Relative-Strength Watchlists: Point-in-Time Ranking and Research Controls

A relative-strength watchlist ranks dated total returns against explicit benchmarks; reproducibility requires point-in-time universes, exact return definitions, benchmark discipline, event review, liquidity controls, and implementation costs.

For educational purposes only; not investment advice. Investing may result in loss.

Direct answer

A relative-strength watchlist is a dated, reproducible ranking of securities against explicit benchmarks over specified return windows. It is a research queue, not a forecast, recommendation, or complete trading strategy. The ranking says which securities outperformed on the chosen historical definition; it does not establish why they moved, whether the move will persist, whether the information is already priced, or whether a position can be implemented at acceptable risk and cost.

Several valid measures answer different questions. Arithmetic excess return is ARᵢ,b = Rᵢ − Rb in percentage points. Geometric relative return is GRᵢ,b = (1 + Rᵢ) ÷ (1 + Rb) − 1, which compares ending wealth exactly. A relative-strength ratio chart can normalize two total-return wealth indexes as RSᵢ,b,t = (Wᵢ,t ÷ Wᵢ,0) ÷ (Wb,t ÷ Wb,0). A cross-sectional percentile or rank compares one security with the full eligible universe. Store the formula, not just a column labeled “relative strength.”

This use of relative strength is not the Relative Strength Index (RSI), a bounded oscillator calculated from one instrument’s own recent gains and losses. It is also not identical to academic cross-sectional momentum, which requires specified formation, skip, holding, universe, breakpoint, weighting, and rebalancing rules, or time-series momentum, which compares an asset with its own past rather than ranking it against peers.

Building a reproducible watchlist

Write the question and version first. Define research purpose, as-of timestamp, decision horizon, formula version, universe, benchmark hierarchy, return type, windows, score weights, eligibility gates, event flags, refresh frequency, and who can change the rules. Freeze them before viewing ranks.
Reconstruct the point-in-time universe. Use securities actually eligible on each historical date, including delisted, bankrupt, acquired, suspended, and later-renamed names. Preserve listing date, share class, primary venue, country, currency, security type, free float, and membership history. Do not backfill today’s survivors into old tests.
Build aligned total returns. Match start and end timestamps, exchange calendars, time zones, currencies, dividends, splits, rights, spin-offs, special distributions, mergers, stale prices, and missing observations. Distinguish price, gross total, net total, and currency-hedged returns. A vendor’s adjusted close is an input that still requires definition and audit.
Calculate transparent signals. Produce absolute return, broad-market arithmetic and geometric relative return, sector or industry relative return, rank or percentile, and any multi-window composite separately. Specify winsorization, volatility scaling, minimum history, ties, missing windows, overlapping observations, standardization universe, and whether larger scores are always better.
Decompose benchmark choice. Compare a security with a broad market, its point-in-time sector or industry, and a defensible peer set without selecting whichever benchmark makes the result look strongest. Separate market, industry, and security-specific relative moves while recognizing that classifications, weights, currencies, and business mixes differ.
Attach primary research and risk. Record the filing or source available at the timestamp, earnings and guidance dates, estimate revisions, corporate actions, one-day gaps, valuation, profitability, leverage, dilution, borrow availability, liquidity, spread, volatility, drawdown, catalyst, invalidation condition, and unanswered question. A score is a pointer to work, not evidence of causality.
Test implementation and governance. Define selection count, buffer, entry, exit, rebalance, position sizing, constraints, order type, turnover convention, spread, commissions, market impact, taxes, borrow, capacity, and no-trade rules. Evaluate out of sample against a timestamped baseline and retain every version, override, rejected name, delisting, and realized cost.

Worked examples

Arithmetic and geometric relative return differ. Over one matched window, Stock A has total return 35.00% and the benchmark has 5.00%. Arithmetic strength is 35.00% − 5.00% = 30.00 percentage points. Exact relative wealth is 1.35 ÷ 1.05 − 1 = 28.5714%, so a normalized relative-strength ratio rises from 1.0000 to 1.2857. The two values use different units and must not share an unlabeled score column.
Positive relative strength can accompany a loss. A stock returns −8.00%, its sector returns −12.00%, and the broad market returns −15.00%. Stock-versus-sector arithmetic strength is −8.00% − (−12.00%) = 4.00 percentage points; sector-versus-market strength is −12.00% − (−15.00%) = 3.00 points; stock-versus-market strength is 7.00 points. Exact stock-versus-market relative wealth is 0.92 ÷ 0.85 − 1 = 8.2353%, even though the stockholder lost 8.00% before costs.
A composite needs frozen standardization. A security’s 20, 60, and 120-day arithmetic excess returns are 4.00%, 10.00%, and 18.00%. Point-in-time universe means are 1.00%, 4.00%, and 8.00%, with standard deviations of 3.00%, 5.00%, and 10.00%, so z-scores are (4.00% − 1.00%) ÷ 3.00% = 1.0000, (10.00% − 4.00%) ÷ 5.00% = 1.2000, and (18.00% − 8.00%) ÷ 10.00% = 1.0000. Weights of 25%, 35%, and 40% give 25% × 1.0000 + 35% × 1.2000 + 40% × 1.0000 = 1.0700. This score is meaningful only under the stated universe, treatment of outliers, and missing-data rules.
Ranking turnover can consume the paper edge. A $10m equal-weight portfolio holds 20 names at 5% each. At rebalance, 7 remain and 13 are replaced. Sells equal 13 × 5% = 65% of starting value and buys equal 65%, so one-way turnover is 65%, two-way traded percentage is 130%, and total traded notional is $10m × 130% = $13m. At an illustrative all-in cost of 35 bp, cost is $13m × 0.35% = $45,500, or $45,500 ÷ $10m = 0.4550% of capital before taxes or market movement.

