SLIM-LQVAR: Post-LASSO Shrinkage for Quantile Connectedness and Tail-Risk Transmission in Cryptocurrency Markets
This study introduces Shrinkage of Lagged Influences in Multivariate LASSO-Quantile Vector Autoregression (SLIM-LQVAR), evaluated as SLIM-LQVAR10: a fixed 10% post-LASSO contraction of quantile vector autoregressive lag coefficients, for generalized variance-decomposition connectedness among BTC, ETH, BNB, XRP and ADA. An archived panel of Binance spot daily returns covers 2018–2026 (2,479 estimation and 620 holdout observations). The analysis reports q05/q50/q95 network estimates and sensitivity to covariance, horizon, lag order and training period. A 100-seed-per-quantile Monte Carlo study spanning five known-oracle DGP families and seven estimators finds a 4.73% q05 Edge MAE reduction relative to LASSO-QVAR; the DGP-stratified paired 95% interval for the absolute reduction excludes zero. Mean Edge MAE is 0.35% and 0.40% higher at q50 and q95, respectively, with intervals covering zero. The nonlinear heteroskedastic upper-tail DGP shows pronounced deterioration. A 100-replicate joint coefficient–covariance moving-block bootstrap provides conditional intervals for TCI, FROM, TO, NET and directed FEVD edges. Holdout quantile coverage is imperfect (q05: 1.97%; q95: 97.65%), and validation-only intercept recalibration does not resolve all asset-level failures. SLIM-LQVAR10 demonstrates statistically supported improvements in lower-tail network recovery relative to LASSO-QVAR, while maintaining competitive forecasting performance across quantiles. However, its network-recovery advantage does not generalize uniformly to the median and upper-tail states. The network estimates are noncausal, and the original tail forecasts are not validated for stand-alone Value-at-Risk use.
Authors
- Tanattrin Bunnag (ORCID: https://orcid.org/0000-0002-3395-5597)
Publication Details
- Journal
- Preprints.org
- Published
- 2026-11-13
- DOI
- https://doi.org/10.20944/preprints202611.1377.v1
- Primary Topic
- Financial Risk and Volatility Modeling
- Type
- preprint