Yuanda | Momentum Filtering in High Dividend Stock Selection and Portfolio Performance — Yield Factor Return Characteristics and a Two-Stage Strategy

Deep News11-12

For stock trading insights, rely on authoritative, professional, timely and comprehensive analyst research to help you uncover potential thematic opportunities! Yuanda Securities Research Report / Financial Engineering Research: Momentum Filtering in High Dividend Stock Selection and Portfolio Performance — Yield Factor Return Characteristics and a Two-Stage Strategy. Analyst: Wu Qidi, Registration No.: A0190523020001, wuqidi@yd.com.cn. Research Assistant: Zhang Zhe, Registration No.: A0190126040002, zhangzhe@yd.com.cn.

Key investment points

Dividend yield possesses a certain return-ranking capability, and pure high-dividend portfolios have achieved historical excess returns. From 2015 to Q3 2026, the average RankIC of the dividend yield factor for the CSI 300, CSI 500 and CSI 1000 was 0.093, 0.061 and 0.063 respectively. The annualized returns of the corresponding pure high-dividend TOP10 portfolios we constructed for the CSI 300, CSI 500 and CSI 1000 were 4.88%, 4.93% and 9.46%, exceeding their respective equal-weighted total return indices by 2.13, 0.34 and 5.34 percentage points. (See Tables 1 and 2.) Comparison of the Shanghai Composite Index and Shenzhen Component Index trends. Source: Wind, Yuanda Information Securities Research Institute.

Combining high dividend stocks with medium-term price momentum forms a two-stage screening portfolio. Considering that falling stock prices may passively inflate dividend yields, we introduce medium-term momentum for a second screening. The pure high-dividend TOP10 directly selects the top 10 by dividend yield; the high-dividend momentum TOP10 selects 10 by momentum from the top 20 by dividend yield. The average RankIC of the representative momentum within the TOP10 candidate pool for the three indices was 0.193, 0.094 and 0.094 respectively. (See Tables 3 and 4.)

Momentum screening enhances after-fee returns, and some TOP10 portfolios also improve drawdown. After considering certain trading costs, all three momentum measures achieved positive return increments in the nine TOP10 comparisons across the three indices. The after-fee annualized returns of the representative portfolios for the three indices were 13.80%, 14.57% and 12.94%, higher than pure high-dividend by 8.92, 9.64 and 3.48 percentage points, and higher than the equal-weighted indices by 11.05, 9.99 and 8.81 percentage points. For the CSI 500, all three momentum measures improved Sharpe and maximum drawdown, with MOM84 raising Sharpe from 0.32 to 0.70 and narrowing maximum drawdown from 58.43% to 42.56%; for the CSI 1000, the drawdowns of all three momentum measures slightly widened. (See Tables 7 and 8.)

After expanding the holding range to TOP50, the momentum enhancement effect diverges. Among 48 comparisons of eight momentum measures in the TOP10/TOP20 of the three indices, 40 showed positive after-fee increments; among 27 comparisons of the three representative momentum measures in TOP30–50, 17 were positive, with only 2 for the CSI 1000. The momentum enhancement effect diverges with the stock pool, window and number of holdings, and historically leading configurations should not be directly extrapolated. (See Tables 5 and 6.)

Investment recommendation: When selecting high dividend stocks, first verify the support of earnings and cash flow for dividends, then use medium-term price trends to assist screening. All nine momentum TOP10 groups across the three indices had historical after-fee return increments, but drawdown, turnover and fee erosion differ; investors are advised to allocate prudently based on the stock pool, number of holdings and actual trading costs, and should not simply copy a single historically optimal window.

Risk warning: Historical dividend data may differ from information available at the time; simulated fees do not equal real costs. Market changes may render the strategy ineffective, and historical performance does not represent future returns.

