Algorithmic Trading Regulation and Stock Market Liquidity: Evidence from China’s A-Share Market

1. Introduction

Algorithmic trading has become an important feature of modern financial markets, but its implications for China’s A-share market remain unsettled. High-frequency trading (HFT), which relies on low-latency algorithms to submit, revise, and cancel orders at high speed has fundamentally transformed the market microstructure landscape. Whether HFT enhances or degrades market quality remains one of the central debates in financial economics (Biais et al., 2015; Menkveld, 2013; O’Hara, 2015). This question takes on particular significance in China’s A-share market, where institutional environments differ markedly from those of developed exchanges.

In this paper, we study the liquidity effects of restricting HFT in China’s A-share market. We exploit the introduction of the Provisions on Program Trading in the Securities Market (Trial), issued by the China Securities Regulatory Commission (CSRC) on May 11, 2024, as a quasi-natural experiment. This regulation imposes explicit constraints on HFT, including restrictions on order submission rates exceeding 300 per second or 20,000 per day. The exogenous nature of this policy shock, combined with cross-sectional variation in HFT exposure across stocks, allows us to implement a difference-in-differences (DID) framework for causal identification.

To identify heterogeneous exposure to the regulation, we construct a multidimensional HFT intensity index (HFTI) from tick-level order data. The index combines six order-level proxies: the order-to-trade ratio, cancellation frequency, trade frequency, order lifetime, best quote change frequency, and orders per shareholder. Stocks in the top 30% of the HFTI distribution are classified as more exposed to HFT and form the treatment group, whereas stocks in the bottom 30% form the control group. We then estimate a difference-in-differences model around the regulatory shock.

Our main findings are as follows. First, restricting HFT leads to a statistically and economically significant improvement in stock liquidity, as measured by both the Roll (1984) bid-ask spread and the Pastor & Stambaugh (2003) price impact measure. Second, the effect is strongly state-dependent: liquidity improvements are concentrated during periods of stock price decline, whereas no significant effect is detected during market rallies. This pattern is consistent with the view that HFT tends to withdraw liquidity precisely when it is most needed, amplifying fragility through procyclical order cancellations and momentum-driven selling (Kirilenko & Lo, 2013; Brogaard et al., 2014). Third, the liquidity benefits of HFT restrictions are disproportionately larger for small- and mid-cap stocks and for stocks eligible for margin trading, reflecting HFT’s preference for targets with thinner order books or greater leverage flexibility.

Our paper contributes to the literature along three dimensions. First, we provide causal evidence on the liquidity effects of HFT regulation from an emerging market context. While prior studies predominantly examine developed markets with deep liquidity pools and sophisticated investor bases (Hendershott et al., 2011; Brogaard et al., 2014; Menkveld & Zoican, 2017), evidence from markets with different institutional features remains limited. China offers a distinct institutional setting characterized by retail participation, a T+1 settlement rule, price limits, and the absence of formal market-making obligations for most stocks. These institutional features make China an informative setting for evaluating the effectiveness of HFT regulation.

Second, we develop a multidimensional HFT measurement approach that combines multiple order-level proxies into a composite index. This mitigates the well-known measurement error inherent in single-proxy approaches (Hasbrouck & Saar, 2013) and provides more reliable treatment assignment for the DID framework.

Third, beyond documenting average treatment effects, we provide new evidence on the state dependence of HFT’s liquidity effects. While the theoretical literature has long recognized the potential for HFT to exacerbate liquidity crises (Kirilenko & Lo, 2013; Biais et al., 2015), direct empirical evidence on how HFT restrictions affect liquidity during market stress remains scarce. Our finding that the benefits of HFT restrictions are concentrated during declining markets offers policy-relevant evidence that targeted regulation can serve as a stabilizing mechanism.

The remainder of the paper is organized as follows. Section 2 reviews the related literature and develops the hypothesis. Section 3 describes the research design. Section 4 presents baseline estimates, mechanism analysis, heterogeneity analysis, and robustness tests. Section 5 concludes.

