---
created: 2026-05-09
tags:
  - FLUX - strategy - alpha
updated: 2026-05-18T11:01
---
# Algo Trading Research — What Actually Works (2026)

## The Data

- **60%** of retail algo traders show positive annual returns (vs 5-10% manual day traders)
- **$543 billion** in quant hedge fund gains in 2025 (highest ever)
- **89-95%** of ALL retail traders (manual) lose money within a year
- Automation eliminates: emotional entries, revenge trades, missed stop-losses, fatigue errors
- "Positive returns" ≠ "beating the market" — many algo traders trail buy-and-hold

## Realistic Return Expectations (Retail)

- Beginner: 5-15% annually
- Experienced with proven strategy: 15-25%
- AI-enhanced sentiment: 28-40% (high end of realistic)
- DCA bots: ~18.7% annualized (3Commas verified users)
- Grid bots: 11% monthly before fees (much less after)
- Best tested AI bot (TradeAlgo): 31.2% annualized, Sharpe 2.14, drawdown <8.4%

## Strategy Performance Benchmarks

### 1. AI-Enhanced Sentiment Trading ★ (Our primary)
- Annual return: 28-40%
- Sharpe: 1.1-1.4
- Max drawdown: 20-25%
- Win rate: 58%
- Sources: Twitter, Reddit, financial news APIs
- Key: sentiment score aggregation + price action confirmation

### 2. Mean Reversion (Our secondary)
- Annual return: 15-20%
- Sharpe: 1.5-2.0
- Max drawdown: 8-10%
- Win rate: 68%
- Key: buy overreactions, profit from reversion to mean

### 3. Momentum / Trend Following
- Annual return: 15-30% (regime dependent)
- Sharpe: 1.0-1.8
- Win rate: 40-50% (but winners 2-3x larger than losses)
- Key: identify trends early, ride them, exit when momentum fades

### 4. Sector Rotation (ETF)
- Real live result: Sharpe 1.02 (vs S&P 0.16) over 3 months
- Monthly rebalancing across sector ETFs
- Works on relative strength ranking

### 5. Statistical Arbitrage / Pairs
- Annual return: 10-20%
- Sharpe: 1.5-3.0
- Market-neutral (works in up AND down markets)
- Key: Z-score divergence between correlated assets

## Institutional Proof (Not Replicable but Proves the Model)

- Renaissance Medallion: 66% annually since 1988, >$100B total gains
- D.E. Shaw Oculus: 36.1% in 2024
- Citadel Tactical: 22.3% in 2024
- Jim Simons: "Being right 50.75% of the time is enough"

## Key Insights for FLUX

### Signal Decay
- Reuters/Benzinga article: useful for 3-5 days
- Social media post: stale within hours
- **Implication:** playbooks should have short expiration times (hours to days, not weeks)

### Combination is Key
- "The most robust models combine news (for direction) and social (for timing/extremes)"
- Mean reversion + momentum + sentiment = diversified alpha
- Multiple strategies reduce correlation to any single market regime

### What Kills Algo Traders
1. Overfitting to backtest (looks great historically, fails live)
2. Ignoring transaction costs (eats small edges)
3. No adaptation (market regimes change)
4. Over-leverage (one bad day wipes months of gains)
5. Treating bots as "set and forget" (they need monitoring)

### What Winners Do
1. Validate that live performance matches backtest BEFORE scaling
2. Use AI to process more info faster (not to predict prices directly)
3. Risk max 1-2% of portfolio per trade
4. Compound — reinvest, scale slowly
5. Treat it like engineering (hypothesis → test → validate → deploy)
6. Combine multiple uncorrelated strategies
7. Continuous monitoring + adaptation

## Applying This to FLUX

**Current strategies and their expected performance:**

| Strategy | FLUX Implementation | Expected Return | Status |
|----------|-------------------|----------------|--------|
| News Sentiment | Pipeline → playbooks → execution | 28-40% | Active (paper) |
| Mean Reversion | VTI/SPY dip-buy playbooks | 15-20% | Active (paper) |
| Event-Driven | Pre-positioned FOMC/earnings playbooks | 15-25% | Active (paper) |
| Momentum | Analyst upgrade → buy SPY | 10-15% | Active (paper) |
| Pairs/Stat-Arb | BTC/ETH, sector pairs | 10-20% | Planned |
| DCA + Rebalance | Ledger monthly contribution | 8-12% | Active (paper) |

**Combined target (blended across strategies): 20-30% annually**

**Key improvements needed:**
1. Add sentiment decay (playbooks expire faster)
2. Add position sizing based on confidence score
3. Add social sentiment signals (not just news)
4. Track Sharpe ratio in real-time (target > 1.0)
5. TensorZero learning loop validates which strategies actually produce alpha

## One Stat That Should Give Us Confidence

"AI-driven algorithms demonstrated 23% higher returns versus traditional strategies" — JP Morgan Research, 2025

We're building exactly this. The edge is real. Execution is what matters now.