Wealth Effect of Asset Securitization in Real Estate and Infrastructure Sectors: Evidence from China

4.1 Samples and Data Sources

Because the rapid evolution of asset securitization in China began in 2014, this study adopts a sample period from 2014 to 2021. We collect issuance information on RE & Infra ABS from the Wind Database and identify 457 RE & Infra ABS. Then, we correlate the sponsors in the issuance information with pertinent listed firms, which are the sponsors or holding parent companies of sponsors, using the equity relationship charts provided by the Wind Database to define our treatmant group.^[3] Specifically, we first get all direct issuing firms’ information in the sample period from archival information on issuing registration of RE & Infra ABS. Second, we investigate which dominant shareholders (holding more than 50% of shares), within at most four layers of the equity relationship with direct issuing firms, are listed. The reason for such a match is that some direct issuing companies of RE & Infra ABS are non-listed project entities, which are exempt from the obligation to disclose detailed financial statements.

To align equity relationships with those at the time of securitization issuances, we update missing and historical equity relationships using the iFinD Database. Updating these historical equity relationships reduces the sample size by four firms, which have unclear or no equity relationships during the ABS issuance. In addition, we source corporate financial statement variables from the CSMAR Database. Finally, we identify 34 relevant listed companies as the treatment group samples.

For a comparative baseline, we identify other listed companies in the same industries as control group samples, based on the China Securities Regulatory Commission’s 2012 industry classification (CSRC 2012), referring to the practice of Jiang et al. (2022).

Following data collection, we implement a standardized sample adjustment procedure. Specifically, we initially exclude firms that fall under the finance sector, under special treatment, and under particular transfer categories. To mitigate the influence of outliers, we winsorize the top and bottom 1% of all continuous variables. Finally, our annual unbalanced panel dataset consists of 859 valid listed Chinese companies as sample firms, with 4,045 observations. Among them, the experimental group has 188 observations, and the control group has 3,857 observations. Given that the sample sizes of the experimental group and the control group exhibit asymmetry, this article will apply multiple propensity score matching methods to improve the comparability between the experimental group and the control group samples when conducting the robustness test. Specifically, by means of statistical techniques, this article will select control group samples from the full sample that are more comparable, and then re-perform the benchmark regression with the experimental group. This approach alleviates the issue of an excessively large sample size of the control group, thereby ensuring that the conclusion of the benchmark regression is more robust.

4.2 Variables

Following established methodologies in corporate finance literature (Cheung et al., 2015; Jiang et al., 2022), we adopt Tobin’s Q as the primary proxy for firm valuation. This metric captures both market performance and operational fundamentals, providing a more robust indicator than stock price movements alone.

To examine the impact of RE & Infra ABS issuance behavior, we construct an interaction term of the treatment indicator Treat i , t and the post-event indicator Post i , t . The binary variable Treat i , t assumes a value of 1 for firms that act as sponsors issuing the asset securitization during the observation window, and 0 otherwise. The variable Post i , t is coded as 1 for the issuance year and subsequent years for participating firms. Otherwise, it is set to 0.

Our empirical model incorporates a set of control variables from the perspective of firm characteristics and corporate management. In view of the corporate characteristics, we control for the firm size (lnasset), return on assets (ROA), and degree of financial leverage (Financialrisk). The valuation implications of financial leverage have been extensively examined in prior investigations (Adenugba et al., 2016; Dimitrov & Jain, 2008). At the corporate governance level, we control for leadership structure through a dummy variable (Dual) indicating CEO/Chairman duality, following Dahya et al.’s (2002) evidence on the governance implications of role separation. Board composition is accounted for through the proportion of independent directors (Independent), while managerial incentives are captured by the percentage of shares held by management (Magholding). Specific definitions and measurement methods of the main variables are shown in Table 2.

Table 3 shows the descriptive statistics of our key variables. The mean Tobin’s Q of 1.908 in our sample contrasts with Xiao and Zhang’s (2013) estimate of 3.197 for A-share listed companies (2007–2010). This discrepancy likely reflects the sample composition, as issuers of RE & Infra ABS predominantly operate in mature sectors with constrained growth prospects, typically commanding lower valuation multiples. Interestingly, while Cheung et al. (2015) report a mean Tobin’s Q of 1.268 for US REITs (1994–2004), our higher estimate may be attributable to the listing premium phenomenon in China’s developing capital markets.