2. Literature Review and Research Hypothesis

The effect of HFT on market liquidity is a subject of active debate. Market microstructure theory suggests that liquidity fundamentally depends on the willingness of limit-order traders to provide depth, with its stability determined by information asymmetry and trader heterogeneity (Foucault et al., 2005; Li et al., 2021). Within this framework, HFT may either improve liquidity by reducing adverse selection costs or amplify fragility through procyclical trading behavior.

A substantial body of evidence documents the liquidity-enhancing role of HFT. By exploiting low-latency technology, high-frequency traders can rapidly update limit orders and increase depth near the best quotes, thereby narrowing bid–ask spreads (Hasbrouck and Saar, 2013; Boehmer et al., 2021). Ke and Zhang (2020) further show that HFT acts as a competitive liquidity supplier: by posting limit orders, high-frequency traders reduce spreads and lower execution costs for arbitrageurs. Algorithmic competition in limit-order placement also weakens the pricing power of traditional market makers and compresses trading costs (Hasbrouck and Saar, 2013; Jones, 2013). These studies suggest that, under normal market conditions, HFT can be an important source of market liquidity.

In contrast, a growing body of work highlights the procyclical and fragile nature of HFT-provided liquidity. Brogaard (2010) argues that high-frequency traders’ liquidity demand is primarily driven by short-lived arbitrage opportunities rather than long-term inventory motives, implying that HFT may consume rather than supply liquidity. Kirilenko and Lo (2013) further highlight the procyclical nature of HFT liquidity provision: high-frequency traders provide liquidity in tranquil markets, but their limited risk-bearing capacity and algorithmic herding can quickly turn them into liquidity demanders during market downturns. Consistent with this view, empirical evidence shows that, during extreme market conditions, high-frequency traders often cancel orders aggressively or trade in the direction of price movements, causing order-book depth to collapse and liquidity to dry up (Jain et al., 2016; Musciotto et al., 2021). Brogaard et al. (2017) also find that HFT can exploit liquidity suppliers through speed advantages and short selling, increasing adverse selection and widening bid-ask spreads.

These findings indicate that HFT’s role in liquidity is state-dependent. While generally enhancing liquidity in stable environments, it may exacerbate liquidity shortages in stressed conditions. This duality is particularly relevant for China’s retail-dominated A-share market, where shallow depth magnifies the impact of HFT withdrawals. The 2024 Provisions on Program Trading in the Securities Market (Trial) therefore provide a rare quasi-natural experiment to test whether restricting HFT can alleviate liquidity risks, especially during stocks decline periods.

Based on the theoretical and empirical literature reviewed above, we propose the following hypothesis:

Hypothesis. Algorithmic trading regulation improves stock liquidity in China’s A-share market, with a stronger effect during periods of stock price decline.

3. Research Design

3.1. Empirical Model

We employ a DID approach to identify the causal effect of restricting HFT on the stock liquidity. The baseline specification is:

$${Liquidity}_{it}=\alpha +{\beta }_1({Treat}_i\times {Post}_t)+\gamma X_{i,t-1}+{\theta }_t+{\mu }_{i,q}+{ϵ}_{it}$$

(1)

where ${Liquidity}_{it}$ denotes stock liquidity measures. ${Treat}_i$ is the treatment indicator, which equals one if stock $i$ is classified into the treatment group and zero otherwise. Treatment group assignment is determined using tick-by-tick order data to identify stocks’ exposure to program trading. ${Post}_t$ is the policy indicator, which equals one for May 2024 and onward, and zero otherwise. The ${\beta }_1$ captures the treatment effect of HFT restrictions on stock liquidity. The ${\theta }_t$ represents monthly time fixed effects. The ${\mu }_{i,q}$ denotes firm-quarter interactive fixed effects controlling for time-invariant firm characteristics and quarterly financial reporting effects. $X_{i,t-1}$ represents control variables including market capitalization, book-to-market ratio, price-earnings ratio, turnover, and volatility. Table 1 shows the definitions of these variables.

3.2. Dependent Variables

Measures of stock liquidity can generally be classified into two categories: bid-ask spread measures and market impact measures (Sarr & Lybek, 2002). Specifically, we employ the Roll_Spread to capture the bid-ask spread, and the Gamma measure to assess price impact.

