Toward a tradeoff model for international market selection

International Business Review 11 (2002) 165–192 www.elsevier.com/locate/ibusrev

Toward a tradeoff model for international market selection N. Papadopoulos a,∗, Hongbin Chen b, D.R. Thomas a a

Eric Sprott School of Business, Carleton University, 1125 Colonel By Drive, Ottawa, Ontario K15 5B6 Canada b Nortelnetworks Inc., Montreal, Canada

Received 1 June 2001; received in revised form 4 September 2001; accepted 30 October 2001

Abstract The literature on International Market Selection (IMS) contains many proposed models which make significant contributions but do not effectively address the IMS problem. A new tradeoff model is proposed that uses two key constructs, demand potential and trade barriers, as well as firm strategy as a contingency construct. Each key construct is measured by only four variables, resulting in simplicity and low application cost, and strategy is used to develop weights for the variables. The model is tested in real market conditions for 17 target countries and three diverse products over a six-year period. The weighting solution that favors demand potential is shown to have strong predictive power for the short to medium term. In addition to being tested and validated, the model introduces a number of advances including the weighting of indicators, an approach to quantifying nontrade barrieers, and validation for two different exporting countries.  2002 Elsevier Science Ltd. All rights reserved. Keywords: International market selection; Foreign market choice; International market segmentation; International marketing; International expansion; Modelling

International market selection (IMS) is the first and most important step in export strategy (Root, 1994), making it a critical success factor for both smaller exporters and mature multinational firms. Systematic IMS contributes to export success while wrong choices can put the firm in an unfavorable strategic position (Papadopoulos &

∗

Tel.: +1-613-731-6065; fax: +1-613-731-7572. E-mail address: [email protected] (N. Papadopoulos).

0969-5931/02/$ - see front matter  2002 Elsevier Science Ltd. All rights reserved. PII: S 0 9 6 9 - 5 9 3 1 ( 0 1 ) 0 0 0 5 4 - 3

166

N. Papadopoulos et al. / International Business Review 11 (2002) 165–192

Denis, 1988). Given its importance, IMS attracted significant research attention from the 1960s to the mid-1980s. However, this interest waned in later years, mainly because of the difficulty in developing IMS models that would be generalizeable to various industries while also having adequate predictive power for business (Douglas & Craig, 1992). As a result, the literature to date is limited to either general qualitative frameworks or operational models that have not been tested sufficiently, offer little or no evidence that they can in fact predict market attractiveness, and/or are too complex to apply in practice. On first review, a promising exception appeared to be the shift-share model (Green & Allaway, 1985), which Douglas and Craig (1992) rightly called “the only new approach” to have been proposed in recent years. The model’s core strength is that it appears to be able to predict the industry-specific attractiveness of target markets using two simple and readily available measures: their import size and import growth rate. The model’s major apparent weakness is its exclusive reliance on import measures. Since exporters compete not only among themselves but also against domestic producers in the target countries, the initial intent of this research was to explore ways for extending the model to include total demand measures, in order to reflect the true attractiveness of potential target markets. However, an in-depth review indicated that the base model itself in fact lacks predictive power and is not well grounded theoretically. As a result, this study was revised to develop a new model that would build on the core logic of shift-share but also attempt to meet Douglas and Craig’s (1992: 301) call for “more sophisticated, parsimonious [market selection] techniques in a multi-country context”.

1. Overview of the IMS literature IMS aims to determine the relative attractiveness of markets within a considered set, prior to the final in-depth assessment of the most appealing one(s) for expansion (Reid, 1981). Among others, Ayal and Zif (1978) and Chetty and Hamilton (1993) stress that effective market choice is a strategic decision that affects export performance. Errors can be costly and may also dampen the firm’s export enthusiasm (Welch & Wiedersheim-Paul, 1980). Cooper and Kleinschmidt (1985) have shown that exporters who select targets from the total set of available countries realize more rapid export growth than those who consider only a few (or no) alternatives. The need for a systematic IMS approach is underscored by the complexity of current markets and the growing importance of global strategic positioning (Douglas & Craig, 1983; Albaum, 1998). Yet the evidence is that markets are usually chosen without much systematic analysis and, in the case of smaller firms, most often as a response to unsolicited orders. Generally, export expansion begins from culturally or geographically near countries (Johansson, 1999). Although most of the earlier research has focused on small and medium sized firms, newer studies and/or those focusing on larger firms have reported similar findings (Papadopoulos & Jansen, 1993). Reasons for the general lack of a systematic approach include the limited experience of managers in export research, difficulties in gathering relevant data,

N. Papadopoulos et al. / International Business Review 11 (2002) 165–192

167

and the absence of proven effective approaches for the task (Papadopoulos & Denis, 1988). In their comprehensive review, Papadopoulos and Denis (1988) classified over 40 proposed IMS models into decision making frameworks, most of which are hierarchical process conceptual models; grouping models, which cluster potential target markets on the basis of similarity; and estimation models, which rank countries by order of preference without attempting to group them. Most of the decision frameworks are strategic (Papadopoulos, 1987) but all are qualitative and none deals with operationalizing the steps in the proposed process. Conversely, most grouping and estimation models are not strategic but they do attempt to operationalize parts of the process using formalized statistical analysis. These models contain macro approaches, which consider only general country factors (e.g., see the pioneering work of Liander, Terpstra, Yoshino, & Sherbini, 1967), and micro or industry-specific approaches (e.g., Douglas & Craig, 1982). Except for a handful of econometric methods based on regression analysis (e.g., Moyer, 1968), most models use multiple criteria indices of target country characteristics. While most approaches focus on total demand, some, including the shift-share model which is discussed separately below, only consider imports (e.g., UNCTAD, 1968). A detailed review of the various models suggests that, notwithstanding their many strengths, none offers a reasonably complete solution to the IMS problem. A key issue with both grouping and estimation models is the indicators used to measure market attractiveness. In a nutshell, there is no agreement on which indicators to use and how they might be weighted to reflect their relative importance. Approaches that limit the analysis to only one or two types of variables (e.g., economic or cultural characteristics) do not capture the full set of influences on demand (Day, Fox, & Huszagh, 1988). Some studies include large numbers of equally-weighted indicators as a potential remedy, but Denis (1978) observed that this does little more than introduce significant redundancy (also see Nachum, 1994). Other researchers used factor analysis to group indicators (e.g., Day et al., 1988), but none went to the next step of actually testing and validating a generalizeable variable set. The use of managerial judgment (Douglas, Lemaire, & Wind, 1972) or “expert poll” can be cumbersome if many indicators are used, or very subjective if they are limited to a pre-selected set (Vogel, 1976). Further, this approach can only help the firm that uses it and does not answer the need for a generalizeable model. Lastly, two newer studies (Kumar, Stam, & Joachimsthaler, 1994; Hoffman, 1997) made significant contributions by focusing on the analytical process for IMS (how to analyze and use input indicators) but did not deal with testing and validating the inputs themselves (which indicators to use). The macro segmentation methods share an additional weakness (Samli, 1977; Helsen, Jedidi, & DeSarbo 1993): the implicit theoretical model on which they are based assumes that industry-specific demand can be predicted from general statistics at the country level. By contrast, micro methods need input data on situation-specific variables (e.g., consumer lifestyles) which would only be available, in the best of circumstances, to a handful of large firms. Econometric models are industry-specific but are mostly suitable for mature products facing relatively stable demand

168

N. Papadopoulos et al. / International Business Review 11 (2002) 165–192

(Douglas & Craig, 1983). Further, Singh and Kumar (1971) identified several inconsistencies in the predictive power of selected independent variables depending on the analytical method and countries used. The import demand models have the additional weakness that they cannot account for the market potential that is “available to exporters who may be able to compete against domestic producers or to [stimulate] primary demand” (Papadopoulos & Denis, 1988; also see Root, 1994). In summary, the literature contains a large number of useful theoretical and applied suggestions but none of the models combines the characteristics of being industryspecific, generalizeable, relatively simple to use, strategic, able to reflect the total demand available to the firm, and empirically validated—except for the shift-share model, which, at least on first review, appeared to meet most of these criteria and was therefore selected for a more in-depth examination.

