Analysis of Technical Efficiency for the Hotel Industry in Vietnam Bach Ngoc Thang

Abstract Following the model of Battese and Coelli (1995), a frontier production model and a technical inefficiency model will be simultaneously estimated in this paper. Through empirical research, the paper estimates scores of technical efficiency at firm level for 474 firms in the hotel industry in Vietnam during the period 2000–2003. In addition, some selected control factors for firms, such as firm size, ownership structure, and geographic location, will be used to explore their impacts on firms’ performance. The estimated results show that these control factors had significant impacts on the technical efficiency scores of the studied firms. Firm size and geographic location were positively related with technical efficiency enhancements. Findings also indicated that state-owned firms were more technically efficient than the other three forms of ownership, i.e., private, joint venture, and foreign direct investment (FDI). Firms located in Hanoi and Ho Chi Minh City tended to be less technically efficient than those located in other areas, and this result might be explained by the fact that firms in these cities were operating under various forms of ownership, so they faced fiercer competition than those in the rest of the country. The time effects are also reported to assess their impacts on technical efficiency within the unbalanced dataset. The paper finally suggests some policy implications for further improvements of technical efficiency in this industry. Key words: data envelopment analysis (DEA), hotel efficiency, likelihood estimation, stochastic frontier production function (SFPF), Vietnam. JEL Classification: C14, L8 1. Introduction The field of technical efficiency analysis at firm level has garnered much interest from the research community around the globe. Since the first empirical work by Farrell (1957), where he provided the framework for the measurement of technical efficiency and allocative efficiency at firm level, many other researchers—both theoretical and empirical—have been involved in this field of study. To estimate technical efficiency accurately, a proper production function must be constructed. A proper production function requires the right specification under given input factors and production technology. Cobb and Douglas (1928) suggested a type of production function with two input factors, including capital and labor, and this function has a property such that it produces constant returns to scale. Other scholars, such as Berndt and Christensen (1973), also devoted their time and effort to the estimation of production functions. The production function provides a frontier on which firms’ technical efficiency can be calculated. The production function, or the frontier, is defined as the maximum possible output attained by a firm, provided that it has a given combination of inputs and a fixed technology. 71

Thus, the gap between the actual (or observed) output and the frontier is defined as the technical inefficiency attached to a particular firm. A firm’s output is assumed to lie either on the frontier (indicating that the firm has demonstrated technical efficiency) or beneath it (indicating that the firm has demonstrated technical inefficiency). A number of approaches are used in estimating the frontier. In his pioneering work on technical efficiency, Farrell (1957) proposed the data envelopment analysis (DEA) method for constructing the frontier. Later on, Aigner and Chu (1968) and Timmer (1971) made significant contributions to the field of efficiency analysis, and these studies led directly to the approach of stochastic frontier analysis. Among these authors, Aigner and Chu (1968) suggested a homogenous Cobb-Douglas production function with the assumption that all observations are on or beneath the frontier. Therefore, the assumption of constant returns to scale made in Farell’s work was relaxed in later approaches. The error term associated with Aigner and Chu (1968)’s mathematical model has a property of non-positive value. A zero error term means a firm is operating on its frontier. It is then said to be technically efficient. A firm is said to be technically inefficient if it is operating beneath the frontier in association with negative values of the error term. From this point of view, the technical inefficiency of a firm may be measured by the extent to which the actual output falls short of the potential output (as computed from the model). Later developments in the field of efficiency analysis have provided a broader framework for measuring technical efficiency and for the analysis of different sources of technical inefficiency at firm level. The constraints of the error terms attached to the frontier have been relaxed, and the inclusion of firms’ control factors in the estimation of the frontier has allowed researchers to track the sources of technical inefficiency. The development of software has eased the task of estimating firms’ technical efficiency. This paper aims to estimate technical efficiency scores and to provide an empirical analysis of different sources of technical inefficiency for firms in Vietnam’s hotel industry during 2000– 2003. The rest of the paper is structured as follows. Section 2 presents a literature review with regard to some theoretical and empirical research focused on the hotel industry. Section 3 describes the sample data, variables, and model specifications, while Section 4 reports the empirical results calculated from the methodology described in Section 3. Section 5 provides concluding remarks and policy implications for further enhancement of productive performance of firms in Vietnam’s hotel industry. 2.

