Skip to content

R package for evaluating linear forecasting models.

License

Notifications You must be signed in to change notification settings

nelson-n/lmForc

Repository files navigation

lmForc

CRAN Version CRAN Posit Mirror Downloads R-CMD-check Lifecycle: stable

The R package lmForc introduces functions for testing forecasting models and a new class for working with forecast data: Forecast. Test models out-of-sample by conditioning on realized values, vintage forecasts, or lagged values. Benchmark against AR models, historical average forecasts, or random walk forecasts. Create performance weighted or states weighted forecast combinations. These functions are built around the Forecast class and support both linear forecasting models and more complex models such as logistic regression, tree based models, and neural networks.

Vignette

For an overview of the lmForc package, please read the vignette: lmForc Vignette

Paper

Accompanying Paper: lmForc: Linear Model Forecasting in R

Abstract: Linear forecasting models are a popular option in many domains due to their simplicity and interpretability. This paper considers the tools a forecaster has for evaluating linear forecasting models and presents lmForc, a package which implements these evaluation functions. Functions in the lmForc package are built around a new S4 class, Forecast, which introduces a dedicated data structure for storing forecasts to the R language and permits the creation of complex forecasting functions. The lmForc package is designed to leverage the simplicity and interpretability of linear models so that a forecaster may easily test model specifications and understand precisely how a model arrives at a forecast.

Installation

To install the stable version from CRAN:

install.packages("lmForc")

To install the development version from GitHub:

# install.packages("remotes")
remotes::install_github("nelson-n/lmForc")

Linear Forecasting Model Example

Produce an out-of-sample forecast conditioned on realized values. Calculates linear model coefficients in each period based on information that would have been available to the forecaster. Coefficients are combined with future realized values to compute a conditional forecast. Evaluates the performance of a linear model had it been conditioned on perfect information.

library(lmForc)

# Stylized dataset.
date <- as.Date(c("2010-03-31", "2010-06-30", "2010-09-30", "2010-12-31",
                  "2011-03-31", "2011-06-30", "2011-09-30", "2011-12-31", 
                  "2012-03-31", "2012-06-30"))
y    <- c(1.09, 1.71, 1.09, 2.46, 1.78, 1.35, 2.89, 2.11, 2.97, 0.99)
x1   <- c(4.22, 3.86, 4.27, 5.60, 5.11, 4.31, 4.92, 5.80, 6.30, 4.17)
x2   <- c(10.03, 10.49, 10.85, 10.47, 9.09, 10.91, 8.68, 9.91, 7.87, 6.63)
data <- data.frame(date, y, x1, x2)

# Linear model out-of-sample forecast.
forecast1 <- oos_realized_forc(
  lm_call = lm(y ~ x1 + x2, data),
  h_ahead = 2L,
  estimation_end = as.Date("2011-03-31"),
  time_vec = data$date
)

forecast1
#> h_ahead = 2 
#> 
#>       origin     future forecast realized
#> 1 2011-03-31 2011-09-30 1.623750     2.89
#> 2 2011-06-30 2011-12-31 2.341664     2.11
#> 3 2011-09-30 2012-03-31 3.415198     2.97
#> 4 2011-12-31 2012-06-30 2.708308     0.99

Produce an out-of-sample forecast based on the historical median. In each period the historical median of the series is calculated based on information that would have been available to the forecaster. Replicates the historical median forecast that would have been produced in real-time and serves as a benchmark for other models.

# Historical Median Forecast
forecast2 <- historical_average_forc(
  avg_function = "median",
  realized_vec = data$y,
  h_ahead = 2L,
  estimation_end = as.Date("2011-03-31"),
  time_vec = data$date
)

forecast2
#> h_ahead = 2 
#> 
#>       origin     future forecast realized
#> 1 2011-03-31 2011-09-30    1.710     2.89
#> 2 2011-06-30 2011-12-31    1.530     2.11
#> 3 2011-09-30 2012-03-31    1.710     2.97
#> 4 2011-12-31 2012-06-30    1.745     0.99

Compare the performance of both models by calculating RMSE forecast error.

rmse(forecast1)
#> [1] 1.09634
rmse(forecast2)
#> [1] 0.9857009

Create a performance weighted forecast combination of forecast1 and forecast2. In each period forecast accuracy is calculated over recent periods and each model is given a weight based on recent accuracy. The forecast for the next period is calculated as a weighted combination of both forecasts.

performance_weighted_forc(
  forecast1, forecast2,
  eval_window = 1L,
  errors = "mse",
  return_weights = FALSE
)
#> h_ahead = 2 
#> 
#>       origin     future forecast realized
#> 1 2011-03-31 2011-09-30       NA     2.89
#> 2 2011-06-30 2011-12-31       NA     2.11
#> 3 2011-09-30 2012-03-31 2.502552     2.97
#> 4 2011-12-31 2012-06-30 2.575770     0.99

General Forecasting Model Example

Produces an out-of-sample forecast conditioned on realized values similar to the example above, but does so using a logistic regression. Functions with _general in the name are designed to work with any model that can be estimated in R, including tree based models, neural networks, or models with hand-built parameters. The user only needs to specify a model_function that estimates the model and the prediction_function which produces predictions from the model.

# Stylized Dataset.
date <- as.Date(c("2010-03-31", "2010-06-30", "2010-09-30", "2010-12-31",
                  "2011-03-31", "2011-06-30", "2011-09-30", "2011-12-31", 
                  "2012-03-31", "2012-06-30", "2012-09-30", "2012-12-31",
                  "2013-03-31", "2013-06-30", "2013-09-30", "2013-12-31"))
y  <- c(1, 0, 0, 0, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 1, 0)
x1 <- c(8.22, 3.86, 4.27, 3.37, 5.88, 3.34, 2.92, 1.80, 3.30, 7.17, 3.22, 3.86, 4.27, 3.37, 5.88, 3.34)
x2 <- c(4.03, 2.46, 2.04, 2.44, 6.09, 2.91, 1.68, 2.91, 3.87, 1.63, 4.03, 2.46, 2.04, 2.44, 6.09, 2.91)
dataLogit <- data.frame(date, y, x1, x2)

# Logit model out-of-sample forecast.
forecast3 <- oos_realized_forc_general(
    model_function = function(data) {glm(y ~ x1 + x2, data = data, family = binomial)},
    prediction_function = function(model_function, data) {as.vector(predict(model_function, data, type = "response"))}, 
    data = dataLogit,
    realized = dataLogit$y,
    h_ahead = 2L,
    estimation_end = as.Date("2012-06-30"),
    time_vec = dataLogit$date,
    estimation_window = NULL
)

forecast3
#> h_ahead = 2 
#> 
#>       origin     future   forecast realized
#> 1 2012-06-30 2012-12-31 0.20301888        0
#> 2 2012-09-30 2013-03-31 0.24452583        0
#> 3 2012-12-31 2013-06-30 0.07931267        1
#> 4 2013-03-31 2013-09-30 0.98707714        1
#> 5 2013-06-30 2013-12-31 0.18387762        0