2. Review and analysis of the shift-share model Shift-share uses only one input variable, industry-specific imports at two time points for a selected basket of countries, and estimates each country’s market potential based on its rate of import growth relative to the others (Green & Allaway, 1985). The model’s logic is summarized in the Appendix, part A. Given its core strengths of simplicity and industry-specific orientation, a more intensive analysis was undertaken to examine whether the model’s major weakness, that it is limited to import-only measures, could be addressed. The first step was a careful examination of the relevant literature within the model’s traditional home discipline, regional economics (Cahill & Kristi, 1979). Somewhat surprisingly, this revealed empirical evidence of unreliability and that the model has in fact been criticized for a number of theoretical shortcomings. For instance, in a test of regional employment James and Hughes (1975) found that shift-share yielded forecasts superior to the simple growth model only in a simple majority of cases, and relatively large inaccuracies in the remaining cases. Brown (1969) and Hellman (1976) found shift-share to be even less reliable and raised substantial doubts about its ability to describe regional economic growth. Williams (1980) found that, even with some improvements, the model could at best be somewhat more accurate than several simple alternatives. Shift-share has also been used in a handful of marketing applications (Huff & Sherr, 1967; Yandle, 1978; Kerin, Mahajan, & Peterson, 1980; Green & Allaway, 1985; Vitali, 1990). The researchers noted overall satisfaction with the model’s performance, but also that the results can be biased depending on the base years chosen and may fluctuate greatly due to outliers. Given the doubts raised in the literature, we undertook further analysis to examine the soundness of the model (Appendix, part B). This revealed a further key weakness: its ranking of countries amounts to ranking essentially uncorrelated random noise in the variables, rather than real differences, and, therefore, may be quite misleading. To re-check this, we carried out empirical tests for three products and 50 importing countries using the procedure of Green and Allaway (1985). The results, in part C

N. Papadopoulos et al. / International Business Review 11 (2002) 165–192

169

of the Appendix, confirm the above concerns: countries that seem promising at one time may perform very poorly in subsequent years, the rankings identified by the model are quite volatile, and the shift-share and simple growth model rankings were highly correlated. In conclusion, the in-depth analysis showed that the model lacks predictive power, and its high correlation with the simple growth model even renders it redundant. However, there are two key strengths in the shift-share approach that provide a good base for developing a new model: (a) its core logic, which attempts to capture both market potential and import barriers that may prevent firms from capitalizing upon it; and (b) its underlying objective of developing a simple and flexible model which emphasizes industry-specific demand estimation rather than generic country rankings. We therefore undertook to develop a new model drawing from these characteristics of shift-share.

3. Model development and method The step-wise IMS modelling approach proposed in the seminal studies of Douglas et al. (1972) and Dichtl and Koglmayr (1987) was used to guide the development of the model. This consists of setting relevant criteria; specifying key constructs and their measures and allocating weights to them; setting rules for aggregating scale values and their weights; and using these to transform the data into an overall score. Because the model attempts to address a major gap in international research and to rekindle interest in IMS (Douglas & Craig 1992), every effort was made to identify and use only those criteria, constructs, and measures which had theoretical support, and, more importantly, which could be tested and validated in real market conditions. Since the development process itself may be seen as a contribution of this paper, it is presented at some length. 3.1. Model development: criteria and constructs, variables, and their measures The development process focused on satisfying two general and three specific criteria: (a) the model should be able to screen many markets at the industry level in order to identify those worthy of more investigation, and testable to confirm its external validity and generalizeability; and, (b) the model should use a multiplevariable approach, since these produce more meaningful results (Baalbaki & Malhotra, 1993); use as few variables as possible to achieve simplicity and low cost (Papadopoulos & Denis, 1988); and be able to “assess ... general environmental conditions [but also be] responsive to the specific product … and capable of accounting for the strategic dimension” (Denis, 1978). IMS and mode of entry (MOE) theory suggest that both the “pluses” and “minuses” of the objects under review (in this case, countries) must be considered for effective decisions. These are commonly expressed as tradeoffs between opportunities vs. risks, costs vs. benefits, or cost vs. control (Douglas & Craig, 1983; Anderson & Gatignon, 1986; Ekeledo & Sivakumar, 1998). In this study, given its nature

170

N. Papadopoulos et al. / International Business Review 11 (2002) 165–192

we expressed them as the demand potential and trade barriers in the countries under review. Interestingly, while many researchers have identified trade barriers as the most important export deterrent, these have not been explicitly accounted for in earlier IMS models (e.g., Liander et al., 1967; Helsen et al., 1993) except for the inclusion of tariffs by Hoffman (1997) and the reference to them by Kumar et al. (1994), who, however, did not include them in their model. This is most likely due to the difficulty in quantifying non-tariff barriers, which may have led to leaving them out on the apparent assumption that they will be dealt with qualitatively during the in-depth market analysis stage. A significant by-product has been that the content of the trade barrier construct, and particularly of non-tariff barriers given the reduction of tariffs particularly among developed nations, has not been dealt with in screening models, representing a research gap which this study attempts to address. In addition to market characteristics, to be realistic and relevant an IMS model must account for the firm’s strategic orientation (e.g., Baalbaki & Malhotra, 1993), a dimension that has been neglected in spite of its importance given the strategic nature of IMS. The theoretical rationale behind the need to account for firm strategy is in fact accentuated by the potential vs. barrier specification of the model: different firms with different needs would attach different degrees of importance to each side of the tradeoff. Since IMS decisions cannot be based solely on universal laws applying to all firms, while, for a screening model, they also do not need to be understood at the level of the single firm, strategic orientation can be treated as a contingency variable (Ekeledo & Sivakumar, 1998) to guide the weighting of the constructs and their measures. In conclusion, for an effective IMS model it seems intuitively logical and theoretically valid to explore an approach that has not been attempted before: the tradeoff between demand potential and trade barriers in the context of firm strategy. Based on the above, the model was specified as shown in Fig. 1. To meet the criterion of efficiency, only four variables were used for each of the two main constructs. Since there can be no single “best” set of variables, the selected set will limit any model’s applicability in certain situations and this holds true in the present case (limitations from the variables used are discussed in the concluding section). However, every effort was made to select the most appropriate variables in light of the objective of developing an industry-level multiple-market screening model. Several criteria have

Fig. 1.

Potential-barrier tradeoff IMS model.

N. Papadopoulos et al. / International Business Review 11 (2002) 165–192

171

been suggested in the literature for this task, and these were used to guide the research based on an a priori procedure as suggested by Armstrong (1970). They include relevance, frequency of use in past research, evidence of satisfactory performance in various settings, data availability, reliability, and comparability, and ability to express qualitative factors where necessary. The variables, the rationale for selecting them, and their measures, are described in Exhibit 1. Concerning firm strategy, Ayal and Zif (1978) identify two main types: offensive and defensive. These terms, while not necessarily used by firms to describe themselves, distinguish between those that seek growth at their competitors’ expense and value opportunities more than being concerned about risks, versus those that focus more on preventing competitors from making inroads on their market share. This is Exhibit 1 Variables and their measures Demand Potential Apparent Consumption Measure: Domestic production plus imports minus exports Import data do not portray the total available market. Root (1994) and many others recommend apparent consumption as the appropriate reflection of true market size in a given industry. Import Penetration Measure: Imports as % of apparent consumption Widely used in industry-specific analyses; a high ratio means import market openness and low domestic producer competitiveness, signaling an attractive target market (UNCTAD, 1968). Origin Advantage Measure: Exporting country’s share in target’s total imports Competitors from countries with a high overall share in the target country’s imports in a given industry enjoy advantages (Bilkey, 1987) stemming from such factors as critical mass (e.g., other firms from the exporter’s country can help with market information); a favorable image of the origin country’s products in the given industry (Papadopoulos, 1999); and strong trade relations between the exporting and importing countries, often leading to greater trade promotion effort and local representation by the former (Alexandrides & Moschis, 1977). Market Similarity Measure: Overall score of four indicators Demand tends to be higher in markets similar to where a product was initially developed (Linder (1961) and Vernon (1966) demand-effect model, confirmed empirically by Davidson (1983)). High similarity reduces risk and uncertainty and enables strategy standardization and scale and scope economies (Davidson, 1983). Sethi (1971) proposed 29 indicators of market similarity, grouped in four categories: health & education, personal consumption, production & transportation, and trade. Given the need for a parsimonious model, we selected one indicator from each group based on its high correlations with others in the group. To obtain an overall score for the variable, we first standardized the indicators and then weighted them by their loadings on the first factor identified by principal components analysis. The indicators used, corresponding to Sethi’s (1971) groups, were (loadings in brackets): life expectancy (0.77), GNP/capita (0.87), electricity production (0.48), and imports-to-GDP ratio (⫺0.48).

