WP46: A-05 multi-start + GARCH(1,1) likelihood floor (degenerate optima), first-pass analysis archived
15 changed files +19,995 −0
added
paper/figures/fig_heatmap_h1_tierA.pdf
+0 −0
Binary file not shown.
added
paper/figures/fig_heatmap_h1_tierB.pdf
+0 −0
Binary file not shown.
added
paper/figures/fig_heatmap_h22_tierA.pdf
+0 −0
Binary file not shown.
added
paper/figures/fig_horizon.pdf
+0 −0
Binary file not shown.
added
paper/figures/fig_mcs_h1_tierA.pdf
+0 −0
Binary file not shown.
added
paper/figures/fig_mcs_h1_tierB.pdf
+0 −0
Binary file not shown.
added
paper/figures/fig_mcs_h22_tierA.pdf
+0 −0
Binary file not shown.
added
paper/tables/tab_accounting.tex
+84 −0
@@ -0,0 +1,84 @@ | ||
| 1 | +\begin{tabular}{lrrrrr} | |
| 2 | +\toprule | |
| 3 | +Model & Failed (\%) & Inadmissible (\%) & Coverage (\%) & Median refit (s) & Iterations \\ | |
| 4 | +\midrule | |
| 5 | +EGARCH(1,1)-GED & 0.2 & 0.2 & 99.8 & 0.018 & 20 \\ | |
| 6 | +EGARCH(1,1)-N & 0.1 & 0.4 & 99.9 & 0.008 & 17 \\ | |
| 7 | +EGARCH(1,1)-skt & 0.2 & 0.5 & 98.8 & 0.030 & 22 \\ | |
| 8 | +EGARCH(1,1)-t & 0.1 & 0.5 & 98.7 & 0.019 & 22 \\ | |
| 9 | +GJR-GARCH(1,1)-GED & 0.0 & 1.8 & 100.0 & 0.014 & 17 \\ | |
| 10 | +GJR-GARCH(1,1)-N & 0.0 & 2.2 & 100.0 & 0.008 & 13 \\ | |
| 11 | +GJR-GARCH(1,1)-skt & 0.0 & 3.5 & 100.0 & 0.023 & 18 \\ | |
| 12 | +GJR-GARCH(1,1)-t & 0.0 & 3.4 & 100.0 & 0.015 & 16 \\ | |
| 13 | +NAGARCH(1,1)-GED & 0.0 & 0.3 & 100.0 & 0.023 & 60 \\ | |
| 14 | +NAGARCH(1,1)-N & 0.1 & 0.2 & 100.0 & 0.011 & 51 \\ | |
| 15 | +NAGARCH(1,1)-skt & 1.2 & 0.2 & 100.0 & 0.072 & 72 \\ | |
| 16 | +NAGARCH(1,1)-t & 0.0 & 0.2 & 100.0 & 0.019 & 60 \\ | |
| 17 | +QGARCH(1,1)-GED & 0.0 & 0.1 & 100.0 & 0.025 & 71 \\ | |
| 18 | +QGARCH(1,1)-N & 0.0 & 0.1 & 100.0 & 0.012 & 58 \\ | |
| 19 | +QGARCH(1,1)-skt & 0.3 & 0.3 & 100.0 & 0.059 & 88 \\ | |
| 20 | +QGARCH(1,1)-t & 0.0 & 0.2 & 100.0 & 0.021 & 72 \\ | |
| 21 | +TGARCH(1,1)-GED & 0.0 & 0.0 & 100.0 & 0.021 & 18 \\ | |
| 22 | +TGARCH(1,1)-N & 0.0 & 0.0 & 100.0 & 0.010 & 16 \\ | |
| 23 | +TGARCH(1,1)-skt & 0.0 & 0.0 & 100.0 & 0.034 & 19 \\ | |
| 24 | +TGARCH(1,1)-t & 0.0 & 0.0 & 100.0 & 0.021 & 18 \\ | |
| 25 | +Constant variance-N & 0.0 & 0.0 & 100.0 & 0.000 & 0 \\ | |
| 26 | +EWMA-N & 0.0 & 0.0 & 100.0 & 0.001 & 3 \\ | |
| 27 | +Historical variance, 22 days & 0.0 & 0.0 & 100.0 & 0.000 & 0 \\ | |
