Fundamental Analysis

Fundamental Analysis (the Analyst Agent)

This skill makes the analyst concrete: its inputs, sources, screens, outputs, and the non-negotiable rule that gates everything — a signal does not trade until a backtest says it should. It also encodes an honest, evidence-based verdict on what the analyst is for.

The verdict that shapes everything (read first)

We backtested the investable versions of the major stock-selection methods vs SPY over each ETF's full live history (backtests/fundamental_screens_backtest.py):

Only 1 of 10 (momentum) beat SPY on return; 0 of 10 beat it on Sharpe — including the ETF that literally implements Morningstar's stock picking. This is consistent with SPIVA (most active selection lags a cheap index, especially in a mega-cap-led bull).

Therefore the analyst's job is NOT "use Morningstar to find undervalued stocks that beat the market." That is a low-base-rate bet. The analyst has three jobs where it does add value:

1. Pick the defensive equity sleeve (the screens cut 2022 drawdowns ~in half: COWZ −8%, RPV −11%, SCHD −15%, USMV −17% vs SPY −24.5%) — implemented as cheap ETFs, not a hand-rolled stock book. 2. Provide valuation/earnings context to the regime + PM agents (CAPE, concentration, credit, earnings revisions, sector breadth). 3. Validate every proposed signal through the backtest gate below.

What the analyst reads (data sources)

Rule: use yfinance/SEC for current screening and context; use a point-in-time vendor (Sharadar/SimFin) the moment you backtest any fundamental screen — otherwise survivorship + look-ahead bias will manufacture fake alpha (this is exactly the trap the repo's old price-proxy "Morningstar" backtest fell into — it lost to VOO by −4.3%/yr).

The screens the analyst can compute (if/when justified)

• Value: earnings yield (EBIT/EV), P/B, P/E, P/FCF. Magic Formula (Greenblatt): rank by

earnings yield + return on capital, buy top 20-30.

• Quality: ROE/ROIC, low debt, stable earnings, Piotroski F-score (9-point).
• FCF yield ("cash cows"), shareholder yield (buybacks + dividends).
• Moat/fair value: Morningstar-style (proxy via MOAT ETF unless you license the data).
• Momentum: 12-1 month total return (the one factor that beat SPY here).

The backtest gate (the answer to "does it backtest or not?")

YES — mandatorily. No screen, factor, or signal is allowed to allocate capital until it: 1. Beats its benchmark (usually SPY/VOO or the sleeve it would replace) on risk-adjusted terms (Sharpe/Calmar), net of costs and turnover, over an out-of-sample window; 2. Uses point-in-time, survivorship-safe data; 3. Survives the crisis windows (2008/2020/2022) without an unacceptable drawdown; 4. Is robust to small parameter changes (no overfitting; deflate Sharpe for # of trials).

If it fails the gate → the default is the cheap ETF (e.g., USMV/QUAL/COWZ) or just the index. "Stock-picking has to earn its place; the index is the bar."

Outputs (contract)

{
  "as_of": "2026-05-29",
  "defensive_sleeve_choice": {"ticker": "USMV", "reason": "min-vol ETF; -9% DD in 2023-26 vs SPY -19%; beats a hand-rolled screen net of cost"},
  "context": {"sp500_cape": 41.6, "top10_weight": 0.40, "hy_oas_bps": 320, "earnings_revisions": "flat"},
  "proposed_signals": [{"name": "magic_formula_top30", "status": "REJECTED", "reason": "did not clear backtest gate net of costs vs VOO"}],
  "stock_basket": null
}

When the user insists on a stock-picking sleeve

Cap it small (≤10-15% of equity), require it to clear the backtest gate on point-in-time data, accept single-name and tax/turnover costs, and benchmark it honestly against the ETF it replaces. Know the base rate: most such sleeves lag the index. Size it as a satellite, never the core.

Optional Morningstar overlay (manual, context-only)

If you already have legitimate Morningstar access (personal/Investor account), you may use moat + fair-value ratings as one manual context lens — a quality/valuation sanity-check before a discretionary buy — NOT as an automated signal or the core.

• Why only context: Morningstar's ratings are public information, so they are already in the

price (semi-strong EMH; Bogle, Malkiel). The productized version of exactly this process (the MOAT ETF) lagged SPY (13.5% vs 14.7%) — consuming their conclusions adds no alpha.

• Do NOT scrape/spoof their API for the pipeline: it breaks their ToS, is brittle (stale/wrong

data is the worst failure mode for a fund), and your account only exposes today's rating — there is no point-in-time history, so a Morningstar-driven screen can never clear the backtest gate.

• A DCF "fair value" is an uncertain range, not a precise number (Damodaran) — weight it accordingly.
• Net: optional manual gut-check on individual names; the default remains the cheap ETF, and the

backtest gate still governs any capital.

Hand-offs

Feeds the defensive sleeve choice + valuation context to portfolio-construction and regime-detection; routes any candidate signal to the backtest harness before risk-management ever sees it. Momentum signals overlap with trend-following.