Sensitivity
Which settings of the trading rules were tested, how big each effect was, how sure we are, and why the chosen value was kept. Values that hurt are shown as clearly as values that helped.
Settings examined
28
Values tested
106
each on 39 blind years
Made things worse
70
shown below, not hidden
Survive multiple testing
2
|t| ≥ 3.50
Champion mean week
+0.31%
How to read this
Each setting is changed alone while everything else stays at the champion. Effect is the change in average weekly return over the champion; the interval comes from the year-by-year effects; stability is the share of years the change pointed the same way. With 106 values tried, some will look good by luck, so a single value only counts when |t| ≥ 3.50 (Bonferroni). Alternatives that lost are listed in full.
Settings ranked by how much they matter
| Setting | Kind | Selected | Range of effect | Sensitivity class | Best tested | Why the selected value stays |
|---|---|---|---|---|---|---|
| w_model | knob | 0.5 | 0.00490 | significant | 1.0 | a tested value beats the champion after multiple-testing correction (t +3.77 vs needed 3.50); a candidate for the next proposal, not yet adopted |
| liq_q | knob | 0.3 | 0.00219 | significant | 0.0 | the best alternative gained +0.0007/week (t +1.73) but that is below the multiple-testing bar of 3.50, so it is not distinguishable from luck; the champion value stays |
| rebalance_weeks | knob | 2 | 0.00136 | significant | 1 | the best alternative gained +0.0002/week (t +0.48) but that is below the multiple-testing bar of 3.50, so it is not distinguishable from luck; the champion value stays |
| k | knob | 8 | 0.00115 | slight | 6 | the best alternative gained +0.0001/week (t +0.51) but that is below the multiple-testing bar of 3.50, so it is not distinguishable from luck; the champion value stays |
| brake | knob | 0.08 | 0.00102 | significant | 0.12 | the best alternative gained +0.0002/week (t +0.89) but that is below the multiple-testing bar of 3.50, so it is not distinguishable from luck; the champion value stays |
| max20 | indicator | -0.8 | 0.00091 | slight | x2 | the best alternative gained +0.0000/week (t +0.11) but that is below the multiple-testing bar of 3.50, so it is not distinguishable from luck; the champion value stays |
| pool_q | knob | 0.98 | 0.00089 | slight | 0.9 | the best alternative gained +0.0004/week (t +0.82) but that is below the multiple-testing bar of 3.50, so it is not distinguishable from luck; the champion value stays |
| ind_mom60 | indicator | 0.6 | 0.00087 | slight | reversed | the best alternative gained +0.0003/week (t +0.83) but that is below the multiple-testing bar of 3.50, so it is not distinguishable from luck; the champion value stays |
| mom_12_1 | indicator | 0.3 | 0.00076 | slight | x0.5 | the best alternative gained +0.0000/week (t +0.18) but that is below the multiple-testing bar of 3.50, so it is not distinguishable from luck; the champion value stays |
| skew60 | indicator | -0.2 | 0.00067 | slight | x2 | the best alternative gained +0.0001/week (t +0.54) but that is below the multiple-testing bar of 3.50, so it is not distinguishable from luck; the champion value stays |
| dist_52wh | indicator | 0.7 | 0.00067 | slight | reversed | the best alternative gained +0.0001/week (t +0.20) but that is below the multiple-testing bar of 3.50, so it is not distinguishable from luck; the champion value stays |
| r5_nonews | indicator | -0.5 | 0.00065 | slight | x0.5 | the best alternative gained +0.0001/week (t +0.18) but that is below the multiple-testing bar of 3.50, so it is not distinguishable from luck; the champion value stays |
| exit_q | knob | 0.8 | 0.00050 | slight | 0.6 | the best alternative gained +0.0003/week (t +0.99) but that is below the multiple-testing bar of 3.50, so it is not distinguishable from luck; the champion value stays |
| trend_filter | knob | off | 0.00046 | significant | -0.1 | every tested alternative did the same or worse than the champion; the champion value stays |
| frog | indicator | 0.4 | 0.00044 | slight | x2 | every tested alternative did the same or worse than the champion; the champion value stays |
