hf Module¶
The hf module provides functions for synchronization of asynchronously
observed multivariate time series and observation noise cancelling. When
possible, functions are parallelized and accelerated via JIT compilation
with Numba. By default all cores of your machine are used. If your pipeline
allows for parallelization on a higher level, it is preferable to do so. You
may manually set the number of cores used by setting numba.set_num_threads(n)
.
Every estimator takes tick_series_list
as the first argument.
This is a list of pd.Series (one for each asset) containing
tick log-prices with pandas.DatetimeIndex. If you want to comute the covariance
of residuals after predictions are subtracted from log-returnsjust cumsum
the residuals. The output is the integrated covariance matrix
estimate as a 2d numpy.ndarray.
Functions¶
|
Ensemble multiple covariance matrix estimates with weights given by |
|
The h-th realized autocovariance. |
|
Compute the optimal bandwidth parameter $H$ for |
|
From a pd.Series of tick prices and predictions get a pd.Series of tick log-prices with zero-mean returns, i.e. |
|
The (pairwise) Hayashi-Yoshida estimator of Hayashi and Yoshida (2005). |
|
The kernel realized volatility matrix estimator (KRVM) of Barndorff-Nielsen et al. |
|
The modulated realised covariance (MRC) estimator of Christensen et al. |
|
The multi-scale realized volatility (MSRV) estimator of Zhang (2006). |
The Parzen weighting function used in the kernel realized volatility matrix estimator ( |
|
|
The preaveraging scheme of Podolskij and Vetter (2009). |
The Quadratic Spectral weighting function used in the kernel realized volatility matrix estimator ( |
|
|
The all-refresh time scheme of Barndorff-Nielsen et al. |
|
The two-scales realized volatility (TSRV) of Zhang et al. |