Keywords

1 Introduction

The term financial distress (or financial failure in some literature) has different meanings, depending on the context. In a broad sense, financial distress can be used to indicate any condition when the business operation and financial circulation of a company fails to proceed normally or stagnates due to internal and external factors. To be specific, the financial distress of a company is a process, typically represented by persistent losses, events of default and insolvency, and finally corporate bankruptcy.

Real estate business, different from the general manufacturing industry, is characterized by large capital investment, long recovery cycle, high debt ratio and etc. In this instance, real estate enterprises in this capital-intensive industry are sometimes confronted with tense capital chain due to slow property sales, and thus leading to the financial distress. In 2008, under the influence of global financial crisis, a lot of property developers suffered from the capital pressure went bankruptcy, leaving the construction in a shutdown state (called as Làn Wěi Lóu in Chinese). The recent example is Top 10 Chinese property developer Greentown China (HKG: 3900). In September 2011, because of the corporate real estate trust business, Greentown triggered the investigation from China Banking Regulatory Commission and then the stock market worried about its tight capital chain. In 1 week from Sep. 22 to Sep. 27, the share price dropped by more than 30 %. The evidences above have preliminarily indicated that the financial distress in the real estate company has its own characteristics. As a consequence, how to establish a financial distress early warning model applicable to listed real estate companies has become the common issue to be solved for both academics and business managers in China.

The research on financial distress early warning model emerged in the 1960s. The application of financial statement analysis in the early warning model at the early stage is univariate models based on the single financial ratio, such as the ratio of cash flow to total debt [1]. Soon after, multivariate models were established with the multiple discriminant analysis (MDA) method, which drove financial distress early warning research into the period of the Z-Score model and its derivative models such as ZETA [2, 3]. Then, other nonlinear statistical techniques are introduced in financial distress prediction, such as Logit [4] and Probit methods [5]. The hybrid model integrated with the MDA and Logit method was also constructed to achieve the longer prediction length [6]. In China, the financial distress analysis and its early warning models developed rapidly along with domestic security market and accounting system. The F-score model (Failure Score Model) was established on the basis of Z-score model by adding the new variable to reflect the variant cash flow [7]. Similarly, with the application of MDA method to modify the Z-score model, another derivative model (referred as Z0 model hereinafter) was proposed, and its predictive accuracy for financial distress is higher based on 120 listed companies randomly selected from A-share market [8].

Existing literature on financial distress early warning model mostly focuses on all listed companies, so the distinctive business model and the accounting feature in real estate companies is omitted in these models, which could be inapplicable. This paper attempts to make contributions by extending the financial distress research to real estate companies and furthermore with the application of the MDA method, the prediction accuracy for financial distress in real estate companies can be improved.

The remainder of the paper is structured as follows: Sect. 117.2 evaluates the prediction accuracy of two general models based on the financial statement of listed real estate companies from the A-share market, and recognizes the better one as the baseline model. Sect. 117.3 establishes a financial distress early warning model through the modification of the baseline model with the MDA method. Conclusions are presented finally in Sect. 117.4.

2 Baseline Model Selection Based on Predictive Accuracy of Financial Distress

Based on the predictive accuracy for financial distress, we can evaluate and compare two different early warning models, including the traditional Z-score model and its derivative Z0 model applied for Chinese listed companies.

2.1 Basic Assumptions

In order to conduct the comparative research, we need distinguish financially distressed companies from listed companies in healthy financial state. Therefore, some basic assumptions need to be made to identify the judging rule and the forecast time limit for financial distress.

  1. 1.

    Take the ST firmFootnote 1 designated as the judging rule for financial distress

    According to the China Securities Regulatory Commission (CSRC) regulation of the special treatment mechanism, a listed company shall be classified as an ST firm if the audited results of the most recent two fiscal years show that it has suffered a loss. Simply speaking, all the ST firms are in terrible or abnormal financial situation, and thereby the use of special treatment as the judging rule for financial distress is reasonable for listed real estate companies in China which is also supported by previous researchers [7, 8].

  2. 2.

    Set the time limit of the financial distress early warning model as 5 years

    The aforementioned literature indicates that, the longest time of the early warning model for financial distress is 5 years. In other words, based on the data of a listed company in the financial year T, the early warning model is just capable to predict its financial situation before the year T+5. When the time limit exceeds 5 years, the forecast accuracy of financial distress sharply drops down as the year increases [6]. In this instance, the time limit of the early warning model is set as 5 years in this paper.

