Broker Order Limitations (Alpaca)

Experiment Overview

Context

Alpaca broker has specific limitations on order types. The most important one for this trading system:

Crypto shorting is NOT supported by Alpaca. Stock shorting IS allowed.

This distinction is critical because:

1. Model generates SELL signals for both crypto and equities
2. SELL signal on crypto = blocked by broker limitation
3. SELL signal on equities = allowed (short selling)

Verified Implementation

Crypto Short Gate (Correct)

# In live_trader.py (~line 1335)
# Only block shorts for crypto, not equities
if desired_side == -1 and asset_type == AssetType.CRYPTO and state.side <= 0:
    # Trying to go short on crypto when not already long
    gate_statuses[symbol] = GateStatus(
        final_status='BLOCKED',
        blocking_gate='crypto_short',
        crypto_short_gate=False,  # Failed
    )
    continue

Key Logic Points

# desired_side == -1: Model wants to SELL
# asset_type == AssetType.CRYPTO: Asset is cryptocurrency
# state.side <= 0: Not currently holding a long position

# All three conditions must be true to block:
# 1. SELL signal (-1)
# 2. Is crypto (not equity)
# 3. Would be opening a short (not closing a long)

What IS Allowed

Gate Status in Dashboard

When crypto short is blocked:

gate_status = {
    'final_status': 'BLOCKED',
    'blocking_gate': 'crypto_short',
    'crypto_short_gate': False,
    # ... other gates may be True
}

Failed Attempts (Critical)

Alpaca Order Types Reference

Supported Order Types

Crypto-Specific Limitations

1. No shorting: Cannot borrow and sell crypto
2. 24/7 trading: No market hours restrictions
3. Symbol format: Ends with USD (e.g., BTCUSD, ETHUSD)
4. Fractional minimum: Varies by crypto (check API)

Equity-Specific Features

1. Short selling: Allowed with margin account
2. Market hours: 9:30 AM - 4:00 PM ET (regular)
3. Extended hours: 4:00 AM - 8:00 PM ET (with flag)
4. Pattern day trader: Rules apply if >3 day trades in 5 days

Key Insights

Detecting Asset Type

# Simple heuristic used in live_trader.py
def is_crypto(symbol: str) -> bool:
    return symbol.endswith('USD')

# Or use Alpaca's asset class
from alpaca.trading.enums import AssetClass
asset = api.get_asset(symbol)
is_crypto = asset.asset_class == AssetClass.CRYPTO

Why Not Block at Model Level?

The model is trained on both crypto and equity data. It learns:

• When to go long (BUY = +1)
• When to go short (SELL = -1)
• When to stay out (HOLD = 0)

Blocking shorts in the model would:

• Reduce training efficiency
• Create inconsistent behavior
• Miss valid equity short signals

Better approach: Let model predict freely, filter at execution gate.

Capital Shift Interaction

When equity markets close, capital shifts to crypto. During this time:

• More crypto signals generated
• All crypto SELL signals blocked
• Only crypto BUY signals executed
• Effective strategy: crypto long-only overnight

Files Modified

scripts/live_trader.py:
  - Line 1335: Crypto short gate check
  - Uses AssetType enum for asset classification
  - Gate status stored in gate_statuses dict

References

• Alpaca Crypto Trading
• Alpaca Short Selling
• scripts/live_trader.py: Order execution logic (lines ~1300-1400)
• alpaca_trading/risk/capital_shift.py: Equity→crypto shift logic