Survival Analysis of Logistics Service Providers: An Empirical Study of Chengdu, Area in China

This paper worked on a sample of 6791 logistics establishments registered in Chengdu, China over the period 1984-2016 to understand the survival status of logistics service providers (LSPs) by non-parametric Kaplan-Meier estimation, together with Cox proportional hazard regression model, to identify factors affecting the failure of LSPs. In particular, it studies the interaction effect between LSPs’ size and entry timing and location. The empirical results show that: 1) Regarding the survival time, 1365 of the 6791 sample LSPs exited from the market by 2017. The exit rate is 20.1%, and the average life of the 6791 LSPs is about 6 years. 2) The survival of LSPs depends on their ty-pology, ownership structure. And there is no significant difference in the probability of survival for both independent LSPs and logistics branches after controlling the effects of other variables. 3) Location and entry timing also play an important role in the survival of small-scale LSPs, but these factors cannot explain large-scale LSPs’ failure.


Introduction
In China, the logistics industry is a basic and strategic industry and is widely valued by governments at all levels. Logistics is always playing an increasing role in the industrial structure transformation, transportation structure adjustment, supply chain innovations and domestic demand stimulus, and provides impor-these firms, the exit numbers were 311 thousand, accounting for 7.9% of the total firms that exited from the market in China, which was ranked fourth in all industries. Hence, it is urgent to analyze the factors influencing the survival of LSPs to help LSPs make correct business decisions and reduce the exit rate.
In this paper, we used a sample of 6791 LSPs to understand the post-entry performance of the logistics industry from 1985 to 2016 in Chengdu, China. As an important logistics hub city in the western region, the number of China-Europe freight trains via Chengdu has approached 1600 in 2018, ranking first in China for three consecutive years. In Chengdu, the logistics industry was considered as a strategic industry in 2004, and became an important factor in attracting foreign enterprises such as Intel and Foxconn. Therefore, it is of great practical significance and value to use Chengdu as a research area.
Existing research on the firm survival is mainly concentrated in the industrial field and rarely involves the service sector [2] [3], particularly the logistics industry. And a small number of studies focusing on the survival of service firms have also found significant differences in size, survival and growth with manufacturing enterprises [4] [5]. Hence, This paper draws upon the Resource-Based View of the firm, which contends firms are a heterogeneous bundle of tangible and intangible resources [6] [7], and applies the Cox proportional hazard regression model, combined with the Kaplan-Meier estimation, to identify the factors influencing the survival of LSPs. In addition, special emphasis is also placed on the interaction effect between LSPs' size and entry timing and location to clarify the impact mechanism efficiently.
The innovations and contributions of the paper are as follows: 1) The most important contribution of it is to indicate the relative importance of LSPs' type, ownership structure, location, entry timing, and size factor to the failure risk of LSPs. Especially through the interactive effects between LSPs' size and entry timing and location, it explains in detail the impact mechanism of LSPs' survival. States.
The rest of the paper is structured as follows. Section 2 reviews the relevant literature and proposes research hypothesis, while Section 3 is devoted to the data and variable description. Section 4 presents the picture of survival in the Chinese logistics industry by Kaplan-Meier estimation. Section 5 presents the empirical results. And the final part is the conclusions and managerial recommendations.

Literature Review and Research Hypotheses
The ultimate criterion for firm performance is the firm's survival [8]. And the length of survival is one of the most widely used measures of firm performance [9]. The earlier studies emphasize the entry and exit of the firms, focusing more on the impact of the firm's entry process, the exit risk and entry on market performance, and treated the survival process between the entry and exit as a "black box", and less on post-entry performance [10]. These researches have exaggerated the impact of entry on market performance [11]. Compared with the empirical data, what happens to firms subsequent to their entry is at least as important as the entry process itself [12]. As a result, the research on the post-entry performance has been increasing. A large number of relevant empirical studies on different manufacturing industries in different countries have been carried out [7] [13], becoming an important perspective to understand the industry dynamics and growth process.
The survival of firms depends on several factors, as summarized by Josef, et al. [14] and Manjón-Antolín and Arauzo-Carod [2], and these factors are mainly

