Low-turnover stocks and subsequent returns
The main signal averages daily share turnover over the 60 trading sessions before selection, excluding the selection date. This page compares subsequent returns and checks what changes after accounting for other stock characteristics.
The historical return gap is positive. Reliable trading gains remain unverified.
In this historical sample, the low-turnover group averaged 1.38 percentage points more per month than the high-turnover group. After accounting for the full listed controls and regrouping on the adjusted score, the gap was about 1.20 points. These are group comparisons. A consistent trading test is still needed to establish after-cost portfolio performance.
The 86 selection dates run from May 2019 through June 2026, with holding periods from June 2019 through July 2026. The common sample requires positive PE and usable control data. It does not represent every A-share stock. This history has been examined repeatedly.
Evidence snapshot · 16 September 2026. Content and source hashes checked · 22 October 2026. This edit adds no new observation months.
A single summary image with dates and limitations. No page URL is included.
Controls and group comparisons
Every row uses the same 86 selection months. Controls are applied sequentially, then stocks are ranked again on the remaining signal and divided into ten groups. The low and high labels refer to the two ends of that adjusted score. Membership changes between rows.
The final row includes size, low volatility, trading activity, Amihud illiquidity, momentum, reversal, beta and valuation. Amihud illiquidity compares absolute price moves with traded value as a rough indicator of how strongly prices move for a given trading amount.
Dots show mean gaps in percentage points. Lines show 95% block-bootstrap intervals, resampling 12-month blocks 2,000 times so adjacent months can move together. They describe uncertainty within this history and do not account for every specification tried during the research.
Sequential controls compare conditional relationships. Correlated features and changing memberships prevent the row differences from identifying one feature’s causal contribution. The plotted group returns do not include a complete execution or borrowing model.
A separate joint cross-sectional regression estimates about 0.60 percentage points more next-period return per one-standard-deviation higher low-turnover score. The coefficient is positive in 76.7% of the 86 months. This conditional historical relationship does not establish a causal effect or an executable return.
What might explain the difference?
The historical comparisons leave several explanations open. The study has not separated their causal effects.
Lower volatility
Adding low volatility narrows this sample’s gap from 1.63 to 0.86 percentage points. Both the score and group membership change, so the decline cannot be assigned entirely to volatility.
Trading activity and liquidity
Share turnover, traded value and price response measure related aspects of trading. Their controls change which stocks rank at each end. A positive remaining gap does not identify a separate economic mechanism.
Check both groups separately
The current raw groups average +1.08% and −0.30% per month. Earlier samples have different leg contributions. The present comparisons do not establish that avoiding crowding or speculation causes the gap.
What still limits the result?
The signal is an average daily share-turnover rate. Portfolio turnover later on this page measures the portfolio’s own trading. A stock with low share turnover need not be cheap or easy to trade.
The revised ranking fixes membership before checking future return availability. Missing future prices still limit which outcomes can be observed. Historical revisions and reconstructed stock eligibility remain unresolved, and the repeatedly examined 2024–2026 history is not an untouched test set.
The archive dated 16 September 2026 contains one completed August window and one pending window. Its manifest names a July cutoff but references a September data product. This does not by itself establish registration before the August holding period. Retain it as a later observation until a verifiable pre-period registration record is recovered.
Theoretical long-short spreads and net long-only simulations use different portfolio directions and accounting rules. Their difference cannot measure execution costs. A matched test must keep the signal, universe and period fixed while changing only the execution assumption.
Does adding the rule improve a comparable portfolio?
Use one universe, ledger and cost model to compare a benchmark with portfolios that exclude the highest-turnover group or favor low turnover. The existing 5.46% versus 5.33% annualized comparison already tried one exclusion rule and found only a small difference. Future tests need timestamped registration before holdings begin and a record of every specification tried.
When to revisit the conclusion
LT-C01 · 2026-10-22
Current evidence
A positive historical gross spread remains after the listed controls. Net trading gains are unverified.
Evidence still needed
Keep the registered adjusted-score spread and raw comparator on identical support, then accumulate the first 24 eligible new completed months. Evaluate investability separately with matched execution assumptions.
When the judgment changes
Adequate new observations fail to reproduce the registered relationship, or source corrections change the historical result. One month is insufficient.
Review trigger
Review each completed frozen interval and any change to inputs or ranking. Preserve old decisions and report missing evidence as inconclusive.
Future observation status
The existing register has zero completed eligible new months at its reviewed snapshot. A candidate October formation still needs a verified official calendar, contemporaneous input archives and a pre-entry timestamped receipt. The retained August window remains a later observation. No scheduler or trading task is installed.
Read the existing protocol and registrationResult version and earlier records
This page uses the reviewed formation-first ranking revision. Earlier figures remain in the methodology as labeled historical records. Source hashes below identify this version; they do not certify revision-safe financial histories.
- Source manifest SHA-256
9638fd5f4a3edc39896ac20dcfb7e47ddb4cd0f0d861e4033a4751e82b5f543f- Effect summary SHA-256
3d52f7713be407752fc0dc36252fb1830f3b214a31ac4720f8314d2a65b243e4