ms_dcf_estim: Markov switching model estimation by the Kim filter (Hamilton...

In opendoor-labs/MarkovSwitchingDCF: Markov Switching Dynamic Common Factor Model using the Kim filter (Hamilton + Kalman filters)

Description

Model priors (initial values) Parameter naming convention DCF AR coefficients: phi1 and phi2 only Error MA coefficients: psi_i1 to psi_i2 only for each series i Error standard deviation: sigma_i only for each series i Observation coefficient on DCF: First gamma (gamma_i, ..., gamma_n) only 1 index number, not i0 Any more gammas per equation: gamma_i1 to gamma_ik Markov switching growth rate: mu_d and mu_u Transition probabilities: p_dd, p_md (or p_mu), p_mm, p_md (or p_mu), p_uu, p_ud (or p_um)

Usage

ms_dcf_estim(y, freq = NULL, panelID = NULL, timeID = NULL,
  level = 0.01, log.vars = NULL, diff.vars = NULL,
  detect.formula = F, detect.growth = F, detect.diff = F,
  diff.lag = 1, n_states = 2, ms_var = F, prior = NULL,
  formulas = c("y ~ c + e.l1 + e.l2"), weighted = F,
  optim_methods = c("BFGS", "CG", "NM"), maxit = 1000, trace = F)

Arguments

y	Multivariate time series of data values. May also be a data frame containing a date column
freq	Seasonality of the data
panelID	Column name that identifies the cross section of the data
timeID	Column name that identifies the date
level	Significance level of statistical tests (0.01, 0.05, 0.1)
log.vars	Character vector of variables to be logged
diff.vars	Character vector of unit root variables to be differenced
detect.formula	Logical, detect lag length of the dynamic common factor to include in each observation equation using the cross correlation function up to a max of 3
detect.growth	Logical, detect which variables to log
detect.diff	Logical, detect which variables to difference
diff.lag	Integer, number of lags to use for differencing
n_states	Number of states to include in the Markov switching model
ms_var,	Logical, T for Markow switching variance, default is F
prior	"estimate", "uninformative" or vector of named prior parameter guesses: DCF AR coefficients: phi1 and phi2 only; Error MA coefficients: psi_i1 to psi_i2 only for each series i; Error standard deviation: sigma_i only for each series i; Observation coefficient on DCF with first gamma (gamma_i, ..., gamma_n) only 1 index number, not i0 and any more gammas per equation: gamma_i1 to gamma_ik; Markov switching growth rate: mu_d and mu_u; Transition probabilities: p_dd, p_md (or p_mu), p_mm, p_md (or p_mu), p_uu, p_ud (or p_um)
formulas	R formula describing the relationship between each data series, the unobserved dynamic common factor, and the erorr structure
weighted	Logical, use weighted maximum likelihood. Weights are the rescaled inverse of the determinant of the forecast error covariance matrix for each observation.
optim_methods	Vector of 1 to 3 optimization methods in order of preference ("NR", "BFGS", "CG", "BHHH", or "SANN")
maxit	Maximum number of iterations for the optimization

Value

List of estimated values including coefficients, convergence code, the panel and time ids, the variables in the data, the variable that were logged, and the variables that were differenced

Author(s)

Alex Hubbard ([e-mail])

Examples

ms_dcf_estim(y = DT[, c("date", "y")])

opendoor-labs/MarkovSwitchingDCF documentation built on Jan. 8, 2020, 12:24 p.m.