Economic policy uncertainty and shifting systematic risk in the Chinese equity market

The Chinese economy is a major driver of global trade and investment, making its market dynamics critical for risk assessment and investment decisions. This study compares the Chinese equity market with the global benchmark, highlighting post-COVID-19 divergence. In the literature it is argued that EPU increasingly shapes corporate and macroeconomic behavior: fiscal policy under uncertainty can produce nonlinear effects on growth and investment (Belke and Goemans, 2022), while heightened uncertainty depresses firm-level investment and constrains strategy (Montes and Nogueira, 2022). Regulatory actions, such as anti-corruption campaigns, also generate systematic and idiosyncratic uncertainty, influencing corporate decision-making (Kim et al., 2023; Ito et al., 2023). Kumar and Prasanna (2024) show that rising uncertainty increases corporate bond yields in emerging Asian economies, including China, affecting financing and investment. Soni et al. (2023) find that EPU causes stock market volatility in India, while gold remains a safe haven. Chinese policy uncertainty also spills over to U.S. asset returns (Lee et al., 2021), and global markets, including ASEAN indices, respond to both American EPU and Chinese policy initiatives like OBOR and COVID-19 (Lean et al., 2024).

Despite extensive research, few studies compare the investability of China’s equity market with the global benchmark over time, as most focus on individual stocks or segments (Burdekin and Weidenmier, 2015). Evidence shows that China’s EPU influences macroeconomic conditions, firm investment, financial markets, and international spillovers, underscoring the role of domestic policy and uncertainty in shaping economic outcomes and informing policymakers, investors, and researchers.

The present study fills a gap in the literature by examining how Chinese equities—especially in the technology sector—perform relative to global stock returns, while considering the moderating effect of EPU. We contribute to literature by employing a comparative approach to quantify the extent of decoupling from global benchmarks. Using four exchange-traded funds (ETFs)—FXI, GXC, CQQQ, and MCHI—this study provides a nuanced view of how Chinese equities align with or diverge from global market trends. This approach extends beyond Burdekin and Weidenmier (2015), who focused on China’s stimulus impact, by incorporating a comparative framework that assesses investment potential relative to global markets.

Building on Ito et al. (2023), this study examines how EPU fluctuations affect Chinese ETF betas over time, using up to 12 months of lagged EPU as threshold variables. Employing threshold and breakpoint regressions with recursive coefficient analysis, we identify structural shifts in ETF betas and assess EPU’s impact on Chinese equity risk. Focusing on the technology sub-sector, we find that post-COVID-19 global markets consistently outperformed Chinese ETFs, with declining betas following heightened EPU, indicating shifts in systematic risk and growing decoupling from the global benchmark.

Many scholars, such as Auer and Schuhmacher (2015) and Liang and Wei (2020), use Eq. (1) to estimate unconditional betas. This method decomposes monthly changes in a fund’s returns into two components: one reflecting global (systematic) comovements and the other capturing unique peculiarities specific to the Chinese market. That is:

Where:

R_t = monthly returns on a given Chinese ETF at time t,

R_Mt = Monthly returns on the vanguard total world ETF at time t,

β = beta coefficient,

D_t = a binary variable set to 1 for month t (t = 1, …, 12) in a calendar year and 0 otherwise,

α_t = average price returns on a given Chinese ETF in the corresponding month of the year during the entire estimation period,

ε_t = stochastic residual term.

Since there is no intercept in Eq. (1), the issue of the dummy variable trap is inapplicable even if all 12 months-specific-intercept dummy variables are incorporated in the model. Following common practice in the literature (e.g. French, 1980; Washer et al., 2016), we use the Newey-West (1987) heteroskedasticity and autocorrelation consistent (HAC) method to obtain robust estimators. A higher coefficient of determination in Eq. (1) suggests that the fortunes of a given Chinese ETF are more closely aligned with global market performance. It is also important to examine the parameter stability for beta in Eq. (1). To make robust inferences on beta’s stability, this study employs three approaches: (1) a discrete threshold regression model, where contemporaneous and lagged values of EPU (up to 12 months prior) are used as optimal triggers (thresholds) for causing significant shifts in beta coefficients; (2) a breaking regression analysis, which identifies endogenously determined break dates using a grid search; and (3) a recursive regression analysis, which observes the evolution of beta over time by adding one observation at a time in the estimation process.