| ear | indicator | 1.0 | 0.00041 | slight | x0.5 | every tested alternative did the same or worse than the champion; the champion value stays |
| stress_k | knob | 4 | 0.00035 | slight | 2 | the best alternative gained +0.0002/week (t +1.41) but that is below the multiple-testing bar of 3.50, so it is not distinguishable from luck; the champion value stays |
| ev_offering | indicator | -1.0 | 0.00031 | slight | off | the best alternative gained +0.0001/week (t +0.75) but that is below the multiple-testing bar of 3.50, so it is not distinguishable from luck; the champion value stays |
| max_per_sector | knob | 2 | 0.00025 | slight | 1 | every tested alternative did the same or worse than the champion; the champion value stays |
| stress_thr | knob | 1.05 | 0.00024 | slight | off | every tested alternative did the same or worse than the champion; the champion value stays |
| ins_buyers30 | indicator | 0.8 | 0.00021 | slight | reversed | the best alternative gained +0.0000/week (t +0.36) but that is below the multiple-testing bar of 3.50, so it is not distinguishable from luck; the champion value stays |
| ins_officer30 | indicator | 0.4 | 0.00012 | negligible | x2 | the best alternative gained +0.0000/week (t +0.30) but that is below the multiple-testing bar of 3.50, so it is not distinguishable from luck; the champion value stays |
| ev_shelf | indicator | -0.3 | 0.00008 | slight | x0.5 | the best alternative gained +0.0001/week (t +1.86) but that is below the multiple-testing bar of 3.50, so it is not distinguishable from luck; the champion value stays |
| ev_red_flag | indicator | -1.0 | 0.00007 | slight | off | every tested alternative did the same or worse than the champion; the champion value stays |
| evidence score (bug fix) | fix | traded as champion | 0.00000 | significant | off, as traded | a tested value beats the champion after multiple-testing correction (t +3.77 vs needed 3.50); a candidate for the next proposal, not yet adopted |
| pick | knob | hivol | 0.00000 | negligible | top | every tested alternative did the same or worse than the champion; the champion value stays |
| vol_filter | knob | False | 0.00000 | slight | True | every tested alternative did the same or worse than the champion; the champion value stays |
| trend_gross | knob | 0.0 | 0.00000 | negligible | 0.25 | every tested alternative did the same or worse than the champion; the champion value stays |
Negative results (never hidden)
| Alternative tried | Effect / week | t | Years same direction | Stability |
|---|---|---|---|---|
| w_model = 0.0 | -0.00168 | -2.87 | 64% | inconsistent |
| liq_q = 0.85 | -0.00149 | -2.69 | 72% | consistent |
| liq_q = 0.7 | -0.00119 | -2.40 | 69% | consistent |
| rebalance_weeks = 3 | -0.00116 | -2.05 | 64% | inconsistent |
| k = 2 | -0.00107 | -1.83 | 59% | inconsistent |
| trend_filter = 0.0 | -0.00088 | -2.57 | 69% | consistent |
| vol_filter = True | -0.00087 | -1.70 | 67% | consistent |
| max20 = x0.5 | -0.00087 | -1.95 | 64% | inconsistent |
| brake = 0.03 | -0.00086 | -3.31 | 69% | consistent |
| trend_filter = -0.05 | -0.00085 | -2.16 | 38% | inconsistent |
| mom_12_1 = reversed | -0.00073 | -1.55 | 64% | inconsistent |
| max20 = off | -0.00071 | -1.46 | 69% | consistent |
| liq_q = 0.5 | -0.00071 | -2.04 | 64% | inconsistent |
| r5_nonews = reversed | -0.00060 | -1.40 | 49% | inconsistent |
| w_model = 0.2 | -0.00057 | -1.20 | 56% | inconsistent |
| ind_mom60 = x2 | -0.00054 | -1.38 | 59% | inconsistent |
| skew60 = reversed | -0.00053 | -1.26 | 59% | inconsistent |
| dist_52wh = x0.5 | -0.00052 | -1.40 | 64% | inconsistent |
| frog = x0.5 | -0.00048 | -1.55 | 56% | inconsistent |
| pool_q = 0.8 | -0.00044 | -0.69 | 51% | inconsistent |
| ear = reversed | -0.00043 | -1.26 | 31% | inconsistent |
| ear = x2 | -0.00043 | -1.70 | 26% | inconsistent |
| trend_filter = -0.1 | -0.00041 | -1.41 | 15% | inconsistent |
| dist_52wh = x2 | -0.00041 | -1.34 | 54% | inconsistent |
| mom_12_1 = off | -0.00040 | -1.10 | 46% | inconsistent |
| pool_q = 0.95 | -0.00037 | -0.84 | 51% | inconsistent |