4.3 Model Setting

To evaluate the wealth effect of RE & Infra ABS on sponsors, we utilize a two-way fixed-effects panel regression as shown in equation (1). This model framework offers two main benefits to our research. First, sponsors issue the asset securitization across different years. Flammer (2021) and Jiang et al. (2022) used similar models for panel data to assess the impact of financing tool issuances on the issuing firms, and this configuration helps mitigate potential confounding effects, as noted by Lin et al. (2021). This method identifies the impact of the events by analyzing the changes in the treatment group after the events compared to the control group and can effectively mitigate the influence of reverse causality.

Second, given that China is a developing nation with a stock market of limited efficiency, traditional event study methods based on stock reactions may not be reliable. It is plausible that information regarding the asset securitization issuances might leak well before official announcements or that the market possibly lags in its response to official announcements.

Our panel regression with two-way fixed effects is as follows, according to Flammer (2021), Lin et al. (2021), and Jiang et al. (2022).

where i indicates firms and t indicates years. The dependent variable Q i , t , representing firm value, is measured by Tobin’s Q. Treat i , t × Post i , t represents the occurrence of the asset securitization issuance behavior. Control i , t is a vector of covariates as described in detail in Section 4.2. δ i and ρ t are firm-fixed effects and year-fixed effects, respectively. ε i , t is the error term. In consideration of potential correlation within industries, we cluster the standard errors at the industry level based on CSRC 2012. We use industry clustering for the following reasons. First, as noted by Cameron et al. (2011), when performing nested two-way clustering, it suffices to cluster at the higher level. Second, the practice of tangible asset securitization is often strongly influenced by industry-specific policies, regulations, or market conditions. Third, our data reinforce the need for industry clustering. There are 4,045 observations across 859 firms, with an average of less than 5 observations per firm. Therefore, it is difficult to reliably estimate firm-level standard errors.

We employed the Hausman test to assess the appropriateness of the fixed-effects model. The test results strongly reject the null hypothesis (p = 0.0000 < 0.01), indicating that the random-effects model is misspecified and necessitating the use of fixed effects. Moreover, we implement a parallel trend test in Section 5.2.1 to validate the assumption that there is no significant difference between the treatment and control groups before RE & Infra ABS take place.

5.1 Benchmark Regression Results and Analysis

To test the impact of RE & Infra ABS on the value of sponsors, we run the regressions in equation (1). Table 4 presents the results. Column 1 displays the empirical outcomes, accounting for two-way fixed effects without any control variables. Column 2 incorporates additional control variables into the regression and clusters the industry-level standard errors.

The regression results are positively significant in both columns 1 and 2. According to column 2, the regression coefficient of the independent variable Treat i , t × Post i , t is 0.4917 at a p-value of 1% significance, which implies that issuances of the asset securitization raise the firm value in comparison to firms without this behavior. Our conclusion supports the Hypothesis developed in Section 3.

Our regression results support the view that asset sales add value to the enterprise (Clayton & Reisel, 2013; Finlay et al., 2018). However, our empirical evidence is contrary to the finding by Lockwood et al. (1996) that corporate asset securitization in the US industrial and automobile industries has no wealth effect on issuers.

5.2 Robustness of Benchmark Regression

5.2.1 Parallel Trend Test

To ensure that there is no interference from confounding factors, this article employs the parallel trend test, as shown in equation (2), referring to Xu et al. (2021), Jiang et al. (2022). The satisfaction of the parallel trend assumption means that there is no significant difference between the treatment group and the control group before issuance of RE & Infra ABS. We focus on the confidence intervals of the D j × Treat i , t ’s regression coefficients before issuance of RE & Infra ABS. Specifically, we first construct a new variable Diff i , t , which represents the number of years relative to the issuance year. Diff i , t is calculated as the current year minus the issuance year of the asset securitization. For the control group samples, which did not experience an issuance event, Diff i , t is undefined. The range of Diff i , t is from −7 to 6, in integers.

(2) Q i , t = τ + β j ∑ j = − 6 6 D j × Treat i , t + γ Control i , t + δ i + ρ t + ε i , t ,

where i indicates firms and t indicates years. D j is a year dummy variable that equals 1 if Diff i , t is exactly j, and 0 otherwise. To avoid multicollinearity, we exclude the baseline period, which is the 7th year before issuance (i.e., j = −7), from the regressions. All other settings follow those in equation (1).

