Evaluation Criteria

This document defines the specific criteria used to evaluate stocks for the Buy-Write / Mean Reversion strategy.

Strategy Overview

Goal: Acquire shares near historical support/lows to sell covered calls. Target stability or moderate upside.

What We Want:

Entry near a floor — Stock unlikely to fall significantly further
High volatility — Fat option premiums compensate for risk
Sideways or up — Not continued decline

Primary Risk: "The Falling Knife" — continued rapid decline after entry. Covered calls provide almost NO downside protection (COVID: covered call strategies dropped 29% vs 32% for the market).

Two-Layer Evaluation

Evaluation happens in two layers:

Screener Layer (Mechanical) — Data filtering, calculations, thresholds. Handled by code.
AI Layer (Judgment) — Context, reasoning, synthesis. Requires intelligence, not formulas.

The AI layer should NOT recalculate data the screener already provides. Its job is to interpret context and form opinions.

Screener Layer: Mechanical Filters

Hard Exclusions

Screener flags these automatically. If ANY are true, requires AI judgment on severity:

Burning cash with <12 months runway — Unprofitable + low cash = dilution spiral
Cluster insider selling — 2+ executives selling within 30 days (see Insider Signal section)

Positive Signals

Screener calculates these — strong evidence of support:

Cluster insider buying — 2+ insiders buying within 90 days. Academic research shows 6-10% annual outperformance.
Insider Signal Score ≥ 3 — Weighted by role (CFO highest signal) and timing proximity.

Neutral Data Points

Screener provides these for AI interpretation:

Price vs 200-day MA percentage
RSI, Beta, Short Interest
Analyst targets and consensus
Smart Score (TipRanks — backtested only, not independently validated)

AI Layer: The Three Questions

The AI must form a judgment on each of these. Not calculate — reason.

Question 1: Is This Done Falling?

This is the most critical question. Look for a "cluster of indicators" — not just one signal.

Bottoming signals (research-backed):

Insider buying — "Broader insider buying signals a market bottom" (strongest signal)
RSI divergence — Price makes lower low, RSI doesn't → "hints selling pressure is fading"
Extreme bearish sentiment — When everyone is panicking, that's often the floor
Exhaustion — "Sellers get exhausted, stocks start going up even after bad news"

AI must reason about:

WHY is this stock down? Is the reason temporary or permanent?
Is bad news already priced in, or are there more shoes to drop?
Are insiders buying into weakness (bullish) or silent (neutral) or selling (bearish)?

Question 2: Will I Get Paid for the Risk?

High volatility = fat premiums. But excessive volatility may signal chaos.

Research context:

Calm market (IV ~25%) → ~7% annualized yield
Volatile market (IV ~55%) → ~21% annualized yield
Sweet spot: "liquid, high-quality stocks with stable or moderate upward movement and consistently high option premiums"

AI must reason about:

Is the volatility from temporary uncertainty (good) or structural problems (bad)?
Are there upcoming catalysts (earnings, FDA, etc.) that could cause unpredictable moves?
Is options liquidity sufficient to exit/roll positions?

Question 3: What's the Likely Direction?

We want sideways or up. Mean reversion works best in range-bound markets.

Research context:

"Covered calls outperform in bear, range-bound, and moderate bull markets but lag during strong bull rallies"
Larry Connors' mean reversion: only buy oversold stocks when price is ABOVE 200-day MA (long-term trend intact)
Below 200-day MA = long-term downtrend, higher risk of continued decline

AI must reason about:

Is the long-term trend still intact, or has it broken down?
Is this a broken stock (temporary) or a broken company (permanent)?
Are there positive catalysts coming that could drive recovery?
Is sentiment overly negative (contrarian opportunity) or appropriately negative (things are actually bad)?

Data Sources

Data Point	Source	Notes
Insider Transactions	SEC Form 4 Filings	Filtered to "informative" (P=purchase, S=sale). Excludes grants, exercises, tax withholding.
Analyst Ratings	TipRanks	Includes price targets. Top analysts tracked at 65-88% success rates.
Smart Score	TipRanks	Backtested score 1-10. Claims 390% return since 2016 for score 10 vs 225% S&P. Not independently validated.
Hedge Fund Activity	13F Filings	Academic research shows ~10% annual excess returns from informed demand.
Price/Technicals	Market Data	200-day MA, RSI, 52-week range, short interest, beta.