It should be noted that EPU can alter a firm’s exposure to aggregate risk—the systematic component measured by β with respect to a benchmark—through several complementary channels. First, by raising the probability and perceived magnitude of economy-wide policy shocks, EPU increases the frequency of common macroeconomic disturbances that simultaneously affect many firms. A higher incidence of such common shocks elevates cross-sectional co-movement and can therefore change estimated β even when firm-specific variance is unchanged (Baker et al., 2016). Second, EPU modifies equilibrium investor preferences and the pricing of systematic exposures. When policy uncertainty rises, investors rebalance across risk classes and demand altered compensation for systematic risk, producing measurable changes in observed β as equilibrium risk premia and factor sensitivities shift (Pástor and Veronesi, 2013; Kelly et al., 2016). These adjustments imply that uncertainty-related βs carry distinct cross-sectional pricing implications.

Third, EPU influences corporate behavior—particularly investment, financing, and cash-holding decisions—which in turn reshape firms’ exposure to aggregate shocks. Firms that defer capital expenditure or alter leverage in response to elevated policy uncertainty change their sensitivity to market-wide fluctuations over subsequent months, producing delayed effects consistent with the lag structures examined in this study (Gulen and Ion, 2016). While EPU also affects volatility and liquidity, these are conceptually distinct channels. Volatility and liquidity reflect changes in the scale of returns or trading frictions, whereas β captures co-movement with a common factor. Hence, although EPU can raise volatility or reduce liquidity, such effects do not subsume the co-movement channel that directly modifies systematic exposure. This distinction justifies modeling time-varying β rather than focusing solely on conditional volatility or liquidity (Baker et al., 2016; Kelly et al., 2016).

A threshold regression model “assumes that the border between the two regimes is given by a specific value of the threshold variable” (Franses and Van Dijk, 2000, p. 772). For m number of threshold values, we will have m+1 number of beta coefficients. In the case of one single shift, we will then have two sets of betas in Eq. (2):

Where:

• 1(·) = the indicator function, which is equal to one if the condition in the parentheses is met and zero otherwise,

• d = the length of the delay, and

• Z_t = the threshold variable, which can be the present or past values of the EPU index (up to 12 lags) to be determined in a grid search.

• β¹ = beta in regime 1 when uncertainty is low, and

• β² = beta in regime 2 when uncertainty is high,

Unlike Eq. (1), beta can now switch from a low uncertainty regime to a high uncertainty regime at any point in time. This data-driven switch depends on the optimal value of the threshold variable, which will be one of the following candidates in our Z_t vector: (EPU_t, EPU_t-1, EPU_t-2, …, EPU_t-12). We select the optimum threshold variable that minimizes the sum of squared residuals (SSR). The “nuisance parameters” (τ, d) are estimated by conducting an iterative grid search. For each pair (τ, d) in the grid, the indicator function is first defined, and Eq. (2) is re-estimated through an iterative process. Then, within a standard 15% trimming region, both a lower and upper bound are established for the threshold parameter (Hansen, 2011). The objective within this range for τ is to minimize the SSR associated with the specified sets of parameters:

Where $Γ=(αi,βs,τ,d)$⁠. In this grid search, we incrementally adjust the lower bound by a small amount (e.g. 0.0001) and calculate the SSR. For instance, we start with (i.e. τ^l + 0.0001) and obtain the SSR. In the next iteration, we increase the threshold value to τ^l +0.0002 and record the new SSR. This iterative process continues until we reach the upper bound τ^u. Ultimately, we select the threshold value that yields the lowest SSR:

Furthermore, we also conduct a breaking regression, similar to the threshold regression in Eq. (2), but with the threshold variable now defined as a time trend (Z_t = T_t) [instead of Z_t = EPU_t] and using a one-month increment instead of a small value for the EPU index. In this method, we focus on the middle 85% of the observations. According to Bai and Perron (2003), threshold and breakpoint models are essentially equivalent and only differ in definition of Z_t-d. In the breaking regression, the break date (year/month or τ) is determined endogenously, splitting the sample into two parts. Initially, we allowed for up to four breaks or threshold points in both models, but only one proved statistically significant for our dataset. As a result, we proceed with a two-regime model instead of a multi-regime approach.