| 28 | +Historical variance, 252 days & 0.0 & 0.0 & 100.0 & 0.000 & 0 \\ | |
| 29 | +Historical variance, 66 days & 0.0 & 0.0 & 100.0 & 0.000 & 0 \\ | |
| 30 | +RiskMetrics 2006-N & 0.0 & 0.0 & 100.0 & 0.003 & 3 \\ | |
| 31 | +ARCH(5)-GED & 0.0 & 1.3 & 100.0 & 0.018 & 16 \\ | |
| 32 | +ARCH(5)-N & 0.0 & 1.2 & 100.0 & 0.011 & 14 \\ | |
| 33 | +ARCH(5)-skt & 0.0 & 2.9 & 100.0 & 0.025 & 17 \\ | |
| 34 | +ARCH(5)-t & 0.0 & 3.5 & 100.0 & 0.018 & 16 \\ | |
| 35 | +GARCH(1,1)-GED & 0.0 & 1.8 & 100.0 & 0.011 & 15 \\ | |
| 36 | +GARCH(1,1)-N & 0.0 & 1.8 & 100.0 & 0.006 & 11 \\ | |
| 37 | +GARCH(1,1)-skt & 0.0 & 4.8 & 100.0 & 0.016 & 16 \\ | |
| 38 | +GARCH(1,1)-t & 0.0 & 6.2 & 100.0 & 0.011 & 14 \\ | |
| 39 | +GARCH(1,2)-GED & 0.0 & 2.0 & 100.0 & 0.015 & 16 \\ | |
| 40 | +GARCH(1,2)-N & 0.0 & 2.1 & 100.0 & 0.008 & 13 \\ | |
| 41 | +GARCH(1,2)-skt & 0.0 & 5.3 & 100.0 & 0.021 & 18 \\ | |
| 42 | +GARCH(1,2)-t & 0.0 & 6.9 & 100.0 & 0.015 & 16 \\ | |
| 43 | +GARCH(2,1)-GED & 0.0 & 2.0 & 100.0 & 0.014 & 16 \\ | |
| 44 | +GARCH(2,1)-N & 0.0 & 2.0 & 100.0 & 0.008 & 13 \\ | |
| 45 | +GARCH(2,1)-skt & 0.0 & 5.2 & 100.0 & 0.020 & 17 \\ | |
| 46 | +GARCH(2,1)-t & 0.0 & 6.4 & 100.0 & 0.014 & 16 \\ | |
| 47 | +GARCH-M(1,1)-GED & 0.0 & 0.0 & 100.0 & 0.023 & 69 \\ | |
| 48 | +GARCH-M(1,1)-N & 0.0 & 0.0 & 100.0 & 0.011 & 53 \\ | |
| 49 | +GARCH-M(1,1)-skt & 0.2 & 0.1 & 100.0 & 0.036 & 81 \\ | |
| 50 | +GARCH-M(1,1)-t & 0.0 & 0.1 & 100.0 & 0.024 & 71 \\ | |
| 51 | +IGARCH(1,1)-GED & 0.0 & 0.0 & 100.0 & 0.009 & 41 \\ | |
| 52 | +IGARCH(1,1)-N & 0.0 & 0.0 & 100.0 & 0.004 & 29 \\ | |
| 53 | +IGARCH(1,1)-skt & 0.4 & 0.0 & 100.0 & 0.039 & 46 \\ | |
| 54 | +IGARCH(1,1)-t & 0.0 & 0.0 & 100.0 & 0.007 & 39 \\ | |
| 55 | +Component GARCH-GED & 0.0 & 2.4 & 100.0 & 0.049 & 111 \\ | |
| 56 | +Component GARCH-N & 0.0 & 2.6 & 100.0 & 0.030 & 93 \\ | |
| 57 | +Component GARCH-skt & 0.2 & 2.7 & 100.0 & 0.091 & 138 \\ | |
| 58 | +Component GARCH-t & 0.0 & 3.0 & 100.0 & 0.041 & 112 \\ | |
| 59 | +FIGARCH(1,d,1)-GED & 0.0 & 0.0 & 100.0 & 0.172 & 16 \\ | |
| 60 | +FIGARCH(1,d,1)-N & 0.0 & 0.0 & 100.0 & 0.125 & 13 \\ | |
| 61 | +FIGARCH(1,d,1)-skt & 0.0 & 0.0 & 100.0 & 0.187 & 17 \\ | |
| 62 | +FIGARCH(1,d,1)-t & 0.0 & 0.0 & 100.0 & 0.153 & 16 \\ | |
| 63 | +APARCH(1,1)-GED & 0.0 & 0.0 & 100.0 & 0.086 & 29 \\ | |
