Analyzer methodology

Every score on the analyzer pages is computed from public market data. This page documents how, with backtest validation.

We blend a heuristic interpretation of William O'Neil's CAN SLIM framework with trend / quality / risk signals to produce a 0-100 composite score per stock. Below is the full methodology and — most importantly — the historical backtest of each underlying signal.

1. The framework

·CAN SLIM (40% weight in composite) — 7 letters: Current EPS, Annual ROE, New highs, Supply/demand, Leadership, Institutional, Market direction.
·Trend score (25%) — EMA50/EMA200 structure, RSI, ADX, recent return momentum.
·Quality score (25%) — ROE, operating margin, debt/equity, revenue growth.
·Risk penalty (−10%) — high volatility (ATR%), deep drawdowns, leverage, negative earnings.
·The grade buckets (STRONG/WATCH/NEUTRAL/CAUTION/AVOID) and weights are author-chosen heuristics — not optimized against a historical objective.

2. Backtest validation

We tested whether each individual signal was predictive of forward returns. Method: monthly snapshots from 2024-06-01 → 2025-11-30, 1134 (date × stock) observations across 69 companies, mean forward returns measured at +30/+90/+180 trading days.

Sector RS percentile → forward return

Bucket	n	Mean +30d	Mean +90d	Mean +180d	Median +180d
RS 1-20	110	+10.55%	+28.28%	+60.10%	+41.13%
RS 21-40	80	+9.57%	+34.32%	+83.36%	+70.38%
RS 41-60	80	+7.32%	+26.67%	+57.77%	+40.59%
RS 61-80	87	+5.91%	+26.16%	+70.15%	+52.81%
RS 81-99	95	+7.89%	+23.85%	+54.00%	+40.51%

Honest finding: sector RS percentile did NOT predict forward returns in our universe. The RS 21-40 bucket actually beat RS 81-99 by 29 percentage points over 180 days. Likely cause: small cohorts (avg 7 stocks per sector) plus a strong bull trend caused mean-reversion to dominate. We reduced RS weight in our Leadership scorer accordingly.

Price vs EMA50/EMA200 → forward return

Bucket	n	Mean +30d	Mean +90d	Mean +180d	Median +180d
↑↑ above both	298	+5.92%	+24.31%	+62.25%	+41.41%
↑↓ mixed (recovering?)	89	+8.39%	+18.84%	+53.29%	+40.65%
↓↑ mixed (rolling over?)	44	+3.84%	+26.93%	+38.49%	+32.97%
↓↓ below both	220	+4.44%	+21.32%	+51.14%	+40.57%

Validated: stocks above both EMA50 and EMA200 outperformed those below both by 11 percentage points over 180 days. This is the strongest single predictor in our universe.

EMA50/EMA200 cross → forward return

Bucket	n	Mean +30d	Mean +90d	Mean +180d	Median +180d
death ↘	350	+5.66%	+20.78%	+53.75%	+40.05%
golden ↗	301	+5.57%	+25.00%	+60.07%	+41.17%

Modest signal: golden cross (EMA50 above EMA200) outperformed death cross by 6.3 percentage points over 180 days. Consistent with classic trend-following.

Backtest generated: 2026-05-17

Why we show what didn't work

A backtest that only shows green checkmarks is suspicious. Real signals fail in real periods; selective reporting is how garbage scoring systems pass casual review. Showing the RS failure publicly is how we earn the right to claim the trend signal works.

3. Data sources

Data	Source	Refresh
Daily prices	pykrx (KR), Yahoo Finance (TW/US/NL)	daily
Market indices	Yahoo Finance (^KS11, ^TWII, ^GSPC, etc.)	daily
Quarterly fundamentals	DART (KR), MOPS (TW), SEC EDGAR XBRL (US)	quarterly + filings
Institutional flow	Naver Finance (KR), TWSE T86 (TW)	daily
13F holdings	yfinance (aggregates SEC Form 13F-HR)	quarterly, ~45d lag
Short interest	yfinance (FINRA biweekly)	biweekly
Sector RS rating	computed from daily prices, intra-sector	daily
Disclosures (filings)	DART, MOPS, SEC EDGAR	real-time
DRAM spot prices	DRAMeXchange (scraped)	daily

4. Known gaps & limitations

·Universe is limited to ~70 semiconductor stocks across KR / TW / US / NL. Not a general-purpose stock screener.
·Composite weights are author-chosen, not statistically optimized. The backtest validates signals individually, not the weighted composite as a single grade.
·Backtest covers a single ~18-month period dominated by a bull trend in semis. Predictive power may differ in other regimes (correction, sideways).
·O'Neil's "A" (annual EPS growth) is approximated by TTM ROE — we don't yet ingest 3-year annual EPS history.
·O'Neil's "N" uses "% of 52-week high" as a proxy — we don't detect proper base/breakout patterns.
·KR / TW short-interest data is not yet integrated (KRX login wall).
·DCF card uses fixed assumptions (10% discount, sector-agnostic growth) — sensitive to inputs by ±30%+ for reasonable assumption changes.
·Distribution-day count requires index volume data, which may be unreliable for KOSPI / TAIEX from Yahoo Finance.