| stress_thr = 1.0 | -0.00035 | -1.51 | 15% | inconsistent |
| frog = reversed | -0.00035 | -0.88 | 56% | inconsistent |
| frog = off | -0.00035 | -0.99 | 46% | inconsistent |
| max_per_sector = off | -0.00029 | -0.89 | 51% | inconsistent |
| pool_q = 0.99 | -0.00028 | -0.73 | 46% | inconsistent |
| ind_mom60 = x0.5 | -0.00027 | -0.82 | 51% | inconsistent |
| brake = 0.05 | -0.00027 | -1.56 | 41% | inconsistent |
| stress_thr = 0.95 | -0.00027 | -1.42 | 18% | inconsistent |
| rebalance_weeks = 4 | -0.00025 | -0.42 | 51% | inconsistent |
| stress_thr = 1.1 | -0.00023 | -1.02 | 10% | inconsistent |
| ev_offering = reversed | -0.00022 | -1.30 | 31% | inconsistent |
| max_per_sector = 3 | -0.00022 | -1.09 | 49% | inconsistent |
| ear = off | -0.00022 | -1.30 | 23% | inconsistent |
| w_model = 0.3 | -0.00020 | -0.47 | 49% | inconsistent |
| dist_52wh = off | -0.00019 | -0.41 | 51% | inconsistent |
| exit_q = 0.9 | -0.00018 | -0.64 | 56% | inconsistent |
| pick = top | -0.00017 | -0.34 | 59% | inconsistent |
| r5_nonews = off | -0.00017 | -0.49 | 51% | inconsistent |
| k = 4 | -0.00017 | -0.73 | 49% | inconsistent |
| stress_k = 6 | -0.00017 | -2.06 | 18% | inconsistent |
| mom_12_1 = x2 | -0.00017 | -0.52 | 49% | inconsistent |
| ins_buyers30 = x2 | -0.00016 | -1.52 | 18% | inconsistent |
| stress_thr = 0.9 | -0.00015 | -0.81 | 18% | inconsistent |
| k = 16 | -0.00014 | -0.72 | 36% | inconsistent |
| ins_buyers30 = x0.5 | -0.00013 | -1.30 | 23% | inconsistent |
| exit_q = 0.5 | -0.00013 | -0.26 | 67% | consistent |
| stress_thr = off | -0.00011 | -1.24 | 13% | inconsistent |
| ins_officer30 = reversed | -0.00010 | -0.92 | 23% | inconsistent |
| ev_red_flag = x2 | -0.00009 | -1.05 | 23% | inconsistent |
| k = 12 | -0.00007 | -0.31 | 36% | inconsistent |
| exit_q = 0.95 | -0.00006 | -0.18 | 46% | inconsistent |
| ev_red_flag = x0.5 | -0.00006 | -0.68 | 18% | inconsistent |
| frog = x2 | -0.00005 | -0.10 | 51% | inconsistent |
| ins_officer30 = off | -0.00005 | -0.91 | 26% | inconsistent |
Every setting in detail
Open a setting to see each tested value, its interval and its year-by-year sign pattern.
w_model (knob) — selected 0.5, 3 better, 3 worse, sensitivity significant
Reason: a tested value beats the champion after multiple-testing correction (t +3.77 vs needed 3.50); a candidate for the next proposal, not yet adopted
| Tested value | Effect: change in mean weekly return | 95% interval (across years) | t | Years same direction | Year by year | Stability | Change in worst drawdown |
|---|---|---|---|---|---|---|---|
| 0.0 | -0.00168 | -0.00284 to -0.00053 | -2.87 | 64% | inconsistent (significant) | -0.016 | |
| 0.2 | -0.00057 | -0.00150 to 0.00036 | -1.20 | 56% | inconsistent (slight) | 0.000 | |
| 0.3 | -0.00020 | -0.00106 to 0.00066 | -0.47 | 49% | inconsistent (slight) | 0.008 | |
| 0.7 | 0.00062 | -0.00056 to 0.00180 | +1.02 | 54% | inconsistent (slight) | -0.013 | |
| 0.85 | 0.00078 | -0.00053 to 0.00209 | +1.16 | 56% | inconsistent (slight) | -0.030 | |
| 1.0 | 0.00321 | 0.00154 to 0.00489 | +3.77 survives correction | 72% | consistent (significant) | -0.008 |
liq_q (knob) — selected 0.3, 2 better, 3 worse, sensitivity significant
Reason: the best alternative gained +0.0007/week (t +1.73) but that is below the multiple-testing bar of 3.50, so it is not distinguishable from luck; the champion value stays
| Tested value | Effect: change in mean weekly return | 95% interval (across years) | t | Years same direction | Year by year | Stability | Change in worst drawdown |
|---|---|---|---|---|---|---|---|
| 0.0 | 0.00070 | -0.00009 to 0.00149 | +1.73 | 64% | inconsistent (slight) | 0.004 | |
| 0.2 | 0.00067 | 0.00024 to 0.00111 | +3.03 | 69% | consistent (significant) | 0.007 | |
| 0.5 | -0.00071 | -0.00139 to -0.00003 | -2.04 | 64% | inconsistent (significant) | -0.011 | |
| 0.7 | -0.00119 | -0.00216 to -0.00022 | -2.40 | 69% | consistent (significant) | -0.007 | |
| 0.85 | -0.00149 | -0.00257 to -0.00041 | -2.69 | 72% | consistent (significant) | -0.019 |
rebalance_weeks (knob) — selected 2, 1 better, 2 worse, sensitivity significant