Watchlist and backtest controls

Record the exact as-of timestamp, time zone, data cutoff, formula version, code version, universe snapshot, and benchmark version.
Build historical universes from point-in-time membership and include delistings, bankruptcies, acquisitions, suspensions, and failed securities.
Keep issuer, security, share class, depositary receipt, primary listing, ticker history, and corporate identifier distinct.
Align exchange calendars, holidays, closing auctions, time zones, stale prices, trading halts, and asynchronous international closes.
Use consistent price, gross total, net total, or hedged returns and audit dividends, splits, rights, spin-offs, and special distributions.
Match currency and hedge basis; do not call an exchange-rate move company-specific relative strength.
Label arithmetic percentage-point excess, geometric relative return, normalized ratio, percentile, ordinal rank, and z-score separately.
Freeze broad-market, sector, industry, and peer benchmarks before ranking and retain point-in-time classification history.
Separate cross-sectional ranking from an asset’s own time-series trend and from RSI or another bounded oscillator.
State lookback endpoints, skip period, minimum observations, compounding, annualization, overlapping windows, and rebalance date.
Specify winsorization, outlier handling, volatility scaling, tie breaking, missing data, IPO seasoning, and negative-price or stale-price rules.
Avoid double counting correlated 20, 60, and 120-day signals as three independent confirmations.
Flag earnings, guidance, mergers, tenders, index changes, litigation, financing, halts, and one-day gaps that dominate a window.
Link each name to the primary filing and information actually available at the ranking timestamp; lag fundamentals and estimates appropriately.
Review absolute return, drawdown, volatility, beta, leverage, liquidity, borrow, valuation, and fundamental change alongside rank.
Define entry, exit, rank buffer, holding period, position size, sector and issuer caps, cash, invalidation, and no-trade rules separately.
Calculate one-way and two-way turnover explicitly and model spread, commissions, impact, borrow, taxes, delay, partial fills, and capacity.
Preserve rejected names, manual overrides, overrides’ timestamps and reasons, and subsequent outcomes to detect discretionary hindsight.
Run rolling and genuinely out-of-sample tests across regimes; control multiple testing and do not choose windows after viewing results.
Compare live decisions with the timestamped watchlist and report rank stability, turnover, realized slippage, attrition, and delisting outcomes.

Common misconceptions

“Relative strength and RSI are the same indicator.” Relative strength here compares instruments or benchmarks; RSI summarizes one instrument’s recent gains and losses on a bounded scale.
“A positive relative score means the security made money.” It can lose less than its benchmark and still have positive relative performance.
“The top rank identifies an undervalued future winner.” A rank summarizes chosen past returns and contains no standalone valuation, causality, or forecast conclusion.
“More windows provide independent confirmation.” Overlapping lookbacks are highly correlated and can repeatedly count the same event or price path.
“A successful ranking backtest is tradable.” Survivorship, look-ahead, benchmark selection, multiple testing, turnover, spreads, impact, borrow, taxes, and capacity can eliminate reported results.

Authoritative sources

Daily Momentum Factor - Kenneth French Data Library formation endpoints, size and prior-return breakpoints, value weighting, and winner-minus-loser construction.
Returns to Buying Winners and Selling Losers - Jegadeesh and Titman evidence, formation and holding horizons, continuation, and later dissipation.
Do Industries Explain Momentum? - Moskowitz and Grinblatt evidence on industry and individual-stock momentum decomposition.
Momentum Crashes - Daniel and Moskowitz evidence on infrequent, persistent momentum losses and rebound-sensitive crash states.
Introduction to Relative Strength or Ratio - CMT Association distinction between comparative relative-strength ratios and RSI.
Index Mathematics Methodology - S&P Dow Jones Indices price and total-return variants, weighting, dividends, corporate actions, and divisor continuity.
EDGAR Application Programming Interfaces - SEC timestamped submissions history, XBRL facts, entity identifiers, update timing, and data limitations.
Trading 101 - SEC Investor Bulletin on market and limit orders, execution-price uncertainty, partial fills, and trading instructions.