| 64 | +APARCH(1,1)-N & 0.1 & 0.0 & 100.0 & 0.049 & 29 \\ | |
| 65 | +APARCH(1,1)-skt & 0.0 & 0.0 & 100.0 & 0.102 & 30 \\ | |
| 66 | +APARCH(1,1)-t & 0.0 & 0.0 & 100.0 & 0.081 & 29 \\ | |
| 67 | +AVGARCH(1,1)-GED & 0.0 & 0.0 & 100.0 & 0.015 & 16 \\ | |
| 68 | +AVGARCH(1,1)-N & 0.0 & 0.0 & 100.0 & 0.007 & 14 \\ | |
| 69 | +AVGARCH(1,1)-skt & 0.0 & 0.0 & 100.0 & 0.027 & 17 \\ | |
| 70 | +AVGARCH(1,1)-t & 0.0 & 0.0 & 100.0 & 0.015 & 16 \\ | |
| 71 | +GARCH-X(1,1) with lagged realized measure-N & 0.0 & 0.6 & 100.0 & 0.056 & 127 \\ | |
| 72 | +GARCH-X(1,1) with lagged realized measure-t & 0.0 & 1.4 & 100.0 & 0.082 & 157 \\ | |
| 73 | +HAR-RV & 0.0 & 0.0 & 100.0 & 0.001 & 0 \\ | |
| 74 | +HEAVY(1,1)-N & 0.0 & 0.1 & 100.0 & 0.030 & 84 \\ | |
| 75 | +HEAVY(1,1)-t & 0.0 & 0.0 & 100.0 & 0.049 & 103 \\ | |
| 76 | +log-HAR-RV & 0.0 & 0.0 & 100.0 & 0.001 & 0 \\ | |
| 77 | +Realized GARCH(1,1), log-linear-N & 0.0 & 0.1 & 100.0 & 0.042 & 92 \\ | |
| 78 | +Realized GARCH(1,1), log-linear-t & 0.0 & 0.0 & 100.0 & 0.074 & 147 \\ | |
| 79 | +Realized-measure random walk & 0.0 & 0.0 & 100.0 & 0.000 & 0 \\ | |
| 80 | +MS-GARCH(2)-N & 0.2 & 0.2 & 99.9 & 0.115 & 192 \\ | |
| 81 | +MS-GARCH(2)-t & 0.4 & 0.1 & 100.0 & 0.162 & 221 \\ | |
| 82 | +GAS-t(1,1) / Beta-t-GARCH-t & 0.0 & 0.0 & 100.0 & 0.014 & 49 \\ | |
| 83 | +\bottomrule | |
| 84 | +\end{tabular} | |
added
paper/tables/tab_crosscheck.tex
+20 −0
@@ -0,0 +1,20 @@ | ||
| 1 | +\begin{tabular}{llrrrl} | |
| 2 | +\toprule | |
| 3 | +Model & R reference & R log-lik. & Ours & Diff. & Note \\ | |
| 4 | +\midrule | |
| 5 | +GARCH(1,1)-$t$ & rugarch sGARCH & -3379.50 & -3379.46 & 0.041 & \\ | |
| 6 | +GARCH(1,1)-skew-$t$ & rugarch sGARCH/sstd & -3379.31 & -3379.28 & 0.039 & \\ | |
| 7 | +GJR(1,1)-$t$ & rugarch gjrGARCH & -3372.99 & -3372.95 & 0.037 & \\ | |
| 8 | +TGARCH(1,1)-$t$ & rugarch fGARCH/TGARCH & -3378.27 & -3377.91 & 0.357 & \\ | |
| 9 | +EGARCH(1,1)-$t$ & rugarch eGARCH & -3376.68 & -3376.67 & 0.013 & \\ | |
| 10 | +APARCH(1,1)-$t$ & rugarch apARCH & -3372.85 & -3372.76 & 0.090 & \\ | |
| 11 | +FIGARCH(1,$d$,1)-$t$ & rugarch fiGARCH & -3381.98 & -3383.45 & -1.471 & \\ | |
| 12 | +IGARCH(1,1)-$t$ & rugarch iGARCH & -3385.52 & -3385.47 & 0.050 & \\ | |
| 13 | +Component GARCH-$t$ & rugarch csGARCH & -3379.76 & -3379.46 & 0.300 & \\ | |
