Custom Factors

Build your own proprietary factors.

Building Custom Factors

Step 1: Hypothesis

Start with an economic intuition: - "Companies with improving margins will outperform" - "High insider buying signals confidence" - "Low short interest means less downside"

Step 2: Implementation

Text Only

// Margin improvement factor
signal margin_improvement:
  current_margin = gross_margin
  past_margin = lag(gross_margin, 4)  // 4 quarters ago

  improvement = current_margin - past_margin

  emit zscore(improvement)

Step 3: Validation

Text Only

portfolio test_factor:
  weights = rank(margin_improvement).long_short(top=0.2, bottom=0.2)
  backtest walk_forward(train_years=5, test_years=1) from 2010-01-01 to 2024-12-31

Example Custom Factors

Earnings Surprise

Text Only

signal earnings_surprise:
  // Actual vs estimate
  surprise = (actual_eps - estimated_eps) / abs(estimated_eps)

  emit zscore(surprise)

Analyst Revision

Text Only

signal revision_momentum:
  current_estimate = analyst_eps_estimate
  past_estimate = lag(analyst_eps_estimate, 21)

  revision = (current_estimate - past_estimate) / abs(past_estimate)

  emit zscore(revision)

Price-to-52-Week High

Text Only

signal nearness_to_high:
  high_52w = rolling_max(prices, 252)
  nearness = prices / high_52w

  // Near high = momentum, far from high = value opportunity
  emit zscore(nearness)

Combining Custom Factors

Text Only

signal proprietary:
  f1 = margin_improvement
  f2 = earnings_surprise
  f3 = revision_momentum

  // Equal weight or IC-weighted
  emit (f1 + f2 + f3) / 3

Testing Checklist

• Economic rationale
• Out-of-sample testing
• Transaction cost analysis
• Factor correlation check
• Capacity analysis

See Also

• Classic Multi-Factor
• Factor Timing