Measuring Efficiency in Hotel Industry: Literature Review

Mullins (1995) considered the hotel sector as one of the four main sectors of the broader hospitality industry. The other three sectors included industrial catering, institutional catering and domestic service, and fast food. Foremost, the author clarified the nature of hotel services with a number of characteristics. Among them were simultaneous production and consumption, labor-intensive production, and difficulties in measuring performance. Though this research was not empirical, it could contribute, to some extent, to efficiency-related arguments in the next section of this paper. To quantitatively assess the efficiency level of the industry, Anderson (2000) made use of DEA to measure the overall, allocative, technical, pure technical, and scale efficiency levels. The results showed that efficient firms allocated more resources to food and beverage operations, while less efficient firms spent more on hotel operation and other expenses, employed too many workers, and were too large in terms of the number of rooms. In the industry, it is argued that competition might be imperfect. Differences in hotel location and hotel quality in a variety of dimensions have been seen as the industry’s prime candidates for focus in an efficiency study. 72

Johns et al. (1999) also used the DEA method in monitoring and benchmarking productivity in a chain of 15 hotels over a 12-month period. In this regard, they were able to identify and study units that showed anomalous behavior in terms of their measured productivity and gross profit. The researchers concluded that size, represented by staffing levels, was statistically associated with technical efficiency. Based on the empirical results, policy initiatives with regard to local management and the optimization of budgetary control were worked out in an attempt to further improve the hotels’ performance. In an attempt to find out the performance of state-owned hotels in Portugal, Barros (2004) made use of the stochastic cost frontier under an assumption that the technical inefficiency effects were time-varying. The empirical results were, however, mixed with low efficiency scores and non-time-varying technical inefficiency effects. In the concluding remarks, the author suggested an alteration of management procedures to enable an increase in efficiency, based on a government-environment framework. To sum up, factors such as size, organizational structure, and local regulations have been used to analyze the productive performance of firms in the hotel industry. The general conclusions drawn from the literature are that size is one of the factors that have significant impacts on technical efficiency, though its magnitude varies across firm size, as do the other factors. To the best of our knowledge, little research has involved the comprehensive study of the effects of ownership structure on technical efficiency. To narrow this gap, in the next section, the inefficiency effects resulting from ownership structure and other firm-specific factors, such as firm size and geographic location, are analyzed in an effort to determine the distinguishing patterns of inefficiency in Vietnam’s hotel industry. Due to the unbalanced nature of the data, time effects of technical efficiency are also included in the regression model. 3. Descriptions of Data, Variables, and Model Specifications 3.1. Data Description The dataset used in this paper is from the Economic Census for Enterprise, which was conducted by the General Statistics Office of Vietnam (GSO). The study period is from 2000 to 2003. Observations obtained in the survey are related to firms’ input factors and their performance during this period. Among these factors are the number of laborers, total wage bills, total assets, total intermediate costs, and depreciation of fixed assets. All these variables are observed on an annual basis. Firms’ annual revenue represents their performances in the survey years. All the firms that were under normal operation by the end of the previous year of each survey year are included in the survey. The survey was conducted nationwide. With regard to firms’ input factors, the number of laborers is observed twice a year, once at the beginning of each sample year and again at the end of that year. The same rule is applied for firms’ total assets. Therefore, their mean values are needed when incorporating them in the regression models. The other productive variables of firms in the sample period, such as total wage bills, intermediate costs, and revenue, are “flow” variables, so averaging is not needed. The number of laborers, which stands for the firms’ labor force, are the permanently employed laborers in the sample year, not taking into account seasonal or temporary employees. Their total wage bills are the total incomes received in the forms of salary, wage, and security contributions paid by the employers. Firms’ total assets include fixed assets and short-term and long-term investments. Their total intermediate costs are composed of the costs for fuel and energy, outsourcing services, debts, and other accountable payments. In this paper, we retain observations that have sufficient information on all the above input factors, i.e., intermediate costs, laborers, wage bills, total assets, and depreciation. Consequently, we were able to obtain data for 474 firms in the study period. 73

