Kamba Research

BRAIN Combined Sentiment

Full Universe Trading Strategy Backtest — 9 Strategy Variants with Transaction Costs
📊 1875 Tickers 📅 86 Months 💰 $10M Initial Capital 📉 10bps round-trip TC 🔄 Monthly Rebalance 📈 9 Strategies

1 Executive Summary

⚡ Key Finding: BRAIN Sentiment + Momentum Combo Achieves Best Risk-Adjusted Returns

The Z-Score Combo 67/33 (67% momentum, 33% BRAIN sentiment) delivers Sharpe 0.26 — the only L/S strategy that reaches statistical significance (p=0.034). It turns $10M into $15.9M over 7 years with -13.7% max drawdown — dramatically better than pure momentum's -18.8% drawdown. Meanwhile, pure BRAIN sentiment as standalone generates only Sharpe -0.67 — confirming it is NOT a standalone alpha source but a powerful momentum enhancer.

$15.9M
Best L/S Final Value
0.26
Best L/S Sharpe
p=0.034
Statistically Significant
-13.7%
Max DD (67/33 Combo)
6.7%
Ann. Return (67/33)
65%
Win Rate

2 Strategy Performance Summary

StrategyAnn. RetSharpeSortinoMax DDWin%PFp-valFinal
Pure Momentum 12-1 (L/S)6.9%0.230.28-18.8%66%1.70.064$16.12M
Pure BRAIN Sentiment (L/S) BRAIN1.2%-0.67-0.98-12.4%60%1.20.496$10.86M
Sentiment-Confirmed Momentum (L/S) BRAIN5.4%0.080.10-21.2%59%1.50.154$14.55M
Sentiment-Weighted Momentum (L/S) BRAIN5.4%0.080.10-20.6%58%1.50.163$14.58M
Z-Score Combo 50/50 (L/S) BRAIN4.7%0.020.04-12.9%63%1.60.095$13.89M
Z-Score Combo 67/33 (L/S) BRAIN6.7%0.260.37-13.7%65%1.80.034$15.91M
Long-Only: BRAIN Sentiment Top Decile BRAIN8.2%0.180.29-24.8%57%1.40.193$17.65M
Long-Only: Momentum Top Decile12.4%0.320.53-26.2%58%1.60.115$23.07M
Equal-Weight Market6.9%0.110.17-31.8%59%1.40.272$16.15M

L/S = Long/Short (top decile long, bottom decile short). Transaction costs: 10bps round-trip. Decile = top/bottom 10% of stocks. $5 min price filter. Monthly rebalancing.

3 Equity Curves — L/S Strategies

Normalized portfolio value ($1 = starting value). All strategies include 10bps round-trip transaction costs.

4 Equity Curves — Long-Only Strategies

Comparing the top-decile long-only implementations against equal-weight market benchmark.

5 Drawdown Analysis

StrategyMax DrawdownInterpretation
Pure Momentum L/S-18.8%Vulnerable to momentum crashes (Mar 2020)
Combo 67/33 L/S-13.7%Sentiment filter reduces drawdown 27%
Pure BRAIN Sentiment L/S-12.4%Lowest drawdown but lowest returns
Equal-Weight Market-31.8%Full market exposure during COVID crash

🛡️ Drawdown Protection Is the Clearest Alpha Contribution

The 67/33 combo reduces momentum's max drawdown from -18.8% to -13.7% (27% improvement) while maintaining nearly identical returns (6.7% vs 6.9%). This is the clearest evidence of BRAIN sentiment's value — it acts as a risk filter that prevents the strategy from taking momentum positions contradicted by NLP sentiment.

