Discussion, limitations and future research

47

48 variable. The second noteworthy finding includes the effect of global cities on ROS which was found to be negative when no control variables were included. To reduce potential bias, this cannot be concluded with certainty but it is worth investigating this in future research. The explanation of the non-significant relationship for the first hypothesis lays in the performance literature. Numerous researchers have tackled the effect of internal and external factors on performance, for example the relationship of the level or speed of internationalization, cultural distance, geographical distance, governance distance, diversification strategy, or ownership mode on performance (e.g. Hennart, 2007; Belkaoui, 1996; Hutzchenreuter et al., 2014; Chan

& Rhee, 2011). The large scope of topics that have an influence on performance might suggest that performance cannot be researched in isolation as is influenced by multiple factors. A second argument which could explain the non-significant hypothesis lays in the nature of firm performance. Firm performance can be both financial as strategic performance (Selvan et al., 2016). It could be that being located in a global city will not enhance firm financial performance, but will increase strategic performance such as improved reputation or diversification performance. Future research should investigate whether strategic improvements could be a consequence of global city locations.

Another argument why this hypothesis might not be supported is based on the limitations of the study. First of all, the dataset was based on the top fifth-teen retailers worldwide. This should have resulted in a dataset of 4009 subsidiaries. However, after the data was retrieved from Orbis only 418 subsidiaries remained. Moreover, the top fifth-teen retailers included a majority of North-America based companies, such as Walmart and Amazon.

However, in the final data set the parent companies originated mostly from Europe. As retailers often focus on a regionalized approach, this might have decreased the overall measured performance as the United States based retailers account for the largest, best-performing retailers in the world (Deloitte, 2021). Moreover, retailers in the United States can enjoy economies of scale, which can positively influence eventual performance (Hennart, 2007).

Future research should focus on selecting a more representative sample to test whether the results are similar when more North-American companies would have been included, and to ensure a representative sample, even in terms of size. Another limitation that explains the non-significant result is the use of a cross-sectional design. To calculate the average return on assets and return on sales, the average of 2019, 2018 and 2017 was computed. According to the World Bank, the economy differs per year. Whereas 2018 and 2017 show somewhat similar patterns, 2019 shows a sharp decrease probably due to the Covid-19 pandemic which hit China early on

49 (World Bank, 2021). This may have affected the computed average number, especially for Asian-based companies. Future research should therefore not use a cross-sectional design, but perform a longitudinal research when investigating the effect of being located in a global city on firm financial performance.

The second hypothesis tested was the moderating effect of being located in a global city within the home region vs outside the home region. It was expected that the benefits of global cities would have less impact on firm financial performance when subsidiaries were located within the home region of the MNE as the multinational had more experience in the market already. By performing a preliminary test, it was confirmed that global city locations were indeed chosen less when the subsidiary was located within the home region. The literature explains this by the lower levels of LOF these retailers face as they already have substantial knowledge about the market (Kudina, 2012). Therefore, these retailers might not need the benefits of global cities. The second hypothesis aimed to explain performance differences within or outside the home region. This hypothesis was not supported, when subsidiaries are located in a global city vs a non-global city within their home region no significant moderation was found on its relationship with performance. However, evidence was found that being located outside the home-region negatively moderates the relationship between global city locations and firm financial performance only when measured by ROA. This can be explained by high prices of assets and rental prices within global cities and the high investments which need to be made to deal with competition which can all reduce the ROA of a company (Statista, 2019; Picone et al., 2008). Moreover, the necessity of being located within a global city might be higher for companies that internationalize outside their home region due to the interconnectedness, the APS businesses, and cosmopolitan character which help them navigate through the new business environment (Goerzen et al., 2013). Therefore, these companies need to face these lower ROA rates. Future research should take the effects of being located in a global city within or outside the home region into consideration as it seems that the advantages of global cities are not equally important between regions. A study that solely focuses on global cities vs non-global cities outside the home region might give an extra explanation about performance. Furthermore, it might be interesting to research the difference in significance of the moderation effect on ROA and ROS. In addition, it should be noted that the moderation effect was tested for a relatively small sample (90 for ROA, 87 for ROS) as the PROCESS function of Andrew Hayes uses a pairwise exclusion to test for moderation. A smaller sample size decreases the power of a study (Raudys & Jain, 1991), which is the reason it is seen as a

50 limitation for this hypothesis specifically. Future research should work with a sample of at least 267 to test moderation to increase the representativity of the study.

The last hypothesis tested the moderating effect of being a food vs non-food company as it was expected that the global city location would help to eliminate the discrimination effect of LOF against food companies. This hypothesis was rejected as not enough evidence was found for the relationship. The results of the test did show that being a non-food company positively moderates the relationship between global city locations and performance. This can be an effect of higher sales or lower costs for the non-food retailer. First, the higher sales.

Retailers often internationalize due to market-seeing purposes (Charavarty et al., 2021). Global cities might help to gain market and legitimacy in a new country (Goerzen et al., 2013).

Therefore, retailers might choose global city locations even though it will decrease their initial performance. This can also be the case for a non-food retailer, which might benefit from the increased market despite facing high costs of a global city which would explain the moderation effect. The moderation effect does not take place for food retailers, which might be explained by the cultural entity of food. This might have resulted in a lower increase in market share and therefore in sales. Second, non-food retailers might face lower costs. It can for example be the case that non-food retailers have less diversification costs in global cities as the customer population has a cosmopolitan nature (Goerzen et al., 2013). This causes costs to drop, but initially might not increase the sales which would explain the moderation effect. For food retailers, diversification costs are initially higher (Hart, 2016) so costs might not drop when locating in a global city which explains why no moderation effect is taking place. It is suggested that future research focuses on the difference between food and non-food retailers when examining global cities. The differences are notable and might explain more about which advantages generate a higher performance for which retailer. Furthermore, an emphasis should be placed on sales and costs to distinguish which element causes the eventual performance. In addition, similar to the previous moderation, it should be noted that the moderation effect was tested for a relatively small sample (90 for ROA, 87 for ROS) as the PROCESS function of Andrew Hayes uses a pairwise exclusion to test for moderation. It is recommended to test this hypothesis again, using a sample of at least 267 to increase the representativity and the power of the study.

51