Table 1: Main Productive Indicators of Observed Firms in Dataset Indicator Revenue Laborers Wage bills Net total assets Total intermediate costs Revenue Laborers Wage bills Net total assets Total intermediate costs Revenue Laborers Wage bills Net total assets Total intermediate costs Revenue Laborers Wage bills Net total assets Total intermediate costs Revenue Laborers Wage bills Net total assets Total intermediate costs

Obs. Mean 2000 (year 1) 13 69,39.462 13 247 13 1,254.846 13 2,233.462 13 4,889.769 2001 (year 2) 15 94,513.2 15 641 15 9,652.733 15 126,899.5 15 3,729

Std. Dev.

Min.

Max.

14,527.36 683.6169 3,868.545 5,873.837 10,931.47

61 3 4 49 11

52,930 2,500 14,083 21,621 39,615

278,219.6 13,38.047 30,184.47 309,967.9 8,015.837

40 3 13 609 8

1,092,408 4,741 117,886 1,058,294 31,205

8 2 2 53 7

1,299,348 6,717 160,591 1,116,349 1,184,623

17 2 7 90 19

334,568 807 35,135 944,887 216,998

8 2 2 49 7

1,299,348 6,717 160,591 1,116,349 1,184,623

2002 (year 3) 243 19,488.84 90,046.22 243 119 474.1032 243 2,957.7 11,630.8 243 54,838 171,008.5 243 16,852.62 81,795.52 2003 (year 4) 203 14,680.3 38,088.28 203 75 121.3442 203 2,496.704 6,074.992 203 56,488.93 168,618.2 203 11,412.77 30,934.07 Total (period 2000–2003) 474 19,459.49 85,189.11 474 120 442.7613 474 2,925.435 10,672.85 474 56,382.73 173,748.2 474 13,779.5 62,042.54

Source: Author’s estimates from the dataset Table 1 displays the summary statistics of some main productive indicators of firms in the hotel industry in Vietnam during 2000–2003. Noticeably, the dataset is unbalanced: the number of observations in each year differs from that in the others. The number of observations in the first year, for example, was 13, and the numbers in the three following years were respectively 15, 243, and 203. All of these productive indicators were measured in millions of Vietnamese dong (VND), except for the number of laborers, which is in people. On average, year 1 (or year 2000 in the sample) and year 2 (or year 2001) had majorities of observations for large-scale operations, although they both had limited numbers of observations. The mean values of productive indicators, which were mostly counted for particular firms and the number of observations in each year, were different. Firms in the second year of the survey period were big in terms of net total assets, number of laborers, and amount of generated revenue, but these did not necessarily mean higher intermediate costs. This type of input cost tended to increase over time, with big jumps in the third and fourth years of the sample period. Other statistics are also recorded in the table, including standard deviation (denoted Std. Dev.), minimum, and maximum values of each indicator. In total for four years, the number of laborers had an average value of 120, and revenue, wage bills, net total assets, and total intermediate costs had the average values of 19,459; 2,925; 56,382; and 13,779 million VND, respectively. In addition to the above variables, the dataset also had information about the firms’ geographic locations and ownership structure. Each firm was marked as being located in a 74

specific province or city where its headquarters were located. In accordance with the nature of asset possession and forms of establishments, the dataset classified firms into different forms of ownership, including state-owned, private, joint stock, and foreign direct investment (FDI). Table 2 provides productive indicators of firms by ownership. Table 2: Productive Indicators across Ownership Indicator Revenue Laborers Wage bills Net total assets Total intermediate costs Number of observations Revenue Laborers Wage bills Net total assets Total intermediate costs Number of observations Revenue Laborers Wage bills Net total assets Total intermediate costs Number of observations Revenue Laborers Wage bills Net total assets Total intermediate costs Number of observations Revenue Laborers Wage bills Net total assets Total intermediate costs Number of observations