1. Return characteristics and portfolio performance of the dividend yield factor

1.1 Definition of dividend yield and return-ranking test

This paper uses the trailing twelve-month dividend yield (DY-TTM) to measure the level of cash dividends over the past year relative to the current stock price, calculated as follows. Trailing dividend yield = total cash dividends per share implemented within the 365 calendar days before the observation date ÷ unadjusted closing price on the observation date × 100%. Cash dividends are counted by ex-dividend date; bonus shares and capital reserve conversions are not counted as cash dividends. A falling stock price may also push up the dividend yield and does not necessarily mean improved dividend capacity. To measure the return-ranking capability of dividend yield, this report ranks all eligible constituent stocks of each index at each quarter-end by dividend yield and by next-quarter individual stock returns, calculates the Spearman rank correlation coefficient (RankIC), and then takes the arithmetic average over 47 valid quarters from Q1 2015 to Q3 2026. The averages for the CSI 300, CSI 500 and CSI 1000 were 0.093, 0.061 and 0.063 respectively (Table 1). This is a full-index cross-sectional statistic and does not change with the final number of holdings such as the pure high-dividend TOP10 or TOP20; the momentum RankIC within high-dividend candidate stocks is discussed separately in Chapter 2. Note: The backtest has 47 quarterly observations in total; ICIR is the ratio of the mean quarterly RankIC to the sample standard deviation, not annualized. Dividend yield has a certain return-ranking capability, but whether it can translate into portfolio returns still depends on the specific stock selection and holding method.

1.2 Construction and return performance of pure dividend yield portfolios

The backtest period of this paper is from December 31, 2014 to September 30, 2026. December 31, 2014 is the signal formation and net value comparison base date, and January 5, 2015 is the first simulated trading day of the period; thereafter, stocks are selected at the end of March, June, September and December each year, rebalanced on the next trading day, and the portfolio is held equally weighted. From the eligible constituent stocks of the CSI 300, CSI 500 and CSI 1000 at each period, 10, 20, 30, 40 and 50 stocks are selected by trailing dividend yield from high to low, forming five tiers of pure dividend yield portfolios. This paper also measures the cases of no fees and fees of 5, 10, 20 and 30 basis points on each side of buy and sell; the main portfolio tables use after-fee results with 30 basis points on each side, and fee differences are compared uniformly in Chapter 4. Note: 2015.01.05–2026.09.30; both types of total return indices include dividend reinvestment, calculated on equal-weighted and market-cap-weighted bases respectively, with the former as the main performance benchmark and the latter for reference. The after-fee annualized returns of the 15 pure high-dividend portfolios across the three indices were all higher than both types of total return indices, but the return advantage varied by stock pool and number of holdings (Table 2).

1.3 Impact of number of holdings on portfolio returns

Pure high-dividend portfolios do not necessarily have higher returns with fewer holdings. For the CSI 500, expanding from TOP10 to TOP50 raised the after-fee annualized return from 4.93% to 7.74%; for the CSI 1000, the TOP10 return of 9.46% was slightly higher than the TOP50 return of 9.01%. These two opposite results indicate that neither concentrated stock selection nor expanding the number of holdings is universally superior (Figure 1). Note: Unit is percentage points; return differences are calculated using full-precision CAGR. Figure 1 shows that the return advantage of pure high-dividend portfolios changes with the number of holdings: the CSI 500 TOP10 exceeded the equal-weighted total return index by 0.34 percentage points, and the TOP50 by 3.15 percentage points; the CSI 1000 TOP10 exceeded it by 5.34 percentage points. Whether there is still room for return enhancement within high-dividend stocks remains to be further compared by adding momentum.

2. Momentum screening and return improvement in high dividend strategies

The returns of high dividend investment come from both cash dividends and stock price changes during the holding period. Looking only at the level of dividend yield makes it difficult to judge whether dividends can continue; if the stock price weakens, capital losses may offset dividend returns, and a higher dividend yield may simply be the result of a price decline. The subsequent performance of high-dividend stocks often diverges: earnings improvement and valuation repair may drive some companies stronger, while declining industry prosperity and weakening operating expectations may drag down others. Therefore, on the basis of dividend yield ranking, we introduce medium-term price momentum, hoping to use price trends for further screening and to test whether portfolio returns can thereby improve.

