Low-Volatility Anomaly on the Select Sector SPDRs: Monthly Hold of the 3 Lowest 252-Day Return-Vol Sector ETFs, Long-Only, Equal-Weight (USEQ 1-DAY, pure OHLCV, 2 parameters)

Source: QuantConnect Strategy Library #18 'Volatility Effect in Stocks' (Quantpedia, 'Volatility Effect in Stocks - Long-Only Version'). The source mechanism is the low-volatility anomaly: assets with lower trailing volatility earn higher risk-adjusted returns than high-volatility ones. Leverage-constrained and benchmark-tied investors bid up high-beta, lottery-like names and neglect dull ones (Frazzini-Pedersen betting-against-beta; Baker-Bradley-Wurgler). The source ranks the top 50 US large caps by market cap, holds the 5 with the lowest 252-day volatility, and rebalances on the first trading day of each month. Three changes make it testable here. (1) UNIVERSE. We have no point-in-time market-cap data, and a hand-typed 50-stock list selected in 2026 is survivorship-biased (L151). It also exceeds the ~10-name compute ceiling (L173). The mechanism is therefore expressed on the 11 Select Sector SPDR ETFs (XLB, XLC, XLE, XLF, XLI, XLK, XLP, XLRE, XLU, XLV, XLY). This universe is defined by a rule, not picked in hindsight: it is the S&P 500's GICS sector partition. A sector enters the ranking only once it has 252 daily bars. That point-in-time rule automatically excludes XLRE before ~2016-10 and XLC before ~2019-06. The low-vol anomaly is documented at the industry and sector level as well as at the stock level (Baker-Bradley-Wurgler 2014 find much of it is between industries). (2) VOLATILITY DEFINITION. The source computes the standard deviation of closing PRICE LEVELS. That is scale-dependent: it ranks cheap stocks as 'low vol' and is a bug. We use the standard deviation of daily log returns over 252 sessions. (3) BREADTH. We hold N=3 names instead of 5 of 50, keeping roughly the source's selectivity ratio (the lowest ~25%). Rebalance on the first session of each month. Each holding is equal-weighted at ~33% of equity, 1x, with no leverage and no shorts. There are no regime, trend or cash gates (L130). A SPY-200d cash gate may be pre-registered later as a separate, separately-scored variant; it is not part of this claim. Flat state: the primary design is fully invested except during the 252-session warm-up. This is declared openly. It is not a never-flat long-short book (L150): there are no sign flips, and turnover is only the names that leave the bottom-3 set, typically 0-1 per month, at ~0.03% round-trip spread cost on commission-free USEQ. Data: USEQ daily bars from ~1999 for 9 of the 11 ETFs, which gives ~26 years and multiple regimes (2000-02, 2008, 2020, 2022). This is the longest clean history available anywhere in the factory. Implementation note (sandbox-timeout lesson from recent failures): keep one fixed-size deque of the last 252 log returns per ETF, updated O(1) per bar. Compute vol only on the monthly rebalance bar, sorting 11 numbers. Never rescan history. Novelty vs corpus: USEQ is 3.3% of experiments. The low-vol/BAB family previously died only as cross-sectional NEUTRAL books on crypto perps (L45/L137) and as a 30-name crypto betting-against-volatility basket. This is a long-only, 11-ETF, rule-defined equity universe with a 2-parameter design.