In addition, the recursive ordinary least squares (ROLS) method is used to estimate betas over time by sequentially updating it as new data points are added to the end of the sample (Brown et al., 1975). The process begins with an initial subset of the minimum data required to form the first estimate of the coefficient vector. New observations are then added one by one sequentially, with each update providing a refined estimate of the beta. This method can track changes in beta coefficients over time, making it valuable for detecting structural shifts or instability in the data.

Chinese equities are highly volatile and currently carry notable downside risk, making it essential to benchmark their risk–return profile against global markets. Risk-adjusted performance can be assessed through the Sharpe ratio—excess return per unit of total volatility (Sharpe, 1964)—though it assumes normally distributed returns and treats upside and downside volatility equally. Complementary measures address these limitations: the Omega ratio (Keating and Shadwick, 2002) captures the entire return distribution, including tail risks; the Martin ratio (Martin and McCann, 1989) incorporates drawdown depth and duration via the Ulcer Index; and the Calmar ratio (Young, 1991) evaluates returns relative to maximum drawdown over extended periods. Together, these metrics provide a fuller picture of downside exposure and tail risk, offering a more informative evaluation of Chinese equities in the current environment.

This study utilizes monthly price returns (calculated as logarithmic changes) for five ETFs—VT, FXI, GXC, CQQQ, and MCHI—over the sample period from April 2011 to November 2023. Table 1 presents key characteristics of these ETFs, including the number of holdings, dividend yields, inception dates, expense ratios, and top ten holdings. VT serves as a broad proxy for the global stock market, encompassing 9,829 stocks from a wide range of countries, with major holdings concentrated in leading U.S. technology firms. In contrast, FXI, the oldest U.S.-listed Chinese ETF, tracks a market-cap-weighted index of only 52 major Chinese companies, including top tech and financial names. These are the largest Chinese stocks listed on the Hong Kong Stock Exchange, including H-shares, P-chips, and Red Chips, but exclude mainland-listed A-shares and US-listed N-shares. It draws from the FTSE All World Index, with stocks screened for liquidity.

In comparison, GXC tracks a comprehensive, market-cap-weighted index of 1,169 investable Chinese shares, spanning all market-cap sizes and including major share classes like A, B, H, Red Chips, P Chips, and foreign listings. The index is float-adjusted, only including shares available to the public, and weights are determined by float-adjusted market capitalization. Arguably, GXC provides a more representative measure of the overall Chinese equity market. As of September 2024, MCHI also tracks a market-cap-weighted index of 651 investable Chinese shares, including H-shares, B-shares, Red Chips, P Chips, and foreign listings. Similar to GXC, it provides broad exposure across large- and mid-cap companies, generally investing at least 90% of its assets in the index’s securities, as can be seen in Table 1. Unlike the above four ETFs, CQQQ has a narrower focus and tracks a market-cap-weighted index of 151 investable Chinese technology stocks (similar to QQQ in the US), covering A-shares, N-shares, and other share classes. It includes all tech companies based on the Industry Classification Benchmark (ICB), offering broad exposure without factor-based screens.

The Chinese EPU index has been available since 1995, but as of September 2024, the most recent data extends only up to November 2023 ^[1]. Based on the inception dates listed in Table 1, the common sample period with available monthly data for all ETFs and the EPU index spans from April 2011 to November 2023. All five ETFs in this study are denominated in U.S. dollars, with data sourced from Yahoo Finance. While numerous other Chinese ETFs are available in the U.S., they were excluded from this study due to their recent inception dates. Incorporating them would have considerably shortened the sample period. For example, KWEB was launched in July 2013, ASHR in November 2013, and CXSE in September 2017.

