portfoliowalkr: The Portfoliowalkr function

In jluby/portfoliowalkr: Characteristic-Matched Random Portfolio Creation

Description

Given an asset universe A and a vector of baseline weights with which to calculate target portfolio characteristics b, portfoliowalkr samples points from the intersection of Ax = b with the n-simplex (∑ x = 1, x_i ≥ 0).

Usage

portfoliowalkr(universe, match, portfolio.weight = "portfolio",
  ret.var = "return", points = 1000, method = "dikin", chains = 1,
  thin = 1, burn = 0.5)

Arguments

universe	is the universe of stocks available for sampling, with dates, returns, weights, and any desired characteristics
match	is the list of columns containing characteristics to be matched
portfolio.weight	is the name of the column containing the weights of the target portfolio
ret.var	is the name of the column containing percentage return over the time period
points	is the number of points we want to sample
method	is the MCMC sampling method. Please enter "hit-and-run" or "dikin"
chains	is the number of chains run
thin	every thin-th point is stored
burn	the first burn points are deleted

Details

portfoliowalkr, per its namesake, makes use of the walkr package to sample randomly from the space of characteristic-matched portfolios. portfoliowalkr returns points vectors of weights, which when applied to the universe A yields a portfolio with the same characteristics as the target.

Per walkr:

1) Hit-and-run is computationally less expensive and also guarantees uniformity asympotically with complexity of O(n^3) points with respect to dimension n. However, in real practice, as dimensions ramp up, the mixing of hit-and-run is poor compared to Dikin. Thus, a lot of thinning would be needed as dimension ramps up.

2) Dikin Walk is a nearly uniform method known for its very strong mixing properties. However, each Dikin step is much more computationally expensive than hit-and-run, so it takes more time to sample every point. Thus, the "dikin" method uses RcppEigen to speed up the core computationally expensive operations in the algorithm.

Value

A list of objects for assessing portfolio performance. $plot returns a histogram of random portfolio performances with target portfolio and universe performances demarcated for comparison. $summary returns a numeric description of these performances. $frame returns the universe frame with matched weights attached. $returns is a frame containing each portfolios' return, including one evenly weighted within the universe for comparison. $explore is a list of chains from the walkr output which can be input into the explore_walkr function to better understand the convergence of the model.

Examples

match_list <- portfoliowalkr(universe = jan, match = c('sector', 'growth', 'size'), points = 100, method = "dikin")       
match_list$frame
                                                                                             

jluby/portfoliowalkr documentation built on April 4, 2020, 1:46 a.m.