predict_momentum_pocket: Predicts momentum using a regularized pocket algorithm
In benSepanski/pfselect:

See regularized_pocket() for a description of the algorithm used. Either uses the weight_elimination constant given or cross-validates at each trading period to choose one of the selected. If there is at least one cv asset, then each stage runs once with previous weights and once from linear regression weights, then picks whichever does best on cv error

predict_momentum_pocket(
  data,
  feature_colname,
  weight_elimination,
  maxit,
  momentum_colname = "momentum",
  max_cv_window = NULL
)

`data`	A tibble of class `@eval{AGG_WINDOW_CLASS_NAME}` as returned from `aggregate_price_and_mean_windows`, with an added column of column name `momentum_colname` holding the momentum.
`feature_colname`	The column name of the features to use. Should be a list column, each list entry holding a double vector.
`weight_elimination`	A double vector, at each stage picks the one which has performed best in the past cv window (or all of the past, whichever is selected)
`maxit`	the maximum number of iterations
`momentum_colname`	The name of the column holding the true momentum values (as a string), defaults to `"momentum"`. Should be (possibly negative) momentum values (see `evaluate_momentum_at_window`)
`max_cv_window`	If `NULL` then checks over all past error, otherwise only checks the past `max_cv_window` time steps to determine error.

A tibble like data but with a pocket_prediction pocket_error, and pocket_weight_elimination, columns holding the absolute value momentum prediction, whether it was an error, and the elimination weight used.

benSepanski/pfselect documentation built on May 1, 2020, 1:57 p.m.