172

N. Papadopoulos et al. / International Business Review 11 (2002) 165–192

Exhibit 1 Continued. Trade Barriers Tariff Barriers Measure: Weighted mean annual tariff rate over the study period Tariffs have a direct effect on the exporter’s prices and pricing strategy discretion. Nontariff Barriers Measure: Composite quantitative index of 20 barrier items Nontariff restrictions often are more important than tariffs as an obstacle to exporting (Alexandrides & Moschis, 1977). A low overall incidence suggests target market openness, and vice versa. Most are qualitative (e.g., labeling rules, price surveillance) and measuring them requires a quantification scheme, unavailable in other research. For this study we developed a composite index consisting of all 20 barrier items in the World Trade Organization’s Trade Policy Review, by weighting each item by its frequency of occurrence in the target countries (per WTO data; there were 112 occurrences in the countries selected to test the model). Geographic Distance Measure: Mileage distance between exporting-target countries Directly related to transport costs; can act as a major barrier through its effect on export price (Alexandrides & Moschis, 1977, Aksoy & Kaynak, 1994). The distance between the two countries’ main ports (if no port, the capital or next closest major city) was used. Exchange Rate Measure: Percent change in official exchange rate vs. previous year Given their volatility, currency exchange rates between the exporting and importing countries are a major risk element in exporting and can have a major impact on pricing and strategy.

a dichotomous representation of a continuous variable, but for purposes of simplicity and given the needs of the model it was adopted as the framework in this case. The external criteria used in model validation were product-specific import size and import market share of the exporting country, and the relative growth rates of these measures. Since these measures are industry-specific, the limitations of import indicators discussed above in relation to the earlier macro models would not apply in this case and so their use was justified. 3.2. Method 3.2.1. Target (importing) markets After considering various alternatives, we selected 17 OECD countries as the target market set for the model illustration. In addition to the benefit of data availability, the similarities among these developed countries help to reduce any importing country effects.

N. Papadopoulos et al. / International Business Review 11 (2002) 165–192

173

3.2.2. Exporting countries A key weakness of earlier IMS models is that they approach country screening by focusing on the importing countries only without any regard to which is the exporting country. As noted, however, exporters have different advantages depending on their origin country (Exhibit 1). To test for potential exporter country effects and improve model validity, we used two greatly different countries: Canada (highly developed mid-sized nation, G7 member, major and experienced exporter) and China (world’s largest country by population, developing nation, earlier stage of internationalization). 3.2.3. Products Three products were chosen to represent industrial goods and consumer durables and non-durables, using data from the SITC (Standard International Trade Classification) and ISIC (International Standard Industry Classification) systems: aircraft, furniture, and beverages. The UN Statistical Bureau (1994) cautions that secondary trade data at the most detailed level of classification may contain relatively large errors, and some data are only available at the 2-or 3-digit SITC level (equivalent to the 3- and 4-digit ISIC level). The SITC-ISIC code match for the selected products is good (respectively 792-3845, 11-313, and 821-332), thus improving comparability and reliability (limitations are discussed in the concluding section). Except for China in aircraft, Canada and China are relatively important exporters of these products with respective world export shares in 1993 of 3.1% and 0.3%, 5.1% and 3.2%, and 3% and 2.8%. 3.2.4. Time period The exact period is of no consequence for the model test, since, in business applications, it would be chosen to fit the firm’s needs. The main criterion for illustrating the model was selecting a long enough period to not be influenced by lagged effects, i.e., the time distance from when managers make an observation about a country to when they react to it. This was necessary since there is no clear guidance in the literature as to the length of this distance, or the reverse, whether the difficulty to forecast over long periods means that the effects of decisions drop off as the lag period gets longer. Based on this rationale, the six-year period 1989 to 1994 was selected for reference, to account for either possibility, with 1988 as the base year. 3.2.5. Data development and model validation Measurement data was collected from nine statistical databases mainly from the OECD, IMF, UN, and WTO, and entered into a matrix for each of the two exporting countries, three products, 17 target countries, and six reference years. Some data editing was done to get around missing data (e.g., missing import data for Austria in 1994 were imputed by regression analysis based on its imports in other years and the 1994 imports of the other 16 countries). The data for each variable was scaled using the procedure of Liander et al. (1967), by subtracting the lowest country value from the highest and dividing the difference by 10, thus forming 10 equal scale intervals and then assigning to each country a score from 0 to 10 based on its absolute

174

N. Papadopoulos et al. / International Business Review 11 (2002) 165–192

value. The scores of each country for each variable were then averaged to get a total score for each of the potential and barriers dimensions. High scores on the 0–10 scale correspond to high potential or low barriers. Next, countries were plotted using SPSS and grouped into four clusters (high-low potential and low-high barriers) in a matrix. Since the two dimensions were identified based on IMS and MOE theory and have been used in various earlier studies, as described, the procedure is well grounded. The tests for the “two-dimensional” approach were carried out first, without assigning weights, as in earlier IMS research. Weights were then assigned to develop “total score” country attractiveness scales that combine the two dimensions. The correlation coefficient was adopted as the criterion for evaluating the model’s predictive power, and correlation tests were conducted between the country scales for the base year and for each successive year from 1989 to 1994. The use of a weighting scheme made it possible to carry out sensitivity analysis to explore the possible influence of different firms’ strategic orientations and different product characteristics.

4. Empirical results and discussion 4.1. “Two-dimensional” approach The target markets were clustered by their values into even quadrants, based on the plotting described above for each of the six [exporting country ∗ product] sets. The results are shown in Fig. 2. Target markets in the upper right (high potential/low barriers) offer the best export opportunities, those in the lower left (low potential/high barriers) are the least promising, and so on.

Fig. 2.

Two-dimensional solutions for Canada and China.