State-owned Year 1 245 6 21 49 174 1 Year 2 266,542 1,232 25,238.4 357,730.4 1,494 5 Year 3 54,446.54 370 7,135.897 73,075.59 42,476.49 39 Year 4 15,864.68 134 2,083.871 24,982.52 7,112.484 31 Total 51,949.67 326 6,172.539 71,225.09 24,798.87 76

Private

Joint Stock

FDI

1352 7 100.3333 377.6667 1,048.667 3

312 9 51.33333 219.3333 251.6667 3

14,843 507 2,591.714 17,079.43 8,670 7

176 13 29 1,370 8 1

314.6667 4 23.66667 1,242.667 257.6667 3

12,988.6 606 238.8 3,476.4 9,297.2 5

1,314.452 22 207.9677 3,499.387 1,175.258 31

532.5814 11 94.61628 1,905.942 433.8953 86

29,032.45 148 4,894.678 117,279.2 2,7182.24 87

4,989 61 877.4 10,575 4,113 5

480.7765 10 90.85882 1,901.165 405.3647 85

29,542.5 121 5,245.378 127,784.4 24,893.72 82

1,748.125 25 279.1 4,096.475 1,503.8 40

500.2712 10 90.87571 1,863.819 414.1186 177

28,271.56 163 4,918.751 115,019.6 24,935.45 181

Source: Author’s estimates from the dataset During the study period, state-owned firms were the biggest in terms of annual revenue, number of laborers, and wage bills; FDI firms were the biggest in terms of net total assets and total intermediate costs. Conversely, the firms of the other two forms of ownership, i.e., private and joint stock, were much smaller in terms of all five productive indicators. In general, the state-owned and FDI firms were large in scale, while the private and joint stock firms were small and medium. For example, on average during the period, state-owned firms had 326 laborers and FDI firms had 163 laborers, while private firms had only 25 laborers and joint stock firms had only 10 laborers.

75

3.2. Variable Description and Model Specifications 3.2.1. Variable Description Table 3 presents detailed descriptions of variables used in the two regression models, i.e., the frontier production function and the technical inefficiency effect model. The specifications of the models will be specifically presented in Section 3.2.2. Table 3: Description of Variables Name of Variables [1] Frontier Model Ln(REVENUE)it Ln(WAGE)it Ln(WAGE)2it Ln(ATOTAL)it Ln(ATOTAL)2it Ln(CINTER)it Ln(CINTER)2it Ln(CINTER)* Ln(WAGE)it Ln(WAGE)* Ln(ATOTAL)it Ln(ATOTAL)*Ln(CINTER)it

Description Natural logarithm of revenue of firm i in year t Natural logarithm of total wage bills of firm i in year t Square of natural logarithm of total wage bills of firm i in year t Natural logarithm of net total assets of firm i in year t Square of natural logarithm of net total assets of firm i in year t Natural logarithm of total intermediate costs of firm i in year t Square of natural logarithm of total intermediate costs of firm i in year t Cross sectional effect of natural logarithm of total intermediate costs and that of total wage bills of firm i in year t Cross sectional effect of natural logarithm of total wage bills and that of net total assets of firm i in year t Cross sectional effect of natural logarithm of net total assets and that of total intermediate costs of firm i in year t

