ETS Weight Optimization Study

Finding the Optimal ADX / Entropy / Hurst / dH/dt Balance

Date: 2026-02-24 Classification: Research — For Professional Review Version: 1.0

Executive Summary

Tested 938 weight combinations for the ETS Trend Strength Indicator against 39 known grid trading outcomes (23 profitable, 16 unprofitable) across 3 rolling windows.

Optimal weights found:

Component	Original	Current	Optimized	Change
ADX	40%	35%	60%	↑
Entropy	35%	30%	25%	↓
Hurst Level	25%	20%	5%	↓
Hurst Derivative	0%	15%	10%	↓

Performance comparison:

Metric	Original	Current	Optimized
Spearman Correlation	0.051	-0.025	0.124
Signal Accuracy	59.0%	59.0%	64.1%
Score Separation	3.2	1.4	10.8
Composite Score	0.217	0.176	0.307

1. Methodology

1.1 Data Points

For each of the 13 assets in each of the 3 rolling windows, we:

Calculated all 4 indicators (ADX, Entropy, Hurst, dH/dt) using only Year 1 data
Ran the Dynamic Grid simulation on Year 2 to get the actual grid trading return
This gives us a labeled dataset: indicator values → known profitability

1.2 Optimization Objective

We seek weights that maximize a composite score combining:

Spearman rank correlation (40%) — Does the ETS score rank assets correctly by profitability?
Score separation (30%) — How far apart are profitable vs unprofitable scores?
Signal accuracy (30%) — Does the GRID_TRADING/TREND_FOLLOWING threshold classify correctly?

1.3 Weight Constraints

All weights sum to 100%
Each weight: 5% minimum, 60% maximum
dH/dt weight can be 0% (testing whether derivative helps at all)
Grid search with 5% step size: 938 combinations tested

2. Top 25 Weight Combinations

Rank	Name	ADX	Entropy	Hurst	dH/dt	Composite	Spearman	Accuracy	Separation
1	A60_E25_H5_D10	60%	25%	5%	10%	0.307	0.124	64.1%	10.8
2	A60_E25_H15_D0	60%	25%	15%	0%	0.297	0.115	64.1%	9.7
3	A60_E25_H10_D5	60%	25%	10%	5%	0.295	0.105	64.1%	10.1
4	A60_E35_H5_D0	60%	35%	5%	0%	0.295	0.115	59.0%	12
5	A60_E30_H5_D5	60%	30%	5%	5%	0.292	0.118	59.0%	11.3
6	A60_E20_H20_D0	60%	20%	20%	0%	0.289	0.115	64.1%	8.4
7	A55_E40_H5_D0	55%	40%	5%	0%	0.288	0.115	59.0%	10.8
8	A60_E30_H10_D0	60%	30%	10%	0%	0.287	0.115	59.0%	10.6
9	A55_E35_H5_D5	55%	35%	5%	5%	0.286	0.118	59.0%	10.3
10	A55_E30_H5_D10	55%	30%	5%	10%	0.285	0.124	59.0%	9.7
11	A55_E25_H20_D0	55%	25%	20%	0%	0.283	0.115	64.1%	7.4
12	A50_E45_H5_D0	50%	45%	5%	0%	0.282	0.115	59.0%	9.8
13	A55_E25_H15_D5	55%	25%	15%	5%	0.281	0.1	64.1%	8
14	A55_E35_H10_D0	55%	35%	10%	0%	0.281	0.115	59.0%	9.6
15	A55_E25_H10_D10	55%	25%	10%	10%	0.28	0.093	64.1%	8.5
16	A50_E35_H5_D10	50%	35%	5%	10%	0.279	0.124	59.0%	8.7
17	A50_E40_H5_D5	50%	40%	5%	5%	0.279	0.118	59.0%	9.2
18	A55_E25_H5_D15	55%	25%	5%	15%	0.278	0.077	64.1%	9.2
19	A45_E50_H5_D0	45%	50%	5%	0%	0.276	0.115	59.0%	8.8
20	A55_E30_H15_D0	55%	30%	15%	0%	0.275	0.115	59.0%	8.6
21	A50_E40_H10_D0	50%	40%	10%	0%	0.274	0.115	59.0%	8.5
22	A55_E30_H10_D5	55%	30%	10%	5%	0.274	0.105	59.0%	9.2
23	A45_E45_H5_D5	45%	45%	5%	5%	0.273	0.118	59.0%	8.2
24	A55_E20_H25_D0	55%	20%	25%	0%	0.272	0.104	64.1%	6.3
25	A40_E55_H5_D0	40%	55%	5%	0%	0.269	0.115	59.0%	7.7
235	Current production ⬅️	35%	30%	20%	15%	0.176	-0.025	59.0%	1.4
90	Original (no dHdt) 📌	40%	35%	25%	0%	0.217	0.051	59.0%	3.2

Current production ranks #235 of 938 combinations. Original (no dH/dt) ranks #90 of 938 combinations.

