<![CDATA[We examine climate variables in a pipeline for dynamic Value-at-Risk (VaR) of the SRI-KEHATI ESG index and the IHSG (Jakarta Composite Index). Using 1,499 daily observations (20142025), we combine the Least Absolute Shrinkage and Selection Operator (LASSO) over 50 candidate predictors, a quantile regression neural network (QRNN) for VaR at three quantile levels, and SHAP for attribution. LASSO removes all climate variables from both indices. For IHSG, the partial correlation of clean-energy returns falls from 0.121 to 0.0005 once domestic equity factors enter, consistent with domestic absorption, though constituent overlap cannot exclude a purely compositional explanation. A diagnostic quantile regression shows climate contributions rising 2.82-fold from VaR 90% to VaR 99%, but the pattern is flat under the QRNN itself (ratio 1.03). In the main backtest, QRNN has lower pinball loss than Historical Simulation and GARCH(1,1)-t in all 12 DieboldMariano comparisons (p 0.001), although it uses same-day predictors the benchmarks lack and the advantage was not reproduced in the ablation. In a common-sample ablation (n = 249 of 300 nominal test days), LASSO pre-selection is associated with 4 of 6 Kupiec passes versus 0 of 6 without it, but IHSG still fails at two of three quantile levels. Whether SRI-KEHATI screening isolates climate risk was not tested; mediation covers IHSG only.]]>
Copyrights © 2026
