The concept of postponement advocates that commitment, as to the form and place of commodities, can be delayed to the latest possible point in the supply chain (Bowersox and Morash 1989). Although this concept was first proposed in the early 1950s (Alderson 1950; Bucklin 1965), increased application and interest have been noted only recently (e.g., Closs et al. 1998; Feitzinger and Lee 1997; Pagh and Cooper 1998; van Hoek, Commandeur, and Vos 1998). This is reflective of the move toward customization and recognition of inventory risks and logistics costs for firms.

Zinn and Bowersox (1988) examined the benefits of implementing form postponement strategies in place of the traditional anticipatory distribution strategy. Their simulation results, based on a normative cost model, suggest that form postponement strategies can reduce the direct cost of production under many circumstances, However, the optimal postponement strategy under each circumstance may be different. Product and demand characteristics will influence the cost function and thus the optimal choice of strategy or mixture of strategies (Bowersox and Morash 1989; Closs et al. 1998; Feitzinger and Lee 1997; Pagh and Cooper 1998; van Hoek, Commandeur, and Vos 1998). For example, Zinn and Bowersox find that products marketed under different brand names benefit from labeling postponement. In contrast, products sold in various configurations benefit from assembly postponement.

Taylor (1996) noted that postponement strategy is especially important for an international enterprise. International enterprises, with diverse markets, tend to have more extensive supply chains and encounter greater degrees of risks resulting from inventory management. These enterprises have to critically examine the trade-offs between reducing inventory risks (postponement strategies) and increasing the economies of scale in productions (anticipatory strategies).

In this empirical paper, Taiwanese information technology (IT) firms are examined. The IT sector was selected because IT products (such as personal computing products, network equipment, and communication devices) are characterized by high product values, short product life cycles, and high demands for customization. These are the attributes identified by Zinn and Bowersox and others that could potentially benefit from form postponements. Taiwan was selected for our study because it is one of the largest producers of IT products and is the largest original equipment manufacturing (OEM) partner for U.S. and Japan. Furthermore, Taiwanese firms have recently taken up the responsibility of providing global logistics. Thus, Taiwanese IT firms provide a rich data set on postponement behaviors, and the conclusions drawn from this data have strong implications for all managers.

The goal of this paper is multi-fold. First, the four types of form postponements proposed by Zinn and Bowersox are empirically examined. The measurement scale is developed from and tested on 102 IT firms in Taiwan. Second, we explore the factors affecting the adoption of different form postponement strategies. Confirmatory factor analysis is used to validate the dimensionality of form postponement strategies, while path analysis is used to examine the relationships between product/demand characteristics and the adoption of postponement strategies.

Following the introduction, section two discusses the role of Taiwanese IT firms in the global IT industry. Section three then introduces briefly the four types of form postponement strategies proposed by Zinn and Bowersox. Hypotheses are proposed to examine the strength of the relationships among the four postponement strategies. Section four discusses the factors influencing the adoption of the different form postponement strategies. Hypotheses are proposed to examine product/demand characteristics that should benefit from form postponements. Section five discusses the empirical methodology. Section six summarizes the results from our confirmatory factor analysis and path analysis. In-depth discussions of these results are presented in section seven. Section eight discusses the limitations of the study and offers directions on future studies. Finally, section nine states the managerial implications and conclusion of the study.

THE IMPORTANCE OF TAIWAN'S INFORMATION TECHNOLOGY INDUSTRY IN THE WORLD MARKET

Taiwan's IT industry ranks third in dollar outputs after the U.S. and Japan. It produces more than 50% of the world's output in several computer components (Table 1). However, Taiwanese firms rarely compete with U.S. or Japanese firms. They are often partners within a global supply chain (Table 2). Taiwanese IT firms are small (comparably) in size and have traditionally relied on U.S. and Japanese production technologies and distribution channels. However, their lower labor costs and environmental costs give them a competitive edge in manufacturing. Consequently, Taiwanese IT firms have specialized as the electronic world's OEM partners. In recent years, Taiwan has become the largest OEM country for personal computing products. For example, in the year 1998, IBM, HP, Dell, and Compaq purchased over 11.5 billion $USD in OEM personal computing products from Taiwan. Also, notable is Taiwan's foundry industry, which produces OEM integrated circuit (IC) products. Together, Taiwan Semiconductor (TSM) and United Microelectronics (UMC) account for more than 60% of the world's OEM foundry production (Business Week 1999).