Type of LSPs
According to the Resource-Based View, resources are both heterogeneous distributed among firms and imperfectly mobile [6]. In the logistics industry, resources can be tangible (e.g. equipment, plants, fleets, hardware), or intangible (e.g. organizational processes, skills, know-how, reputation) [20]. Those heterogeneous resources across different LSPs may lead to different logistics performance and competitive advantage in the same market. And Lai [20] found that different types of LSPs do have significant differences in service performance. As service performance is related to the survival of firms, we hypothesized a link between the type of LSPs and the probability of survival.
H1. The likelihood of survival varies with the type of LSPs.

Branch
According to Wang [21], logistics enterprises are composed of enterprise attribute elements and logistics attribute elements. In general, the enterprise attribute elements are usually including information center, command center, negotiation coordination, etc. which is common within the headquarters of logistics enterprises. While the logistics attribute elements are including equipment, warehouse, fleets, etc. tangible resources, which is more involved in logistics branches. Following this logic, the Resource-Based View suggests that distinct resources and capabilities between logistics headquarters and branches will lead to different levels of competitive performance. On the other hand, in literature on firm survival, there is also empirical evidence showing that non-branch entrants face lower exit risks than branch entrants [22] [23]. And many LSPs in this research sample have set up branches. This produced the following hypothesis: H2. Non-branch LSPs have a better survival chance than their branch counterparts.

Ownership
There is empirical evidence showing that the ownership structure of firms matters in survival chances. Some authors have found that foreign-owned plants have a higher probability of exit than their domestic-owned counterparts in Ireland and Spanish manufacturing firms [24] [25] [26]. So, we proposed the following hypothesis: H3. Foreign-owned LSPs have better survival prospects than their domestic-owned counterparts.

Location
Another dimension expected to affect the duration of firms, in the long run, is the location [23]. However, researches on the impact of location on the survival of firms have not yet reached a consistent conclusion. Fotopoulos and Louri [27] found that manufacturing companies located in Greater Athens have better survival prospects than others in the rest of the country, while Strotmann [22] found that manufacturing companies in rural areas have a lower risk of exit than their urban areas counterparts. However, in logistics literature, many studies have found logistics sprawl phenomenon, especially for new logistics establishments [28] [29]. This suggested a preference in location determine for LSPs. Based on those, we hypothesized a link between the location of LSPs and the probability of survival.
H4. LSPs in rural areas have better survival prospects than their urban area counterparts.

Timing of Entry
There is evidence that the timing of entry plays an important role in firm survival. According to Klepper [9] [30], enterprises that enter earlier are more likely to survive longer, because they can achieve higher profits in the early stages of the

Size
The positive relationship between survival rate and size has been validated in numerous of empirical studies [31] [32]. Firms' size and age represent the efficiency differences arising from differences in experience, managerial abilities, production technology and firm organization. At the same time, larger firms have a lot of financial resources, and have advantages in raising social funds, obtaining tax incentives, and gathering high-quality human resources [7] [18] [33].
Therefore, we hypothesized a link between size and the probability of survival.
H6. The probability of survival positively depends on LSPs' size.