[2] Inefficiency Model Ln(LABOR)it

OWN0i

OWN1i

OWN2i

OWN3i LOC1i

LOC2i

t

Natural logarithm of number of laborers of firm i in year t Dummy variable that represents effects of ownership structure on firms’ technical inefficiency; equal to 1 for the state (both central and local management) firms and otherwise equal to 0. In regression model, refers to the base group of ownership structure Dummy variable that represents effects of ownership structure on firms’ technical inefficiency; equal to 1 for private (in different forms of establishments, such as limited, partnership, and joint stock with no shares possessed by the government) firms and otherwise equal to 0 Dummy variable that represents effects of ownership structure on firms’ technical inefficiency; equal to 1 for joint stock firms (SOE have been equitized with a portion of shares still possessed by government) and otherwise equal to 0 Dummy variable that represents effects of ownership structure on firms’ technical inefficiency; equal to 1 for foreign-directed firms (both join venture and wholly foreign invested firms) and otherwise equal to 0 Dummy variable that represents effects of geographic locations on firms’ technical inefficiency; equal to 1 for firms located in two big cities (Hanoi and HCMC) and otherwise equal to 0 Dummy variable that represents effects of geographic location on firms’ technical inefficiency; equal to 1 for firms located in tourist spots, such as Hai Phong, Quang Ninh, Thua Thien Hue, Da Nang, Khanh Hoa, Lam Dong, and Ba Ria-Vung Tau, and otherwise equal to 0 Time effects on firms’ technical inefficiency; have value of 1 for firms observed in year 2000, 2 in 2001, 3 in 2002, and 4 in 2003

The paper uses firms’ annual revenue as the output variable in the estimation of the frontier production function. The other variables, including firms’ total intermediate costs, wage bills, and net total assets, serve as input variables in the frontier model. Firms’ number of laborers 76

stands for the effects of firm size on technical inefficiency. In the regression models, all the variables are expressed in the natural logarithm, except for dummy variables representing the ownership structure, geographic location, and time effects. As the models are regressed with the panel data, it is suggested that all the variables in value terms be deflated to exclude the bias caused by the annual changes in prices. In this paper, the annual CPI will be used as the deflator. Though this index is not the best deflator for the intermediate factors, such as labor and assets, it is the only available one. In addition, this kind of index is relatively stable over the study period, and thus it will not cause many changes in the estimation results. To capture the technical inefficiency effects of ownership structure, three dummy variables will be used in the regression models: OWN1, OWN2, and OWN3. The technical inefficiency effects caused by these variables will be compared with the base group of the state-owned firms. In this regard, if there are no differences in the technical inefficiency effects across ownership structure the condition OWN1 = OWN2 = OWN3 will hold. Geographic location is another specific factor that might influence technical inefficiency. In the regression model, we will use two dummy variables, LOC1 and LOC2, to represent the firms located in the two big cities (Hanoi and Ho Chi Minh city [HCMC]) and tourist spots, respectively. The time effects might result in productivity change or technical changes. The productivity changes might take place in the short term, while the technical changes might take place in a longer term. The variable t denotes the time effects on technical efficiency. As the dataset spans four years, it is expected that productivity changes will be recorded. 3.2.2. Model Specifications In this section, the stochastic frontier production function under the two forms of CobbDouglas and the transcendental-logarithm (or translog) will be estimated simultaneously in an attempt to choose the most appropriate model to suit the production technology of firms in the hotel industry. It should be noted that we take firms’ laborers (represented by wage bills), capital (represented by net total assets), and total intermediate costs as the three input factors in the estimation of the production function. In this regard, it is suggested that firms’ revenue will act as the output factor in estimating the production function. This is a reasonable variable in such a case that the data have the information on firms’ intermediate costs. Model 1A: Stochastic Frontier Production Function with the Cobb-Douglas Form Ln(REVENUE)it = β0 + β1*Ln(CINTER)it + β2*Ln(WAGE)it + + β3*Ln(ASSET)it + (Vit – Uit)

