The sufficient forecasting (SF) method is implemented by this package for a single time series forecasting using many predictors and a possibly nonlinear forecasting function. Assuming that the predictors are driven by some latent factors, the SF first conducts factor analysis and then performs sufficient dimension reduction on the estimated factors to derive predictive indices for forecasting. The package implements several dimension reduction approaches, including principal components (PC), sliced inverse regression (SIR), and directional regression (DR). Methods for dimension reduction are as described in: Fan, J., Xue, L. and Yao, J. (2017)
The goal of sufficientForecasting is to forecast a single time series when there is a large number of predictors and a possible nonlinear effect.
You can install the development version of sufficientForecasting like so:
# The easiest way to install sufficientForecasting
install.packages("sufficientForecasting")
# OR
devtools::install_github("JingFu1224/sufficientForecasting")
The following example uses SF.CI to solve a problem: forecast a single time series, and its upper bound and lower bound
library(sufficientForecasting)
## basic example code
SF.CI(y=dataExample$y,X=dataExample$X,newX=dataExample$newX,type="LLM",alpha = 0.05)
#> yhat ci_lower ci_upper
#> -0.3568 -2.4740 1.6076