In response to intensified global competition, Taiwanese IT firms have devoted a substantial portion of their research dollars to improve logistics and manufacturing efficiency. We believe logistics research, in combination with trial-and-error experience, have provided these IT firms with insights into the implementations of form postponement strategies. Their successes in the marketplace as demonstrated by their market shares, profitability, and stock price appreciation suggest that their production/distribution strategies are effective. Therefore, their actions should provide recommendations for managers interested in implementing postponement strategies.

FOUR TYPES OF FORM POSTPONEMENT STRATEGIES

Postponement is the (partial or full) delaying of the movement or formulation of a product until purchase orders are received to reduce the risk of inventory (Feitzinger and Lee 1997). For example, Dell Computer Corporation keeps inventories of computer components at a few centralized locations. Final products are assembled and shipped to customers only after the orders are received. The opposite concept of postponement is anticipatory distribution (or speculative distribution). Anticipatory distribution says that the production and the movement of goods to the forward inventories should be made at the earliest possible time to reduce the costs of manufacturing and the time to delivery (Bucklin 1965).

Unfortunately, the two cost reducing methods are mutually exclusive. Anticipatory production and distribution necessarily imply higher inventory depreciation. On the other hand, postponing production and distribution inevitably lower the economies of scale in production and shipping. The optimal strategy or mixture of strategies under each circumstance can be different. Zinn and Bowersox first attempted to shed light on the issue by classifying form postponement strategies into four types: labeling, packaging, assembly, and manufacturing. A simulation was then performed to compare the performances of the different postponement strategies versus the traditional anticipatory distribution strategy in a pair-wise fashion. Our study seeks to extend their work with empirical evidence by using a path model, which examines the four postponement strategies and four antecedent factors of postponements simultaneously. First, we assess the validity of Zinn and Bowersox's classification of postponement strategies. Then, we examine the product/demand characteristics (the antecedent factors of postponements), which may benefit from the adoption of postponement strategies.

Zinn and Bowersox's classification of form postponements is modified in the current research to address the global logistics issues from the Taiwanese IT manufacturers' perspectives. Specifically, a firm is said to practice labeling postponement if it satisfies the following four criteria: 1) It markets/produces products under different brand names; 2) The final products are not labeled before receipt of purchase orders; 3) When orders are received, the corresponding warehouse units label the products with the appropriate brand name logos and include the corresponding instruction manuals and warranty documents; and 4) The labeling activities are conducted at its overseas unit. For example, Compal Computers produces notebook PCs for Dell Corp. and Specter Computers. Essentially, identical machines are sold under two brand names and model numbers. When orders are received, Compal labels the notebook PCs at its local facilities accordingly and ships the final products with the proper printed materials. We expect labeling postponement to reduce the inventory depreciation for products, which are marketed under different brand names. Correspondingly, the cost of labeling should increase.

A firm is said to practice packaging postponement if it satisfies the following four criteria: 1) It markets/produces products in different bundles or package sizes; 2) The final products are not packaged before the reception of purchase orders; 3) When orders are received, the proper bundles/bulks are packaged and shipped; and 4) The packaging activities are conducted at its overseas unit. For example, D-Link markets a no-frill Ethernet card product and a small office local area network (LAN) solution product. The former includes only an Ethernet card with the appropriate software driver. The latter includes two Ethernet cards, all the necessary cabling, and peer-to-peer networking software. Ethernet cards are shipped to local warehouses in unpackaged bulks; then, depending on the orders received, D-Link packages its Ethernet cards into the appropriate product bundles for delivery. We expect packaging postponement to reduce the inventory depreciation for products marketed under different bundles. Correspondingly, the cost of packaging should increase.