N. Papadopoulos et al. / International Business Review 11 (2002) 165–192

175

4.1.1. Product effect Fig. 2 shows that target countries perform quite differently in different products for the same exporting country. For Canada, for example, the US dominates in all three products but Australia, Austria and Finland appear as attractive aircraft markets, as mediocre in beverages, and as inferior in furniture. For China, Norway appears as an attractive furniture market but as having low potential in aircraft or beverages. 4.1.2. Exporting country effect Target countries also offer different opportunities for different exporting countries in the same industry. In beverages, Japan was attractive for China but mediocre for Canada. In aircraft, Australia offered high potential for Canada but low for China. This validates the need for exporter-specific research and shows that failure to account for the exporter’s home country may be misleading: a good market for one is not necessarily good for another. 4.1.3. Individual indicator effect Countries in the same cluster may have different implications for exporting firms. In beverages, for example, both Japan and Sweden were low potential-low barrier markets for Canada. However, the tabulation of individual indicators showed that Japan had high scores on apparent consumption (7.04) and country-of-origin advantage (3.32) but lower ones on import penetration (0.0) and tariff barriers (0.40). Conversely, Sweden had high scores on import penetration (4.82) and tariff barriers (10.0) but low ones on country-of-origin advantage (0.77). Thus Canada had a higher import market share in Japan but was faced with strong domestic competition and high import barriers. Conversely, in Sweden it had a low import market share in spite of low barriers. Therefore, Canadian exporters would need to focus on competing against domestic producers in Japan but against exporters from other countries in Sweden. Thus the individual indicator values can help firms in developing more effective strategies for individual markets. To summarize, the two-dimensional approach indicates that international opportunities will differ for firms (a) from the same home country but in different sectors, or (b) in the same sector but from different home countries. 4.2. “Total score” approach Many users may prefer to rank countries on a single overall score, rather than plotting them on two separate dimensions. The tradeoff approach makes it possible to assign weights based on the strategy dichotomy, and thus to address what has been one of the main impediments in IMS theory to date. Firms with a defensive strategy would focus more on markets that appear easier to penetrate and be more likely to be deterred from countries with high trade barriers, while those with an offensive strategy would be more likely to focus on opportunities and put more effort into markets that may be difficult to penetrate but present strong potential. Accordingly, hierarchical weighting schemes were designed by assigning different weights to the potential and barrier constructs, with a sum equal to one, and then to each

176

N. Papadopoulos et al. / International Business Review 11 (2002) 165–192

indicator in each dimension, also with a sum equal to one. The scores of the two dimensions were then summed to generate an overall composite score for each country. Three weighting schemes were used: Weq, weighting both dimensions and all indicators equally; Wdp, favoring demand potential; and Wtb, favoring trade barriers. This essentially amounts to choosing the spatial direction onto which one projects the two-dimensional scores. The logic was to downweigh certain barrier variables in the Wdp scheme, and vice-versa, since the target markets are quite similar in structure and distance from either China or Canada (most OECD countries are in Europe) and most are open economies (EU or NAFTA members). Market similarity was also downweighted in Wdp since more aggressive firms are also likely to be more adventurous and less driven by psychic distance. The specific weights used are shown in Table 1. The composite country scales for the base year corresponding to the three weighting schemes were then correlated with the validation criteria for the reference years. The correlation tests, shown in Table 2, showed a great degree of consistency. For all weighting schemes, the correlations between the base and reference years on import growth and market share growth (Table 2a and b) were very low or even negative. The growth rates are volatile and impossible to forecast, as was the case with the findings from our empirical tests of the shift-share model (Appendix A). Conversely, the correlations from scheme Wdp were high on import size and market share except for furniture for China (Table 2c and d). The consistency between these two measures illustrates that a high share in a given market is often accompanied by large imports, confirming that market size and the origin country’s competitiveness are an integral part of market potential. The performance of the other two weighting schemes (Weq and Wtb) showed great inconsistencies and was always poorer than Wdp. 4.2.1. Weighting effect Not surprisingly, the relative standings of the 17 target countries differ depending on the weights. The overall performance of the model based on Wdp was quite good and consistent across the product categories, contradicting the expectation that different approaches are needed depending on product and firm strategy. Since Wdp emphasizes demand potential, the results would seem to suggest that opportunities rather than barriers predominate in firms’ actual IMS decisions (i.e., most firms are aggressive/optimistic and thus weigh market potential more heavily). 4.2.2. Product and export life cycle effect The ratios of total imports in 1994 vs. 1988 show that the imports of all 17 countries from Canada were quite stable in all industries (beverages 1.39, aircraft 1.35, furniture 1.80), as were beverage imports from China (1.35), suggesting products in the maturity stage of the export life cycle. The high import size and market share correlations for these products suggest that for mature industries the model may not be very sensitive to weighting. On the other hand, there was rapid growth in aircraft (14.00) and furniture (6.86) imports from China. The former were characterized by

Demand potential

0.75 0.45

Weight scheme

Wdp Wtb

Constructs

0.25 0.55

Trade barriers 0.33 0.22

Apparent cons.

Potential

0.33 0.22

0.27 0.22

Origin Import advantage penetr.

Table 1 Weighting of constructs and variables for total score approach

0.07 0.34

Market similarity 0.20 0.27

Tariff barriers

Barriers

0.20 0.27

Non-tariff barriers

0.20 0.27

Geogr. distance

0.40 0.19

Exchange rates

N. Papadopoulos et al. / International Business Review 11 (2002) 165–192 177

’90

a. Import growth Canada Weq 0.02 −0.55b Wdp −0.02 −0.39 Wtb 0.02 −0.55b China Weq −0.12 −0.52c Wdp −0.16 −0.36 Wtb −0.16 −0.54c b. Market share growth Canada Weq 0.06 −0.57b Wdp 0.04 −0.37 Wtb 0.07 −0.57b China Weq −0.30 −0.50c Wdp −0.34 −0.33 Wtb −0.32 −0.53c c. Import size Canada Weq 0.71a 0.70a Wdp 0.88a 0.87a Wtb 0.63a 0.63a China Weq 0.64a 0.62a Wdp 0.91a 0.90a Wtb 0.48b 0.47c

’89

Product Beverages

0.06 −0.77 0.35 −0.43 0.01 −0.85

0.09 0.17 0.14 0.30 0.04 0.10

0.19 −0.79b 0.32 −0.43 0.19 −0.88b

0.71a 0.71a 0.87a 0.88a 0.63a 0.63a

0.62a 0.55b 0.90a 0.89a 0.46c 0.39c

−0.37 −0.51 −0.30

−0.18 −0.14 −0.15

−0.24 −0.41 −0.15

0.71a 0.87a 0.63a

0.60a 0.90a 0.44c

’93

0.06 0.27 0.15 0.37 0.01 0.18

’92

−0.21 −0.19 0.17

’91

0.61a 0.88a 0.45c

0.71a 0.88a 0.63a

0.11 −0.06 0.16

−0.05 −0.18 −0.01

0.22 0.02 0.28

−0.05 −0.19 −0.01

’94

0.53b 0.81a 0.44c

−0.08 −0.22 −0.03

0.18 0.22 0.10

−0.03 −0.25 0.04

0.23 0.19 0.19

’90

0.53b 0.81a 0.44c

−0.07 −0.09 −0.05

−0.03 −0.17 0.05

0.15 0.01 0.24

−0.07 −0.20 0.02

’91

0.16 0.08 0.09 0.65a 0.58b 0.59b 0.01 −0.05 −0.03a

0.53b 0.80a 0.44c

0.04 0.00 0.07

−0.37 −0.29 −0.33

−0.17 −0.21 −0.14

−0.37 −0.30 −0.33

’89

Furniture

Table 2 Total country scores vs. validation criteria using three weighting schemesa