Research References

Key findings that inform this framework:

Insider buying outperformance: 6-10% annually (multiple academic studies)
Cluster buying signal: 3+ insiders yields ~2% monthly excess returns vs ~0.5% baseline (TipRanks research)
Insider selling NOT predictive: Individual selling does not predict underperformance (InsideArbitrage, academic research)
Hedge fund 13F: ~10% annual excess returns from informed demand (ScienceDirect study)
Covered call protection myth: COVID showed 29% drop vs 32% for market — minimal protection (ProShares)
Mean reversion timing: Larry Connors 2-period RSI — only enter when above 200-day MA (long-term trend intact)

Verdict Definitions

Verdicts require AI judgment, not just threshold checks. The AI must explain its reasoning.

Verdict	Meaning	AI Must Justify
ACCUMULATE	Strong conviction — insider buying provides floor signal	Cluster insider buying (2+ insiders) AND AI believes downside is limited AND long-term trend intact or stabilizing
BUY-WRITE	Good setup — no major red flags, decent premium opportunity	AI answers YES to all three questions No cluster insider selling Volatility sufficient for meaningful premium
AVOID	Too risky — likely to keep falling or fundamental problems	AI must explain WHY — not just cite data points Examples: "Founders cashing out after 10x run while business fundamentals deteriorating" "Falling knife with no bottoming signals and more bad news likely"
WATCH	Interesting but not ready — needs more evidence or better entry	AI explains what would change the verdict Example: "Would upgrade to BUY-WRITE if insider buying emerges or price stabilizes above support"

Insider Signal Research

Academic research on what insider transactions actually predict:

Signal	Predictive?	Research Finding
Insider Buying	YES — Strong	6-10% annual outperformance. "Single strongest signal of a bottom."
Cluster Buying (2+ insiders)	YES — Very Strong	~2% monthly excess returns vs ~0.5% baseline = 4× multiplier
Individual Selling	NO — Weak	Not predictive. Insiders sell for many reasons: taxes, diversification, planned sales (10b5-1), liquidity needs.
Cluster Selling (2+ executives)	YES — Meaningful	Multiple executives selling close together IS a red flag. Requires AI interpretation of context.

Key insight: Dollar amounts alone are meaningless. A CEO selling $5M after a 10× run is very different from a CEO selling $5M into a decline. Context matters.

Key Definitions

Term	Definition
Informative Transaction	Open market buy (P) or sell (S) by an insider. Signals conviction. Excludes grants, tax withholding, option exercises.
Uninformative Transaction	Grants (A), tax withholding (F), option exercises (M), gifts (G). Routine activity, not a conviction signal.
Cluster Buying/Selling	2+ unique insiders transacting in the same direction within a short period (30-90 days).
Falling Knife	Stock trading significantly below its 200-day moving average. Severity matters — 10% vs 40% are very different.
Mean Reversion Setup	Stock oversold (low RSI) but long-term trend intact (above 200-day MA). Larry Connors research.

Screener Data Points

The screener provides these. AI interprets them — they are not automatic pass/fail.

Data Point	Bullish	Neutral	Bearish
Price vs 200-Day MA	Above (trend intact)	Slightly below (oversold?)	Far below (falling knife?)
Cluster Insider Buying	2+ buyers (strong floor signal)	1 buyer (some signal)	None
Cluster Insider Selling	None	1 seller (probably routine)	2+ sellers (red flag — AI must interpret context)
RSI	Oversold + divergence	Neutral	Oversold, no divergence
Short Interest	<10%	10-20%	>20% (crowded short)
Smart Score	8-10 (institutions bullish)	5-7	1-4 (institutions bearish)

Note: Smart Score is backtested data from TipRanks — not independently validated. Treat as one input, not gospel.

What Good AI Analysis Looks Like

Bad (mechanical): "Insider selling of $12M. AVOID."

Good (reasoning): "CEO sold $12M after stock ran from $0.57 to $11 (19×). This looks like a founder taking chips off the table after a massive win, not a bearish signal. The question is whether the valuation is justified — at $4.7B market cap for a wireless/drone company, this needs real revenue growth to sustain. Watching for Q1 earnings to validate."

Bad (mechanical): "Smart Score 2. AVOID."

Good (reasoning): "Smart Score of 2 means institutions, analysts, and technicals are all negative. But TTD is still the leading independent DSP — is this a broken stock or a broken company? Need to understand if the ad-tech headwinds are cyclical (opportunity) or structural (avoid). The 39% drop below MA with no insider buying suggests waiting for stabilization."