The Roll_Spread measure, proposed by Roll (1984), is calculated as follows:

$${Roll\_Spread}_t=\left\{\begin{array}{ll}2\sqrt{-cov\left(\Delta P_d,\Delta P_{d-1}\right)},& \mathrm{if}cov\left(\Delta P_d,\Delta P_{d-1}\right)<0,\\ 0,& \mathrm{if}cov\left(\Delta P_d,\Delta P_{d-1}\right)⩾0.\end{array}\right.$$

(2)

where ${Roll\_Spread}_t$ represents the monthly spread measure calculated from daily price data, $\Delta P_d$ denotes the price change between day d and day d-1. This metric indirectly estimates the bid-ask spread by measuring the reversal in consecutive price changes. A higher Roll_Spread indicates greater transaction costs and poorer liquidity.

The Gamma measure, proposed by Pastor & Stambaugh (2003), is specified as:

$$R_{i,d+1}^e={\theta }_i+{\phi }_i{R}_{i,d}+{\gamma }_{i}sign\left(R_{i,d}^e\right)\times {Volume}_{i,d}+{\epsilon }_{i,d+1}$$

(3)

where $R_{i,d+1}^e$ is the excess return of stock i on day d+1; $R_{i,d}$ is the stock’s return on day d; $\mathrm{s}\mathrm{i}\mathrm{g}\mathrm{n}\left(\right)$ is the sign function that takes the value 1 for positive values, -1 for negative values, and 0 for zero; and ${Volume}_{i,d}$ represents trading volume. The coefficient ${\gamma }_i$ captures the impact of stock returns and trading volume on day d on the excess return of day d+1. When liquidity is high, current trading volume has minimal impact on subsequent returns. Thus, the absolute value of ${\gamma }_i$, denoted as $\left|{\gamma }_i\right|$, is employed as a proxy for stock liquidity, where a larger value indicates poorer liquidity.

3.3. Data and Descriptive Statistics

We use A-share data from both the Shanghai and Shenzhen stock exchanges from January 2023 to May 2025. We exclude ST stocks and financial firms from our sample. The tick-by-tick data and minute-level price data are sourced from a Chinese securities firm. All other data are obtained from the CSMAR database. All variables are winsorized at the 1% and 99% levels to mitigate the influence of outliers. Table 2 reports the descriptive statistics.

3.4. Identification Strategy

Accurately identifying HFT remains challenging. A key contribution of our study is the construction of a multi-dimensional HFT index based on multiple measures, which can mitigate the measurement error inherent in single-proxy approaches.

3.4.1. Measurement Proxies

Proxy 1: Order-to-Trade Ratio. The order-to-trade ratio (OTR) is calculated as:

$$OTR={\displaystyle \frac{TotalOrderValue}{TotalTradeValue}}$$

A higher OTR indicates a smaller proportion of orders resulting in trades, reflecting greater HFT influence.

Proxy 2: Cancellation Frequency. Cancellation frequency (CF), measured as the number of order cancellations within one-second intervals, is a key indicator of HFT behavior:

CF = Number of 1-second cancellations per day

The pattern of order cancellations within one second can sensitively reflect this dynamic adjustment behavior. Higher cancellation rates indicate more frequent strategic adjustments, demonstrating stronger HFT presence.

Proxy 3: Trade Frequency. Trade frequency (TF), measured as orders per minute, is calculated as:

$$TF={\displaystyle \frac{DailyOrderCount}{Dailytradingminutes}}$$

With A-share trading hours from 9:30-11:30 and 13:00-15:00 (240 minutes total), this metric directly captures order submission intensity. HFT’s algorithmic decision-making generate exceptionally high order submission rates, making trade frequency a clear indicator of HFT participation intensity.

Proxy 4: Order Lifetime. Order lifetime (OL) measures the average duration from order submission to either execution or cancellation, calculated as:

$$OL={\displaystyle \frac{\sum (OrderExecutionTime-OrderSubmissionTime)}{TotalOrders}}$$

HFT requires ultra-fast execution, typically resulting in extremely short order lifetimes. Smaller values of this metric better reflect the rapid entry and exit characteristics of HFT, making it a crucial identifier of HFT.