Table 2 presents the descriptive statistics for the monthly returns of five ETFs over the common sample period from April 2011 to November 2023, encompassing 152 monthly observations. The total world stock return (VT) shows a positive mean monthly return of 0.449%. All Chinese ETFs, except for CQQQ, exhibit negative mean returns: FXI at −0.381%, GXC at −0.084%, and MCHI at −0.141%. Among the ETFs, CQQQ has the highest standard deviation at 8.25%, reflecting the highest volatility within the tech sector. The average mean return per unit of risk (standard deviation) for VT is the highest among all five ETFs listed in Table 2, indicating that VT offers superior risk-adjusted returns. All return series, except for FXI, are skewed to the left, suggesting a tendency for more frequent downward price movements. The kurtosis values for all return series exceed 3, with FXI and MCHI having the highest kurtosis at 5.08 and 4.51, respectively, indicating a higher concentration of extreme returns. The Jarque-Bera test results reject the normality assumption at the 1% significance level for all return series, highlighting significant deviations from normality and a propensity for extreme returns. The EPU index exhibits a positive skewness of 0.36 and a lower kurtosis of 1.88, indicating that its distribution also significantly deviates from normality.

Figure 1 illustrates the time plot of the EPU index over the sample period from April 2011 to November 2023, showing an overall upward trend. Economic uncertainty surged in December 2017, peaking towards the end of 2019, before declining until 2022, when it began to rise again. This sharp increase in EPU after 2017–2018 can be attributed to several factors, including escalating trade tensions between the United States and China, significant political events such as Brexit negotiations, and elections in Europe.

Figure 2 presents four individual graphs comparing the performance of each Chinese ETF against the global stock market ETF (VT) over the sample period from April 2011 to November 2023. The Chinese ETFs are shown by dashed red lines, while VT is represented by solid blue lines. The graphs reveal a pronounced trend: since the post-COVID-19 pandemic period, the performance gap between the Chinese ETFs and the global benchmark has widened significantly, particularly in the cases of FXI, CQQQ, and MCHI. This underperformance underscores the divergence in recovery and growth between the Chinese equity market and the global stock market.

Table 3 presents the results of the estimated Eq. (1) for four Chinese ETFs (CQQQ, FXI, GXC, and MCHI) using monthly data from April 2011 to November 2023, with the Newey and West (1987) HAC method applied to adjust standard errors for both heteroskedasticity and autocorrelation of unknown form. The estimated beta coefficients in Table 3 are all statistically significant at the 1% level and are fixed during the sample period, irrespective of fluctuations in the EPU index. The highest beta is associated with CQQQ, representing the technology sector (1.125), while the lowest beta belongs to FXI, which comprises only the largest 52 companies (0.857).

Based on the adjusted R² statistics, it can be argued that overall, the performance of the Chinese market is closely aligned with global market dynamics, particularly in the technology sector, where the adjusted R² is the highest. Interestingly, nearly all seasonal dummy variables are statistically insignificant at the 5% level, with the exception of a positive effect in January, observed only for CQQQ. Calendar anomalies are typically evident in the US and Europe markets (Valadkhani and O’Mahony, 2024). The Durbin-Watson (DW) statistics indicate that the residuals from the estimated equations are well-behaved and free from serial correlation. The Akaike Information Criterion (AIC), Schwarz Information Criterion (SIC), and Hannan-Quinn Information Criterion (HQIC) are reported at the bottom of Table 3. As discussed later, when the assumption of beta constancy is relaxed in threshold or break regression models (Tables 4 and 5, respectively), these three model selection criteria consistently point to the inadequacy of Eq. (1) in capturing the true systematic risk in the Chinese equity market.

Although the dependent variable is in logarithmic differences, the estimated equations in Table 4 perform better than those in Table 3 in terms of goodness-of-fit statistics (adjusted R²) and the three model selection criteria (i.e. AIC, SIC, and HQIC). In other words, when the moderating impacts of uncertainty are incorporated into the beta estimates, the threshold regressions outperform the non-threshold equations in Table 3, where betas were forced to be constant. The reported DW statistics once again indicate no sign of serial correlation. The estimated betas in Regime 1 (low uncertainty) and Regime 2 (high uncertainty) are all positive and statistically significant at the 1% level.