0.08 0.58b −0.04a

0.53b 0.81a 0.44c

0.41 0.48c 0.35

0.05 −0.03 0.06

0.40 0.37 0.41

0.19 −0.03 −0.22

’92

−0.06 0.06 −0.37

’94

0.09 0.26 0.00

0.15 0.64a 0.02

0.53b 0.81a 0.44c

0.18 0.65a 0.04

0.53b 0.81a 0.44c

0.35 −0.09 0.15 −0.17 0.43c −0.02

−0.39 −0.33 −0.34

0.43c −0.14 0.47c −0.11 0.39 −0.12

−0.16 0.08 −0.36

’93

’90

0.22 0.22 0.82a 0.82a 0.12 0.12

0.46c 0.46c 0.81a 0.80a 0.40 0.40

0.27 0.09 0.24 -0.11 0.25 0.10

0.04 0.18 0.16 −0.13 0.02 0.21

’89

Aircraft

−0.27 −0.18 −0.29

’92

0.15 0.75a 0.05

0.45c 0.80a 0.39

∗

0.12 0.73a 0.02a

0.44c 0.79a 0.38

−0.39 −0.22 −0.02 −0.11 −0.42c −0.23

∗

−0.20 0.16 −0.24

’91

0.45c 0.79a 0.39

−0.19 −0.07 −0.18

−0.38 −0.18 −0.38

’94

0.20 −0.26 0.79a 0.22 0.10 −0.30a

0.41c 0.76a 0.36

0.24 0.01 0.28

0.25 −0.01 0.28

’93

178 N. Papadopoulos et al. / International Business Review 11 (2002) 165–192

0.76a 0.88a 0.69a

0.58b 0.81a 0.43c

0.75a 0.91a 0.67a

063a 0.84a 0.49b

share size

’90

0.54b 0.79a 0.40

0.76a 0.87a 0.69a

’91

’93

0.56b 0.41 0.79a 0.73a 0.42 0.26

0.76a 0.76a 0.89a 0.90a 0.69a 0.69a

’92

−0.24 0.02 −0.39

0.48 −0.15 0.72a 0.08 0.35 −0.30

’90

0.50b 0.79a 0.41

’89

0.50b 0.79a 0.41

0.76a 0.90a 0.68a

’94

Furniture

−0.28 0.00 −0.43c

0.51b 0.80a 0.42c

’91

0.19 0.13 −0.35

0.51b 0.80a 0.42c

’92

−0.01 0.29 −0.18

0.50b 0.79a 0.41c

’93

0.09 0.39 −0.08

0.51b 0.79a 0.42c

’94

’90

0.51b 0.88a 0.45c

’91

0.50b 0.85a 0.44c

’92

0.19 0.21 0.04 0.04 0.79a 0.79a 0.44c 0.62a 0.09 0.12 −0.03 −0.04

0.55b 0.60b 0.88a 0.89a 0.48c 0.54b

’89

Aircraft

0.60 0.72a 0.58b

’94

0.24 −0.37 0.80a −0.15 0.14 −0.36

0.59b 0.82a 0.54b

’93

a Significance: a: ⬍0.01; b: ⬍0.05; c: ⬍0.10. ∗ China had no aircraft exports to many OECD countries in 1988, and so the analysis of import growth rate and import share growth rate for this country was limited to the other two industries.

d. Market Canada Weq Wdp Wtb China Weq Wdp Wtb

’89

Product Beverages

Table 2 Continued

N. Papadopoulos et al. / International Business Review 11 (2002) 165–192 179

180

N. Papadopoulos et al. / International Business Review 11 (2002) 165–192

small volume, high market concentration, and rapid annual growth, while the latter were relatively higher, diversified across many countries, and with high growth. This points to industries respectively at the introduction and growth stages. The model’s predictive power for the growth product was relatively low, suggesting that in this stage high market share does not necessarily relate to large imports at first. However, in markets where the exporting country has a large share imports are likely to grow faster. This can be seen in the increasing market share size coefficients in 1989– 1994 for furniture for China (Table 2d), which enjoyed greater export growth in those countries where it already had a relatively higher market share in 1988 (e.g., US, UK). 4.2.3. Exporting country effect As shown in Table 2c and d, the model showed similar high predictive power for both Canada and China in beverages and, except for a few cases in 1994, aircraft. Conversely, a large difference in predictive power between Canada and China was found in furniture. This could be attributed to differences in life cycle stages, as above, and so, excluding this effect, the model has relatively similar predictive power for both countries. The model may display different predictive power for different countries (especially developed vs. developing) for which exports of the same product are likely to be in different stages. 4.2.4. Planning horizon The predictive power of the model over time is strong, with high correlations between the base and all reference years on country-specific imports and market share size. Thus lagged effects, which were identified above as a potential concern, do not appear to be an issue. With suitable caution for product exports in the growth stage, the model can be used for short- to medium-term IMS planning provided the market does not suffer major external shocks. In summary, the model performed very well with weighting scheme Wdp, showing that some variables are more important in IMS regardless of product characteristics. Excluding the life cycle effect, no significant difference in model power was found over the two different exporting countries, showing that the model may be generalizeable for use by various exporting countries. 4.3. Solution of overall country scaling Since one weighting scheme (Wdp) consistently points to a more accurate forecast, IMS can be based on the overall country scales generated by Wdp. The target countries were clustered into groups by their total scores to reflect four levels of opportunity: high, medium-high, medium-low, and low. The relative standings of the target countries are illustrated in Table 3. 4.3.1. Product effect Similarly to the two-dimensional approach, target markets offer different opportunities for the same country in different products (e.g., Austria is in the medium-

US J S,UK,NL,B

G,D,F,I,AU,N,NZ, FI,SP,P,A

Group 1 (⬎6) Group 2 (4-6) Group 3 (3-4)

Group 4 (⬍3)

I,B,SP,AU,A, FI,P,NZ

US J,NL S,N,G,UK,F,D

Furniture US A AU,SP,N,FI,G, NL,D,B,UK S,NZ,P,J,I,F

Aircraft

UK,NL,S,AU,G,F,D,I,NZ, P,N,A,FI,SP

J US B

Beverages

China

Individual countries within cells are in declining order by total score. Country abbreviations as in Fig. 2.

Beverages

(score levels)

a

Canada

Groups

Table 3 Relative market opportunities for Canada and Chinaa

J,US NL,S,N,UK, G,F,D B,I,NZ,SP,AU, P,A,FI

Furniture

US AU,SP,A,FI,N,B, G,D NZ,P,NL,J,UK,S,F,I

Aircraft

N. Papadopoulos et al. / International Business Review 11 (2002) 165–192 181

182

N. Papadopoulos et al. / International Business Review 11 (2002) 165–192

high group for Canada in aircraft, but offers fewer opportunities in beverages and furniture). 4.3.2. Country effect Japan was identified as the best market for China in both beverages and furniture, while the US was the dominant market for Canada in all products. However, the relative standings of the other markets showed great consistencies for both exporters. This is somewhat surprising compared to the above findings from the two-dimensional approach. The country-specific advantages of Canada and China in markets other than the US and Japan may not be diverse enough to distinguish among them, or this may be an artifact of the weighting scheme in the total score method. 4.3.3. Individual indicator effect The individual indicators also provide additional insights for firms using the total score method. For instance, Holland and the UK seem to offer similar strengths for Canadian beverages, but Holland had a higher score on import penetration and lower ones on apparent consumption and origin advantage. That is, this is a small market with weak domestic firms where Canadian exporters would encounter strong competition from other exporters to it. 4.4. Comparison of two-dimensional and total score approaches A comparison between Fig. 2 and Table 3 shows several differences between the two-dimensional and total score results (e.g., France is a high potential/high barrier market for Canada in beverages in Fig. 2, but part of the lowest opportunity group in Table 3). Such differences mainly reflect the absence of indicator weighting in the two-dimensional approach. To confirm, the Wdp weighting of the total score method was applied separately to each of the potential and barriers constructs, and the importing countries were re-plotted into high-low potential-barriers quadrants, as before. The results illustrated a great degree of consistency between the two methods, with both now classifying the importing countries in the same opportunity groups. This emphasizes the importance of using a weighting scheme in IMS, and raises confidence in the model in line with the argument by Liander et al. (1967) that consistent results from two methods help to confirm the validity of both. 4.5. Two extensions in model application 4.5.1. Outliers One problem of scaling techniques is the potential influence of outliers, which may define the 0–10 scale so as to reduce the degree of differentiation among other markets. A potential solution, which we selected for testing using beverages from Canada, is to re-apply the model by excluding outliers. The US and Portugal were identified as outliers at the upper and lower ends, respectively, and the total score model (using Wdp) was re-tested without them. The results showed high correlations (0.80 to 0.92) between the total score scales and import size and

N. Papadopoulos et al. / International Business Review 11 (2002) 165–192

183

only somewhat lower correlations with market share (0.66–0.83 for most years). Therefore, the influence of outliers in market screening may not be as large as is commonly thought. 4.5.2. Optimal weights The tests showed that the weighting scheme was crucial to the model. The “best” weights, or the linear combination of different variables that maximally correlate with the validation variable, would be given by the coefficients from a regression of the validation variable on the predictor variables. We therefore examined whether weights established by regression would improve the model compared to those established by the theoretical reasoning. Since only 17 countries were used and regression analysis cannot reliably identify best weights for eight individual indicators, the scales weighted per Wdp on potential and barriers were treated as the independent variables, and the country scales on import size as the dependent variables. The results suggested that regression models in separate years may generate different coefficients. In such cases, best weights could be taken as the simple or weighted average of the sets of coefficients identified in each separate model, obtained by a generalized least square regression as proposed by Zellner (1962). Further, the coefficients were very sensitive to outliers (17- vs. 15-country solutions) and also differed depending on the product. Therefore, regression can only identify best fits for an individual case. A comparison between Wdp in Table 2 and the coefficients in the regression model showed no great drop in the predictive power of Wdp. In most cases, the multiple correlation coefficient of Wdp was above 0.8. Thus, Wdp may not be optimal but its use is justified by its simplicity. Of course, for users willing to do the additional work, regression analysis can help to identify the best weights for each variable.