(1)

where i is the identity of observations in each sample year, and the same firm has the same identity in different years; t = 1, 2, 3, 4 indicates the time period with the years 2000, 2001, 2002, 2003, respectively; Ln(REVENUE)it, Ln(CINTER)it, Ln(WAGE)it, and Ln(ASSET)it are variables defined in Table 3; β0, β1, β2, and β3 are parameters to be estimated; and (Vi-Ui) is the composed error term attached to the stochastic frontier model, in which Vis are the error terms representing statistical noise and other external shocks beyond firms’ control (these are assumed to be independently and identically distributed [i.i.d.] random errors, which have normal distribution with zero mean and unknown variance σ2v); and Uis are non-negative random variables showing technical inefficiency effects, which are assumed to be i.i.d. such that they follow half-normal distribution with the mean Mi = Zi*σu + Wi. Moreover, that wage bills are used as a proxy for the labor input factor in the production function (or the frontier model) is pursuant with the corresponding capital input factor, represented by firm’s net total assets, in such a common feature that they are both measured in value terms (in VND million). Also, the paper makes use of firms’ revenue as the output factor in the estimation of the frontier model. This variable is also measured in VND million. 77

Model 1B: Stochastic Frontier Production Function with the Translog Form Ln(REVENUE)it = α0+ α1*Ln(CINTER)it + α2*Ln(WAGE)it + α3*Ln(ATOTAL)it + + α4*Ln(CINTER)2it + α5*Ln(WAGE)2it + α6*Ln(ATOTAL)2it + + α7* Ln(CINTER)it* Ln(WAGE)it + α8* Ln(WAGE)it* Ln(ATOTAL)it + + α9* Ln(ATOTAL)it* Ln(CINTER)it + (Vit – Uit)

(2)

where i and t are the identities denoted for the sequence of observations and the time periods as specified in the previous part of this section; αk (k = 1, 2,..., 9) are the parameters to be estimated; the composed error term (Vi-Ui) has the properties indicated in Model 1A; and the endogenous and exogenous variables are already defined in Table 3. There are several choices of specification for the production frontier. The most popular functional forms are those of Cobb-Douglas (as specified in Model 1A), constant elasticity of substitution (CES), and the translog model. It has been widely accepted that the translog frontier model permits more general substitution, scale, and technical change possibilities, and it does not impose any a priori restrictions on the production structure. This functional form allows variability in the elasticity of substitution among factors of production. However, it does have some drawbacks, including a potentially high correlation between cross-term variables. However, these drawbacks will be expected to be minimum when the sample is large and the correlation among variables is not too high. In this paper, a test of functional form is needed to determine whether Model 1A or Model 1B is more appropriate. Model 2: Technical Inefficiency Effect Model With regard to the technical inefficiency effect model, the component of technical inefficiency effects in the frontier production function is defined to rely on firm-specific factors, such as firm size, geographic location, and ownership structure. The following is the specification for this type of model. Uit = θ0 + θ1*Ln(LABOR)it + θ2*OWN1it + θ3*OWN2it + θ4*OWN3it + + θ5*LOC1it + θ6*LOC2it + θ7*t + Wi

(3)

where i and t are the identities denoted for the sequence of observations and the time periods as specified in the previous part of this section; θi (i = 0, 1, 2,…, 7) are the parameters to be estimated; Ui is the non-negative random variable that has properties as indicated in Model 1A; Wi is the random variable defined by the truncation of the normal distribution with zero mean and an unknown variance, δw2, such that Ui is non-negative and the point of truncation is -Ziθ, or, alternatively Wi ≥ - Ziθ; in which Zi is a vector representing firm-specific factors that may have effects on a firm’s technical inefficiency; and θ is a vector of parameters to be estimated. Model 2 has variables that represent firm size (i.e., number of laborers), ownership structure (i.e., three forms of ownership in reference to state-owned firms, which are the base sub-group of ownership in the regression model), geographic location (i.e., big cities (Hanoi and HCMC) and tourist spots in reference to the rest of the country), and the time effects on technical inefficiency. 3.2.3. Estimation Procedures and Usage of Software Package This paper uses FRONTIER Version 4.1 to do the estimates for the frontier model and the technical inefficiency effect model. The two models are estimated simultaneously to avoid statistical biases inherent in two-stage estimation methods. Battese and Coelli (1995) argued that factors under firm control, such as firm size, location, age, and ownership structure, might have potential effects on technical inefficiency, and thus they should all be incorporated in the 78