The optimum threshold variables are also shown in Table 4, clearly revealing that uncertainty influences the market on average between approximately 3 months (for large-cap companies) to 12 months (total market and the tech sector). Therefore, one can posit that uncertainty influences the four Chinese ETFs differently over time. For example, when EPU_t-11 exceeds 488, both GXC and MCHI, which are broad-based ETFs representing the entire market, enter the high uncertainty regime approximately after one year. This threshold for CQQQ is slightly lower at 413, but for the largest 50 Chinese companies, the optimum threshold is observed when EPU_t-3 is less than 632, indicating a faster response to uncertainty but with a greater threshold. Table 4 also shows that EPU affects the β only with significant lags of three to twelve months. This establishes a recursive dynamic in which past EPU influences current βs, thereby ruling out simultaneity. Since the influence operates purely through lagged effects, the concern that market conditions simultaneously drive policy uncertainty is eliminated under our identified lag structure.

The results consistently indicate that the performance of the Chinese market is more in sync with global benchmarks when uncertainty is lower. However, in Regime 2, when uncertainty is elevated, all betas for the Chinese ETFs fall three to four times. A Wald test was also carried out, revealing that β² is greater than β¹ at the 1% level (the results available from the authors upon request). This suggests that during periods of heightened uncertainty, China becomes more insular and less responsive to global market trends, failing to experience the same level of prosperity as the global economy, which continues an upward trajectory. For instance, when uncertainty is low, or the EPU_t-12 falls below 413, CQQQ increases by 1.70% in response to a 1% rise in the global market. Conversely, when uncertainty exceeds 413, a 1% increase in global returns results in only a modest 0.58% rise. Similar conclusions can be drawn from analyzing the estimated β¹ and β² coefficients in Table 4, suggesting that uncertainty causes China to fall behind significantly, making it difficult for the country to get back on track. These findings are not counterintuitive; rather, they are reinforced by the widening disparities between Chinese ETFs and VT, as shown in Figure 2, which further supports similar conclusions.

We also re-estimated the results reported in Table 4 using the Economic Policy Uncertainty (EPU) index developed by Huang and Luk (2020). A comparison between the results in Table 4 (based on Baker et al., 2016) and those presented in this  Appendix indicates that, according to the R², adjusted R², AIC, SIC, and HQC criteria, the EPU measure proposed by Baker et al. (2016) employed in the present study outperforms the alternative index by Huang and Luk (2020). One may argue that newspapers in mainland China do not operate under the same level of press freedom as those in Hong Kong. Nonetheless, the results derived from the two indices are broadly consistent for CQQQ and MMCHI (i.e. β¹ > β²) while some discrepancies are evident for FXI and GXC, as shown in Table 4 and the table in  Appendix. In this study, we therefore rely on the Baker et al. (2016) EPU index, as it consistently yields superior model selection statistics and exhibits greater explanatory power.

Table 5 presents the results from a breaking regression model, where a time trend represents the threshold variable Z_t. Similar to the results in Table 3, the findings in Tables 4 and 5 indicate that the seasonal dummy variables are once again statistically insignificant at the 5% level, with the exception of just a positive January effect for CQQQ and a negative February effect for the other three ETFs. This is surprising, given the widespread prevalence of calendar anomalies in U.S. and European markets (Valadkhani and O’Mahony, 2024).

The estimated equations in Table 5 have higher goodness-of-fit statistics (R² and adjusted R² statistics) and outperform the estimated equations in Table 3 pairwise based on all three model selection criteria (AIC, SIC, and HQIC). This reaffirms that the assumption related to the constancy of betas is not statistically defensible. However, it is worth noting that the results from Table 4 (threshold regressions) still outperform those from Table 5 (breaking regressions). Table 5 shows that the endogenously and iteratively determined break date for the betas of all four Chinese ETFs occurred in January 2020. Before this date (referred to as Regime 1), the betas for CQQQ, FXI, GXC, and MCHI were estimated to be 1.617, 1.319, 1.397, and 1.356, respectively. However, in the post-COVID era, the estimated betas experienced a significant decline, reaching values of 0.655, 0.414, 0.457, and 0.449, respectively. These results are not surprising, given the overall increase in uncertainty during this period (as depicted in Figure 1). Based on these observations, we can conclude that the Chinese market is becoming increasingly insular and less responsive to the prosperity enjoyed by the rest of the world. Furthermore, the estimated beta for CQQQ remained consistently higher than all three other ETFs both before and after January 2020, suggesting that in an upside market, the tech sector could outperform the overall Chinese market.