Data Source
This paper utilizes a source of data set derived from the National Enterprise

Variable Description
Our variable of interest is the duration of an LSP, defined as the time elapsed between the entry and the exit of the LSP, that is, the period between the foundation of the LSP (entry) and the end of its activities [5]. Information on entrants is based on the commercial registration date of the LSP. Likewise, information on exits is based on the dissolution date. For some LSPs, this period can be subject to right censoring (i.e. when the exit does not take place or still active). However, survival models can account for right censoring [35] [36]. Although the data is available up to December 2016, we prolonged the observation time to the next year (December 2017) in order to observe more exits of those opened in the last 3 years, since it is difficult to observe "deaths" for these young LSPs.
For H1, we worked with one categorical variable, Firm type, capturing attributes of the type of LSPs-transportation service providers (TSPs), warehousing service providers (WSPs), freight forwarding companies (FFCs) and integrated logistics service providers (ILSPs) are represented by 1, 2, 3 and 4 respectively. This classification is based on China's "Classification and Evaluation Index for Logistics Enterprise" and related scholars' research [20] [37].
For H2, we used one dummy variable, Branch, to measure whether an LSP is a branch or not-the variable that takes the value 1 when an LSP is a branch and 0 elsewhere.
For H3, we used one dummy variable, Ownership, to measure whether an LSPs belongs to foreign-owned or domestic-owned, the variable that takes the value 1 when an LSP is foreign-owned and 0 elsewhere. "Foreign-owned LSPs" in this research are defined as all types of foreign-funded LSPs, including LSPs from Hong Kong, Macao, and Taiwan, which provide logistics services for other manufacturing or commercial companies.
For H4, We worked with one dummy variable, Location, to capture different attributes of locations of LSPs, the variable that takes the value 1 when the firm is located at rural area and 0 elsewhere. For H6, as we were interested in testing the effect of size, we used current registered capital (current refer to observation end date December 2017), to measure the size of the LSPs. Compared with initial size, the current size is found to be a better predictor of firm survival [31]. Then we followed Agarwal and Audretsch [38], distinguish between "small" and "large" firms by classifying firms as small if their current registered capital is less than the 60 th percentile of the registered capital distribution for all sample LSPs. Finally, we used one dummy variable, Size, to represent the size of LSPs. The variable that takes the value 1 for large scale and 0 for small scale.

Kaplan-Meier Estimation of the Survival Time of the Samples
The Kaplan-Meier estimator is the most widely used method for estimating survival functions, as it is a nonparametric maximum likelihood estimator with extremely few restrictions [10]. Indeed, the only restriction to consider is that the observed companies, if the data are censored, are assumed to have continued behaving the same way as they did until the death event occurred [5]. the survivor function ( ) S t is the probability of survival past time t or, equivalently, the probability of failing after t [39]. The survival function is shown below: there are j n LSPs who are supposed to be at risk of an exit. Being at risk means they have not experienced an exit nor have they been censored prior to time j t . If any cases are censored at exactly j t , they are also considered to be at risk at j t . Let j d be the number of LSPs who die on time j t [36].
Based on the Kaplan-Meier estimation method, this paper uses Stata14 to estimate the survival function of LSPs. The estimated survival curve by Kaplan-Meier is shown in Figure 2. In order to investigate whether the survival function of LSPs between G groups corresponding to each variable are significantly different, we perform the log-rank test, which is the most widely used test for differences in the survival function [36]. Under the null hypothesis, the different groups of LSPs have the equivalent survival function. The test statistic is approximately chi-square in large samples with G-1 degrees of freedom, where G denotes the number of groups corresponding to each covariable [40].