To facilitate understanding, we further classify the interaction terms in Figure 1 into three categories: Pre ‘i’, Cur, and Post ‘i’. Pre ‘i’ denotes the years prior to the issuance year, Cur refers to the year of the issuance, and Post ‘i’ represents the years following the issuance year. Figure 1 demonstrates that the 95% confidence interval (the area between the two changing thin lines) always includes 0 before the issuance year, verifying that the experimental and control group samples satisfy the parallel trend assumption.

Figure 1

Parallel trend test graph. This figure depicts the parallel trend test and is created by software Stata 16. The horizontal axis denotes the years relative to the asset securitization issuance. Pre ‘i’ denote the years before the debut year. Cur denotes the current year of initial issuance. Post ‘i’ denote the years after the debut year. The vertical axis denotes the coefficient of the independent variable D j × Treat i , t . The bold solid line is the line connecting the coefficient of the independent variable in each period. The area between the two lines above and below the bold solid line represents the 95% confidence area.

The dynamic changes in firm performance after issuance of RE & Infra ABS are also illustrated. After the issuance year, the effect exerts an overall positive upward trend. The lower bounds of the 95% confidence intervals for the coefficients in the first and second periods after RE & Infra ABS are 0.04 and 0.03, respectively, above 0. The sustained response supports the viewpoint that financing through asset securitization is beneficial for corporate development in the long run.

5.2.2 Placebo Test

We conduct a placebo test for the samples, aiming to guarantee that the benchmark regression result does not arise from some random factor generated in the statistical process.

Referring to Chetty et al. (2009), we randomly allocate 34 firms from our samples as the treatment group who issue asset securitization in a random year for the first time. The number of random issuing firms, 34, accurately reflects reality within the sample period. This article re-runs the baseline model in pseudo samples for estimating the coefficient of Treat i , t × Post i , t in equation (1) and repetitions are 500.

Figure 2 illustrates the placebo test results, where the X-axis and the Y-axis denote the estimated coefficient and p-value, respectively. Figure 2 shows that the random simulation results approximate a normal distribution with a mean of 0, and most of them are above a p-value of 0.1. The coefficient for Treat i , t × Post i , t (0.4917) in our benchmark regression significantly deviates from the mean value of the random simulation process, suggesting that the benchmark regression outcomes are not merely coincidental.

Figure 2

Placebo test graph. This figure depicts a placebo test and is created by software Stata 16. The horizontal axis denotes the estimated coefficient of the independent variable Treat × Post. The vertical axis denotes the p-value. The hollow points represent the regression results of the coefficient of the independent variable Treat × Post in pseudo-samples. The horizontal dashed line indicates that 0.1 p-value. The vertical dashed line is the value of 0.4917, which is the coefficient of Treat × Post in our benchmark regression.

5.2.3 Propensity Score Matching Method

To address the sample self-selection issue, we employ the propensity score matching (PSM) method on the samples and then execute the benchmark regression.

Given that the significantly larger size of the control group samples can be utilized, we use the kernel function as the weight calculation method during the matching process. This ensures the construction of a control group that closely resembles the treatment group.

We employ 2013, the year before the first year in the sample period, as the matching year when issuances of the asset securitization did not happen. In the PSM process, we use a binary choice model to calculate the propensity score that is the conditional probability of entering the treatment group given a set of explanatory variables. Treat is defined as the explained variable, while the control variables in the benchmark model are set as the explanatory variables. Ultimately, we get a total of 2,950 observations after PSM based on the kernel function.

Table 5 demonstrates the balance test of firm characteristics and management between the experimental group and the control group before and after the PSM process. Table 5 indicates that the mean values of firm characteristics and management variables in the experimental group remain relatively consistent before and after applying PSM. Firms issuing asset securitization tend to be larger in size. Moreover, there is no significant difference in the other five covariates between the two sample sets. Following the PSM process, the disparity in all covariates narrows, and firms in both the experimental group and the control group exhibit greater similarity.

Table 5

Balance test of covariates

Variable	Before/After PSM	Mean		p-value of t-test	Significance
		Treatment	Control
Lnasset	Before	23.463	22.4	0.000	***
	After	23.32	22.91	0.371
ROA	Before	0.03829	0.03995	0.802
	After	0.03832	0.03937	0.879
Financialrisk	Before	1.1916	1.5602	0.141
	After	1.1978	1.3092	0.525
Dual	Before	0.09091	0.18924	0.246
	After	0.09524	0.13095	0.723
Independent	Before	0.36865	0.37299	0.727
	After	0.37033	0.3686	0.919
Magholding	Before	7.9231	8.4754	0.886
	After	8.2665	8.9732	0.904

This table depicts the balance performance of covariates before and after the kernel-based PSM method. The samples are the same as what we discussed in Section 4.1. p-value of the t-test refers to outcome derived from the balance test of covariates between samples in the experimental group and the control group. Null hypothesis of the balance test: There is no significant difference between the experimental and control groups on all covariates. *** denotes significance at the 1% level.