2.1 Return-ranking characteristics of price momentum

Price momentum (MOM) reflects the price change of a stock over a past period. This paper compares 21, 42, 63, 84 and 126 trading days, as well as 4, 5 and 6 month momentum excluding the most recent month, for a total of eight indicators. We first observe the RankIC of each index and its high dividend stocks, then test actual portfolio performance. Note: The high dividend group is the top 20% of stocks by dividend yield in each index. The return-ranking capability of momentum varies with the stock range. For the CSI 300, the average RankIC of MOM84 across the full index was −0.005, rising to +0.060 among the top 20% by dividend yield; for the CSI 1000, MOM42 improved from −0.072 to −0.050 but remained negative (Table 3). This shows that the performance of momentum signals among high-dividend stocks differs from that of the full index, and not all windows exhibit positive ranking capability. The data show that MOM84, MOM126 and MOM6_1 deserve further testing, but whether the RankIC improvement can bring after-fee return increases still must be judged through portfolio backtests.

2.2 Two-stage stock selection combining high dividends and momentum

We construct two types of portfolios, pure high-dividend and high-dividend momentum, both equally weighted. Pure high-dividend directly selects the top K by dividend yield; high-dividend momentum selects K by medium-term price momentum from the top 2K by dividend yield. This paper compares five tiers of holdings with K equal to 10, 20, 30, 40 and 50. The "high dividend group" in Table 3 (top 20% of stocks by dividend yield) is used only for ranking diagnosis and differs from the actual top 2K candidate stocks screened. Taking TOP10 as an example, pure high-dividend directly holds the 10 stocks with the highest dividend yield, while high-dividend momentum selects 10 from the top 20; TOP20 correspondingly compares directly holding the top 20 with selecting 20 from the top 40. The second screening is conducted only within high-dividend candidate stocks, using price trends to provide reference information different from dividend yield; whether it improves returns should be compared with pure high-dividend portfolios of the same stock pool and same number of holdings. In the TOP10 and TOP20 portfolios of the three indices, MOM84, MOM126 and MOM6_1 each had positive after-fee return increments in all six cases, so they were selected as representative windows for further analysis.

2.3 Momentum ranking and after-fee portfolio returns

Note: The initial screening RankIC is calculated within the top 2K candidate stocks by dividend yield, where K is the final number of holdings; it differs from the full-pool RankIC and after-fee CAGR. After momentum enters the high-dividend candidate pool, the return ranking of some windows turns from negative to positive. For example, for the CSI 1000, the average RankIC of MOM84 across the full index pool was −0.065, rising to +0.023 among the top 20 high-dividend candidate stocks, corresponding to a TOP10 after-fee annualized return 0.89 percentage points higher than pure high-dividend (Table 4). However, ranking capability and portfolio returns are not one-to-one: for the CSI 500, the RankIC of MOM126 in the high-dividend group was higher than that of MOM84, yet the TOP20 return increment was smaller, and the portfolio effect still needs to be evaluated in combination with return and risk. The TOP10 results in Table 4 are consistent with the conclusions on the first page: the MOM6_1 of the CSI 300, the MOM84 of the CSI 500 and the MOM126 of the CSI 1000 increased returns by 8.92, 9.64 and 3.48 percentage points respectively compared with the pure high-dividend TOP10 with the same number of holdings. After expanding to TOP20, the corresponding increments for the three groups were 3.60, 4.61 and 4.83 percentage points. The results for both holding tiers were positive, but the return increments were not the same; other momentum windows and wider holding numbers are further compared in Chapter 3.