Model Design and Proportional Hazard Test
In this section, we use the cox model [41] [42] to capture the effects of explanatory variables upon death (hazard rates) rather than upon times to death [43]. In addition, it corrects for the problem of censored data, which uses the following hazards model specification: where ( ) , h t X represents the hazard at time t for an LSP with a given specification of a set of explanatory variables denoted by X. That is, the ( ) 1 2 , , , p X x x x  represents a vector of predictor variables that are being modeled to predict an LSP's hazard [40]. Where 1 2 , , , p β β β  is a set of unknown regression coefficients and ( ) 0 h t is an unknown non-negative baseline hazard function [43].
Through the partial maximum likelihood estimation, we estimate the value of Due to the fact that Firm type is a categorical variable, we introduce four dummy variables: TSPs, FFCs, WSPs and ILSPs (Dummy variable that takes the value 1 if an LSP belongs to that type and 0 otherwise), to evaluate the exit risk between different groups for each given covariate. To avoid multiple collinearities, we use ILSPs as a reference category. In the same way, we deal with the Entry timing variable, introducing three dummy variables: ES (early-stage), MS (medium-stage) and LS (late-stage), dummy variable that takes value 1 if an LSP's commercial registration date belongs to corresponding period and 0 elsewhere. We also set LS as the reference category. Therefore, the Cox proportional hazard model is initially set as follows: That is, uj r is the difference between the covariate value for the failed observation and the weighted average of the covariate values (weighted according to the estimated relative hazard from a Cox model) over all those subjects at risk of failure when subject j failed [39].
The test results are shown in Table A2 (See Appendix). Based on Grambsch and Therneau [44], the null hypothesis that the hazard rates are proportional over time for Size variable is violated at the five percent level, with the p-value is 0.000 (The P values of other variables were all higher than 0.05). Therefore, it is necessary to adjust for the Model (3).

Model Adjustment and Analysis Results
Since the PH assumption is not satisfied in the Model (3)   After control for other variables and at the 5% level of significance, the four types of LSPs face the risk of exit is ISLPs < TSPs < WSPs < FFCs. The hazard ratio for the FFCs are 3.1919, implying that the odds of the FFCs of exiting the market are more than three times that of the ISLPs, controlling for all other variables. A possible explanation is that, unlike the other three types of LSPs, ISLPs have differentiated competitive advantages. In addition to providing basic transportation, warehousing, and freight forwarding services, they also provide more diversified service offerings, such as logistics system design and information management. Finally, the results support that the performance of integrated service logistics providers is the best while freight forwarders are worst [20]. Hence, Hypothesis 1 is accepted.
Hypothesis 2 pays attention to the difference between the survival of branch and non-branch LSPs. Although Kaplan-Meier estimation indicates that the branch group has a shorter survival time than their non-branch group, there is no significant difference in the survival experience of the two groups after controlling the effects of other variables. From the perspective of the supply chain, the branch can be regarded as a partner in the outsourcing of an independent LSP. That is to say, the branch is a link in the logistics process, and its operating results are directly related to the performance of the entire supply chain members. In addition, Ono [45] found that due to the dependence of the branch on headquarters or other affiliates, the branch has a lower dependence on external suppliers and has less operational risk than independent enterprises. Therefore, the survival status of the branch is not necessarily worse than that of the independent LSPs. This is consistent with Audretsch and Mahmood [43] who found that the difference in survival experience between the branch and the non-branch is caused by the characteristics of the firm and the external environment, rather than its own reasons. Hence, hypothesis 2 is rejected.
Hypothesis 3 focuses on the impact of the ownership structure of the LSPs on survival. At the 5% level of significance, the exit risk of foreign-owned LSPs is nearly half of domestic-owned, after controlling for other variables, which shows that foreign-owned LSPs have better survival prospects, which echoes with the findings of Hull, et al. [46]. The possible explanation is that the foreign-owned LSPs entering the Chinese market are large-scale and have strong anti-risk capabilities. Secondly, foreign enterprises have rich management experience and stable customers. Therefore, the survival and development of foreign-funded enterprises are better.
Considering the interactions effect of location, entry timing and size in hypotheses 4, 5 and 6, we find that the Model (4) had the best fitting effect, and its corresponding p-value was 0.0004 in 3 degrees of freedom through the likelihood ratio test. Therefore, we use the results of the Model (4) to verify these three assumptions.
Hypothesis 4 concerns on the impact of location on survival. After controlling other variables, the exit risk of LSPs located in the rural area is about ( ) times that of urban areas. The results are divided into two categories: 1) When 1 Size = (large-scale LSPs), at the 10% level of significance, the exit risk in rural area is 0.5119 times that of the urban area; 2) When 0 Size = (small-scale LSPs), at the 1% level of significance, the exit risk in the rural area is 0.6910 times that of the urban area. This shows that the exit risk of LSPs located in the suburbs is lower than that of urban LSPs, especially for small-scale ones (1% significance level). A possible explanation is that due to the high land price and increased congestion, land shortage, and lack of logistics infrastructures, such as parking and loading space, the urban core area is no longer suitable for organizing logistics services [47]. Furthermore, social conditions in the suburbs are more attractive to LSPs, such as lower staff wages and higher acceptance of tough logistics jobs due to lack of job opportunities and higher unemployment [29]. Finally, many suburban local governments have adopted favorable land-use policies, environmental standards, and financial incentives to attract logistics investment, which also contributes to the survival of LSPs to some degree. However, for large-scale LSPs, the impact of location on their survival remains to be further tested, since this conclusion is only established at a level of significance of 10%. One possible interpretation is that large enterprises are usually more self-sufficient and therefore less dependent on external resources. In addition, small-scale LSPs are more sensitive to labor costs than large ones [48].  [49]. Since then, the market mechanism has begun to play a greater role in the Chinese logistics industry. Therefore, economies of scale are beginning to work. Hypothesis 6 focuses on the impact of size on survival. Because the size variable is a hierarchical variable that is not directly incorporated into the model. Therefore, it is impossible to directly compare the exit risks of large-scale and small-scale LSPs. We can only find that large scale LSPs are much more likely to survive than small scale ones in Kaplan-Meier estimation without controlling other variables. Hence, the impact of size on LSPs' survival remains to be further confirmed.