Figure 3 portrays the distribution of propensity scores for the treatment group samples and the control group samples before and after PSM, which visualizes the matching quality change. According to Figure 3, in the situation without matching, the control group samples gather in a low propensity score location below 0.4, and the density peak of the control group samples is nearly two times higher than that of the treatment group samples. To some extent, the distinctive distribution of density in the control group and the treatment group means a relatively significant difference in their probabilities of issuing the asset securitization. However, after sample matching, the probability density distribution in the two groups aligns closely, suggesting that the sample selection issue has been significantly mitigated.

Figure 3

Propensity scores’ density distribution graph. This figure depicts the propensity scores’ core density distribution before and after the kernel-based PSM process and is created by software Stata 16. The upper one is the distribution before the PSM process and the lower one is the distribution after the PSM process. The horizontal axis denotes the propensity score. The vertical axis denotes the density. The bold line denotes the treatment group, and the thin line denotes the control group.

After PSM, we re-regress equation (1) using treatment group samples and the matched control group samples based on the matched weights.

Column 3 of Table 4 demonstrates that the coefficient of Treat i , t × Post i , t is still positively significant at the 1% p-value after kernel-based PSM. This implies that issuance of RE & Infra ABS contributes to the sponsors’ value increment by 0.2396, compared to the matched control group, which further enhances the robustness of our benchmark regression.

Moreover, we employ PSM specification using one-to-one and one-to-two nearest neighbor matching methods. The regression coefficient of the independent variable Treat i , t × Post i , t remains positively significant. This indicates that our conclusion based on the PSM specification is robust and does not depend on the choice of matching methods. The regression results are shown in Table A1.

6.1 Debt Repayment Mechanism

In this section, we study whether the issuance of RE & Infra ABS is due to the motivation of relieving debt pressure. Under the finance hypothesis, firms sell assets through securitization behavior for financing purposes. Debt repayment pressure is often a significant factor driving corporate financing (Clayton & Reisel, 2013; Lang et al., 1995; Pang et al., 2024).

We introduce the leverage ratio (DebtRatio) as a proxy variable for total debt level and current ratio (CurrRatio), quick ratio (QuickRatio), and cash ratio (CashRatio) as proxy variables for short-term debt level. Equations (3) and (4) present our study settings for mediation effect testing. Equation (3) is designed for testing the influence of RE & Infra ABS on the debt level of sponsors, where the coefficient of interest is Treat i , t × Post i , t . Equation (4) is set for testing the influence of mediating variables on firms’ value, where we focus on the coefficient of Med i , t .

where Med i , t represents mediating variables. Calculation of these mediating variables is demonstrated in Table 6. Other settings are defined identical to those in equation (1).

Columns 1, 3, 5, and 7 in Table 7 report the regression results of equation (3) when debt ratio, current ratio, quick ratio, and cash ratio are used as mediating variables, respectively. As illustrated in these four columns, the coefficient of Treat i , t × Post i , t is significantly negative when the mediating variable is the debt ratio, and the coefficients of Treat i , t × Post i , t are all significantly positive when mediating variables are current ratio, quick ratio, and cash ratio. These results suggest that the issuance of RE & Infra ABS relieves the debt-overhang problem for sponsors.

Columns 2, 4, 6, and 8 in Table 7 illustrate the results of equation (4), and the coefficients of current ratio, quick ratio, and cash ratio are all significantly positive at the 1% level, which implies that the short-term debt reduction significantly benefits corporate value enhancement.

Our empirical findings support the perspective that asset securitization provides valuable liquidity to non-financial firms, as evidenced by U.S. data (Clayton & Reisel, 2013; Lemmon et al., 2014; Riachi & Schwienbacher, 2013, 2015). Additionally, Lemmon et al. (2014) find that a significant portion of non-financial asset securitization proceeds in the USA is used by firms for debt repayment. The reduced debt levels can help relevant firms that hold low liquidity and heavy assets lower their financial distress risk, thereby enhancing the wealth effect.

