model: Direct Model
In nelson16silva/wavdrcast: Wavelet-Based Direct Multistep Forecast

Description Usage Arguments Details Value See Also Examples

This function estimates a direct least square horizon-specific model.

model(
  df,
  lags,
  .h,
  ic = BIC,
  .var = "all",
  .diff = FALSE,
  wav = FALSE,
  xreg = NULL,
  ...
)

`df`	A data frame. The first column of the data frame must be the base variable for constructing the dependent variable, the multiperiod ahead value to be forecasted, x_{t + h}. If `wav = TRUE`, `ncol` of `df` can be equal to one (just the dependent variable), otherwise, `ncol(df)` must be greater than one, even if `.var = "ar"`.
`lags`	An integer vector defining the lags of the regressors. If `wav = FALSE`, the length of the vector `lags` have to be the same as the number of columns in `df`. However, if `wav = TRUE`, an additional element in `lags` must be add such that the last element in the vector `lags` is the lag associatd with the wavelet regressor.
`.h`	h step-ahead.
`ic`	Information criterion, `BIC` or `AIC`. When searching for the best model the dataset is adjusted so that every model have the same data length for appropriate comparasion.
`.var`	A string to determine how the model will be specificated: `"all"` (default), `"ar"` or `"ar_out"`.
`.diff`	Logical `FALSE` or `TRUE`. If `TRUE`, the dependent variable is differentiated. See `wavdrcast-package` for the implication on the model specification and wavelet variable.
`wav`	Logical. If `TRUE`, a wavelet-based signal is add to `df`, where the signal is related to the explained variable, the first column of `df`.
`xreg`	Data frame. Exogeunous variable not subjected to be lagged. The number of rows must be the same as in `df`.
`...`	Further arguments passed to `wavsigmap::signal`.

General OLS regression model is writting as:

x_{t + h} = α + \mathbf{{β}}'\mathbf{x}_t + \mathbf{{γ}}'\mathbf{z}_t + \mathbf{{θ}}'\mathbf{d}_t + ε_{t + h},

where variables are explained in wavdrcast-package details section. Arguments .var, .diff and wav allow one to estimate subsets of this general model, as discussed in wavdrcast-package. If .var = "all", the general model is estimated (unless xreg = NULL, so that \mathbf{d} is out). If .var = "ar", \mathbf{z} terms are excluded (not \mathbf{d}, if included). Finally, if .var = "ar_out", the model is estimated without \mathbf{x} regressors.

A list with the following elements:

xreg: A tibble with the data regressors used to run the model choosed by the information criterion. This can be a different data set from that one returned by the lm object.
model: An object of class lm.

fcast

model(df = inf, lags = c(2, 0), .h = 2, .var = "ar")
model(df = inf, lags = c(2, 2), .h = 2)
model(df = inf[1], lags = c(2, 2), .h = 2, wav = TRUE)
model(df = gdp, lags = c(2, 2), .h = 2, .diff = TRUE)
model(df = gdp[1], lags = c(2, 2), .h = 2, .diff = TRUE, wav = TRUE)
model(df = gdp[1], lags = c(0, 0), .h = 2, .diff = TRUE, wav = TRUE, .var = "ar_out")