Extreme Quantile Regression Neural Networks for Risk Forecasting
Check if Torch Backend Libraries are Installed
Default batch size (internal)
Check directory existence
Generalized Pareto likelihood loss of a EQRN_iid predictor
Generalized Pareto likelihood loss of a EQRN_seq predictor
Performs a learning rate decay step on an optimizer
Default torch device
End the currently set doFuture strategy
Ensure Torch Backend Libraries are Installed
Tail excess probability prediction using an EQRN_seq object
Tail excess probability prediction using an EQRN_iid object
Wrapper for fitting EQRN with restart for stability
EQRN fit function for sequential and time series data
EQRN fit function for independent data
Load an EQRN object from disc
Internal predict function for an EQRN_seq fitted object
Internal predict function for an EQRN_iid
GPD parameters prediction function for an EQRN_seq fitted object
GPD parameters prediction function for an EQRN_iid fitted object
Predict function for an EQRN_seq fitted object
Predict function for an EQRN_iid fitted object
Save an EQRN object on disc
EQRN: Extreme Quantile Regression Neural Networks for Risk Forecasting
Tail excess probability prediction method using an EQRN_iid object
Tail excess probability prediction method using an EQRN_iid object
Excess Probability Predictions
MLP module for GPD parameter prediction
Self-normalized fully-connected network module for GPD parameter predi...
Maximum likelihood estimates for the GPD distribution using peaks over...
(INTERNAL) Corrects a dimension simplification bug from the torch pack...
Get doFuture operator
Computes rescaled excesses over the conditional quantiles
Tail excess probability prediction based on conditional GPD parameters
Compute extreme quantile from GPD parameters
Install Torch Backend Libraries
Instantiates the default networks for training a EQRN_iid model
Covariate lagged replication for temporal dependence
Last element of a vector
Internal renaming function for back-compatibility
Convert a list to a matrix
GPD tensor loss function for training a EQRN network
Generalized Pareto likelihood loss
Create cross-validation folds
Mean absolute error
Mean squared error
Dataset creator for sequential data
Multilevel 'quantile_exceedance_proba_error'
Multilevel quantile MAEs
Multilevel quantile MSEs
Multilevel prediction bias
Multilevel 'proportion_below'
Multilevel quantile losses
Multilevel 'quantile_prediction_error'
Multilevel R squared
Multilevel residual variance
Alpha-dropout module
Dropout module
(DEPRECATED) On-Load Torch Backend Internal Install helper
Performs feature scaling without overfitting
Predict semi-conditional extreme quantiles using peaks over threshold
Predict unconditional extreme quantiles using peaks over threshold
Predict method for an EQRN_iid fitted object
Predict method for an EQRN_seq fitted object
Predict method for a QRN_seq fitted object
Prediction bias
Prediction residual variance
Feature processor for EQRN
Proportion of observations below conditional quantile vector
Wrapper for fitting a recurrent QRN with restart for stability
Recurrent QRN fitting function
Sigle-fold foldwise fit-predict function using a recurrent QRN
Foldwise fit-predict function using a recurrent QRN
Predict function for a QRN_seq fitted object
Recurrent quantile regression neural network module
Quantile exceedance probability prediction calibration error
Tensor quantile loss function for training a QRN network
Quantile loss
Quantile prediction calibration error
R squared
Recurrent network module for GPD parameter prediction
Mathematical number rounding
Safe RDS save
Semi-conditional GPD MLEs and their train-validation likelihoods
Self-normalized separated network module for GPD parameter prediction
Set a doFuture execution strategy
Instantiate an optimizer for training an EQRN_seq network
Instantiate an optimizer for training an EQRN_iid network
Square loss
Unconditional GPD MLEs and their train-validation likelihoods
Convert a vector to a matrix
Insert value in vector
This framework enables forecasting and extrapolating measures of conditional risk (e.g. of extreme or unprecedented events), including quantiles and exceedance probabilities, using extreme value statistics and flexible neural network architectures. It allows for capturing complex multivariate dependencies, including dependencies between observations, such as sequential dependence (time-series). The methodology was introduced in Pasche and Engelke (2024) <doi:10.1214/24-AOAS1907> (also available in preprint: Pasche and Engelke (2022) <doi:10.48550/arXiv.2208.07590>).
Useful links