset Pricing”, (2015 wp)

Attributes

theta_	(DataFrame) Estimated thetas
weights_	(DataFrame) Estimated weights
pfs_	(DataFrame) The time series of the estimated stochastic discount factor (sdf) and information portfolio (ip)

fit()

Fit this model.

Returns:

self : object

Returns the instance itself.