varGuidTS

Variance-Guided Time-Series Modeling for Temporal Risk Detection

varGuidTS fits a balanced panel ARX-GARCHX model with subject-specific intercepts, shared population coefficients, and joint $\ell_1$ penalization on the conditional mean and variance covariate vectors. The fitted object also exposes an exceedance-based risk score

$$ \hat\pi_{s,t}(c) = P!\left(Y_{s,t} > c ;\middle|; \mathcal{F}_{t-1}\right), $$

useful as a personalized, scale-normalized temporal risk score from high-frequency multimodal panel data such as wearable-sensor streams.

Installation

# from GitHub
install.packages("remotes")
remotes::install_github("zionwzz/variance-guided-risk-demo")

# or from a local source tree
remotes::install_local("path/to/varGuidTS")

Project URL: https://github.com/zionwzz/variance-guided-risk-demo

Quick start

library(varGuidTS)

set.seed(1)
sim <- simulate_scenario(scen = 2, S = 4, T = 60,
                         d_noise = 4, noise_kind = "ar1", seed = 1)
df  <- sim[, c("s", "t", "y",
               grep("^X|^Noise|^Xbin", names(sim), value = TRUE))]

fit <- lmvt(df, p = 1, q = 0, r = 1, s_ord = 1,
            lambda_beta = 0.05, lambda_gamma = 0.05,
            use_x_in_variance = TRUE, maxit = 8)
print(fit)

pred <- predict(fit, df, threshold = quantile(df$y, 0.9))
head(pred)

# Coefficient summary table
sm <- summary(fit, include_zero = FALSE, sort_by = "abs_estimate")
head(sm$table)

# Subject-specific thresholds for personalized exceedance probabilities
subject_cutoffs <- tapply(df$y, df$s, quantile, probs = 0.75, na.rm = TRUE)
pred_subject <- predict(fit, df, threshold = subject_cutoffs,
                        innov_g = TRUE, innov_t = TRUE, df_t = 8)
head(pred_subject)

The README quick start gives the main workflow; the simulation function can be used to generate built-in scenarios for examples, tests, and method demonstrations.

What the package provides

Function	Purpose
lmvt()	Fit penalized panel ARX–GARCHX
predict.lmvt()	Conditional mean, variance, and exceedance probability
summary.lmvt()	Model diagnostics and coefficient summary table
simulate_scenario()	Generate panels under the four canonical scenarios

Maintainer

Min Lu luminwin@gmail.com.

License

MIT.

Citation

Please cite the package repository or the CRAN package record when using varGuidTS.