| 14 | +NAGARCH(1,1)-$t$ & rugarch fGARCH/NAGARCH & -3373.35 & -3373.31 & 0.038 & \\ | |
| 15 | +Realized GARCH(1,1)-N & rugarch realGARCH & -- & -4989.21 & -- & R fit did not converge \\ | |
| 16 | +GARCH-X(1,1)-N (return eq.) & rugarch sGARCH + vxreg & -3439.50 & -3438.59 & -- & per-observation comparison (R drops one obs.) \\ | |
| 17 | +MS-GARCH(2)-N & MSGARCH FitML & -3379.75 & -3380.15 & -0.397 & 22-step variance forecasts within 7.2\% \\ | |
| 18 | +GAS-$t$(1,1) & GAS (scale parameterization; reference) & -- & -3382.32 & -- & different parameterization, not comparable \\ | |
| 19 | +\bottomrule | |
| 20 | +\end{tabular} | |
added
paper/tables/tab_pooled_tierA.tex
+47 −0
@@ -0,0 +1,47 @@ | ||
| 1 | +\begin{tabular}{lccccccccc} | |
| 2 | +\toprule | |
| 3 | + & \multicolumn{3}{c}{$h=1$} & \multicolumn{3}{c}{$h=5$} & \multicolumn{3}{c}{$h=22$} \\ | |
| 4 | +\cmidrule(lr){2-4}\cmidrule(lr){5-7}\cmidrule(lr){8-10} | |
| 5 | +Model & ratio & 90\% CI & W/L/N & ratio & 90\% CI & W/L/N & ratio & 90\% CI & W/L/N \\ | |
| 6 | +\midrule | |
| 7 | +APARCH(1,1)-N & 1.142*** & [1.10, 1.19] & 0/3/42 & 1.180*** & [1.12, 1.24] & 0/1/42 & 1.166*** & [1.10, 1.23] & 0/1/42 \\ | |
| 8 | +APARCH(1,1)-t & 1.489*** & [1.32, 1.61] & 0/1/42 & 1.847*** & [1.51, 2.09] & 0/0/42 & 2.087*** & [1.58, 2.39] & 0/0/42 \\ | |
| 9 | +ARCH(5)-N & 1.114*** & [1.09, 1.14] & 0/23/42 & 1.150*** & [1.11, 1.19] & 0/19/42 & 1.199*** & [1.14, 1.26] & 0/10/42 \\ | |
| 10 | +ARCH(5)-t & 1.104*** & [1.07, 1.13] & 1/22/42 & 1.147*** & [1.10, 1.19] & 1/16/42 & 1.208*** & [1.12, 1.27] & 1/8/42 \\ | |
| 11 | +AVGARCH(1,1)-N & 1.001 & [0.99, 1.01] & 3/0/42 & 0.999 & [0.98, 1.03] & 2/0/42 & 0.981 & [0.96, 1.00] & 0/1/42 \\ | |
| 12 | +AVGARCH(1,1)-t & 0.999 & [0.98, 1.01] & 1/0/42 & 0.987 & [0.97, 1.01] & 1/0/42 & 0.964*** & [0.95, 0.99] & 1/0/42 \\ | |
| 13 | +Component GARCH-N & 0.988** & [0.98, 1.00] & 0/1/42 & 1.003 & [0.99, 1.01] & 0/0/42 & 1.027** & [1.01, 1.05] & 0/0/42 \\ | |
| 14 | +Component GARCH-t & 0.980*** & [0.97, 0.99] & 1/1/42 & 0.992 & [0.98, 1.01] & 1/1/42 & 1.019 & [0.99, 1.04] & 1/0/42 \\ | |
| 15 | +Constant variance-N & 2.056** & [1.58, 2.77] & 0/27/42 & 2.203** & [1.62, 3.08] & 0/19/42 & 1.575*** & [1.33, 1.97] & 0/6/42 \\ | |
| 16 | +EGARCH(1,1)-N & 1226951.881*** & [619178.80, 1776382.65] & 0/7/42 & 8125780.064*** & [4079951.78, 11624359.68] & 0/7/42 & 32613850.182*** & [14831429.01, 49384651.31] & 0/0/42 \\ | |