3. Does the Hurst Derivative Actually Help?

Config	Best Composite	Best Spearman	Best Accuracy
Best 3-component (no dH/dt)	0.297	0.115	64.1%
Best 4-component (with dH/dt)	0.307	0.124	64.1%

Verdict: Yes — the Hurst derivative improves the ETS signal. The best 4-component weights outperform the best 3-component weights.

4. Component Importance Analysis

Average weights across the top 20 combinations:

Component	Avg Weight	Min	Max	Interpretation
ADX	55.5%	45%	60%	High importance
Entropy	31.5%	20%	50%	High importance
Hurst Level	9.0%	5%	20%	Minor contributor
Hurst Derivative	4.0%	0%	15%	Negligible

5. Recommendation

Update recommended. The optimized weights (60/25/5/10) improve the composite score by 13.1 points over current production weights.

The optimized weights should be applied to:

TrendStrengthIndicator.ts — the live API indicator
calculate-ets-standalone.cjs — the batch ETS calculator
BenchmarkTrendIndicator.ts — if crypto weights are used for benchmarks

Adding dH/dt was the right call. The best weights with dH/dt improve by 9.0 points over the original 3-component weights — confirming the Regime Transition Detection Study finding.

6. Dataset — Indicator Values vs Actual Returns

Window	Asset	ADX	Entropy	Hurst	dH/dt	Grid Return	Profitable?
W1	BTC	27.7	1.00	0.57	+0.40	4.7%	✅
W1	ETH	26.8	1.00	0.60	+0.15	20.2%	✅
W1	SOL	17.0	1.00	0.60	-0.31	13.5%	✅
W1	BNB	13.9	1.00	0.58	+0.02	16.7%	✅
W1	ADA	28.9	1.00	0.56	+1.00	70.0%	✅
W1	XRP	15.9	1.00	0.57	-0.41	98.8%	✅
W1	DOT	24.8	1.00	0.56	-0.20	6.1%	✅
W1	AVAX	13.2	1.00	0.56	-0.67	7.7%	✅
W1	POL	26.4	1.00	0.48	-0.68	-8.1%	❌
W1	LINK	15.6	1.00	0.57	+0.05	57.6%	✅
W1	UNI	14.6	1.00	0.54	-0.05	37.5%	✅
W1	LTC	21.8	1.00	0.58	+0.36	184.6%	✅
W1	DOGE	18.8	1.00	0.53	-0.14	4.4%	✅
W2	BTC	46.2	1.00	0.58	+1.00	9.7%	✅
W2	ETH	40.0	1.00	0.56	+0.44	30.6%	✅
W2	SOL	49.2	1.00	0.59	+0.39	14.1%	✅
W2	BNB	40.0	1.00	0.56	+0.67	12.8%	✅
W2	ADA	22.8	1.00	0.47	-0.75	-3.3%	❌
W2	XRP	40.3	1.00	0.59	+1.00	62.4%	✅
W2	DOT	50.0	1.00	0.64	+1.00	-50.6%	❌
W2	AVAX	49.3	1.00	0.63	+1.00	-22.6%	❌
W2	POL	56.3	1.00	0.64	+1.00	-23.0%	❌
W2	LINK	54.6	1.00	0.63	+1.00	26.4%	✅
W2	UNI	47.9	1.00	0.45	-1.00	-7.0%	❌
W2	LTC	23.9	1.00	0.56	+1.00	99.3%	✅
W2	DOGE	20.0	1.00	0.50	-1.00	-51.2%	❌
W3	BTC	24.7	1.00	0.52	-0.49	8.6%	✅
W3	ETH	34.5	1.00	0.56	-0.43	25.9%	✅
W3	SOL	31.3	1.00	0.51	-1.00	-15.9%	❌
W3	BNB	36.9	1.00	0.53	-0.19	7.6%	✅
W3	ADA	47.1	1.00	0.51	+0.39	-37.7%	❌
W3	XRP	42.4	1.00	0.49	-0.68	-5.2%	❌
W3	DOT	24.5	1.00	0.52	+0.29	-10.6%	❌
W3	AVAX	31.8	1.00	0.52	-0.08	-44.5%	❌
W3	POL	45.3	1.00	0.54	+0.30	-2.1%	❌
W3	LINK	38.1	1.00	0.49	-0.20	-8.8%	❌
W3	UNI	25.5	1.00	0.45	-1.00	-42.1%	❌
W3	LTC	25.1	1.00	0.52	-0.50	20.4%	✅
W3	DOGE	33.2	1.00	0.54	+0.12	-22.4%	❌

7. Limitations

Small dataset. 39 observations (13 assets × 3 windows). Risk of overfitting.
Indicators calculated at one point. Year 2 start only — no intra-period recalculation.
Grid search resolution. 5% step size means the true optimum may be between tested points.
In-sample optimization. Weights are optimized on the same data we validate on. True out-of-sample would require more windows.
Signal threshold fixed. The 50%/75% thresholds for GRID_TRADING/NEUTRAL/TREND_FOLLOWING are not optimized here.

Generated by ETS Weight Optimization Study 938 weight combinations, 39 data points, 2026-02-24