5. Implications and conclusions Research of this type is limited primarily by the deficiencies of secondary data. Key challenges include the lack of direct conversion schemes between the trade coding systems (the SITC codes alone have been revised three times since 1965 and some countries still report using earlier versions), varying data definitions, and the unavailability, unreliability, or aging of data for some countries, particularly less developed ones (e.g., see Nachum, 1994). Greater product specificity would also be preferable to manufacturers of product sub-categories (e.g., lawn vs. upholstered furniture), for whom data aggregated at the SITC-ISIC levels used here may be misleading. Lastly, the nature of some of the selected variables may also limit model applications. For example, apparent consumption measures would be of little interest to firms seeking first mover advantages in countries where the market for their product is presently nonexistent, and the country-level variables reflect factors that affect some industries more than others (e.g., geographic distance has a greater effect on the transport of bulky goods, and nontariff barriers would ideally be measured at the industry level).

184

N. Papadopoulos et al. / International Business Review 11 (2002) 165–192

In this study we attempted to address as many of these limitations as possible, mainly by following various approaches proposed in earlier studies (e.g., use of the UN Statistical Office guidelines for code matching); applying a variety of theoretically-grounded ad hoc solutions where necessary; using variables that past research has confirmed as the most appropriate and important, given the screening nature of the model, and as being relevant to a wide a range of industries; and using a very parsimonious set of indicators. This, coupled with the model validation in real market conditions, makes us reasonably confident that potential negative effects were minimized. Larger multinational firms, which would need to apply the model separately for each of the industries in which they compete, also have more resources at their disposal and may in fact be able to obtain more product-specific information. As well, modern information technology should further help to address the deficiencies of secondary data. The situation is improving even in the case of less developed countries where, more than 10 years ago, Day et al. (1988: 17) were able to find “enough data [18 indicators] on enough countries [96] to permit a meaningful analysis.” Carefully selected proxy variables remain a promising alternative for other cases (e.g., in another phase of this research we found that the imports-to-GDP ratio is a good substitute for import penetration, for which data is often unavailable). Therefore, notwithstanding its limitations, the model provides a significant improvement over earlier ones since (a) it captures total rather than import-only demand; (b) it is industry-specific and efficient, unlike most multiple criteria models; (c) it was tested using three different products and, unlike any previous model, using two very different exporting countries; (d) unlike econometric methods, it is generalizeable across industries, and (e) it was externally validated. The performance of the total score model was good except for a few cases, suggesting that careful selection of constructs, measures, and weights can pay off. With the proper weighting scheme, which in this case reflects opportunity-oriented business behavior, the model’s ability to predict future imports and market shares in selected target markets was high. Excluding product exports at the growth stage, the lack of significant differences in predictive power between the two exporting countries suggests that the model may be used by both developed and developing countries. Long-term shifts in the structure of a country’s GDP or the ratio of domestic production to imports, and/or major events causing such shifts in the near-term, cannot of course be predicted and would limit any model’s applicability. However, this model showed no sign of sharply declining predictive power over the six-year period, suggesting the absence of major lag effects and, therefore, that it may be used with reasonable confidence for short- to medium-term market planning. The consistency of findings between the two-dimensional and total score approaches, and the predictive power of Wdp, point to several practical and theoretical benefits from this research. For business, statistics can only tell a small part of the story. Detailed sales potential analysis would be needed, and firm and product characteristics would need to be taken into account, before a decision to

N. Papadopoulos et al. / International Business Review 11 (2002) 165–192

185

expand. Therefore, what a model such as this can provide is a systematic, relatively inexpensive, and effective method for scanning a large number of markets, whose potential usefulness is underscored by the high cost of wrong IMS decisions. The individual indicators can also be used to help develop different strategies even for target countries which appear to offer similar overall opportunities. In addition to having more resources, and thus likely being in a better position to obtain more product-specific data, large multinational firms may find the model particularly useful since they typically export several related products in each sector, making the SITC-ISIC data level less of a concern. For public policy makers and export promotion agencies, the finding that firms in different industries from the same country should explore different export markets indicates that generic export assistance programs, which are common in most countries (Seringhaus & Rosson, 1991), will likely be less effective than programs tailored to the needs of particular sectors. From the theory standpoint, the in-depth review of the shift-share model may lead to further research in marketing and/or regional economics aimed at rechecking its assumptions and perhaps revising it. In addition, the study provides a theoretical framework for IMS through a tradeoff model that draws from and improves upon the tradition of multi-criteria and demand estimation approaches. Instead of providing a “universal list” of market opportunities, which can be highly misleading for different industries, the model can identify unique industry-specific opportunities. Given that the more specific the segmentation basis, the less stable the segment (Wind, 1978), the model uses a reasonable analytical level (SITC 2- or 3-digit) while also representing a step forward compared to general country rankings. The predictive power of the Wdp weighting system is, in itself, an interesting comment on the export behavior of firms, as well as being the first validated IMS model that accounts for firm strategy. Lastly, the research suggests a potentially useful approach for further work on IMS model development and validation. Future studies may focus on at least three areas that call for a renewed research effort. First, replications of the tradeoff model to test its validity using different importing country sets, exporting countries, products, and time periods. Second, efforts to expand the model beyond the exporting domain, since with suitable adjustments the underlying logic should be applicable to market selection for other entry modes, including higher control ones (such as direct investment) which would be of greater interest to larger MNEs. Finally, future studies may improve the model by examining its various individual aspects and limitations in more detail. This may include, for example, testing of different variable sets, confirmatory factor analysis using a larger country sample to validate the two dimensions, refinements that allow for the differential weighting of critical factors such as geographic distance, depending on their relevance at the firm level (Hoffman, 1997), or a search for proxy variables to further simplify the procedure or to enable its application to countries for which the kind of data needed for this model may not be available.

186

N. Papadopoulos et al. / International Business Review 11 (2002) 165–192

Appendix A. Logic and tests of the shift-share model A.1. Mathematical basis For a set of I mutually exclusive markets, let Vi.t (i = 1, 2 ...I) represent imports for market i at the end of time period t, and let ⌬Vi = Vi.t ⫺ Vi.t⫺1 be the actual change in imports of market i over time t. The sum of imports by all markets at the beginning of the period is ⌺Vi, t⫺1, where the summation is taken over i = 1,......I. The average growth rate for all markets, k, will be k ⫽ ⌺Vi,t / ⌺Vi,t⫺1

(1)

Had market i grown at the average rate for all markets, its expected imports at the end of the period would be E(Vi,t) = kVi,t⫺1. The difference [actual - expected change] is the net shift: Ni ⫽ Vi,t⫺E(Vi,t) ⫽ Vi,t⫺kVi,t⫺1

(2)

Per eq. (1), the sums of all positive and all negative net shifts must be equal, and the sum of the net shifts for all markets is zero, i.e. ⫺ ⌺N+i ⫽ ⫺⌺N⫺ i ⫽ ⌺兩Ni 兩

(3)

Let S represent the total sum of positive net shifts, S = ⌺|Ni⫹ |. The relative gain or loss of imports for market i is the percent net shift, and the sums of the positive and negative percent net shifts are 100 and ⫺100 [i.e., Pi = Ni/S (100%)]. The percent net shift reflects the relative import gain or loss. Countries with a high positive percent net shift are identified as attractive for exports. A.2. Statistical analysis The simple linear regression model in this case would be specified as follows: Vi,t ⫽ b0 ⫹ b1Vi,t⫺1 ⫹ e

(4)

It can be seen that the shift-share model explored by Green and Allaway (1985) is closely related to a special case of this model obtained by forcing b0 to equal zero. From the theory of linear regression, it is known that such a model is not necessarily a good fit (Stevens, 1992). The most flexible model for prediction is one that includes b0, that is, one that minimizes the sum of the squared residuals (Vi,t ⫺ b0 ⫺ b1Vi,t⫺1)2 . If the technique is applied to the basic linear regression model, we get percent net shift from the best prediction model as follows: PRi ⫽ NRi / SR(100%)

(5)

where NRi ⫽ Vi,t⫺b0⫺b1Vi,t⫺1, or the regression residual errors, which take the place of the net shifts of Eq. (2). (R superscripts used to separate the regression from the shift-share model.)