To check the sensitivity and robustness of our findings, Figure 3 presents the recursive coefficient estimates for the betas of the four Chinese ETFs. This plot demonstrates how the beta coefficients are updated sequentially with each new observation, enabling us to track changes and detect any structural shifts or instability in the data over time. As illustrated in Figure 3 and Table 5, we consistently observe a significant drop in all betas around January 2020, coinciding with the outbreak of COVID-19, while the betas were relatively stable prior to this date.

In addition, Table 6 provides a comparative analysis of four risk-adjusted return measures—Calmar, Martin, Omega, and Sharpe ratios—evaluating the performance of four Chinese ETFs against the global market (VT) across two distinct periods: pre-COVID-19 (9/09/2014 to 1/02/2020) and post-COVID-19 (1/02/2020 to 6/09/2024). A higher ratio across all metrics reflects more favorable risk-adjusted returns. For example, before COVID-19, the Calmar ratio of VT (0.971) significantly outperformed all Chinese ETFs, with CQQQ (0.587) being the closest competitor. Following the onset of the pandemic, VT (0.970) maintained its superior position, while the Chinese ETFs exhibited consistently lower ratios. Similarly, the Sharpe ratio before COVID-19 showed VT (0.646) once again leading the Chinese ETFs, with CQQQ (0.561) as the nearest competitor. Post-COVID-19, VT improved its performance, whereas Chinese ETFs experienced a decline. The same inferences can be made based on both the Martin and Omega ratios. Zhang et al. (2024) propose that effectively managing both intertemporal total risk and downside risk can significantly enhance the performance of Chinese equity funds. By focusing on these risk management strategies, fund managers can improve key risk-adjusted performance measures, such as the Sharpe ratio.

The results suggest that long-term market trends are more likely to be upward, but China risks missing significant gains without stronger integration with the global economy, restored confidence, and mitigation of wealth effects from declining property prices and investor uncertainty. The consistent outperformance of the Vanguard Global ETF relative to Chinese ETFs, especially post-COVID-19, highlights the importance of aligning with global markets. Challenges such as falling FDI, a fragile real estate sector, yuan depreciation, strict COVID policies, and rising economic policy uncertainty have undermined domestic and international confidence. While China has relied on fiscal and monetary stimulus, the current focus on long-term sustainability over short-term growth, combined with geopolitical tensions and trade disputes, has heightened market uncertainty, dampened innovation, and slowed recovery. Effective policy balancing short-term stabilization with long-term reform is crucial, as China’s economic trajectory has significant implications for global stability, trade, and investment.

Contagion refers to cross-market co-movements beyond fundamentals, while interdependence reflects persistent linkages (Forbes and Rigobon, 2002). Market integration measures the influence of global versus local factors on returns, often using time-varying β or latent-factor approaches (Bekaert et al., 2005). Our threshold and break regressions indicate that the post-2020 decline in β reflects genuine decoupling rather than volatility-driven correlation changes. Alternative metrics, such as the fraction of returns explained by global factors or R²-based measures, reinforce this interpretation and are more robust for assessing diversification benefits (Pukthuanthong and Roll, 2009). To further strengthen our results, we propose robustness checks comparing structural β shifts with risk-adjusted performance metrics (Sharpe, Omega, Martin, Calmar), confirming that observed changes in systematic risk are substantive rather than artefactual.

This study evaluates the Chinese equity market, particularly its technology sub-sector, by comparing its risk-adjusted performance to global benchmarks and examining the impact of Economic Policy Uncertainty (EPU) on systematic risk. Using monthly data from April 2011 to November 2023 and four ETFs (FXI, GXC, CQQQ, MCHI), the study employs threshold regression, breakpoint analysis, and recursive coefficient estimation to capture both the magnitude and timing of EPU’s effect on market betas. Findings show a declining trend in betas, accelerating after COVID-19, indicating increasing decoupling of the Chinese market from the global benchmark. Lagged effects of EPU on betas are observed over a 3–12-month horizon, with the technology ETF (CQQQ) most sensitive. Structural breaks around January 2020 highlight post-pandemic disconnection from global risk pricing. Risk-adjusted metrics (Sharpe, Omega, Martin and Calmar) reveal consistent underperformance of Chinese equities versus the global benchmark, particularly during periods of elevated uncertainty.