2.4 Supplementary test on the short sample of the CSI 2000

The CSI 2000 was launched in August 2023. This paper studies its current Shanghai and Shenzhen constituent stocks, with the first signal formed on December 29, 2023 and January 2, 2024 as the unified account base date. MOM84, MOM126 and MOM6_1 are calculated using individual stock historical quotes available before the formation date, and do not use the index launch date as the starting point for price lookback. The observation period of the CSI 2000 Shanghai and Shenzhen sub-pool account is from January 2, 2024 to September 30, 2026. The after-fee annualized return of the TOP10 pure high-dividend was 3.65%, and with MOM84, MOM126 and MOM6_1 it was 10.83%, 10.87% and 13.91% respectively; there was return enhancement within the short sample, but it should not be directly extrapolated. The after-fee annualized return of the TOP20 pure high-dividend portfolio was 9.47%, and the corresponding three momentum portfolios were 10.12%, 9.69% and 12.57%, with MOM6_1 exceeding pure high-dividend by 3.10 percentage points.

3. Comparison of momentum windows and number of holdings

The following first compares eight momentum windows, then tests whether the return enhancement can continue after expanding the number of holdings.

3.1 Comparison of return increments of eight momentum windows

Table 5 presents the after-fee CAGR increments of eight momentum measures in the TOP10 and TOP20 of the three indices relative to pure dividend yield portfolios. Note: 2015.01.05–2026.09.30; 30 basis points on each side of buy and sell; the increment is the after-fee CAGR difference under the same index and same number of holdings. The eight momentum measures formed 48 comparisons in the TOP10 and TOP20 of the three indices, of which 40 had after-fee returns higher than the pure high-dividend portfolio (Table 5). Positive increments were not limited to the three aforementioned windows: the shortest MOM21 also improved five groups; however, MOM42 showed negative increments in the TOP20 of the CSI 300 and CSI 500. Different momentum windows and stock pools lead to divergent results, and historical leadership cannot be directly equated with future optimality. The number of holdings also affects return performance. For the CSI 500, MOM84's return increment fell from 9.64 to 4.61 percentage points when expanding from TOP10 to TOP20; for the CSI 1000, MOM6_1 rose from 2.97 to 3.86 percentage points. Concentrated holdings do not always yield higher momentum returns. These data comparisons show that momentum windows cannot be selected only by the highest historical return, but should be judged comprehensively in combination with the stock pool, number of holdings and risk performance.

3.2 Portfolio performance after expanding the number of holdings

Table 6 further compares the after-fee performance of three representative momentum measures in the TOP30, TOP40 and TOP50 of the three indices, totaling 27 items. Each item uses the pure high-dividend portfolio of the same index and same number of holdings as the control, examining how much annualized return can be gained after adding price trend screening. Source: Wind, Yuanda Information Securities Research Institute. Note: 2015.01.05–2026.09.30; 30 basis points on each side of buy and sell; the increment is the CAGR difference under the same index and same number of holdings, in percentage points. Among the 27 wide-holding comparisons, 17 had positive return increments and 10 were negative. All nine items for the CSI 300 were positive, among which MOM6_1 TOP40 was 4.36 percentage points higher than pure high-dividend with the same number of holdings; the CSI 500 had six positive and three negative, and the CSI 1000 had only two positive and seven negative. The enhancement effect differs greatly across stock pools. The data show that increasing the number of holdings does not necessarily stably thicken returns: for the CSI 1000, MOM126 was 1.63 percentage points higher than pure high-dividend in TOP30, but 0.02 and 1.01 percentage points lower in TOP40 and TOP50 respectively. Expansion also changes the initial screening range: TOP30 screens from the top 60 by dividend yield, while TOP50 screens from the top 100. We find that after expanding holdings, momentum screening is not necessarily persistently superior. For example, most wider CSI 1000 portfolios lagged behind pure high-dividend with the same number of holdings. The historical advantages of portfolios such as the TOP10 in our backtest should not be copied directly; investors should comprehensively evaluate multiple factors such as specific index constituents, candidate range and momentum window.

4. Portfolio risk-return and trading cost evaluation

High-dividend momentum should not focus only on how much annualized return is improved, but should also consider factors such as drawdown changes and additional turnover costs. This chapter further compares the risk-return performance and trading fees of the main portfolios.