Realistically, postponing product packaging may also force labeling to be delayed. For instance, some products simply must be packaged and labeled at the same time. However, this need not always be the case. Consider our previous example on Ethernet card products. The Ethernet cards have been labeled with the proper model numbers and logo prior to purchase order receipt. The firm does not practice labeling postponement, but it does practice packaging postponement. We examine the strength of the causal relationship between packaging postponement and labeling postponement in the following hypothesis.

H1a: If a firm adopts a packaging postponement strategy, it is more likely to adopt a labeling postponement strategy as well.

In Zinn and Bowersox's original definition, manufacturing postponement differs from assembly postponement only in the degree of warehouse assembly performed. However, to operationalize the two concepts, we use stronger and more restrictive definitions. Specifically, a firm is said to practice assembly postponement if it satisfies the following four criteria: 1) It markets/produces products which are configured/customized from a base product using a number of common parts; 2) The assembly of the final products from the base product is not performed before the reception of purchase orders; 3) When orders are received, the proper final products are assembled/customized from the base product to meet the configuration requirements; and 4) The assembly activities are conducted at its overseas unit. For example, Acer Computer sells a range of desktop PCs with different processing powers and storage capacities. Computer components such as CPUs, hard drives, video cards, and dynamic random access modules (DRAM) are warehoused alongside barebones systems (which include only the motherboards and the floppy drives preinstalled in standard desktop cases) at the local country. Final products are assembled from these components to meet the exact configuration requirements specified in the purchase orders. We expect assembly postponement to reduce the cost of depreciation on the inventory value of intermediate products. Correspondingly, the cost of assembly is increased.

As before, we expect assembly postponement to prompt packaging and/or labeling postponement. We address these empirical relationships in the following hypotheses.

H1b1: If a firm adopts an assembly postponement strategy, it is more likely to adopt a packaging postponement strategy as well.

H1b2: If a firm adopts an assembly postponement strategy, it is more likely to adopt a labeling postponement strategy as well.

A firm is said to practice manufacturing postponement if it satisfies the following criteria: 1) It delays fully or partially manufacturing processes until the receipt of purchase orders; 2) The manufacturing activities are conducted at its overseas unit. Note that this does not preclude the firm from purchasing the necessary raw materials/components for production in anticipation of orders. For example, United Microelectronics (UMC) produces dynamic random access memory chips for its affiliate company in Japan by using its Japanese factory capacity. UMC does not produce DRAM chips in advance of orders received because of volatile DRAM prices. However, UMC does maintain an anticipatory inventory of semiconductor wafers necessary to produce IC chips. Manufacturing postponement is expected to reduce the cost of depreciation on the inventory value of intermediate products. Correspondingly, the cost of manufacturing should increase.

Again, realistically, postponing manufacturing may force assembly, packaging, and/or labeling to be delayed. However, this need not always be the case. For example, many IT products do not have to be customized with additional components after they are manufactured. The previous example about DRAM production illustrates this point. The final products are DRAM chips; the firm performs no additional assembly or customization. So manufacturing postponement does not necessarily induce assembly postponement. As mentioned earlier, the natural causal relationship from manufacturing postponement to assembly, packaging, and labeling postponements is noted. The strength of these relationships is examined in the following hypotheses.

H1c1: If a firm adopts a manufacturing postponement strategy, it is more likely to adopt an assembly postponement strategy as well.

H1c2: If a firm adopts a manufacturing postponement strategy, it is more likely to adopt a packaging postponement strategy as well.

H1c3: If a firm adopts a manufacturing postponement strategy, it is more likely to adopt a labeling postponement strategy as well.

In sum, H1a1 through H1c3 indicate that the experience in implementing one postponement strategy may have lowered the cost of implementing other postponement strategies. In addition, the experience should provide better forecasts of the benefits and the costs of implementing postponement strategies, thus lowering the risk of implementation. More importantly, these hypotheses are also necessary to complete our structural model. The omission of these relationships will incorrectly specify the relationships among the set of simultaneous equations giving rise to estimation errors in other variables.