Reason: the best alternative gained +0.0002/week (t +0.48) but that is below the multiple-testing bar of 3.50, so it is not distinguishable from luck; the champion value stays
| Tested value | Effect: change in mean weekly return | 95% interval (across years) | t | Years same direction | Year by year | Stability | Change in worst drawdown |
|---|---|---|---|---|---|---|---|
| 1 | 0.00020 | -0.00061 to 0.00100 | +0.48 | 21% | inconsistent (negligible) | -0.015 | |
| 3 | -0.00116 | -0.00227 to -0.00005 | -2.05 | 64% | inconsistent (significant) | -0.011 | |
| 4 | -0.00025 | -0.00144 to 0.00093 | -0.42 | 51% | inconsistent (slight) | 0.005 |
k (knob) — selected 8, 2 better, 4 worse, sensitivity slight
Reason: the best alternative gained +0.0001/week (t +0.51) but that is below the multiple-testing bar of 3.50, so it is not distinguishable from luck; the champion value stays
| Tested value | Effect: change in mean weekly return | 95% interval (across years) | t | Years same direction | Year by year | Stability | Change in worst drawdown |
|---|---|---|---|---|---|---|---|
| 2 | -0.00107 | -0.00222 to 0.00007 | -1.83 | 59% | inconsistent (slight) | -0.094 | |
| 3 | 0.00002 | -0.00060 to 0.00064 | +0.06 | 44% | inconsistent (negligible) | -0.046 | |
| 4 | -0.00017 | -0.00064 to 0.00029 | -0.73 | 49% | inconsistent (negligible) | -0.026 | |
| 6 | 0.00008 | -0.00022 to 0.00037 | +0.51 | 38% | inconsistent (negligible) | -0.014 | |
| 12 | -0.00007 | -0.00049 to 0.00036 | -0.31 | 36% | inconsistent (negligible) | 0.005 | |
| 16 | -0.00014 | -0.00051 to 0.00024 | -0.72 | 36% | inconsistent (negligible) | 0.005 |
brake (knob) — selected 0.08, 2 better, 3 worse, sensitivity significant
Reason: the best alternative gained +0.0002/week (t +0.89) but that is below the multiple-testing bar of 3.50, so it is not distinguishable from luck; the champion value stays
| Tested value | Effect: change in mean weekly return | 95% interval (across years) | t | Years same direction | Year by year | Stability | Change in worst drawdown |
|---|---|---|---|---|---|---|---|
| off | 0.00013 | -0.00044 to 0.00070 | +0.45 | 26% | inconsistent (negligible) | -0.014 | |
| 0.03 | -0.00086 | -0.00136 to -0.00035 | -3.31 | 69% | consistent (significant) | 0.004 | |
| 0.05 | -0.00027 | -0.00062 to 0.00007 | -1.56 | 41% | inconsistent (slight) | 0.002 | |
| 0.12 | 0.00016 | -0.00019 to 0.00051 | +0.89 | 23% | inconsistent (negligible) | -0.007 | |
| 0.2 | -0.00001 | -0.00057 to 0.00056 | -0.03 | 15% | inconsistent (negligible) | -0.014 |
max20 (indicator) — selected -0.8, 1 better, 3 worse, sensitivity slight
Reason: the best alternative gained +0.0000/week (t +0.11) but that is below the multiple-testing bar of 3.50, so it is not distinguishable from luck; the champion value stays
| Tested value | Effect: change in mean weekly return | 95% interval (across years) | t | Years same direction | Year by year | Stability | Change in worst drawdown |
|---|---|---|---|---|---|---|---|
| off | -0.00071 | -0.00166 to 0.00024 | -1.46 | 69% | consistent (slight) | -0.005 | |
| x0.5 | -0.00087 | -0.00175 to 0.00000 | -1.95 | 64% | inconsistent (slight) | -0.011 | |
| x2 | 0.00004 | -0.00071 to 0.00079 | +0.11 | 51% | inconsistent (negligible) | 0.005 | |
| reversed | -0.00002 | -0.00101 to 0.00096 | -0.04 | 44% | inconsistent (negligible) | 0.009 |
pool_q (knob) — selected 0.98, 1 better, 3 worse, sensitivity slight
Reason: the best alternative gained +0.0004/week (t +0.82) but that is below the multiple-testing bar of 3.50, so it is not distinguishable from luck; the champion value stays
| Tested value | Effect: change in mean weekly return | 95% interval (across years) | t | Years same direction | Year by year | Stability | Change in worst drawdown |
|---|---|---|---|---|---|---|---|
| 0.8 | -0.00044 | -0.00170 to 0.00081 | -0.69 | 51% | inconsistent (slight) | -0.026 | |
| 0.9 | 0.00044 | -0.00061 to 0.00149 | +0.82 | 56% | inconsistent (slight) | -0.003 | |
| 0.95 | -0.00037 | -0.00124 to 0.00050 | -0.84 | 51% | inconsistent (slight) | -0.009 | |