Proxy 5: Best Quote Change Frequency. Best quote change frequency (BQ) captures the number of changes per minute to the best bid or ask price, calculated as:

$$BQ={\displaystyle \frac{DailyChangesatBestQuote}{Dailytradingminutes}}$$

HFT exploits minute price discrepancies by frequently triggering changes to the best quotes. Higher values indicate more frequent quote updates, reflecting more intensive HFT.

Proxy 6: Orders per Shareholder. Orders per shareholder (OS) represent the ratio of daily order volume to the number of shareholders, calculated as:

$$OS={\displaystyle \frac{DailyOrderVolume}{NumberofShareholders}}$$

Active HFT generates frequent order submissions, increasing this ratio. Generally, higher values indicate greater HFT influence on a stock, as algorithmic trading typically generates more orders relative to the shareholder base.

3.4.2. Standardization

To control for the confounding influence of firm size, we grouped A-share stocks from the Shanghai and Shenzhen exchanges (excluding financial and ST stocks). Specifically, we evenly divided the sample into five groups based on market capitalization, with each group containing 879 stocks.

Within each capitalization quintile, we standardize all six HFT proxies using the z-score method:

$$Z_{t,group}={\displaystyle \frac{x_t-{\mu }_{group}}{{\sigma }_{group}}}$$

(4)

where $x_t$ represents the raw indicator value, ${\mu }_{group}$ denotes the group mean, and ${\sigma }_{group}$ is the group standard deviation.

3.4.3. Calculate the High-Frequency Trading Index

We aggregate the six standardized proxies into the high frequency trading index (HFTI) using equal weights:

$$HFTI={w_{1}Z}_{OTR}+{w_{2}Z}_{CF}+w_3{Z}_{TF}-w_4{Z}_{OL}+{w_{5}Z}_{BQ}+{w_{6}Z}_{OS}$$

(5)

where $w_1$=$w_2$=$w_3$=$w_4$=$w_5=w_6=$1/6. Given the difficulty in distinguishing the relative importance of each proxy in capturing HFT, we adopt an equal-weighting approach. $Z_{OTR}$, $Z_{CF}$, $Z_{TF}$, $Z_{OL}$, $Z_{BQ}$ and $Z_{OS}$ represent the standardized values of the OTR, CF, TF, OL, BQ and OS. This composite index provides the basis for partitioning stocks into treatment and control groups.

3.4.4. Identification

We use tick-by-tick order data for A-share stocks on the Shanghai and Shenzhen exchanges from January 2023 to May 2024 as the basis for calculating the relevant indicators and the HFTI. Based on this dataset, we compute high-frequency trading metrics and derive each stock’s average HFTI through a weighted aggregation.

We then rank the sample stocks by HFTI, designating the top 30% as the treatment group and the bottom 30% as the control group. The former is more strongly affected by high-frequency trading and can better capture policy-induced changes, while the latter is less influenced and thus serves as a benchmark. This grouping provides a solid foundation for identifying the net effect of restricting high-frequency trading on stock liquidity.

4. Empirical Results

4.1. Baseline Results

Table 3 reports the estimated effects of restricting HFT on stock liquidity. Columns (1) and (2) use the bid-ask spread measure (Roll_Spread) as the dependent variable, while columns (3) and (4) use the price impact measure (Gamma). The treatment interaction term is consistently negative and statistically significant for both Roll_Spread and Gamma, indicating that restricting HFT significantly improves stock liquidity.

4.2. Parallel Trends Test

We use the following dynamic specification to conduct the parallel trends test:

$${Roll\_Spread}_{i,t}=\alpha +{\displaystyle \sum _{k=-7}^6}{\beta }_k({Treat}_i\times {Post}_{t,k})+\phi X_{i,t-1}+{\theta }_t+{\mu }_{i,q}+{ϵ}_{i,t}$$

(6)

where ${Post}_{t,k}$ denotes event-time dummies, with $k=0$ representing the policy month.