| 17 | +EGARCH(1,1)-t & 875909.810** & [273220.93, 1155773.68] & 0/0/42 & 5774588.346** & [1742986.85, 7959122.87] & 0/2/42 & 23019331.736** & [6500505.52, 31614311.59] & 0/2/42 \\ | |
| 18 | +EWMA-N & 1.046*** & [1.02, 1.07] & 1/8/42 & 1.064** & [1.01, 1.10] & 1/5/42 & 1.144** & [1.05, 1.22] & 0/1/42 \\ | |
| 19 | +FIGARCH(1,d,1)-N & 0.977*** & [0.97, 0.99] & 4/0/42 & 0.959** & [0.93, 0.98] & 5/0/42 & 0.972 & [0.93, 1.01] & 0/1/42 \\ | |
| 20 | +FIGARCH(1,d,1)-t & 0.973*** & [0.96, 0.98] & 2/0/42 & 0.950** & [0.92, 0.97] & 3/0/42 & 0.966* & [0.92, 1.00] & 1/2/42 \\ | |
| 21 | +GARCH(1,1)-N & 1.000 & [--, --] & 0/0/42 & 1.000 & [--, --] & 0/0/42 & 1.000 & [--, --] & 0/0/42 \\ | |
| 22 | +GARCH(1,1)-t & 0.992*** & [0.98, 1.00] & 1/2/42 & 0.984** & [0.97, 0.99] & 1/1/42 & 0.977* & [0.95, 0.99] & 1/1/42 \\ | |
| 23 | +GARCH(1,2)-N & 1.010* & [1.00, 1.02] & 0/0/42 & 1.008 & [1.00, 1.02] & 0/0/42 & 1.009 & [1.00, 1.02] & 1/0/42 \\ | |
| 24 | +GARCH(1,2)-t & 0.994* & [0.99, 1.00] & 1/2/42 & 0.987*** & [0.97, 0.99] & 1/0/42 & 0.981** & [0.96, 0.99] & 1/1/42 \\ | |
| 25 | +GARCH(2,1)-N & 1.013*** & [1.01, 1.02] & 0/2/42 & 1.003 & [1.00, 1.01] & 0/0/42 & 1.006* & [1.00, 1.01] & 0/0/42 \\ | |
| 26 | +GARCH(2,1)-t & 1.000 & [0.99, 1.01] & 1/3/42 & 0.984** & [0.97, 0.99] & 1/1/42 & 0.981* & [0.95, 1.00] & 1/1/42 \\ | |
| 27 | +GARCH-M(1,1)-N & 1.002 & [1.00, 1.00] & 5/2/42 & 1.002 & [1.00, 1.00] & 3/2/42 & 1.003 & [1.00, 1.01] & 0/0/42 \\ | |
| 28 | +GARCH-M(1,1)-t & 0.994* & [0.99, 1.00] & 1/2/42 & 0.986** & [0.97, 0.99] & 1/1/42 & 0.977* & [0.95, 0.99] & 1/1/42 \\ | |
| 29 | +GAS-t(1,1) / Beta-t-GARCH-t & 1.025 & [0.99, 1.06] & 2/3/42 & 1.008 & [0.97, 1.05] & 2/0/42 & 0.977 & [0.95, 1.00] & 1/0/42 \\ | |
| 30 | +GJR-GARCH(1,1)-N & 1.006 & [0.99, 1.02] & 1/0/42 & 1.003 & [0.98, 1.03] & 1/0/42 & 0.994 & [0.98, 1.01] & 1/0/42 \\ | |
| 31 | +GJR-GARCH(1,1)-t & 1.003 & [0.98, 1.02] & 1/1/42 & 0.994 & [0.97, 1.02] & 1/0/42 & 0.983 & [0.95, 1.00] & 1/1/42 \\ | |
| 32 | +Historical variance, 22 days & 1.270*** & [1.23, 1.30] & 0/37/42 & 1.380*** & [1.31, 1.43] & 0/28/42 & 1.485*** & [1.35, 1.56] & 0/20/42 \\ | |
| 33 | +Historical variance, 252 days & 1.667*** & [1.39, 2.06] & 1/41/42 & 1.768** & [1.41, 2.30] & 1/17/42 & 1.415** & [1.22, 1.69] & 1/3/42 \\ | |