| 0.99 | -0.00028 | -0.00104 to 0.00047 | -0.73 | 46% | inconsistent (slight) | -0.013 |
ind_mom60 (indicator) — selected 0.6, 2 better, 2 worse, sensitivity slight
Reason: the best alternative gained +0.0003/week (t +0.83) but that is below the multiple-testing bar of 3.50, so it is not distinguishable from luck; the champion value stays
| Tested value | Effect: change in mean weekly return | 95% interval (across years) | t | Years same direction | Year by year | Stability | Change in worst drawdown |
|---|---|---|---|---|---|---|---|
| off | 0.00007 | -0.00068 to 0.00082 | +0.18 | 44% | inconsistent (negligible) | 0.002 | |
| x0.5 | -0.00027 | -0.00093 to 0.00038 | -0.82 | 51% | inconsistent (slight) | -0.008 | |
| x2 | -0.00054 | -0.00130 to 0.00023 | -1.38 | 59% | inconsistent (slight) | -0.012 | |
| reversed | 0.00034 | -0.00045 to 0.00113 | +0.83 | 59% | inconsistent (slight) | 0.004 |
mom_12_1 (indicator) — selected 0.3, 1 better, 3 worse, sensitivity slight
Reason: the best alternative gained +0.0000/week (t +0.18) but that is below the multiple-testing bar of 3.50, so it is not distinguishable from luck; the champion value stays
| Tested value | Effect: change in mean weekly return | 95% interval (across years) | t | Years same direction | Year by year | Stability | Change in worst drawdown |
|---|---|---|---|---|---|---|---|
| off | -0.00040 | -0.00111 to 0.00031 | -1.10 | 46% | inconsistent (slight) | -0.007 | |
| x0.5 | 0.00003 | -0.00031 to 0.00038 | +0.18 | 56% | inconsistent (negligible) | -0.005 | |
| x2 | -0.00017 | -0.00081 to 0.00047 | -0.52 | 49% | inconsistent (negligible) | 0.000 | |
| reversed | -0.00073 | -0.00166 to 0.00019 | -1.55 | 64% | inconsistent (slight) | -0.010 |
skew60 (indicator) — selected -0.2, 3 better, 1 worse, sensitivity slight
Reason: the best alternative gained +0.0001/week (t +0.54) but that is below the multiple-testing bar of 3.50, so it is not distinguishable from luck; the champion value stays
| Tested value | Effect: change in mean weekly return | 95% interval (across years) | t | Years same direction | Year by year | Stability | Change in worst drawdown |
|---|---|---|---|---|---|---|---|
| off | 0.00006 | -0.00044 to 0.00056 | +0.23 | 54% | inconsistent (negligible) | -0.002 | |
| x0.5 | 0.00006 | -0.00030 to 0.00043 | +0.34 | 54% | inconsistent (negligible) | -0.002 | |
| x2 | 0.00014 | -0.00036 to 0.00064 | +0.54 | 51% | inconsistent (negligible) | -0.002 | |
| reversed | -0.00053 | -0.00135 to 0.00029 | -1.26 | 59% | inconsistent (slight) | -0.010 |
dist_52wh (indicator) — selected 0.7, 1 better, 3 worse, sensitivity slight
Reason: the best alternative gained +0.0001/week (t +0.20) but that is below the multiple-testing bar of 3.50, so it is not distinguishable from luck; the champion value stays
| Tested value | Effect: change in mean weekly return | 95% interval (across years) | t | Years same direction | Year by year | Stability | Change in worst drawdown |
|---|---|---|---|---|---|---|---|
| off | -0.00019 | -0.00108 to 0.00071 | -0.41 | 51% | inconsistent (negligible) | -0.009 | |
| x0.5 | -0.00052 | -0.00124 to 0.00021 | -1.40 | 64% | inconsistent (slight) | -0.014 | |
| x2 | -0.00041 | -0.00102 to 0.00019 | -1.34 | 54% | inconsistent (slight) | -0.006 | |
| reversed | 0.00015 | -0.00126 to 0.00156 | +0.20 | 51% | inconsistent (negligible) | -0.015 |
r5_nonews (indicator) — selected -0.5, 1 better, 3 worse, sensitivity slight
Reason: the best alternative gained +0.0001/week (t +0.18) but that is below the multiple-testing bar of 3.50, so it is not distinguishable from luck; the champion value stays
| Tested value | Effect: change in mean weekly return | 95% interval (across years) | t | Years same direction | Year by year | Stability | Change in worst drawdown |
|---|---|---|---|---|---|---|---|
| off | -0.00017 | -0.00087 to 0.00053 | -0.49 | 51% | inconsistent (negligible) | -0.004 | |
| x0.5 | 0.00005 | -0.00051 to 0.00061 | +0.18 | 46% | inconsistent (negligible) | -0.003 | |
| x2 | -0.00004 | -0.00077 to 0.00068 | -0.12 | 54% | inconsistent (negligible) | -0.009 | |