Figure 1 presents the estimated coefficients ${\beta }_k$ with their 95% confidence intervals. The pre-treatment coefficients are insignificant, indicating no notable differences in stock liquidity between the HFT group and the non-HFT group before the policy. This confirms that the model satisfies the parallel trends assumption.

4.3. State-Dependent Effects

We next examine whether the policy effect depends on market states. If algorithmic trading regulation improves liquidity primarily by limiting procyclical liquidity withdrawal during stress, the effect should be concentrated during declining markets. To test this prediction, we partition the sample into declining-market observations (monthly stock return < 0) and rising-market observations (monthly stock return > 0) and re-estimate the baseline model within each subsample.

Table 4 presents the results. During periods of stock price decline, restricting HFT significantly reduces both bid-ask spreads and price impact. In contrast, the coefficients are statistically insignificant during market rallies. This asymmetry is consistent with the theoretical prediction that HFT amplifies liquidity fragility during stressed conditions through rapid position adjustments, massive order cancellations, and momentum-driven selling (Kirilenko & Lo, 2013). By constraining these procyclical behaviors, the regulation preserves market depth and mitigates liquidity dry-ups precisely when liquidity provision is most valuable.

4.4. Cross-Sectional Heterogeneity

4.4.1. Market Capitalization

If HFT disproportionately affects stocks with thinner order books, the benefits of restricting HFT should be more pronounced for smaller stocks. To test this, we partition stocks into large-cap (market cap ≥ ¥50 billion) and small- and mid-cap groups (market cap < ¥50 billion).

Table 5 confirms this prediction. HFT restrictions significantly reduce both bid-ask spreads and price impact for small- and mid-cap stocks, whereas the effects on large-cap stocks are statistically insignificant.

This divergence reflects fundamental differences in market depth. Smaller stocks typically exhibit lower market depth and higher price sensitivity, making their liquidity more vulnerable to external shocks and HFT. Consequently, they benefit more significantly from HFT restrictions. Table 6 further shows that these liquidity improvements for smaller stocks are concentrated during market declines, reinforcing the mechanism identified in Section 4.3.

4.4.2. Margin-Trading Eligibility

We further examine whether the policy effects vary by margin-trading eligibility. Chinese HFT frequently utilizes margin trading and short selling to implement short-term arbitrage strategies and to achieve de facto T+0 trading. These mechanisms provide greater leverage and flexibility, enabling HFT to exert stronger influence on market liquidity, particularly during downturns when momentum short selling through margin accounts may exacerbate liquidity evaporation.

Table 7 shows that HFT restrictions significantly improve liquidity for margin-tradable stocks, while the effects on non-margin-tradable stocks are insignificant. This heterogeneity reflects HFT’s strategic preference for stocks with greater leverage flexibility. Table 8 further confirms that the liquidity improvements for margin-tradable stocks are concentrated during market declines, consistent with the mechanism analysis.

5.5. Robustness Tests

4.5.1. Placebo Test

To ensure the robustness of our DID results, we conduct a placebo test following Li et al. (2016). This approach simulates a “false” scenario without actual policy impact by randomly assigning virtual treatment groups. Specifically, we perform 1,000 random samplings to construct placebo treatment groups and re-run our regression model. This random assignment ensures the virtual treatment groups are not based on actual HFT exposure, thus theoretically should not exhibit any genuine policy effect.

Figure 2 presents the placebo test results. The distribution of placebo coefficients clusters tightly around zero, and most placebo p-values exceed 0.10. The actual treatment coefficient lies well outside the placebo distribution, confirming that the baseline finding reflects a genuine policy effect rather than statistical artifact.

4.5.2. Alternative Liquidity Measures

Following Chung et al. (2020), Eaton et al. (2021), and Amihud (2002), we employ three alternative liquidity measures: the relative quoted spread (QSP), relative effective spread (ESP), and the Amihud illiquidity ratio (AMIHUD).

