Our Process

Data arrives from Bloomberg (consensus estimates, trailing fundamentals, bank-specific line items like loan loss provisions) and Financial Modeling Prep (6 years of historical statements, prices, insider activity). We force USD currency normalization on ADRs. Bloomberg returns local currency by default, which silently corrupts forward multiples on names like TSM and BABA for anyone who doesn't correct for it. We don't trust the inputs until they're cross-validated.

Quality is the weighted average of 4 pillar scores: Profitability (45%), Growth (20%), Safety (15%), Shareholder Returns (20%). Blends roughly 53% forward-looking signals (analyst consensus) with 47% realized results. Valuation combines 10 forward and trailing multiples, weighted ~67% forward. Both are z-scored within 5 peer groups (standard, financial, energy, utility, REIT) so banks are compared to banks, not technology companies. Banks get their own factor set: net interest margin, efficiency ratio, and provisions-to-NII replace non-bank metrics that don't apply. Most factor models skip this step; it's where we refuse to cut corners.

Pure Euclidean Q+V scoring treats a stock at 'minus-3 sigma expensive with plus-1 quality' the same as 'plus-1 cheap with plus-3 quality.' That math is sign-indifferent, and it's a trap. It rewards the momentum-driven speculative premium as heavily as disciplined cheap quality. Our composite explicitly penalizes expensive stocks linearly, rewards cheap stocks quadratically, and caps the net score with a quality-floor filter. This is the rule that separates us from most systematic strategies: we decline the crowd-favorite trade every time quality-at-any-price wins the backtest.

Every zone candidate receives a qualitative assessment. Is the CEO a proven capital allocator? Is the competitive moat widening or narrowing? Are there regulatory or macro headwinds the model can't see? This is where three decades of market experience earns its weight. Equally important: we've written rules for when we override the model and when we don't. The rule-adoption matters as much as the rules themselves.

Position sizes range from 5-10% per stock with a 13% maximum. We do not use leverage. Net exposure is typically 60-100% long. Concentration is an advantage at our scale. We can enter and exit 10 positions without the market impact that a $10B fund would suffer. The portfolio is designed to compound steadily, not swing for the fences.

Each morning, our system produces a daily research note reviewing every portfolio holding against the refreshed model. When a company exits the Alpha Zone or its pillar scores deteriorate materially, the position is flagged for review. We also subject our weight choices to walk-forward cross-validation, optimizing on one window of history and testing on the next, so our process isn't fitting to a single lucky sample period. Turnover is approximately 15% per year; patient, but not passive.