| reversed | -0.00060 | -0.00145 to 0.00024 | -1.40 | 49% | inconsistent (slight) | -0.013 |
exit_q (knob) — selected 0.8, 2 better, 3 worse, sensitivity slight
Reason: the best alternative gained +0.0003/week (t +0.99) but that is below the multiple-testing bar of 3.50, so it is not distinguishable from luck; the champion value stays
| Tested value | Effect: change in mean weekly return | 95% interval (across years) | t | Years same direction | Year by year | Stability | Change in worst drawdown |
|---|---|---|---|---|---|---|---|
| 0.5 | -0.00013 | -0.00107 to 0.00082 | -0.26 | 67% | consistent (negligible) | 0.011 | |
| 0.6 | 0.00031 | -0.00031 to 0.00094 | +0.99 | 51% | inconsistent (slight) | 0.011 | |
| 0.7 | 0.00009 | -0.00054 to 0.00073 | +0.29 | 62% | inconsistent (negligible) | 0.000 | |
| 0.9 | -0.00018 | -0.00075 to 0.00038 | -0.64 | 56% | inconsistent (negligible) | -0.012 | |
| 0.95 | -0.00006 | -0.00072 to 0.00060 | -0.18 | 46% | inconsistent (negligible) | -0.018 |
trend_filter (knob) — selected off, 0 better, 3 worse, sensitivity significant
Reason: every tested alternative did the same or worse than the champion; the champion value stays
| Tested value | Effect: change in mean weekly return | 95% interval (across years) | t | Years same direction | Year by year | Stability | Change in worst drawdown |
|---|---|---|---|---|---|---|---|
| -0.1 | -0.00041 | -0.00099 to 0.00016 | -1.41 | 15% | inconsistent (slight) | 0.009 | |
| -0.05 | -0.00085 | -0.00162 to -0.00008 | -2.16 | 38% | inconsistent (significant) | 0.009 | |
| 0.0 | -0.00088 | -0.00155 to -0.00021 | -2.57 | 69% | consistent (significant) | 0.038 |
frog (indicator) — selected 0.4, 0 better, 4 worse, sensitivity slight
Reason: every tested alternative did the same or worse than the champion; the champion value stays
| Tested value | Effect: change in mean weekly return | 95% interval (across years) | t | Years same direction | Year by year | Stability | Change in worst drawdown |
|---|---|---|---|---|---|---|---|
| off | -0.00035 | -0.00103 to 0.00034 | -0.99 | 46% | inconsistent (slight) | -0.012 | |
| x0.5 | -0.00048 | -0.00110 to 0.00013 | -1.55 | 56% | inconsistent (slight) | -0.009 | |
| x2 | -0.00005 | -0.00098 to 0.00088 | -0.10 | 51% | inconsistent (negligible) | -0.002 | |
| reversed | -0.00035 | -0.00113 to 0.00043 | -0.88 | 56% | inconsistent (slight) | -0.005 |
ear (indicator) — selected 1.0, 0 better, 4 worse, sensitivity slight
Reason: every tested alternative did the same or worse than the champion; the champion value stays
| Tested value | Effect: change in mean weekly return | 95% interval (across years) | t | Years same direction | Year by year | Stability | Change in worst drawdown |
|---|---|---|---|---|---|---|---|
| off | -0.00022 | -0.00054 to 0.00011 | -1.30 | 23% | inconsistent (slight) | -0.001 | |
| x0.5 | -0.00002 | -0.00031 to 0.00027 | -0.14 | 23% | inconsistent (negligible) | 0.002 | |
| x2 | -0.00043 | -0.00092 to 0.00007 | -1.70 | 26% | inconsistent (slight) | -0.004 | |
| reversed | -0.00043 | -0.00111 to 0.00024 | -1.26 | 31% | inconsistent (slight) | -0.006 |
stress_k (knob) — selected 4, 2 better, 1 worse, sensitivity slight
Reason: the best alternative gained +0.0002/week (t +1.41) but that is below the multiple-testing bar of 3.50, so it is not distinguishable from luck; the champion value stays
| Tested value | Effect: change in mean weekly return | 95% interval (across years) | t | Years same direction | Year by year | Stability | Change in worst drawdown |
|---|---|---|---|---|---|---|---|
| 2 | 0.00018 | -0.00007 to 0.00044 | +1.41 | 10% | inconsistent (slight) | -0.002 | |
| 3 | 0.00003 | -0.00004 to 0.00010 | +0.76 | 8% | inconsistent (negligible) | -0.003 | |
| 6 | -0.00017 | -0.00034 to -0.00001 | -2.06 | 18% | inconsistent (slight) | -0.001 |
ev_offering (indicator) — selected -1.0, 2 better, 2 worse, sensitivity slight
Reason: the best alternative gained +0.0001/week (t +0.75) but that is below the multiple-testing bar of 3.50, so it is not distinguishable from luck; the champion value stays