Conclusions and Recommendations
The exit of firms in multiple fields, such as mining, manufacturing, medicine and technology industry, has been widely investigated. However, it seems to be the first study to analyze the exit of firms in logistics industry. Little is conse-  Conclusions as below: 1) The survival of LSPs depends on their typology, ownership structure. There is no significant difference in the probability of survival for both independent LSPs and logistics branches after controlling the effects of other variables.
2) There is no significant difference in survival probability between earlier and later entrants of large-scale logistics enterprises. This is different from Klepper [9], who did not consider interaction effect between firm size and entry timing had found the earlier manufacturers entered the US automobiles, tire, television, and penicillin industries, the lower hazards they faced. The possible reason is that the earlier entrants with large-scale in logistics were mostly state-owned enterprises, which underwent a change of ownership during the reform and opening-up process and became new enterprises.
3) There is also a difference regarding the effect of location on firm survival.
For example, for small-scale LSPs, rural area LSPs have lower hazards than the corresponding risk in urban areas, while this trend cannot be accepted at 5% significance for large-scale LSPs. This finding is different from Strotmann [22], who found that manufacturing companies in rural areas have a lower risk of exit than their urban areas counterparts. This is closely related to the trend of suburbanization of Chinese logistics enterprises [28] [50]. Due to the lack of fixed resources, such as high-standard warehouses and railway dedicated lines, small LSPs are the main objects to be relocated under the pressure of environment and congestion in central urban areas. In addition, the Chinese government has established many public logistics parks in the suburbs and has provided many supporting policies for enterprises moving into the park. This has driven small logistics enterprises to continuously relocate to suburban logistics parks and achieve better survival.
Based on the statistical research of survival of LSPs, this paper puts forward the following recommendations: 1) The survival prospects of integrated logistics enterprises are the best, which shows that diversified operations are conducive to reducing operating risks.
However, the excessive degree of diversification will distract the attention of LSPs and have a negative impact on survival. Therefore, LSPs need to pay attention to the degree of diversification.
2) The empirical results in this paper have confirmed the importance of ownership structure on LSPs' survival. Hence, the government should lower the barriers to entry for foreign-owned logistics enterprises so that they can enter a wider market, thereby promoting comprehensive and healthy competition be-