N. Papadopoulos et al. / International Business Review 11 (2002) 165–192

187

If the regression coefficients b0 and b1 were known, the percent net shift PRi of Eq. (5) would be proportional to the NRi, which is equal to the random errors ei. When the regression model is the true model, these errors are usually assumed to have an independent normal distribution with mean zero and common variance ␴. Thus, applying the shift-share technique under a known regression model is equivalent to ranking un-correlated normal random variables with common variance ␴, i.e., equivalent to ranking noise. In practice, the coefficients will be estimated by least ⵩ ⵩R ⵩ ⫽ Vi,t⫺b⵩ squares, yielding estimates b⵩ 0 and b1 and residuals Ni 0 ⫺b1 Vi,t⫺1. The net shift will again be proportional to these residuals, which, according to linear regression theory, will be normally distributed with mean zero. Unlike the ei, the are correlated but these correlations will be small for moderestimated residuals N⵩R i ate to large samples. Their variance will depend on Vi,t⫺1, with the largest variance associated with markets whose values of Vi,t⫺1 are furthest from the mean (Vi,t⫺1). However, the information in their variances can be regarded as second order, since the expected values of the residuals are all zero. In general regression situations, the effect of correlation between residuals is usually negligible, except when the ratio (n⫺p)/n, (degrees of freedom in residuals)/(no. of residuals), is quite small (Anscombe & Tukey, 1963). Thus, the previous argument still applies when the coefficients are estimated from the data: the ranking of countries by percent net shift is equivalent to ranking essentially uncorrelated random variables which all have a zero mean. A.3. Empirical testing Three products selected randomly from the International Trade Statistical Yearbook, and the top-50 countries for which import data for 1986–1993 was available, were used. The products were reaction engines (23 countries), paper product machines (33 countries) and nitro-phosphate fertilizers (18 countries). To test the model, correlation tests were conducted between the country rankings by percent net shift using 1986–1990 as the base period and two other overlapping time periods (Table 4). Table 5(1) shows that the correlations were low and even negative (e.g., reaction engines—US 1st in 1986–90 vs. 15th in 1990–93). An analysis of country rankings over different periods was also done using two simple models. The first, based on the import growth rate, yielded relatively low or negative correlations, illustrating the volatility of this measure (Table 5(2)). Conversely, the second, based on import size, showed great stability over time with all coefficients at 0.69 to 0.98 with significance levels below 0.01 (Table 5(4)). Lastly, the country rankings identified by shift-share were compared to those from the simple growth model over the same time period and also produced relatively high correlations (0.73 or higher and at low significance levels in most cases; Table 5(3)). These very consistent findings, using different country sets and different products at two different SITC digit levels, show the effect of unreliable country rankings by ordering random errors. More importantly, the shift-share and simple growth model rankings were highly correlated, and further mathematical analysis (not shown here due to lack of space) shows that this should be an expected

9

Turkey

a

12 4 11 8 3 19 9 5 10 7 16 6 21 23

13

1

2 20 22 17 15 14 18

11 3 12 9 2 18 10 6 20 8 14 4 21 23

7

1

15 16 22 19 17 13 5

Belgium/Lux Portugal Singapore Sweden Ireland Tunisia Switzerland Finland Egypt Venezuela Argentina Greece Hong Kong Israel Canada Norway N. Zealand Turkey

Indonesia

Thailand

France Spain Italy Malaysia Germany Japan Brazil

10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27

9

8

1 2 3 4 5 6 7

86–90

12 18 3 28 23 8 29 27 21 17 13 15 11 22 32 24 25 14

2

7

31 10 19 1 20 26 9

88–92

24 26 8 27 21 15 20 25 13 14 5 18 4 10 28 16 12 9

6

3

33 29 31 19 32 30 17

90–93

∗ SITC codes: reaction engines 7144; paper product machines 7252; nitro-phosphate fertilizers 56292.

10 11 12 13 14 15 16 17 18 19 20 21 22 23

8

France

Portugal Greece Sweden Iceland Denmark Australia Finland Austria Belgium/Lux Norway Italy Switzerland Japan UK

1 2 3 4 5 6 7

US Ireland Germany Spain Canada N. Zealand Holland

90–93

Country

89–93

Country

86–90

Paper Product Machines∗

Reaction Engines∗

Table 4 Country rankings by percent net shift in various time periodsa

28 29 30 31 32 33

86–90

Canada Spain France Japan Belgium/Lux Australia Portugal UK Greece N. Zealand Indonesia Turkey Denmark Holland US Ireland Germany Italy

Country

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

86–90

Nitro-Phosphate Fertilizers∗

Denmark Austria Holland UK Australia US

Country

… (cont’d) paper machines

14 7 10 16 3 4 12 15 5 2 8 9 6 11 1 17 13 18

88–92

4 5 16 33 30 6

88–92

8 14 15 10 17 18 9 6 5 2 12 16 11 7 1 13 4 3

90–93

11 23 22 2 7 1

90–93

188 N. Papadopoulos et al. / International Business Review 11 (2002) 165–192

a

0.8004a

1989

Significance: a: ⬍0.01b: ⬍0.05c: ⬍0.10

1986 1988 1990

Years

4. Import size

0.5620a

A

Period

A B A B A B C

1986–90

Years

1. Net shift % (diff. periods) 2. Import growth rate 3. Import growth rate & net shift % (same per.)

Measures

0.7945a

1990 0.8043a 0.8804a 0.8992a

1993

0.4733b

−0.0672

−0.2204

0.4298

−0.0227

−0.0198

b

C

1990–93

B

1989–93

Reaction Engines

Table 5 Correlations between country rankings in various time periodsa

0.9271a

1990

0.8249a

A

1986–90

0.8610a

1992

0.8867

a

0.3453b

0.13

B

1988–92

Paper Product Machines

0.7711a

0.6932a

1993

0.8366a

−0.5434a 0.3319c −0.3429c 0.2848

C

1990–93

1988

0.7358a

A

1986–90

0.8287a

1990

0.8989a

0.3478

0.1352

B

1988–92

Nitro-Phosphate Fert.

0.9835a 0.9794a

1992

0.8453a

0.162

−0.0898

C

1990–93

N. Papadopoulos et al. / International Business Review 11 (2002) 165–192 189

190

N. Papadopoulos et al. / International Business Review 11 (2002) 165–192

outcome. Lastly, further analysis focusing on outliers, or countries which grow or decline much faster than others in the set (in this case US, Japan, and UK for reaction engines, and US for paper machines), also showed that the rankings of such countries (Table 4) may also fluctuate greatly over time. Therefore, even export candidates identified via outlier analysis would still need to be subjected to further validation.