4.1 Returns and drawdowns of representative portfolios

Table 7 focuses on comparing the after-fee returns, volatility, Sharpe, drawdown and turnover of the pure high-dividend TOP10 of the three indices with the TOP10 of the three momentum measures MOM84, MOM126 and MOM6_1. Note: 2015.01.05–2026.09.30; all are TOP10, with buy and sell each set at 30 basis points; momentum selects 10 from the top 20 high-dividend stocks. The risk-free rate for Sharpe is set at 0%, and annualized one-way turnover is in times. Across the three indices, the after-fee CAGR and Sharpe of the three TOP10 momentum portfolios MOM84, MOM126 and MOM6_1 were all higher than the pure high-dividend TOP10 of the same index (Table 7). However, the return ranking and risk performance of different momentum measures differ, and a single window cannot be determined solely by historical returns. The after-fee CAGR of the CSI 300's MOM126 and MOM6_1 were 13.91% and 13.80% respectively, with Sharpe both around 0.77; the maximum drawdown of MOM126 was 34.45%, and that of MOM6_1 was 30.28%, the latter lower than the pure high-dividend's 32.89%. Higher historical returns do not mean smaller drawdowns. The after-fee CAGR of the CSI 500's MOM84 was 14.57%, with a Sharpe of 0.70, and maximum drawdown narrowed from the pure high-dividend's 58.43% to 42.56%; the three momentum measures of the CSI 1000 returned 10.35%–12.94%, with Sharpe all improved, but maximum drawdowns all slightly higher than the pure high-dividend's 44.69% (Table 7).

4.2 Impact of trading fees on portfolio returns

Tables 8 and 7 use the same 12 TOP10 accounts, comparing the annualized returns of pure high-dividend and three momentum portfolios under fees of 0, 5, 10, 20 and 30 basis points on each side of buy and sell, in order to observe the impact of trading costs on return enhancement. Note: 2015.01.05–2026.09.30; all are TOP10, and the rate is basis points on each side of buy and sell; the last column is the CAGR difference between 0bp and 30bp, in percentage points. When the fee on each side of buy and sell rose from 0 to 30 basis points, the CAGR erosion of the pure high-dividend TOP10 of the three indices was 1.09–1.22 percentage points, and that of the nine momentum TOP10 portfolios was 1.39–1.70 percentage points, all higher than the corresponding pure high-dividend portfolios (Table 8). Momentum portfolios are more sensitive to fees, but after setting costs of 30 basis points on each side of buy and sell, the after-fee CAGR of the nine momentum TOP10 portfolios was still higher than the pure high-dividend portfolios of the same index. Among them, the CSI 1000's MOM84 increased by only 0.89 percentage points; compared with some windows of the CSI 300 and CSI 500, the return increment was smaller. The backtest results show that the portfolios we screened using multiple momentum measures generally showed positive effects, but the specific enhancement magnitude still depends on multiple factors such as the stock pool, specific window width and momentum signal.

4.3 Comparison of returns of representative portfolios of the three indices

Figure 2 compares the after-fee CAGR of the pure high-dividend TOP10 of the three indices with their respective representative high-dividend momentum TOP10. Source: Wind, Yuanda Information Securities Research Institute. Note: The momentum measures of the three indices are MOM6_1, MOM84 and MOM126 in order; TOP10, 30 basis points on each side of buy and sell. Figure 2 shows that the after-fee CAGR of the representative high-dividend momentum TOP10 of the three indices was 13.80%, 14.57% and 12.94%, while the pure high-dividend TOP10 was 4.88%, 4.93% and 9.46%. Correspondingly, they exceeded the equal-weighted total return index by 11.05, 9.99 and 8.81 percentage points, and exceeded the market-cap-weighted total return index by 9.67, 10.34 and 10.37 percentage points. Return differences are calculated by CAGR over the same period. Historical return improvement does not equal drawdown control: the maximum drawdowns of the three momentum TOP10 of the CSI 1000 were all slightly higher than pure high-dividend (Table 7), and fee sensitivity also varies by momentum window (Table 8). By segment, from 2018 to 2025 the annualized return differences of the representative TOP10 of the three indices relative to the equal-weighted total return index were 8.44, 8.04 and 4.65 percentage points; for the CSI 1000 from 2023 to 2025 it was −2.63 percentage points. These results come from continuous account history segments, not independent out-of-sample validation. High-dividend momentum should be used as an auxiliary signal for return enhancement rather than a direct substitute for a low-drawdown strategy; allocation still needs to weigh stage performance, drawdown and trading fees.