FACTORS AFFECTING POSTPONEMENT STRATEGY

Based on the interviews with industry professionals and evidences from past studies (Bowersox and Morash 1989; Closs et al. 1998; Feitzinger and Lee 1997; Pagh and Cooper 1998; van Hoek, Commandeur, and Vos 1998), four major product/demand characteristics are identified as candidate drivers of form postponement strategies. These drivers are demand for customization, modularity in construction, product value, and product life cycle.

Feitzenger and Lee noted that to successfully perform assembly postponement, a product should be comprised of independent modules. While assembly postponement could potentially reduce transportation costs and inventory depreciation, production cost will increase due to loss of scale. However, modularity in product design will moderate the increase in production costs. Thus, we should expect modularity to induce assembly postponement.

Unfortunately, a modular design may not be appropriate for all products; in fact, modular designs often impose additional production costs. Therefore, product characteristics are examined that may benefit from modular designs. Modularity, as pointed out by Weng in his recent theory paper, has the effect of risk-pooling over demand uncertainty across products sharing the same modules (Weng 1999). Since customizable products often experience great demand uncertainty over the various configurations, they should benefit from the risk-pooling effect of modular designs. Therefore, demand for customization should inspire modular designs. The above observations are thus interpreted as two simultaneous relationships.

H2a: Firms whose products are characterized by strong customer demand for customization are more likely to use modular product designs.

H2b: Firms whose products are more modular in design are more likely to practice assembly postponement.

Zinn and Bowersox's simulation shows that consumers' demand for customization should also have a direct and positive effect on the implementation of assembly postponement. This suggests that the result of the reduction in depreciation and transportation costs, for products with many configurations, often outweighs the increase in assembly costs. The rationale for this claim is provided below. Products with many final configurations likely do not experience high scale of economy in the assembly stage of production. Thus, the implementation of assembly postponement is less likely to impose additional costs, while the benefits of postponement are still fully extracted. We test the claim with the following hypothesis.

H2c: Firms whose products are characterized by higher consumer demand for customization are more likely to practice assembly postponement.

Zinn and Bowersox found in their simulation that, overall, product value is the most important determinant of whether postponement should be practiced. Products with high inventory carrying costs should benefit from all forms of postponement. In Zinn and Bowersox's simulation, product value is introduced primarily to capture the inventory carrying cost. In practice, product value (the market value) may not closely proxy inventory carrying cost. In this study, the focus is on whether the cost of carrying the component is so high that it will tie up large amounts of working capital. As mentioned in the introductory section, Taiwanese IT firms are small in size and primarily serve as OEM or original design manufacturers (ODM) partners in the global supply chain. They have to purchase different components from different suppliers all over the world. If the cost of the component is expensive relative to the total cost of producing the final product, postponing stocking the key component at the last possible point becomes very important. Postponement strategies can reduce pressure on working capital. For example, the Central Processing Unit (CPU) represents a substantial portion of the cost of a computer motherboard (normally more than 60% of the final cost). The cost of carrying CPUs normally will tie up large amounts of working capital. Therefore, motherboard manufacturers have to delay the installment of the CPU in the motherboard to the latest possible point. Usually Taiwanese motherboard manufacturers do not install the CPU until the motherboards are shipped to the local destination.

Therefore, expensiveness of the component is used as the proxy product value. The carrying cost of the key component may severely tax the firm's working capital.

H3a: Finns whose products' key components are more expensive to carry are more likely to practice labeling postponement.

H3b: Firms whose products' key components are more expensive to carry are more likely to practice packaging postponement.

H3c: Firms whose products' key components are more expensive to carry are more likely to practice assembly postponement.

H3d: Firms whose products' key components are more expensive to carry are more likely to practice manufacturing postponement.