We draw a conclusion contrary to Pang et al. (2024) regarding the effect of asset securitization on issuer leverage. It is important to note that Pang et al. (2024) focus on all corporate ABS, while we spotlight RE & Infra ABS. Moreover, their study does not account for the fact that publicly listed companies may use subsidiaries exempt from disclosure requirements to issue ABS, potentially resulting in missing samples.

To ensure robustness, we employ another test of the mediation effect – the bootstrap method, referring to Preacher and Hayes (2008) and Karanika-Murray et al. (2015). The control variables are the same as those in the benchmark model. This method can overcome the limitations of the distribution assumption for samples. The key to determining the mediating effect under this method lies in whether the confidence interval excludes 0. Based on 500 bootstrap resamples and 95% bias-corrected (BC) confidence intervals, we estimate the indirect effect of issuing RE & Infra ABS on Tobin’s Q mediated by debt repayment capacity.

The results indicate that when mediating variables are the current ratio, quick ratio, and cash ratio, their respective point estimate values for the indirect effect are 0.024, 0.0367, and 0.0183 – all with 95% BC confidence intervals that lie entirely above zero. Specifically, the BC confidence intervals are [0.0068, 0.0516] for the current ratio, [0.0208, 0.0646] for the quick ratio, and [0.0065, 0.0394] for the cash ratio. These findings suggest that the debt repayment represented by these three variables significantly mediates the relationship between the issuance of RE & Infra ABS and Tobin’s Q in the bootstrap framework. These results align with preceding stepwise mediation regression test results, providing further evidence to support the mediation effect of debt repayment under the finance hypothesis. However, the BC confidence interval for the debt ratio straddles zero, that is [−0.0104, 0.0028], which does not support the debt ratio’s mediating role as the proxy variable for debt repayment between issuance and Tobin’s Q.

6.2 Moderation Effect of Tightening Financing Environment

In this section, we investigate the moderation effect of the tightening financing environment. If the finance hypothesis is valid, the wealth effect of RE & Infra ABS will be amplified under tighter financing conditions because of the increase in the scarcity of funds.

The New Asset Management Regulation, passed in 2018, is one of the most significant laws concerning China’s financing environment in recent years. This regulation aims to regulate non-compliant financing activities, greatly limiting financing availability in the financing market. Qiu et al. (2022) discovered that the size of the shadow banking business contracted following the introduction of the New Asset Management Regulation in China. The prohibition of non-compliant financing methods has compelled firms, in the short term, particularly those in real estate and infrastructure sectors – historically the primary stakeholders, to explore alternative financing channels.

To examine the moderation effect of the tightening financing environment, we introduce a dummy variable R of the New Asset Management Regulation as a proxy variable for the financing environment variation and construct the following model.

where the dummy variable R i , t equals 1 for years from 2018 onward, following the introduction of the new regulation, and 0 otherwise. Referring to Zhang and Zhang (2022), we incorporate the variable R i , t and its interaction term with Treat i , t × Post i , t into the benchmark model, as illustrated in equation (5). Other settings are the same as those in the benchmark model.

Columns 1 and 2 of Table 8 exhibit the regression results. Column 1 shows the coefficient of R i , t ’s cross-product term is significantly positive at the 5% level, and the value is 0.4617, which indicates that the wealth effect of RE & Infra ABS on the sponsors is pronounced after the tightening financing environment, and supports the finance hypothesis. This implies that finance by RE & Infra ABS can help firms in a more rigorous financing environment to receive greater firm value. The coefficient of R i , t is significantly negative at the 1% level, suggesting that the tightening financing environment induced by the New Asset Management Regulation indeed leads to a shock to the firm value of enterprises in the real estate and infrastructure sectors in the short term.

Our findings complement those of Gao et al. (2023), who find that larger financing constraints increase the issuance of corporate asset securitization in China. Our study further argues that firms under financing constraints issue non-financial asset securitization to derive value creation benefits, thereby providing further evidence on the issuance incentives of nonfinancial asset securitization in China.

Furthermore, our study diverges from Finlay et al. (2018)’s study. Finlay et al. (2018) demonstrate that asset divestitures by UK firms encountering industry-level financial difficulties resulted in a poor wealth effect due to the fire-sale effect. Conversely, our findings indicate that firms’ divestitures yield a positive wealth effect during times of industry distress.

To guarantee the robustness of the abovementioned finding, we employ the kernel-based PSM setting, similar to what we did in Section 5.2.3. Column 2 presents the result and derives the consistent conclusion.

6.3 Ownership and Industry Heterogeneity