| 34 | +Historical variance, 66 days & 1.323*** & [1.20, 1.49] & 1/30/42 & 1.392*** & [1.21, 1.64] & 0/16/42 & 1.383** & [1.14, 1.71] & 0/3/42 \\ | |
| 35 | +IGARCH(1,1)-N & 0.999 & [0.98, 1.01] & 1/0/42 & 0.988 & [0.94, 1.02] & 1/1/42 & 1.001 & [0.86, 1.10] & 0/6/42 \\ | |
| 36 | +IGARCH(1,1)-t & 1.006 & [0.99, 1.02] & 1/4/42 & 0.999 & [0.95, 1.03] & 1/3/42 & 1.032 & [0.88, 1.12] & 1/5/42 \\ | |
| 37 | +MS-GARCH(2)-N & 1.012* & [1.00, 1.02] & 1/2/42 & 1.022* & [1.00, 1.05] & 0/2/42 & 1.028 & [0.99, 1.08] & 0/0/42 \\ | |
| 38 | +MS-GARCH(2)-t & 1.009 & [0.99, 1.04] & 1/0/42 & 1.015 & [0.99, 1.05] & 1/0/42 & 1.033 & [0.99, 1.08] & 1/0/42 \\ | |
| 39 | +NAGARCH(1,1)-N & 1.055* & [1.01, 1.11] & 0/1/42 & 1.040 & [1.00, 1.09] & 0/0/42 & 0.987 & [0.95, 1.02] & 1/0/42 \\ | |
| 40 | +NAGARCH(1,1)-t & 1.033 & [1.00, 1.07] & 1/1/42 & 1.017 & [0.98, 1.05] & 1/1/42 & 0.984 & [0.94, 1.02] & 1/2/42 \\ | |
| 41 | +QGARCH(1,1)-N & 0.995 & [0.98, 1.01] & 3/1/42 & 0.987* & [0.97, 1.00] & 2/0/42 & 0.979 & [0.95, 1.00] & 0/0/42 \\ | |
| 42 | +QGARCH(1,1)-t & 0.988* & [0.97, 1.00] & 2/0/42 & 0.975** & [0.95, 0.99] & 2/0/42 & 0.967** & [0.93, 0.99] & 1/1/42 \\ | |
| 43 | +RiskMetrics 2006-N & 0.981 & [0.96, 1.01] & 4/4/42 & 0.955* & [0.91, 1.00] & 1/0/42 & 0.980 & [0.91, 1.04] & 1/0/42 \\ | |
| 44 | +TGARCH(1,1)-N & 0.996 & [0.98, 1.02] & 3/0/42 & 0.986 & [0.96, 1.01] & 2/0/42 & 0.957** & [0.93, 0.98] & 1/1/42 \\ | |
| 45 | +TGARCH(1,1)-t & 1.003 & [0.98, 1.03] & 1/0/42 & 0.988 & [0.96, 1.02] & 1/0/42 & 0.952*** & [0.92, 0.98] & 1/0/42 \\ | |
| 46 | +\bottomrule | |
| 47 | +\end{tabular} | |
added
paper/tables/tab_pooled_tierB.tex
+17 −0
@@ -0,0 +1,17 @@ | ||
| 1 | +\begin{tabular}{lccccccccc} | |
| 2 | +\toprule | |
| 3 | + & \multicolumn{3}{c}{$h=1$} & \multicolumn{3}{c}{$h=5$} & \multicolumn{3}{c}{$h=22$} \\ | |
| 4 | +\cmidrule(lr){2-4}\cmidrule(lr){5-7}\cmidrule(lr){8-10} | |
| 5 | +Model & ratio & 90\% CI & W/L/N & ratio & 90\% CI & W/L/N & ratio & 90\% CI & W/L/N \\ | |
| 6 | +\midrule | |
| 7 | +GARCH-X(1,1) with lagged realized measure-N & 0.870*** & [0.82, 0.91] & 28/0/42 & 0.865** & [0.76, 0.94] & 7/0/42 & 1.077 & [0.95, 1.18] & 0/5/42 \\ | |
| 8 | +GARCH-X(1,1) with lagged realized measure-t & 0.865*** & [0.82, 0.90] & 26/0/42 & 0.847** & [0.74, 0.92] & 8/1/42 & 1.030 & [0.89, 1.13] & 0/5/42 \\ | |
