Kelly criterion sizing silently returns 0 when no historical performance data is available

[bradsmithmba]

Summary

ScoringEngine._calculate_kelly_sizing() in src/models/scoring_engine.py defaults to avg_win=1.0 and avg_loss=1.0 (1:1 win/loss ratio) and win_rate=0.5 when no historical data is provided. With these values, Kelly = (0.5 * 1 - 0.5) / 1 = 0. Position sizing is silently disabled for every strategy that has no backtest data, which is the common case since _performance_cache starts empty on every run.

Location

src/models/scoring_engine.py, score_strategies() call site (~line 108):

kelly_size = self._calculate_kelly_sizing(
    metrics.get('win_rate', 0.5),     # default 0.5
    metrics.get('avg_win', 1.0),      # default 1.0
    metrics.get('avg_loss', 1.0)      # default 1.0
)

With those defaults: b = 1.0/1.0 = 1, kelly_full = (0.5*1 - 0.5)/1 = 0. After fractional Kelly: still 0.

Impact

• ScoredStrategy.kelly_size is 0 for all strategies unless the caller explicitly provides avg_win/avg_loss in the risk_metrics dict
• RecommendationEngine uses kelly_size directly for position sizing recommendations
• The user is told "recommended position size: 0%" — which looks like no position should be taken, not like a missing data condition
• There is no warning, no fallback size, no indicator that the 0 is a data artifact

Recommended fix

Two options:

1. Return a conservative default (e.g., max_position_size * 0.5) when historical data is absent, with a flag in ScoredStrategy indicating it is a default:

   if not metrics.get('avg_win') or not metrics.get('avg_loss'):
       return self.max_position_size * 0.5  # conservative default, no historical data

2. Return None and have RecommendationEngine handle the missing-data case explicitly, surfacing it to the user rather than silently outputting 0.

Option 2 is cleaner: make missing data visible rather than masking it with a zero that reads as "don't trade."