Product obsolescence may be the largest contributing factor to depreciation. The risk of obsolete inventory, as pointed out by van Hoek, Commandeur, and Vos (1998), is characteristic of products with short life cycles. Products with short life cycles should benefit from form postponement. However, delaying labeling or packaging of a product with a short life cycle would not reduce the risk of obsolete inventory. Also, delaying assembly will not appreciably reduce the risk of obsolete inventory. The units produced in anticipation of the base product are as likely to suffer from obsolescence as the fully configured product. For example, a barebones computer, with only the motherboard and the floppy drive installed in a desktop computer case, becomes as obsolete as a fully configured computer when a new processor/chipset technology is introduced. Thus, we hypothesize that:

H4: Firms whose products are characterized by a shorter product life cycle are more likely to practice manufacturing postponement.

METHODOLOGY

Six hundred Taiwanese IT firms were selected as targets for this study. The firms are IT hardware producers with primary revenues from exporting products overseas. All 600 firms targeted have overseas operating and production/warehouse units.

Survey questions were designed to identify the product/demand characteristics and the type of form postponements associated with the firms' primary products. They were pre-tested on a selected group of professional managers; ambiguities in the wordings were noted, and clarifications made to the questions as appropriate. Of the 600 questionnaires distributed, 102 were completed and returned (17%).

Measurement

The survey consisted of statements that managers were asked to express degree of agreement or disagreement (measured on a five-point Likert scale). The statements identify the four types of form postponements developed from Zinn and Bowersox's original definitions. In order to perform the Cronbach's alpha test on dimensionality (Dunn, Seaker, and Waller 1994; Mentzer and Flint 1997; Peter 1979), 16 statements (four statements for each type of postponement) were developed. A pre-test was conducted to check the internal consistency of the scale items (Churchill 1979). The results show that one item out of four for each of the form postponement constructs was not highly correlated with the other items; low correlation items were dropped. As a result, only twelve of the original sixteen items are included in the final questionnaire (see Appendix A for measurement items).

The following statements were used to measure the level of the four antecedent factors of form postponements discussed in the previous section. Statements have been translated into English from the original Chinese survey. The demand for customization is measured by the Likert scale (from strongly agree to strongly disagree) response to the statement "The final product must be flexible enough to accommodate different demands from customers." The modularity in construction is measured by the Likert scale response to the statement "The final product is produced in a modular fashion from common components." The expensiveness of product components is measured by the Likert scale response to "The key components of the product are expensive in the sense that the carrying value of the key components puts pressure on working capitals." During the development of the survey statements, some respondents interpreted the expensiveness of the components as relative to the total cost (including process costs) of producing the product; others interpreted expensiveness relative to the amount of corporate resources tied up in the inventory of the components. While both interpretations capture the concern of component carrying cost, the second interpretation better captures the cash constraint imposed by maintaining the inventory of components. Therefore, the survey statement was modified to reflect that interpretation. Finally, the duration of the product life cycle is measured by the Likert scale response to "New generation products are introduced to the market quickly; the product life cycle is very short."

RESULTS

Sample statistics are presented in Table 3. The average annual revenue for firms in the sample is 87.5 million $USD, and the average number of employees is 736. In addition, on average, 47% of the total revenue comes from export orders, and 32% of the total production is attributed to foreign subsidiaries. This confirms that the degree of internationalization is extremely high for Taiwanese IT firms.

Data Analysis

The models are tested using the two-step structural equation procedure proposed by Anderson and Gerbing (1988). First, Confirmatory Factor Analysis (CFA) was performed on the latent constructs. After confirming the validity of the Zinn and Bowersox postponement specification, path analysis was performed on the structural model to test the hypothesized relationships.

