Symbol Selection Asset-Type Filters

Experiment Overview

Item	Details
Date	2024-12-26
Goal	Fix symbol selection using wrong filters for asset types
Environment	alpaca_trading/selection/universe.py, config.py
Status	Success

Context

User reported error:

code

No symbols passed filters.
Stats: 0/30 passed hard filter
Qualified: 0

When running crypto-only selection with:

python

selection_config.equities.enabled = False
selection_config.crypto.enabled = True

Root Cause: Hard filters in universe.py used self.config.min_price (equity default $5.0) for ALL assets, including crypto. Most cryptocurrencies trade below $5.

The Bug

python

# universe.py - BEFORE (broken)
hard_result = apply_hard_filters(
    symbol=symbol,
    df=data,
    min_daily_volume_usd=self.config.min_avg_volume,  # Global setting
    min_price=self.config.min_price,  # $5.0 - wrong for crypto!
    max_price=self.config.max_price,
    min_bars=self.config.min_data_points,
)

SelectionConfig has both:

•Global settings: min_price = 5.0, min_avg_volume = 500_000
•Asset-specific: crypto.min_price = 0.01, crypto.min_daily_volume_usd = 50_000_000

But the hard filter used GLOBAL settings, ignoring asset-specific config!

Verified Fix

1. Detect Asset Type and Use Correct Config

python

# universe.py - AFTER (fixed)
# Determine asset type and use appropriate settings
is_crypto = symbol.endswith('USD') or '/' in symbol
if is_crypto:
    asset_config = self.config.crypto
else:
    asset_config = self.config.equities

# Apply hard filters with asset-type-specific settings
hard_result = apply_hard_filters(
    symbol=symbol,
    df=data,
    min_daily_volume_usd=asset_config.min_daily_volume_usd,
    min_price=asset_config.min_price,
    max_price=self.config.max_price,
    min_bars=self.config.min_data_points,
)

2. Relax Crypto Defaults

python

# config.py - AssetTypeConfig for crypto
crypto: AssetTypeConfig = field(default_factory=lambda: AssetTypeConfig(
    enabled=True,
    max_allocation=0.20,
    max_positions=3,
    min_volatility=0.15,   # 15% annual volatility minimum
    max_volatility=2.00,   # 200% max (crypto is volatile)
    min_daily_volume_usd=1_000_000,   # $1M (was $50M - too strict!)
    min_price=0.0001,      # Allow very low prices (was $5!)
))

3. Add Diagnostic Output

python

# When selection fails, show WHY
print(f'Diagnostic info - Sample of failed symbols:')
for sym, analysis in list(result.analyses.items())[:10]:
    reasons = analysis.exclusion_reasons
    print(f'  {sym}: {reasons[0]}')

Failed Attempts (Critical)

Attempt	Why it Failed	Lesson Learned
Crypto min_price = $5.0	Most cryptos trade below $5	Use asset-type-specific settings
Crypto min_volume = $50M	Many valid cryptos under $50M daily volume	$1M is reasonable minimum
Global config.min_price	Applied equity settings to crypto	Detect asset type first
No diagnostic output	Users couldn't see WHY selection failed	Always show failure reasons

Asset-Type Detection

python

# Simple heuristic for crypto detection
is_crypto = symbol.endswith('USD') or '/' in symbol

# Examples:
# BTCUSD  -> crypto (ends with USD)
# BTC/USD -> crypto (contains /)
# AAPL    -> equity (neither)

Recommended Settings by Asset Type

Equities

python

selection_config.equities.min_price = 5.0             # $5 minimum
selection_config.equities.min_daily_volume_usd = 10_000_000  # $10M
selection_config.equities.min_volatility = 0.05       # 5% annual
selection_config.equities.max_volatility = 0.60       # 60% annual

Crypto

python

selection_config.crypto.min_price = 0.0001            # Allow any price
selection_config.crypto.min_daily_volume_usd = 1_000_000  # $1M
selection_config.crypto.min_volatility = 0.15         # 15% annual
selection_config.crypto.max_volatility = 2.00         # 200% annual

Key Insights

Why Crypto Needs Different Settings

Parameter	Equity	Crypto	Why Different
min_price	$5.00	$0.0001	Many cryptos < $1
min_volume	$10M	$1M	Smaller crypto market
max_volatility	60%	200%	Crypto is volatile

Alpaca Crypto Availability

•Alpaca has ~30 crypto pairs
•All end with USD (e.g., BTCUSD, ETHUSD)
•Some trade at fractions of a cent (e.g., SHIBUSD)

Common Selection Errors

Error	Likely Cause	Fix
`0/N passed hard filter`	Wrong asset-type settings	Check min_price for crypto
`Insufficient data`	min_data_points too high	Lower to 300 for hourly data
`volume` filter fails	min_daily_volume_usd too high	$1M for crypto, $10M equity

Files Modified

code

alpaca_trading/selection/universe.py:
  - Lines 260-275: Detect asset type, use correct AssetTypeConfig

alpaca_trading/selection/config.py:
  - Lines 85-93: Relaxed crypto defaults

notebooks/VSCode_Colab_Training_NATIVE.ipynb:
  - cell-16: Added diagnostic output on failure

References

•alpaca_trading/selection/config.py: AssetTypeConfig dataclass
•alpaca_trading/selection/filters/hard_filters.py: Filter implementations
•alpaca_trading/selection/universe.py: Selection orchestration
•.skills/plugins/trading/symbol-selection-statistical/: Statistical selection guide