2.2 Sample Companies

According to the industry classification standard defined by the CSRC, we select 99 real estate companies listed on A-Share market as the sample. At the end of 2011, 99 listed real estate companies consist of 9 ST firms and 90 non-ST firms. Given that the predictive power is restrained within 5 years for early warning models according to previous research [2, 8], the basic dataset is sourced from the financial statement of the fiscal year 2006. In order to evaluate the predictive accuracy for listed real estate companies, the 99 companies are divided into two groups as ST and non-ST, containing 9 and 90 companies respectively. The descriptive statistics of the whole sample is provided in Table 117.1.

Table 117.1 Descriptive statistics (unit: billion yuan)
Full size table

2.3 Baseline Model Selection

In order to assess the predictive accuracy of financial distress early warning model, the discriminant accuracy matrix is constructed, as shown in Table 117.2.

Table 117.2 Discriminant accuracy matrix
Full size table

Table 117.2 describes a matrix used to assess the discriminant accuracy. H1 and H2 represent the number of correct discriminant samples; M1 denotes the number of Type I error, which misclassifies ST firms as non-ST firms; M2 denotes the number of Type II error, which misclassifies non-ST firms as ST firms. The sum of H1, H2, M1 and M2 represents the total sample. The ratio R that the sum of H1 and H2 accounts for the total sample can be sued to assess the model accuracy of the early warning, as below.

$$ \mathrm{ R}={{{\left( {{{\mathrm{ H}}_1}+{{\mathrm{ H}}_2}} \right)}} \left/ {{\left( {{{\mathrm{ H}}_1}+{{\mathrm{ H}}_2}+{{\mathrm{ M}}_1}+{{\mathrm{ M}}_2}} \right)}} \right.}\times 100\% $$
(117.1)

Following the discriminant analysis standards of Z-score model, the Z scores of ST firms should be less than 1.81, and that of non-ST firms should be larger than or equal to 1.81. In the case of Z0 model, the value of ST firms should be less than 0.5, and that of non-ST firms should be larger than or equal to 0.5. Therefore, we can calculate the discriminant accuracy for each model based on the sample data of listed real estate companies, and the matrix is listed separately in Tables 117.3 and 117.4.

Table 117.3 Discriminant accuracy analysis based on Z-score model
Full size table
Table 117.4 Discriminant accuracy analysis based on Z0 model
Full size table

From Table 117.3, the discriminant accuracy of Z-Score model for listed real estate companies in China is \( {{\mathrm{ R}}_1}={{{\left( {5+48} \right)}} \left/ {{\left( {5+48+4+42} \right)}} \right.}=54\% \).

From Table 117.4, the discriminant accuracy of Z0 model for listed real estate companies in China is \( {{\mathrm{ R}}_2}={{{\left( {8+60} \right)}} \left/ {{\left( {8+60+1+30} \right)}} \right.}=76\% \). By contrast, the accuracy of early warning for financially distressed real estate companies using Z0 model is higher than using Z-score model. Accordingly, we use Z0 model as the baseline model, and modify it with the financial statement of sample companies, and thus establish a financial distress early warning model applicable to listed real estate companies in China.

3 Financial Distress Early Warning for Listed Real Estate Companies

3.1 Early Warning Setup with the MDA Method

The baseline Z0 model applies the MDA method in its modeling process, we follow this methodology to modify Z0 model in line with the financial characteristics of listed real estate companies. The composition of Z0 model is listed as in Eq. (117.2).

$$ {{\mathrm{ Z}}_0}=0.517-0.460\;{{\mathrm{ X}}_1}-0.388\;{{\mathrm{ X}}_2}+9.320\;{{\mathrm{ X}}_3}+1.158\;{{\mathrm{ X}}_4} $$
(117.2)

The definition and the calculation of each independent variable Xi (i = 1, 2, 3, 4) in Eq. (117.2) are described in Table 117.5.

Table 117.5 Definition and calculation of each variable in Z0 model
Full size table

In general sense, higher Z0 value indicates for healthier financial situation of the company, and further results in lower possibility of having the financial distress. Then, we modify the Z0 model with the MDA method, in the following steps:

  1. 1.

    Select predictive variables and group variables

    Select Xi (i = 1, 2, 3, 4) contained in Z0 model as the predictive variables and define the group variable as 1 and 2, respectively for ST firms and non-ST firms. The descriptive statistics of two groups are listed in Table 117.6.

    Table 117.6 Descriptive statistics of predictive variables in two groups
    Full size table

    From Table 117.6, we can find that, except the variable of capital/debt ratio (X1), the average values of the rest predictive variables in non-ST firm group are greater than the ST firm group. Thus, it can be concluded, non-ST real estate companies perform better than ST companies, in terms of long-term, short-term solvency, current and accumulative profitability, and the sample data is to some extent proved to be appropriate.