The Dimensionality of Form Postponement - Confirmatory Factor Analysis

CFA was first performed on the data to assess the construct validity and the dimensionality of form postponements (carver and Mentzer 1999). Using the LISREL program (Joreskog and Sorbom 1993), we compared Zinn and Bowersox's specification of four form postponements against the null specification of one type of form postponement (Hoyle and Lennox 1991). The results are presented in Table 4. The chi-square from Model 2 (form postponement as four constructs) suggests acceptable model fit, while the chi-square for Model 1 suggests that the model fit is unsatisfactory at any level of confidence. Model 2 also has a goodness-of-fit index at .89 vs. Model l's .72. Therefore, it is concluded that form postponement is better modeled as four distinct constructs: Labeling, Packaging, Assembly, and Manufacturing postponements. Further, reliability coefficients are also estimated for the statements representing the constructs (Table 5); they indicate reliabilities at above .80. Therefore, it is also concluded that the survey statements do identify the four forms of postponements as proposed by Zinn and Bowersox.

Factors Affecting Form Postponement - Path Analysis

Path analysis was applied to test the hypotheses proposed. The path model specification is displayed in Figure 1.

Coefficients estimated by path analysis are displayed in Tables 6 and 7. Overall, the model fit statistics show a chi-square of 16.69 (df = 14) with a p-value of .27, indicating a good model fit (Bollen 1989). The goodness-of-fit index (GFI) is also high at .96. Root mean square error of approximation is low at .044; the normed fit index (NFI) is .87; and the Tucker-Lewis index (TLI) and the comparative fit index (CFI) are high at .95 and .97, respectively. These model fit statistics confirm a good model specification.

Based on the individual path coefficients, hypotheses Hla, HIM, and Hlb2 are accepted. Hypothesis Hic 1 is also accepted while Hlc2 and Hlc3 are not. The rejection of Hlc2 and Hc3 indicates that the average Taiwanese IT firm practicing manufacturing postponement, is not more or less likely to also practice labeling or packaging postponement. Overall, the results confirm the suspicion that there is a natural causal relationship between postponement strategies in the earlier stages of production and the later stages of production. Moreover, the strength of these relationships is often quite strong. As mentioned, these results indicate that the experience in implementing one postponement strategy may have lowered the cost of implementing another postponement strategy. In addition, the experience should provide better forecasts of the benefits and costs of implementing postponement strategies, thus lowering the risk of implementation.

The results support hypotheses H2a and H2b at the 95% significance level, indicating that customization is likely to induce modular product designs and modular product designs are likely to induce assembly postponement. Hypothesis H2c is also supported at 95% significance indicating that customization is also likely to induce assembly postponement directly. Hypothesis H3a is accepted at 95% significance while H3b is accepted at 90% significance; H3c and H3d are not accepted. Products with high component costs appear to benefit from labeling and/or packaging postponement strategy but not from manufacturing postponement or assembly postponement. Hypothesis H4a is supported at 90%, indicating that product life cycle has a significantly positive effect on manufacturing postponement implementation. Elaboration of the interpretations of these results is presented in the next section.

DISCUSSIONS

As pointed out earlier, hypotheses Hla-Hlc4 are quite intuitive, but they are essential for specifying the model correctly. The remainder of this section discusses the path coefficients between the four postponement constructs and the four antecedent variables.

The successes of Taiwanese IT firms are due, in no small part, to their ability to manage inventory/logistics costs. Their dominance in the OEM marketplace suggests that their implementations of postponement strategies are quite effective. Thus their actions serve as useful references for process managers everywhere. Next, managerial recommendations are addressed.

As predicted by Feitzenger and Lee, both H2a and H2b were accepted. The acceptance of H2a confirms that products with strong customer demand for customization benefit from the risk-pooling effect of implementing modular designs. The acceptance of H2b confirms that modular product designs will benefit from assembly postponement since modularity reduces the cost of assembly postponement implementation. H2c was also accepted indicating products with high customer demand for customization can also benefit from assembly postponement even if they are not modular in design.