| 9 | +HAR-RV & 7243516.013 & [0.94, 19288893.66] & 4/0/42 & 1176553.963 & [0.93, 3490173.28] & 2/0/42 & 68650.504 & [1.00, 209115.99] & 1/1/42 \\ | |
| 10 | +HEAVY(1,1)-N & 0.873*** & [0.82, 0.91] & 20/0/42 & 0.866** & [0.76, 0.94] & 5/0/42 & 1.078 & [0.95, 1.18] & 0/5/42 \\ | |
| 11 | +HEAVY(1,1)-t & 0.867*** & [0.82, 0.90] & 23/0/42 & 0.847** & [0.74, 0.92] & 6/0/42 & 1.028 & [0.89, 1.13] & 0/4/42 \\ | |
| 12 | +log-HAR-RV & 0.867*** & [0.82, 0.90] & 31/2/42 & 0.845*** & [0.80, 0.87] & 20/1/42 & 0.930*** & [0.90, 0.96] & 4/1/42 \\ | |
| 13 | +Realized GARCH(1,1), log-linear-N & 0.889*** & [0.85, 0.91] & 28/2/42 & 0.869*** & [0.81, 0.91] & 19/2/42 & 0.950*** & [0.92, 0.97] & 2/1/42 \\ | |
| 14 | +Realized GARCH(1,1), log-linear-t & 0.909*** & [0.86, 0.93] & 27/2/42 & 3.311*** & [2.61, 3.00] & 3/20/42 & 7.062*** & [5.71, 6.37] & 2/28/42 \\ | |
| 15 | +Realized-measure random walk & 10638.864* & [922.65, 21612.19] & 0/18/42 & 16062.606* & [823.48, 30688.48] & 0/28/42 & 35019.573* & [1274.24, 52901.58] & 0/26/42 \\ | |
| 16 | +\bottomrule | |
| 17 | +\end{tabular} | |
modified
progress.md
+12 −0
@@ -54,3 +54,15 @@ All times ET. "Done" means executed with the listed command and artefact; "plann | ||
| 54 | 54 | single node's manifest, and `cpu_s` was the cumulative CPU of the worker process. Compute accounting in the paper therefore uses |
| 55 | 55 | the complete per-refit (`time_s`) and per-origin (`time_fc_s`) timings stored in every fits/forecast table (exact, 3,276 jobs); |
| 56 | 56 | `maxrss` is reported as the per-worker peak (≈ 0.4 GB). `run_benchmark.py` now records per-job CPU deltas in `_runs_<host>.jsonl`. |
| 57 | +- **A-05 (2026-11-09, after inspecting the first test-period analysis):** the first pass showed pooled QLIKE ratios of order 10^6 | |
| 58 | + for EGARCH and 3–7 for Realized GARCH-t, traced to converged-but-degenerate optima of arch's default fit (e.g. SI, origin | |
| 59 | + 2019-03-06: EGARCH-t alpha = −641, beta = 0, nu at its bound, log-likelihood −6294 versus −1613 from alternative starting | |
| 60 | + values and −1621 for rugarch; 81/168 EGARCH and 54/168 APARCH jobs had in-sample variance ranges above 10^6). The failure | |
| 61 | + policy of D-10 ("bounded retry sequence, alternative starting values, optimizer diagnostics") is made concrete: every arch fit | |