1. The framework

·CAN SLIM (40% weight in composite) — 7 letters: Current EPS, Annual ROE, New highs, Supply/demand, Leadership, Institutional, Market direction.

·Trend score (25%) — EMA50/EMA200 structure, RSI, ADX, recent return momentum.

·Quality score (25%) — ROE, operating margin, debt/equity, revenue growth.

·Risk penalty (−10%) — high volatility (ATR%), deep drawdowns, leverage, negative earnings.

·The grade buckets (STRONG/WATCH/NEUTRAL/CAUTION/AVOID) and weights are author-chosen heuristics — not optimized against a historical objective.

2. Backtest validation

Sector RS percentile → forward return

Bucket	n	Mean +30d	Mean +90d	Mean +180d	Median +180d
RS 1-20	110	+10.55%	+28.28%	+60.10%	+41.13%
RS 21-40	80	+9.57%	+34.32%	+83.36%	+70.38%
RS 41-60	80	+7.32%	+26.67%	+57.77%	+40.59%
RS 61-80	87	+5.91%	+26.16%	+70.15%	+52.81%
RS 81-99	95	+7.89%	+23.85%	+54.00%	+40.51%

Price vs EMA50/EMA200 → forward return

Bucket	n	Mean +30d	Mean +90d	Mean +180d	Median +180d
↑↑ above both	298	+5.92%	+24.31%	+62.25%	+41.41%
↑↓ mixed (recovering?)	89	+8.39%	+18.84%	+53.29%	+40.65%
↓↑ mixed (rolling over?)	44	+3.84%	+26.93%	+38.49%	+32.97%
↓↓ below both	220	+4.44%	+21.32%	+51.14%	+40.57%

Validated: stocks above both EMA50 and EMA200 outperformed those below both by 11 percentage points over 180 days. This is the strongest single predictor in our universe.

EMA50/EMA200 cross → forward return

Bucket	n	Mean +30d	Mean +90d	Mean +180d	Median +180d
death ↘	350	+5.66%	+20.78%	+53.75%	+40.05%
golden ↗	301	+5.57%	+25.00%	+60.07%	+41.17%

Modest signal: golden cross (EMA50 above EMA200) outperformed death cross by 6.3 percentage points over 180 days. Consistent with classic trend-following.

Backtest generated: 2026-05-17

3. Data sources

Data	Source	Refresh
Daily prices	pykrx (KR), Yahoo Finance (TW/US/NL)	daily
Market indices	Yahoo Finance (^KS11, ^TWII, ^GSPC, etc.)	daily
Quarterly fundamentals	DART (KR), MOPS (TW), SEC EDGAR XBRL (US)	quarterly + filings
Institutional flow	Naver Finance (KR), TWSE T86 (TW)	daily
13F holdings	yfinance (aggregates SEC Form 13F-HR)	quarterly, ~45d lag
Short interest	yfinance (FINRA biweekly)	biweekly
Sector RS rating	computed from daily prices, intra-sector	daily
Disclosures (filings)	DART, MOPS, SEC EDGAR	real-time
DRAM spot prices	DRAMeXchange (scraped)	daily

4. Known gaps & limitations

·Universe is limited to ~70 semiconductor stocks across KR / TW / US / NL. Not a general-purpose stock screener.

·Composite weights are author-chosen, not statistically optimized. The backtest validates signals individually, not the weighted composite as a single grade.

·Backtest covers a single ~18-month period dominated by a bull trend in semis. Predictive power may differ in other regimes (correction, sideways).

·O'Neil's "A" (annual EPS growth) is approximated by TTM ROE — we don't yet ingest 3-year annual EPS history.

·O'Neil's "N" uses "% of 52-week high" as a proxy — we don't detect proper base/breakout patterns.

·KR / TW short-interest data is not yet integrated (KRX login wall).

·DCF card uses fixed assumptions (10% discount, sector-agnostic growth) — sensitive to inputs by ±30%+ for reasonable assumption changes.

·Distribution-day count requires index volume data, which may be unreliable for KOSPI / TAIEX from Yahoo Finance.