| Tested value | Effect: change in mean weekly return | 95% interval (across years) | t | Years same direction | Year by year | Stability | Change in worst drawdown |
|---|---|---|---|---|---|---|---|
| off | 0.00008 | -0.00013 to 0.00030 | +0.75 | 28% | inconsistent (negligible) | 0.001 | |
| x0.5 | 0.00003 | -0.00008 to 0.00014 | +0.50 | 18% | inconsistent (negligible) | 0.000 | |
| x2 | -0.00002 | -0.00018 to 0.00013 | -0.30 | 26% | inconsistent (negligible) | -0.001 | |
| reversed | -0.00022 | -0.00056 to 0.00011 | -1.30 | 31% | inconsistent (slight) | -0.005 |
max_per_sector (knob) — selected 2, 0 better, 3 worse, sensitivity slight
Reason: every tested alternative did the same or worse than the champion; the champion value stays
| Tested value | Effect: change in mean weekly return | 95% interval (across years) | t | Years same direction | Year by year | Stability | Change in worst drawdown |
|---|---|---|---|---|---|---|---|
| off | -0.00029 | -0.00094 to 0.00035 | -0.89 | 51% | inconsistent (slight) | -0.002 | |
| 1 | -0.00004 | -0.00078 to 0.00070 | -0.11 | 51% | inconsistent (negligible) | -0.016 | |
| 3 | -0.00022 | -0.00062 to 0.00018 | -1.09 | 49% | inconsistent (slight) | -0.002 |
stress_thr (knob) — selected 1.05, 0 better, 5 worse, sensitivity slight
Reason: every tested alternative did the same or worse than the champion; the champion value stays
| Tested value | Effect: change in mean weekly return | 95% interval (across years) | t | Years same direction | Year by year | Stability | Change in worst drawdown |
|---|---|---|---|---|---|---|---|
| off | -0.00011 | -0.00029 to 0.00007 | -1.24 | 13% | inconsistent (slight) | 0.001 | |
| 0.9 | -0.00015 | -0.00051 to 0.00021 | -0.81 | 18% | inconsistent (negligible) | -0.008 | |
| 0.95 | -0.00027 | -0.00064 to 0.00010 | -1.42 | 18% | inconsistent (slight) | -0.004 | |
| 1.0 | -0.00035 | -0.00082 to 0.00011 | -1.51 | 15% | inconsistent (slight) | -0.007 | |
| 1.1 | -0.00023 | -0.00068 to 0.00022 | -1.02 | 10% | inconsistent (slight) | 0.002 |
ins_buyers30 (indicator) — selected 0.8, 1 better, 3 worse, sensitivity slight
Reason: the best alternative gained +0.0000/week (t +0.36) but that is below the multiple-testing bar of 3.50, so it is not distinguishable from luck; the champion value stays
| Tested value | Effect: change in mean weekly return | 95% interval (across years) | t | Years same direction | Year by year | Stability | Change in worst drawdown |
|---|---|---|---|---|---|---|---|
| off | -0.00002 | -0.00022 to 0.00019 | -0.17 | 26% | inconsistent (negligible) | 0.001 | |
| x0.5 | -0.00013 | -0.00034 to 0.00007 | -1.30 | 23% | inconsistent (slight) | -0.003 | |
| x2 | -0.00016 | -0.00037 to 0.00005 | -1.52 | 18% | inconsistent (slight) | -0.004 | |
| reversed | 0.00005 | -0.00020 to 0.00029 | +0.36 | 21% | inconsistent (negligible) | 0.001 |
ins_officer30 (indicator) — selected 0.4, 1 better, 3 worse, sensitivity negligible
Reason: the best alternative gained +0.0000/week (t +0.30) but that is below the multiple-testing bar of 3.50, so it is not distinguishable from luck; the champion value stays
| Tested value | Effect: change in mean weekly return | 95% interval (across years) | t | Years same direction | Year by year | Stability | Change in worst drawdown |
|---|---|---|---|---|---|---|---|
| off | -0.00005 | -0.00015 to 0.00006 | -0.91 | 26% | inconsistent (negligible) | -0.001 | |
| x0.5 | -0.00002 | -0.00011 to 0.00007 | -0.39 | 21% | inconsistent (negligible) | -0.000 | |
| x2 | 0.00002 | -0.00013 to 0.00017 | +0.30 | 13% | inconsistent (negligible) | -0.001 | |
| reversed | -0.00010 | -0.00031 to 0.00011 | -0.92 | 23% | inconsistent (negligible) | -0.001 |
ev_shelf (indicator) — selected -0.3, 4 better, 0 worse, sensitivity slight
Reason: the best alternative gained +0.0001/week (t +1.86) but that is below the multiple-testing bar of 3.50, so it is not distinguishable from luck; the champion value stays
| Tested value | Effect: change in mean weekly return | 95% interval (across years) | t | Years same direction | Year by year | Stability | Change in worst drawdown |
|---|---|---|---|---|---|---|---|