H3a and H3b are supported as predicted by Zinn and Bowersox, however, H3c and H3d are not. Expensive components do not induce firms to adopt either assembly or manufacturing postponement. To understand this result fully, recall that in the empirical study only component costs were measured; process costs were not considered. This was done for a very specific reason. Pagh and Cooper (1998) suggest that anticipatory manufacturing and assembly may be more beneficial, even for high cost products, if a significant portion of the cost is accrued prior to these stages of production. This makes sense because form postponement seeks to reduce inventory depreciation. If the necessary components are high in cost, thus subjected to high depreciation cost, then postponing manufacturing or assembly may gain the firm very little in terms of reducing inventory depreciation while still significantly reducing the economy of scale in production and increasing the time to delivery. In the current study, a product with high component costs would have accrued a significant portion of the total production cost before any manufacturing or assembly activity took place. Thus, it should not be too surprising to find that products with high component costs do not induce manufacturing and assembly postponement.

H4a is supported at 90% significance, which says that products with shorter product life cycles are more likely to benefit from manufacturing postponement. We note that the hypothesis is only weakly supported in terms of significance and the size of the coefficient. There is a tacit assumption embedded in the claim that manufacturing postponement is beneficial for products with short life cycles. The assumption is that the resulting delay in time to delivery must be negligible. It is understandable why firms marketing products characterized by short life cycles will demand fast delivery. Manufacturers that are able to deliver on demand will be more competitive in the marketplace and earn a higher margin. Thus, the postponement benefit arising from the reduction in inventory risks is moderated by the delay in delivery.

LIMITATIONS AND FUTURE STUDIES

As stated in section two, IT products are characterized by high product value, short product life cycle, and high demand for customization. While there is certainly variation in each of the variables, the variance may still be too low to provide very robust results. Different industries in different countries may not produce the same results, which limits the ability of this study to be generalized across a wide range of product categories. Future studies may consider replicating similar studies with data from different industries and countries.

As mentioned earlier, to address the global logistics issues from the Taiwanese manufacturers' perspectives, sharper definitions of the postponement strategies are needed. The United States is the major market of the Taiwanese IT firms. Because of the long distance between the main factory and the market, conducting postponement activities in the local market is necessary to execute form postponement strategy effectively. Therefore, form postponement activities were specified as taking place toward the market in this study. However, it should be noted that form postponement does not always take place toward the local market.

It would be helpful to have multiple measures of the antecedent constructs such as product life cycle and value. Both constructs may have several dimensions. As discussed, product value may imply inventory carrying cost and depreciation cost, while product life cycle may need more indicators to capture the whole concept Future studies should develop multiple items to capture the complete domain of the constructs. In addition, field data can be refined into sub-categories to perform more in-depth analysis. In particular, differences or similarities in the application of postponement strategies between up-stream, mid-stream, and down-stream producers or between OEM, ODM, and OBMs (own brand manufacturers), or between firms with their own retail channels and firms without should be examined. This could shed more light on the benefits and costs of adopting postponement strategies. The size of the sample in this study limits the possibility of this exploration.

Finally, it should be noted that a sample size of 102 is a small number to perform LISREL analysis. And also, it would have been preferable to develop the scales and test the relationships with different data. A larger sample is necessary to extend these exploratory results.

CONCLUSIONS AND MANAGERIAL IMPLICATIONS

Taiwan's information technology industry ranks third in total U.S. dollar output per year. The industry suffers from substantially lower profit margins than its U.S. and Japan counterparts due to its lack of advanced research capability. However, the industry thrives on its manufacturing efficiency and is notably characterized by its ability to manage costs. To this end, the industry's practice of postponement provides evidence on the product/demand characteristics which could benefit from implementing form postponement strategies in terms of direct cost savings.

This study provides evidence that form postponement strategies are practiced widely by IT firms in Taiwan. Furthermore, the four types of form postponements as proposed by Zinn and Bowersox fit the postponement strategies practiced by Taiwan's IT firms.

In addition, products characterized by high customer demand for customization appear to benefit from assembly postponement as well as implementing more modular designs. Products, which are modular in design, appear to benefit from assembly postponement as well. Also products with expensive key components appear to benefit from labeling and packaging postponement, but not necessarily assembly and manufacturing postponement. Finally, products which have short product life cycles, appear to benefit from manufacturing postponement. Managers are advised to reference these empirical conclusions when choosing to adopt postponement strategies in their own production processes.