The relative quoted spread (QSP) is calculated as:

$${\mathrm{Q}\mathrm{S}\mathrm{P}}_{i,d}=\sum w_{i,∆\tau }\times {\displaystyle \frac{a_{i,\tau }-b_{i,\tau }}{m_{i,\tau }}}$$

(7)

where ${\mathrm{Q}\mathrm{S}\mathrm{P}}_{i,d}$ represents the day d relative quoted spread , $a_{i,\tau }$ denotes the best ask price, $b_{i,\tau }$ indicates the best bid price, $m_{i,\tau }$ is the average of $a_{i,\tau }$ and $b_{i,\tau }$, and $w_{i,∆\tau }$ represents the time interval between consecutive quotes. This time-weighted bid-ask spread better captures the price formation characteristics of China’s limit order market. Higher values of the relative quoted spread indicate greater transaction costs and poorer liquidity. After computing daily relative quoted spreads, we obtain monthly measures by calculating their averages.

The relative effective spread (ESP) is calculated as follows:

$${ESP}_{i,d}=\sum w_{i,∆\tau }\times {\displaystyle \frac{2\left|p_{i,\tau }-m_{i,\tau }\right|}{m_{i,\tau }}}$$

(8)

where $p_{i,\tau }$ is the transaction price. Higher ESP values indicate poorer liquidity. Daily ESP values are averaged to obtain monthly measures.

The Amihud illiquidity ratio (AMIHUD) is calculated as:

$${AMIHUD}_{i,d}={\displaystyle \frac{\left|R_{i,d}\right|}{T_{i,d}}}$$

(9)

where $R_{i,d}$ represents the daily return and $T_{i,d}$ denotes the day d trading volume. This measure captures the price impact per unit of trading volume—higher values indicate poorer liquidity. After computing daily Amihud measures, we obtain monthly values by calculating their averages.

Table 9 presents regression results using these alternative measures. The policy is shown to significantly reduce all three indicators, indicating that restricting HFT significantly improves liquidity across all alternative measures, reinforcing the robustness of our main findings.

5. Conclusions

Whether algorithmic trading enhances or impairs market quality remains an important question in China’s financial market. This paper exploits the May 2024 Provisions on Program Trading in the Securities Market (Trial) issued by the CSRC as a quasi-natural experiment to examine how algorithmic trading regulation affects stock liquidity in China’s A-share market. Using tick-level order data, we construct a multidimensional HFT intensity index and implement a difference-in-differences design based on cross-sectional variation in stocks’ pre-policy exposure to HFT.

The central finding is that algorithmic trading regulation significantly improves stock liquidity. Treated stocks experience lower bid-ask spreads and lower price impact after the regulation, and the result is robust to placebo tests, alternative liquidity measures, and the inclusion of firm-quarter interactive fixed effects. These findings suggest that, in retail-dominated China’s A-share market, some forms of high-speed algorithmic trading may impose liquidity costs that are not fully captured by evidence from developed markets.

The results also reveal substantial state dependence. Liquidity improvements are concentrated during periods of stock price decline, whereas no significant effect is detected during market rallies. This asymmetric pattern is consistent with the theoretical prediction that HFT can amplify liquidity fragility during stress through procyclical order cancellations and momentum-driven trading. Regulation appears to mitigate these destabilizing behaviors and preserve liquidity when market resilience is most needed.

Cross-sectional analyses further show that the liquidity improvement is stronger for small- and mid-cap stocks and for margin-tradable stocks. These segments are more vulnerable to liquidity shocks because of thinner order books, higher price sensitivity, or greater leverage and short-selling flexibility. The evidence therefore supports a targeted regulatory approach that focuses on market segments where algorithmic trading may generate larger externalities.

The findings have implications for the policy debate on algorithmic trading in emerging markets. Evidence from developed markets often shows that HFT compresses spreads and reduces transaction costs under normal conditions. Our results do not contradict this view. Instead, they show that the net effect of HFT depends on institutional context and market state. In markets characterized by retail participation, T+1 settlement, price limits, and limited market-making obligations, targeted restrictions on HFT can improve liquidity and strengthen market resilience without eliminating the broader informational role of algorithmic trading.