estimation models. In this paper, the frontier and the technical inefficiency effect models will be simultaneously estimated, and then the scores of technical efficiency will be obtained. The parameters of the stochastic frontier production function are estimated with the use of the maximum likelihood method. This method requires numerical maximization of the likelihood function. FRONTIER Version 4.1 easily automates the maximum likelihood method for estimation of the parameters of the frontier model. It is even applicable for unbalanced crosssectional data, an ability that is necessary due to the nature of the dataset. In the following sections, we will present separate estimates of the frontier models under the forms of Cobb-Douglas and the translog to test the functional form. Then, we will do a simultaneous estimation of the chosen frontier and the technical inefficiency effect model under the specifications in order to capture firm-specific factors. 4.

Empirical Results and Analysis

4.1. Identification of Functional Form In order to estimate efficiency of the studied hotels, we need to specify the production function. Suppose that the Cobb-Douglas frontier production function is appropriate. Thus, we have the null hypothesis H0 that these hotels follow Model 1A, and the alternative hypothesis H1 that these hotels follow Model 1B (translog frontier model). The generalized likelihood-ratio test statistic is defined as λ = −2[ L( H 0 ) − L( H1 )] , in which L(H0) is the log-likelihood value of a restricted frontier model under the null hypothesis H0, and L(H1) is the log-likelihood value of the general frontier model under the alternative hypothesis H1. This test statistic has an approximately Chi-square (or mixed Chi-square) distribution with degrees of freedom equal to the difference between the parameters involved in the null and alternative hypothesis tests. The test statistics can be calculated as λ = -2[-403.19- 334.916] = 136.548. This value is much higher than the critical value χ2(6, 0.05) = 12.591. Therefore, the null hypothesis can be rejected at the significance level of 5 percent, and the test statistics are in favor of the translog frontier model. 4.2. Likelihood Estimates for Parameters According to the test results in the previous section, the production technology is in favor of the frontier production function under the translog form. Table 4 provides the likelihood estimates of all the parameters incorporated in the two models, i.e., the frontier model (under the translog form) and the inefficiency model. Table 4: Likelihood Estimates for the Parameters ML value = -304.867 Variable Frontier model CONS.*** Ln(CINTER)** Ln(WAGE)** Ln(ATOTAL) Ln(CINTER)2* Ln(WAGE)2 Ln(ATOTAL)2*** Ln(CINTER)*Ln(Ln(WAGE)* (Ln(WAGE)*Ln(ATOTAL)*** Ln(ATOTAL)*Ln(CINTER) Inefficiency model 79

Coefficient

s.d.

t-ratio

0.720 0.457 0.599 0.065 -0.214 0.073 -0.063 0.126 0.067 0.001

0.436 0.187 0.243 0.144 0.050 0.046 0.036 0.020 0.040 0.016

1.651 2.437 2.463 0.451 -4.316 1.590 -1.763 6.191 1.702 0.047

CONS.* -0.726 0.076 Ln(LABOR)* -0.074 0.023 OWN1** 0.464 0.202 0.639 0.134 OWN2* OWN3* 0.613 0.145 LOC1* 0.190 0.080 LOC2 0.085 0.068 t* 0.144 0.038 0.214 0.015 σ2*(=σ2u+σ2v) 0.030 0.005 γ*(=σ2u /(σ2u+σ2v)) Notes: (*) stands for the significance level of 1%, (**) for 5%, and (***) for 10%.