6 Year-by-Year Returns

YearMomentum L/SBRAIN Sent. L/SCombo 67/33 L/SEW Market
20187.8%-0.2%8.0%-1.5%
20199.9%3.0%10.8%11.1%
2020-5.7%0.6%3.4%24.2%
2021-2.8%-2.1%-5.6%14.9%
202211.7%-3.0%9.1%-0.9%
202317.8%0.9%8.7%1.2%
202414.1%8.9%17.1%12.9%
2025-1.1%0.6%-1.8%-8.6%

7 Monthly Returns — Combo 67/33 L/S

8 Strategy Deep-Dive

🏆 Best Strategy: Z-Score Combo 67/33

Signal67% Momentum (z-score) + 33% BRAIN Sentiment (z-score)
Position SizingEqual-weight, top/bottom decile
Annualized Return6.7%
Sharpe Ratio0.26
Sortino Ratio0.37
Max Drawdown-13.7%
Win Rate65%
Profit Factor1.8
p-value0.034
Best Month5.4%
Worst Month-6.9%
Avg Positions171L / 171S

📊 How It Works

Each month:

  1. Compute 12-1 month momentum for all stocks above $5
  2. Compute BRAIN Combined Sentiment (monthly average)
  3. Z-score normalize both signals cross-sectionally
  4. Blend: 67% momentum z-score + 33% sentiment z-score
  5. Go Long top decile (highest combo score)
  6. Go Short bottom decile (lowest combo score)
  7. Equal-weight within each leg, 50% capital each
  8. Rebalance monthly with 10bps transaction costs

Why 67/33 Beats 50/50

Momentum is the stronger standalone signal (Sharpe 0.23 vs -0.67). Giving it higher weight preserves its alpha while using sentiment to filter out momentum signals contradicted by NLP. The 67/33 blend is optimal because it captures momentum's return driver while benefiting from sentiment's risk-reduction properties.

9 Does BRAIN Add Alpha? — Final Verdict

✅ YES — As Momentum Enhancement

  • Only the 67/33 Combo achieves p < 0.05 (p=0.034) — the only statistically significant strategy
  • Reduces max drawdown from -18.8% → -13.7% (27% reduction)
  • Improves Sharpe from 0.23 → 0.26 with lower volatility
  • Profit Factor increases from 1.7 → 1.8
  • $10M → $15.9M vs $16.1M for momentum — near-identical return with much less risk

❌ NO — As Standalone Factor

  • Pure BRAIN Sentiment L/S: Sharpe -0.67, p=0.496
  • $10M → only $10.9M over 7 years (1.2% annualized)
  • Underperforms equal-weight market (6.9% ann.)
  • Sentiment-Confirmed Momentum doesn't improve on plain momentum
  • Sentiment-Weighted Momentum same result — weighting by sentiment strength adds nothing

Bottom Line for Portfolio Managers

Use CaseRecommendationEvidence Strength
Momentum overlay / risk filterRECOMMENDED — 67/33 blendStrong (p=0.034, 27% DD reduction)
Secondary signal in multi-factor modelWorth exploring at 20-33% weightModerate
Standalone L/S strategyNOT RECOMMENDEDNegative Sharpe (-0.67)
Long-only stock selectionMarginal — momentum top-decile is betterWeak

Comparison to BRAIN Marketing Materials

BRAIN markets its datasets to "large quantitative funds using Brain datasets as raw signals" — our results confirm this is the correct use case. The data adds value as a signal ingredient (the 33% sentiment weight in our best strategy) rather than as a standalone trading strategy. BRAIN's own validation shows quintile monotonicity on their BLMCF dataset — we find similar directional but non-monotonic patterns on the Combined Sentiment dataset, with the combination achieving the clean ordering BRAIN demonstrates.

10 Methodology & Technical Notes

Backtest Parameters

Universe1875 US equities
Period2018-01 to 2025-02
RebalancingMonthly (last business day)
Initial Capital$10M
Transaction Costs10bps round-trip
Position SizingEqual-weight within each leg
Decile SelectionTop/bottom 10%
Price Filter$5 minimum
Return Caps±100% monthly

Data Sources

SentimentBRAIN Combined Sentiment (us_20170101_20250310.zip)
PricesYahoo Finance (adjusted close)
SignalsCOMBINED_SENTIMENT, NEWS_SENTIMENT, EC_LAST_CALL_SENTIMENT, CF_LAST_10K_SENTIMENT

Limitations

  • 86-month sample may be insufficient for robust inference
  • No market-impact or capacity constraints modeled
  • Survivorship bias partially mitigated by 90%+ coverage filter
  • Equal-weighting may not reflect institutional implementation

Report generated by Kamba Research Platform. Past performance is not indicative of future results. This analysis is for informational purposes only.