| 62 | + now runs from arch's default plus three alternative starting vectors and keeps the best converged optimum; a fit whose | |
| 63 | + log-likelihood is more than 50 units below GARCH(1,1) with the same innovation law (models at least as flexible; not applied | |
| 64 | + to ARCH(5), IGARCH, GAS-t, the baselines or the realized-measure models with measurement equations) is retried from perturbed | |
| 65 | + starts and otherwise recorded as failed (reason 'degenerate'), after which the driver keeps the last good parameters (flag | |
| 66 | + stale_fit). The affected jobs (all arch-backed models except the fixed baselines, and the custom daily models) are re-run; | |
| 67 | + losses, analyses and exhibits are regenerated. The first-pass analysis is kept in results/metrics/analysis_main_test_pass1/ | |
| 68 | + for the record. No change to the universe, periods, losses, proxies or inference. | |
modified
results/manifests/figure_audit.csv
+7 −0
@@ -4,3 +4,10 @@ fig_coverage.pdf,results/manifests/panels_manifest.json,42, | ||
| 4 | 4 | fig_volseries.pdf,panels/SPY.parquet;panels/CL.parquet;panels/EURUSD.parquet;panels/BTC.parquet,4, |
| 5 | 5 | fig_timeline.pdf,configs/model_registry.yaml,26, |
| 6 | 6 | fig_lineage.pdf,results/manifests/ingest_manifest.json;results/manifests/panels_manifest.json;results/forecasts/main/_runs.jsonl,0, |
| 7 | +fig_heatmap_h1_tierA.pdf,results/metrics/analysis_*/per_symbol/*.json,1638,• = DM p<0.10 (HAC) | |
| 8 | +fig_heatmap_h22_tierA.pdf,results/metrics/analysis_*/per_symbol/*.json,1638,• = DM p<0.10 (HAC) | |
| 9 | +fig_heatmap_h1_tierB.pdf,results/metrics/analysis_*/per_symbol/*.json,420,• = DM p<0.10 (HAC) | |
| 10 | +fig_mcs_h1_tierA.pdf,results/metrics/analysis_*/per_symbol/*.json,2898, | |
| 11 | +fig_mcs_h22_tierA.pdf,results/metrics/analysis_*/per_symbol/*.json,2898, | |
| 12 | +fig_mcs_h1_tierB.pdf,results/metrics/analysis_*/per_symbol/*.json,420, | |
| 13 | +fig_horizon.pdf,results/tables/pooled_*.csv,18, | |
renamed
results/metrics/analysis_main_test/per_symbol/AAPL.json → results/metrics/analysis_main_test_pass1/per_symbol/AAPL.json
+0 −0
added
results/metrics/analysis_main_test_pass1/per_symbol/AMZN.json
+19808 −0
Large diff (19,808 lines) is not loaded by default to keep the page fast.Load diff
Diff truncated — file too large.