-9.552 -3.282 2.292 4.775 4.218 2.383 1.260 3.773 14.620 6.005

Source: Author’s estimates As can be seen in the translog frontier model, all of the estimated parameters—except for Ln(ATOTAL), Ln(WAGE)2, and Ln(ATOTAL)*Ln(CINTER)—are statistically significant at the required levels. In addition, the signs of the parameters are theoretically acceptable. Almost all the parameters have a positive sign and are statistically significant. In the technical inefficiency model, the overall picture of the sign and significance of the parameters is good. We found that firm size, represented by the number of laborers, was negatively associated with technical inefficiency effects. In other words, the higher the number of laborers was, the more technical efficiency enhancements firms could attain. The firm size impacts were significant at the level of 1 percent. Significant impacts were also recorded for the three dummy variables, which stand for ownership structure. Private, joint stock, and FDI firms tended to have positive impacts on firms’ technical inefficiency. In other words, these three types of firms were less technically efficient than the base group of state-owned firms of both central and local levels. An interesting finding is that firms in the two big cities, Hanoi and HCMC, were statistically less technically efficient than those in the rest of the country. However, this kind of impacts on firms in tourist spots was not statistically significant. Detailed arguments on the differences in the impacts of firm size, ownership structure, and geographic locations on technical efficiency will be presented in Section 4.4. The time trend, which is represented by the variable t in Table 4, was positively related with technical inefficiency effects. Firms in the sample had a trend of becoming less technically efficient over time. This might be attributed to technical changes that shifted the frontier after a particular period of time and thus widened the gap between firms’ output and the frontier. 4.3.

Estimated Technical Efficiency Level

With the above specifications of the frontier and technical inefficiency models, and under such a requirement that the two models are simultaneously estimated, the estimated scores of technical efficiency at firm level are presented in Table 5. Table 5: Summary of the Estimated Technical Efficiency Level Time Year 1 (2000) Year 2 (2001) Year 3 (2002) Year 4 (2003) Whole period

Obs. 13 15 243 203 474

Mean 0.967 0.951 0.848 0.757 0.815

Std. Dev. 0.020 0.057 0.104 0.128 0.126

Min. 0.926 0.803 0.633 0.554 0.554

Max. 1.000 1.000 1.000 0.990 1.000

Source: Author’s estimates The average score of technical efficiency in the first and second year periods were 0.967 and 0.951, respectively. These efficiency levels were very high, but they did not reveal much 80

about the overall picture of technical efficiency in the hotel industry. The reason might be that there were limited numbers of firms in the first two years of the sample period. The scores of technical efficiency in the last two years of the period were around 0.848 and 0.757, respectively. Over the entire study period, firms recorded an average level of technical efficiency at 0.815 (or 81.5 percent), meaning that firms were 18.5 percent below their maximum attainable efficiency. The firm with the lowest technical efficiency had the score of 0.554 (or 55.4 percent). In terms of technical inefficiency, this is a score of 44.6 percent. Some firms in the sample obtained 100 percent efficiency. Remarkably, firms of this kind all fell into the sub-group of state ownership. In general, the annual average scores of technical efficiency tended to decrease over time; the annual average score of technical efficiency declined to 0.757 in year 4 from 0.848 in year 3. This empirical evidence indicates that the time trend had a significant impact on technical efficiency. Firms with technical efficiency levels of greater than 0.9 (or 90 percent) accounted for 32.49 percent of the total. The firms with technical efficiency levels ranging from 70 to 80 percent made up 21.1 percent of the total. The results also showed that there were no firms that had technical efficiency levels of less than 50 percent; the technical efficiency interval ranging from 50 to 60 percent accounted for merely 2.95 percent of the total. Generally, the frequency distribution of technical efficiency in the sample was skewed to the right; almost all of the observations had technical efficiency levels of more than 60 percent (Figure 1). Figure 1: Frequency Distribution of Technical Efficiency Level 35.00%

32.49%

30.00% 24.89%

25.00%

21.10% 18.57%

20.00% TE 15.00%

Frequency

10.00% 5.00%

2.95%

0.00%