References Aksoy, S., & Kaynak, E. (1994). Export behavior of fresh produce marketers: Towards a co-ordination with general theory of exporting. International Marketing Review, 11(2), 16–32. Albaum, J. (1998). Global marketing. Boston: Allyn & Bacon. Alexandrides, C. G., & Moschis, G. P. (1977). Export marketing management. New York: Praeger. Anderson, E., & Gatignon, H. (1986). Model of foreign entry: Transaction cost analysis and propositions. Journal of International Business Studies, Fall, 1–25. Anscombe, F. J., & Tukey, J. W. (1963). The examination and analysis of residuals. Technometrics, 5, 141–160. Armstrong, J. S. (1970). An application of econometric models to international marketing. Journal of Marketing Research, VII(May), 190–198. Ayal, I., & Zif, J. (1978). Competitive market choice strategies in multinational marketing. Columbia Journal of World Business, 13(Fall), 72–81. Baalbaki, I. B., & Malhotra, N. K. (1993). Marketing management bases for international market segmentation: An alternate look at the standardization/customization debate. International Marketing Review, 10(1), 19–44. Bilkey, W. J. (1987). Profitable export marketing practices: An exploratory inquiry. In P. J. Rosson, & S. D. Reid (Eds.), Managing export entry and expansion (pp. 373–386). New York: Praeger. Brown, H. J. (1969). Shift and share projections of regional economic growth: An empirical test. Journal of Regional Science, 9(1), 1–18. Cahill, W. D., & Kristi, C. C. (1979). The shift-share technique of economic analysis: An annotated bibliography. Vance Bibliographies, Public Administration Series, B-P384 Chetty, S. K., & Hamilton, R. T. (1993). Firm-level determinants of export performance: A meta-analysis. International Marketing Review, 10(3), 26–34. Cooper, R. G., & Kleinschmidt, E. J. (1985). The impact of export strategy on export sales performance. Journal of International Studies, Spring, 37–55. Davidson, W. H. (1983). Market similarity and market selection: Implications for international marketing strategy. Journal of Business Research, 11, 439–456. Day, E., Fox, R. J., & Huszagh, S. M. (1988). Segmenting the global market for industrial goods: Issues and implications. International Marketing Review, 5(3), 14–27. Denis, J. E. (1978). La classification et l’evaluation des marches etrangers. Montreal: CETAI. Dichtl, E., & Koglmayr, H. G. (1987). Country risk evaluations and export market entry. In P. J. Rosson, & S. D. Reid (Eds.), Managing export entry and expansion: Concepts and practice (pp. 111–127). NY: Praeger. Douglas, S. P., Lemaire, P., & Wind, Y. (1972). Selection of global target markets: A decision-theoretic approach. In Proceedings of the XXIII ESOMAR Congress. Cannes, France: European Society for Opinion and Market Research. Douglas, S. P., & Craig, C. S. (1982). Information for international marketing decision. In I. Walter, & I. Murray (Eds.), Handbook of international business. New York: John Wiley. Douglas, S. P., & Craig, C. S. (1983). International marketing research. Englewood Cliffs, NJ: Prentice-Hall. Douglas, S. P., & Craig, C. S. (1992). Advances in international marketing. International Journal of Research in Marketing, 9(December), 291–318.

N. Papadopoulos et al. / International Business Review 11 (2002) 165–192

191

Ekeledo, I., & Sivakumar, K. (1998). Foreign market entry mode choice of service firms: A contingency perspective. Journal of the Academy of Marketing Science, 26(4), 274–292. Green, R. T., & Allaway, A. W. (1985). Identification of export opportunities: A shift-share approach. Journal of Marketing, 49(Winter), 83–88. Hellman, D. A. (1976). Shift-share models as predictive tools. Growth and Change, 7(July), 3–8. Helsen, K., Jedidi, K., & DeSarbo, W. S. (1993). A new approach to country segmentation utilizing multinational diffusion patterns. Journal of Marketing, 57(October), 60–71. Hoffman, J. (1997). A two stage model for the introduction of products into international markets. Journal of Global Marketing, 11(1), 65–86. Huff, D. L., & Sherr, L. A. (1967). Measure for determining differential growth rates of markets. Journal of Marketing Research, 4(November), 391–395. James, F. Jr, & Hughes, J. (1975). A test of shift and share analysis as a predictive device. Journal of Regional Science, 15(2), 223–231. Johansson, J. (1999). Global marketing management. New Jersey: Prentice-Hall. Kerin, R. A., Mahajan, V., & Peterson, R. A. (1980). Shift-share analysis as a diagnostic tool for multiproduct performance appraisal. Proceedings, American Inst. of Decision Science, 285–287. Kumar, V., Stam, A., & Joachimsthaler, E. A. (1994). An interactive multicriteria approach to identifying potential foreign markets. Journal of International Marketing, 2(1), 29–52. Liander, B., Terpstra, V., Yoshino, M. Y., & Sherbini, A. A. (1967). Comparative analysis for international marketing. Boston: Allyn and Bacon. Linder, S. B. (1961). An essay on trade and transformation. New York: Wiley. Moyer, R. (1968). International market analysis. Journal of Marketing Research, 3, 353–360. Nachum, L. (1994). The choice of variables for segmentation of the international market. International Marketing Review, 11(3), 54–67. Papadopoulos, N. (1987). Approaches to international market selection for small and medium sized enterprises. In P. J. Rosson, & S. Reid (Eds.), Managing export entry and expansion: Concepts and practice (pp. 128–153). New York: Praeger. Papadopoulos, N. (1999). Product-country images. In M. Baker (Ed.), The encyclopedia of marketing (2nd ed.) (pp. 719–731). London, UK: International Thomson Business Press. Papadopoulos, N., & Denis, J. E. (1988). Inventory taxonomy and assessment of methods for international market selection. International Marketing Review, Autumn, 47–60. Papadopoulos, N. D., & Jansen, D. (1993). Target countries for export. Ottawa, Canada: Dimensions of Int. Business International Business Study Group, 11, 31–52. Reid, S. (1981). The decision-maker and export entry and expansion. Journal of International Business Studies, Fall, 101–112. Root, F. R. (1994). Entry strategies for international markets. Lexington, MA: Lexington Books. Samli, A. C. (1977). An approach for estimating market potential in East Europe. Journal of International Studies, Fall/Winter, 49–53. Seringhaus, F. H. R., & Rosson, P. J. (1991). Export development and promotion: The role of public organizations. Boston, MA: Kluwer Academic Publishers. Sethi, S. P. (1971). Comparative cluster analysis for world markets. Journal of Marketing Research, VIII(August), 348–354. Singh, B., & Kumar, R. C. (1971). The relative income hypothesis—a cross country analysis. The Review of Income and Wealth, December, 341–352. Stevens, J. (1992). Applied multivariate statistics for the social sciences (2nd ed.). Mahwah, NJ: Lawrence Erlbaum Associates. UNCTAD/GATT (1968), The Compilation of Basic Information on Export Markets, Geneva Vernon, R. (1966). International investment and international trade in the product cycle. Quarterly Journal of Economics, 80, 190–207. Vitali, G. (1990). Determinants of international trade of machine tools during the eighties: A shift-share analysis. Technovation, 10(8), 507–519. Vogel, R. H. (1976). Uses of managerial perceptions in clustering countries. Journal of International Business Studies, Spring, 91–99.

192

N. Papadopoulos et al. / International Business Review 11 (2002) 165–192

Welch, L. S., & Wiedersheim-Paul, F. (1980). Initial exports—A marketing failure? Journal of Management Studies, 17(October), 333–344. Williams, R. B. (1980). New tests of shift-share techniques. Growth and Change, April, 19–25. Wind, Y. (1978). Issues and advances in segmentation research. Journal of Marketing Research, XV(August), 317–337. Yandle, B. (1978). Identifying brand performance by shift-share analysis. Journal of the Academy of Marketing Sciences, 6(Winter), 126–137. Zellner, A. (1962). An efficient method of estimating seemingly unrelated regression and tests for aggregation bias. Journal of the American Statistical Association, 57, 348–368.