5. Investment recommendations

High dividend stock selection should first verify the support of earnings and cash flow for dividends, then use medium-term momentum to assist in identifying the market pricing trend of stocks, and formulate allocation plans in combination with drawdown tolerance, holding concentration and trading costs.

5.1 Portfolio selection for different stock pools

CSI 300: All three momentum TOP10 improved returns and Sharpe. MOM126 had an after-fee CAGR of 13.91%, slightly higher than MOM6_1's 13.80%, but its maximum drawdown reached 34.45%, higher than pure high-dividend's 32.89%; MOM6_1 narrowed it to 30.28%. The momentum window should not be determined solely by return level (Table 7). CSI 500: All three momentum windows improved the after-fee returns and Sharpe of the TOP10. MOM84 had a CAGR of 14.57% and a Sharpe of 0.70, with maximum drawdown narrowed from 58.43% to 42.56%; however, implementation still requires balancing turnover and individual stock risk (Tables 7 and 8). CSI 1000: All three momentum TOP10 achieved positive after-fee return increments, but their maximum drawdowns all widened compared with pure high-dividend; although MOM126 had a CAGR of 12.94%, it still cannot be regarded as a defensive strategy. Most TOP30–50 allocations did not improve returns, and capacity and liquidity need attention (Tables 6 and 7). CSI 2000: Since 2024, the short sample of Shanghai and Shenzhen constituent stocks shows certain momentum enhancement, but it is not directly comparable with the long-term accounts of the previous three indices, and a fixed proportion should not be set based on it.

5.2 Portfolio implementation: number of holdings and trading costs

Increasing the number of holdings can reduce single-stock concentration but does not necessarily bring higher returns. Table 6 shows that the three representative momentum measures of the CSI 300 all achieved positive increments in TOP30–50, while most of the CSI 1000 did not improve the pure high-dividend performance. Investors should weigh diversification against return enhancement. After the fee on each side of buy and sell rose from 0 to 30 basis points, the annualized CAGR erosion of the nine momentum TOP10 groups was 1.39–1.70 percentage points, higher than the pure high-dividend TOP10's 1.09–1.22 percentage points (Table 8). Turnover, liquidity and market impact should be taken into account, and historical after-fee returns cannot be equated with achievable returns. Actual stock selection should first verify the earnings and cash flow basis of dividends, then use momentum to assist ranking, and continuously observe the return increments relative to pure high-dividend and equal-weighted indices. For high-turnover, concentrated portfolios, more attention should be paid to execution costs.

6. Risk warning

Historical performance risk. The role of dividend yield and momentum factors may change with market style, and leading results of historical windows and number of holdings do not guarantee future superiority. Trading execution risk. Actual transactions are affected by liquidity, price limits, slippage and market impact; simulated fee rates may differ from real costs, and historical backtest returns do not represent actually obtainable returns. Extreme market risk. Under extreme market conditions, historical statistical relationships may deviate significantly; investors should not overly rely on historical experience, and quantitative models and portfolio strategies may fail in extreme markets.

Disclaimer: Investing carries risk. This is not financial advice. The above content should not be regarded as an offer, recommendation, or solicitation on acquiring or disposing of any financial products, any associated discussions, comments, or posts by author or other users should not be considered as such either. It is solely for general information purpose only, which does not consider your own investment objectives, financial situations or needs. TTM assumes no responsibility or warranty for the accuracy and completeness of the information, investors should do their own research and may seek professional advice before investing.

Comments

We need your insight to fill this gap
Leave a comment