predloglikWB: Evaluates the predictive log likelihood using the Woodbury...

In factorstochvol: Bayesian Estimation of (Sparse) Latent Factor Stochastic Volatility Models

Evaluates the predictive log likelihood using the Woodbury identity

Description

predloglikWB approximates the predictive log likelihood exploiting the factor structure and using the Woodbury idenitity and the corresponding matrix determinant lemma. This is recommended only if many series and few factors are present.

Usage

predloglikWB(x, y, ahead = 1, each = 1, alldraws = FALSE)

Arguments

x	Object of class 'fsvdraws', usually resulting from a call to fsvsample.
y	Matrix of dimension length(ahead) times m where the predictive density should be evaluated.
ahead	Vector of timepoints, indicating how many steps to predict ahead.
each	Single integer (or coercible to such) indicating how often should be drawn from the posterior predictive distribution for each draw that has been stored during MCMC sampling.
alldraws	Should all the draws be returned or just the final results? (Can be useful to assess convergence.)

Value

Vector of length length(ahead) with log predictive likelihoods.

Note

Currently crudely implemented as a triple loop in pure R, may be slow.

See Also

Uses predprecWB. If m is small or many factors are used, consider also using predcov.

Other predictors: predcond(), predcor(), predcov(), predh(), predloglik(), predprecWB()

Examples

set.seed(1)

# Simulate a time series of length 1100:
sim <- fsvsim(n = 1100, series = 3, factors = 1)
y <- sim$y

# Estimate using only 1000 days:
res <- fsvsample(y[seq_len(1000),], factors = 1)

# Evaluate the 1, 10, and 100 days ahead predictive log
# likelihood:
ahead <- c(1, 10, 100)
scores <- predloglikWB(res, y[1000+ahead,], ahead = ahead, each = 10)
print(scores)

factorstochvol documentation built on Nov. 24, 2023, 5:08 p.m.