Historical market research · evidence firstQuant Market Research Archive

A public archive of historical market research, with dates, progress, and limitations made explicit.

Public research · see dates belowAggregate results · historical description only
Long-run style research · FACTOR EXPLORER

18-year A-share Market Style Analysis

Each month, the study compares high and low score groups across 19 A-share style factors.

Current conclusion

Historical archive and descriptive study

Nineteen archived factor series differ widely and overlap. Their history and tradability still need verification.

Evidence boundary · Historical long-short series are not account returns. Factor coverage differs, and the original generation chain and formula mappings are not fully verified.

2008-01-02 — 2026-09-04
Long-run index and ETF performance18-year A-share Market Style Analysis
Selected-factor diagnosticSizeHistorical returns, period metrics, definitions, and calculations update with the selected factor.

Annual composite returns and period performance

Historical series · formula pending audit
Size · size

Size

Annual composite return · %
Geometric annualizednot availableHistorical composite series
Max drawdownnot availableSame historical series
Sample yearsnot availablenot available trading days
Annual returns: Loading

Reading the public derived snapshot.

Returns compound daily long-short spreads; shorter samples are shown over their available periods. Sample windows can differ by factor, so do not rank factors directly from these figures.

What do these return and risk measures mean?

Annualized return expresses a result on a one-year scale; maximum drawdown is the largest fall in this return series from a high to a later low. These figures come from a historical long-short series, not an actual account.

Factor definition, features, and calculation

Linked to selected factor
Size · size

Size

What it is · included features
Size is documented as a historical return series alongside its current recomputable proxy; the historical underlying feature may not be fully preserved.
Historical long-short direction
Large-cap minus small-cap. This retains the legacy page label; the direction of the original raw score has not been confirmed from source code.
How it is calculated · reviewed implementation
ln(total_mv + 1). total_mv is the source total-market-capitalization field; higher scores indicate larger capitalization.
Common processing
Final scores are winsorized at the 1st and 99th percentiles within each cross-section. When first-level Shenwan industry data is available, scores are demeaned by industry and then z-scored across the market. This does not guarantee sector-neutral long-short portfolio weights.
Current core-dictionary mapping
The size proxy in the current core dictionary should be read together with its source, version, and scope.

Overview of 19 factor results

Searchable · sortable

Full figures are provided for reference. Short and long samples are shown together but do not cover the same periods. These returns are not validated tradable returns.

Switch the measure to inspect each factor's historical summary. Sample dates differ, so use this chart to locate contrasts rather than rank like-for-like results.

What do these measures and formula terms mean?

Geometric annualized return converts a compounded multi-period result into an equivalent yearly rate. Annualized volatility scales return variation to a year using the page's frequency. The Sharpe ratio divides average excess return by annualized volatility. Maximum drawdown is the largest fall from a previous high to a later low. Positive daily-return ratio is the share of trading days with positive returns. These describe the historical series and do not establish future performance on their own.

In the formulas, `ln` is the natural logarithm; `clip(x, a, b)` bounds x between a and b; `Cov` measures how two return series move together, and `Var` measures variation in one series; `rᵢ` is a stock return and `rₘ` is the equal-weight average return of this study's sample. Names such as `total_mv`, `turnover_rate`, and `PE_TTM` retain the source field identifiers.

Factor overview: Loading

Reading the public derived snapshot.

Factor correlations

Daily return spread · non-causal

The eight factors most correlated with Size are shown. Positive correlation indicates co-movement; negative correlation indicates opposite movement. Select a factor to inspect it. Paired-sample dates are incomplete, so do not use this matrix directly to construct a portfolio.

Correlations: Loading

Reading the public derived snapshot.

Supplement: size deciles and stability diagnostics
Historical research archive · market style factors

Long-run A-share style-factor performance

Review factor definitions, long-run performance, annual returns, correlations, and sample coverage, followed by a separately computed market-cap decile study.

Historical sample: 2008-01-02 to 2026-09-04

What the archive tells us today

In the archived 2008-01-02 to 2026-09-04 sample, the value factor has a geometric annualized composite return of 10.64% and a maximum drawdown of 20.61%. The archived size series, large minus small, has a geometric annualized return of -22.16%. This documents a historical advantage for the small-cap side under the archive rules. It leaves the return available to a small-cap account unresolved.

Today, these results help identify historical style exposure and choose comparisons for new tests. A current portfolio needs formation-date financial versions and a matched execution ledger before the historical ranking can guide trading. Narrative and published figures were checked on 2026-10-30 without rerunning data or adding observations after 2026-09-04.

Archived numerical summary

Historical long-short composite returns

The chart shows a composite daily return series for the high-minus-low score groups. It describes factor direction and historical regimes, not the return of an actual account. market style factors can support risk interpretation and cross-sectional comparisons; their use as predictive signals requires separate testing.

Research design

Research question and methods

Coverage

The study covers 19 A-share style factors. Each month, stocks are split into five groups; the highest- and lowest-scoring groups are compared with equal weights until the next month. Daily returns use high minus low. Aligning data by announcement date alone does not prove that financial information was available at the time; historical revisions and full point-in-time treatment remain to be verified.

Industry adjustment

Scores are first demeaned by historical Shenwan Level-1 industry and then standardized cross-sectionally. Securities without an industry classification are retained separately rather than treated as if their industry adjustment were zero.

Sample coverage

Most factors cover about 18.6 years. Holdings, shareholder-positioning, and fund data cover shorter periods; fund-flow data currently span about 0.6 years. Their results should not be compared with full-history factors at the same evidentiary strength.

Do scores work, and can they be traded?

IC measures the linear correlation between factor scores and future returns; Rank IC measures rank agreement. An IC near zero indicates little linear relationship between score levels and future returns. Q1–Q10 in this study are decile portfolios ordered from low to high scores, the reverse of some historical long-short charts. Use each chart title and factor definition to confirm direction.

Research limitations Factor direction and portfolio construction are described above. Transaction costs, suspensions, price-limit execution, and a complete tradability audit still need a consistent treatment. Some fund, shareholder-positioning, and flow series cover shorter periods, so those results remain exploratory.

High minus low refers to factor scores, not uniformly to raw descriptors. The archived size factor is large minus small; the separately recomputed size diagnostic is small minus large. The current descriptor dictionary is a separate definition and does not establish the archive's original formulas.

Read full factor definitions

How to read IC and the decile study

IC measures the cross-sectional relationship between a score and later returns; Rank IC measures whether their order agrees. A value near zero indicates little relationship in that sample. The separate size study sorts into ten groups and measures next-session returns; its direction and sample contract differ from the monthly five-group long-short archive above.

Sources, dates, and research progress
Research question

Which style characteristics appear more consistently, and which results are concentrated in certain periods?

Evidence status

Historical archive and descriptive study

Next step

First verify historical formulas and missing-return rules, then recompute using provable formation-date information and add costs, suspensions, and price-limit handling.