  2. 2.

    Test the significance of the discriminant function

    By testing the eigenvalue of the discriminant function and its variance ratio, we can evaluate the explanation power of the discriminant function (see Table 117.7).

    Table 117.7 Eigenvalue and variance ratio of the MDA function
    Full size table

    Seen from Table 117.7, the number of the function is one and its eigenvalue is 0.585. The variance ratio (to what extent total variance can be explained) reaches 100 %, which reflects the discriminant function has strong explanation power for the sample data. In the MDA method, the significance test is also required, as shown in Table 117.8. Wilks’ λ is 0.231, χ2 statistics is 43.785, demonstrating the overall significance of the discriminant function.

    Table 117.8 Significance test of the MDA function
    Full size table
  3. 3.

    Calculate coefficients in MDA function

    The coefficients in the MDA function are listed in Table 117.9.

    Table 117.9 MDA function coefficients
    Full size table

    Based on the coefficients listed in Table 117.9, the specific form of financial distress early warning model can be given as in Eq. (117.3).

    $$ \mathrm{ F}=0.236-0.178\;{{\mathrm{ X}}_1}+0.195\;{{\mathrm{ X}}_2}+0.610\;{{\mathrm{ X}}_3}+1.139\;{{\mathrm{ X}}_4} $$
    (117.3)

    To clarify, the definitions of predictive variables in Eq. (117.3) are exactly the same as in Eq. (117.2).

  4. 4.

    Determine the discriminant standard

    After reaching the early warning model, we need further determine the discriminant standard for financial distress. Based on Fisher’s linear discriminant model [2, 9], the mean values of both groups can be calculated as F1 and F2. The mean values of non-ST firm group and ST firm group are 0.254 and -2.260 respectively. According to the symmetry classification rule, the turning point F* is the average value of F1 and F2, so F* is computed as \( {{\mathrm{ F}}^{*}}={{{\left( {{{\mathrm{ F}}_1}+{{\mathrm{ F}}_2}} \right)}} \left/ {2} \right.}=-1.0 \).

Following the discriminant standard, we can substitute the financial statement of listed real estate company into the early warning model denoted as Eq. (117.3). If the score F is greater than –1.0, the company is financially healthy under normal status. On the contrary, if the score F is less than or equal to –1.0, the company is classified into financially distressed company under poor financial status.

3.2 Reverse Test Using Sample Data

By substituting the original sample data into Eq. (117.3) to calculate F values, we can re-group the sample companies based on the discriminate standard (see Table 117.10).

Table 117.10 Predictive accuracy of discriminant analysis model
Full size table

Table 117.10 shows that, the predictive accuracy of financial distress for listed real estate companies reaches \( \mathrm{ R}={{{\left( {8+90} \right)}} \left/ {{\left( {8+89+1+1} \right)}} \right.}=98\% \), and improves by 22 % compared with the Z0 model (76 %). The result here indicates that the early warning model modified based on the MDA method can provide more accurate prediction for listed real estate companies than Z0 model.

The model robustness is very important for financial distress prediction. In sought to test the robustness of our model, we use 2007 financial statement of the same 99 listed real estate companies, re-calculate its predictive accuracy, the result as shown in Table 117.11.

Table 117.11 Robustness test based on predictive accuracy
Full size table

Seen from Table 117.11, the predictive accuracy \( \mathrm{ R}={{{\left( {8+87} \right)}} \left/ {{\left( {8+87+1+3} \right)}} \right.}=96\% \). Thus, it can be concluded that, this model keeps its high accuracy for financial distress, and the robustness is testified.

4 Conclusions

This paper provides a financial distress early warning model applicable to listed real estate companies in China, by using the multiple discriminant analysis method to modify a baseline Z0 model. The model established in this paper has provided higher forecast accuracy for financial distress than other models. In the context of deepening adjustments in real estate industry and fast-changing global economy, the model can help business managers and market investors to recognize the potential financial risks in listed real estate companies, so as to take actions at the earlier stage to prevent from running into the bankruptcy or considerable losses.

This paper is mainly focused on the application of financial distress early warning research to analyze the listed real estate companies in China. However, the specific mechanism of the financial distress generating in real estate companies is not completely covered in this paper, the government policy on property sales, for instance. The research in the future can make in-depth explorations into this issue based on the characteristics in real estate business model, financial operation and accounting system, so as to further improve the applicability of the financial distress early warning research in real estate companies.