This report, published by Yuanda Information Securities Research, presents a comprehensive financial engineering study examining the size factor through a decade-long backtest. The research defines the size factor as the natural logarithm of total market capitalization, calculated as ln(share price × total shares), with the economic foundation rooted in the well-documented small-cap premium phenomenon. Taking the logarithm helps mitigate the skewness and heteroscedasticity issues arising from the long-tail distribution of A-share market capitalizations, thereby enhancing the efficiency of linear regression fitting.
Key characteristics of the size factor
The size factor exhibits an overall negative stock-selection ability, with a significant small-cap premium that is particularly pronounced in small and mid-cap indices. Within the CSI 1000 index, the factor demonstrates its strongest negative predictive power, achieving an average RankIC of -0.0773, which further improves to -0.5085 in RankICIR after industry neutralization, indicating excellent strength and stability. Conversely, within the CSI 300 index, the factor displays positive stock-selection ability with an average RankIC of 0.0521. The negative selection effect is strongest in the commercial trade, textile and apparel, and light manufacturing sectors, while only the banking and non-bank financial sectors exhibit positive effects. Industries such as commercial trade, textile and apparel, and light manufacturing all record average RankIC values below -0.13, with small-cap stocks in these sectors consistently outperforming their large-cap counterparts. The banking sector, however, shows an average RankIC of 0.1288 with a 64% positive ratio, highlighting a significant large-cap advantage. Furthermore, the negative effectiveness of the size factor diminishes monotonically as market capitalization increases, with the most pronounced effect observed in the smallest market-cap group.
Multi-factor portfolio strategy construction
The size factor captures the small-cap premium, but small-cap stocks contain a substantial number of companies with poor earnings quality and inflated valuations. To address this, two auxiliary constraints are introduced: a net profit threshold of no less than 100 million yuan in TTM to exclude loss-making, marginal-profit, and shell companies, and a price-to-book ratio cap of 3 to avoid overpaying for the size premium. The specific screening process includes three steps: first, eliminating ST and delisted companies; second, excluding companies with TTM net profit below 100 million yuan; and third, removing companies with a price-to-book ratio greater than 3. Portfolio rebalancing occurs on April 30 and October 30 each year based on periodic report disclosure dates, with adjustments carried forward to the next trading day if these dates fall on non-trading days. The strategy selects the top five stocks ranked by the size factor in ascending order and holds them with equal weights.
Between April 30, 2014 and August 25, 2026, this strategy achieved a total return of 708.50% with an annualized return of 18.38%. Compared to the CSI 300 index, the portfolio generated excess returns of 598.14%.
IC testing of the size factor
The size factor is a classic measure of company scale, with its economic foundation rooted in the small-cap premium phenomenon first identified by Banz in 1981. Extensive subsequent research has confirmed that small-cap stocks systematically deliver higher long-term returns than large-cap stocks, a relationship captured by the SMB factor in the Fama-French three-factor model. Common construction methods for the size factor include total market capitalization, float market capitalization, free-float market capitalization, and logarithmic market capitalization. Given that A-share market capitalizations range from billions to trillions with a long-tail distribution, directly using raw market value is susceptible to extreme value influence. Taking the natural logarithm effectively alleviates data skewness and heteroscedasticity issues while improving linear regression fitting efficiency.
Using semi-annual frequency data for CSI All-Share constituents during 2014-2026, IC testing conducted on April 30 and October 31 data reveals several notable characteristics. The factor exhibits an overall negative stock-selection ability across indices, with an average RankIC of -0.0835 and RankICIR of -0.3771, reflecting the overall small-cap premium in the A-share market. The most prominent negative performance appears in the CSI 1000, where the average RankIC is lowest and the absolute RankICIR is highest, strengthening further to -0.5085 after industry neutralization. The CSI 300 shows an average RankIC of 0.0521, indicating positive stock-selection ability, though the RankICIR is relatively low at 0.2964, suggesting weaker stability. Both the CSI 500 and CSI 800 exhibit generally weak factor effectiveness with unstable signal direction.
Industry-level IC testing demonstrates that the factor provides negative stock-selection ability across most sectors, with the small-cap effect being widespread. The strongest negative predictive power appears in commercial trade, textile and apparel, light manufacturing, conglomerates, and media industries, where average RankIC values all fall below -0.13. In contrast, the banking and non-bank financial sectors show positive stock-selection ability, with banking recording an average RankIC of 0.1288 and a 64% positive ratio, making it the sector with the most significant large-cap advantage.
Grouping stocks by market capitalization reveals that factor effectiveness monotonically decreases as market cap increases. Market cap group 1 through group 4, with median market capitalizations ranging from approximately 2.4 billion to 10.9 billion yuan, show average RankIC values of -0.1038, -0.0505, -0.0302, and -0.0103 respectively, all negative with absolute values narrowing in a stepwise manner. The positive ratios are 0.16, 0.24, 0.28, and 0.44 respectively. Group 1 exhibits the highest negative intensity, making the small-cap premium most significant. Market cap group 5, with a median of approximately 30.5 billion yuan, sees the average RankIC turn positive at 0.0402 with a 68% positive ratio, indicating that large-cap stocks take the advantage. The size factor thus demonstrates its strongest negative performance in the small-cap range, combining both predictive strength and stability.
From a time-series perspective, the effectiveness of the size factor is highly dependent on market style rotation. In years when the large-cap style dominates, such as 2017, the factor performs positively. Conversely, in years characterized by small-cap leadership, such as 2015, 2021, and 2025, the factor turns significantly negative as small-cap stocks substantially outperform. When the size factor is divided into ten groups from lowest to highest, the cumulative average return curve confirms that the low size factor group, representing small-cap stocks, delivers superior overall returns compared to the high factor group, corroborating the small-cap premium identified in the IC testing.
Combined strategy construction approach
While the size factor captures the small-cap premium, small-cap stocks represent a highly heterogeneous group containing both high-quality companies in early growth stages with earnings elasticity and a substantial number of fundamentally deteriorating, low-profitability entities with shell-like characteristics. Relying solely on the size factor risks significant drawdowns during market style shifts or liquidity tightening. The research therefore introduces two types of auxiliary constraints centered on earnings quality and valuation safety margin. The net profit TTM threshold of 100 million yuan ensures continuous profitability and a certain operational scale, while the price-to-book ratio cap of 3 confines the portfolio to a reasonable valuation range, preserving adequate safety margin against excessive speculation-driven valuation fluctuations.
The strategy achieved a total return of 708.5% with an annualized return of 18.38% during the April 30, 2014 to August 25, 2026 period, generating excess returns of 598.14% over the CSI 300 index.
Risk disclosure
Historical performance does not guarantee future results. The backtest model does not account for actual transaction costs. There may be statistical errors in the data used for this research.