| off | 0.00007 | -0.00003 to 0.00017 | +1.41 | 18% | inconsistent (slight) | 0.001 | |
| x0.5 | 0.00009 | -0.00001 to 0.00018 | +1.86 | 15% | inconsistent (slight) | 0.000 | |
| x2 | 0.00001 | -0.00007 to 0.00008 | +0.21 | 10% | inconsistent (negligible) | 0.000 | |
| reversed | 0.00007 | -0.00008 to 0.00023 | +0.95 | 21% | inconsistent (negligible) | -0.000 |
ev_red_flag (indicator) — selected -1.0, 0 better, 4 worse, sensitivity slight
Reason: every tested alternative did the same or worse than the champion; the champion value stays
| Tested value | Effect: change in mean weekly return | 95% interval (across years) | t | Years same direction | Year by year | Stability | Change in worst drawdown |
|---|---|---|---|---|---|---|---|
| off | -0.00002 | -0.00024 to 0.00020 | -0.17 | 18% | inconsistent (negligible) | 0.000 | |
| x0.5 | -0.00006 | -0.00023 to 0.00011 | -0.68 | 18% | inconsistent (negligible) | -0.002 | |
| x2 | -0.00009 | -0.00025 to 0.00008 | -1.05 | 23% | inconsistent (slight) | -0.001 | |
| reversed | -0.00004 | -0.00034 to 0.00026 | -0.27 | 23% | inconsistent (negligible) | -0.001 |
evidence score (bug fix) (fix) — selected traded as champion, 1 better, 0 worse, sensitivity significant
Reason: a tested value beats the champion after multiple-testing correction (t +3.77 vs needed 3.50); a candidate for the next proposal, not yet adopted
| Tested value | Effect: change in mean weekly return | 95% interval (across years) | t | Years same direction | Year by year | Stability | Change in worst drawdown |
|---|---|---|---|---|---|---|---|
| off, as traded | 0.00321 | 0.00154 to 0.00489 | +3.77 survives correction | 72% | consistent (significant) | -0.008 |
pick (knob) — selected hivol, 0 better, 1 worse, sensitivity negligible
Reason: every tested alternative did the same or worse than the champion; the champion value stays
| Tested value | Effect: change in mean weekly return | 95% interval (across years) | t | Years same direction | Year by year | Stability | Change in worst drawdown |
|---|---|---|---|---|---|---|---|
| top | -0.00017 | -0.00118 to 0.00083 | -0.34 | 59% | inconsistent (negligible) | 0.030 |
vol_filter (knob) — selected False, 0 better, 1 worse, sensitivity slight
Reason: every tested alternative did the same or worse than the champion; the champion value stays
| Tested value | Effect: change in mean weekly return | 95% interval (across years) | t | Years same direction | Year by year | Stability | Change in worst drawdown |
|---|---|---|---|---|---|---|---|
| True | -0.00087 | -0.00188 to 0.00013 | -1.70 | 67% | consistent (slight) | 0.002 |
trend_gross (knob) — selected 0.0, 0 better, 0 worse, sensitivity negligible
Reason: every tested alternative did the same or worse than the champion; the champion value stays
| Tested value | Effect: change in mean weekly return | 95% interval (across years) | t | Years same direction | Year by year | Stability | Change in worst drawdown |
|---|---|---|---|---|---|---|---|
| 0.25 | 0.00000 | 0.00000 to 0.00000 | +0.00 | 0% | inconsistent (negligible) | 0.000 | |
| 0.5 | 0.00000 | 0.00000 to 0.00000 | +0.00 | 0% | inconsistent (negligible) | 0.000 | |
| 0.75 | 0.00000 | 0.00000 to 0.00000 | +0.00 | 0% | inconsistent (negligible) | 0.000 |
Champion configuration
| Setting | Value |
|---|---|
| k | 8 |
| exit_q | 0.8 |
| rebalance_weeks | 2 |
| brake | 0.08 |
| max_per_sector | 2 |
| w_model | 0.5 |
| pick | hivol |
| pool_q | 0.98 |
| liq_q | 0.3 |
| vol_filter | False |
| stress_thr | 1.05 |
| stress_k | 4 |
| trend_filter | off |
| trend_gross | 0.0 |
Indicator weights
| Indicator | Weight |
|---|---|
| ear | 1.00 |
| ev_offering | -1.00 |
| ev_red_flag | -1.00 |
| ins_buyers30 | 0.80 |
| max20 | -0.80 |
| dist_52wh | 0.70 |
| ind_mom60 | 0.60 |
| r5_nonews | -0.50 |
| ins_officer30 | 0.40 |
| frog | 0.40 |
| mom_12_1 | 0.30 |
| ev_shelf | -0.30 |
| skew60 | -0.20 |
Built 2026-09-29 12:40 UTC. Sources: experiments: 106 tested values. Generated by scripts/site_build.py; sealed test windows and credentials are never read. Paper trading only; nothing here is investment advice.