The impact of retail format diversification on retailers’ financial performance

Abstract

For retailers, format portfolio management is a core marketing operation, but has received little attention in the marketing literature. This study analyzes the relationship between format diversification and retailer performance in a global setting, where retailers as part of their geographic expansion process often employ format diversification. The dual strategies of geographic diversification and format diversification substantially complicate the diversification-performance relationship. Using a six year panel data set for leading global retailers, we find a positive impact for geographic diversification, a negative impact for format diversification and a negative interaction for the dual strategies, supporting a single focus diversification strategy. We further show the consistency of our findings using a series of model robustness checks.

Appendix

Calculation of metrics

Cultural, economic, and physical distances Many different cultural dissimilarity measures have been proposed in the academic literature. We employ a traditional yet widely used approach. Our measure of cultural dissimilarity is based on the four cultural dimensions developed by Hofstede (1980): individualism, masculinity, power distance, and uncertainty avoidance. To calculate cultural dissimilarity, we created a composite index, which is a variant of the index used by Kogut and Singh (1988). We first divided the country rating for each of the four Hofstede dimensions by its variance. We then calculated the average dimension dissimilarity between all countries for each retailer’s portfolio. Next, we took the average of the squared average dimension dissimilarity across the four dimensions. We substituted missing values with the historical mean. If data were missing on all four cultural dimensions and there was no historical mean value to substitute, we dropped the observation. However, such cases requiring a complete drop were rare (less than 10 cases). Thus, this index is:

$$ \mathrm{Cultural}\ \mathrm{dissimilarity}=\sum_i\left({\left(\underset{k<j}{\varSigma \varSigma}\left({H}_{i,k}-{H}_{i,j}\right)/n\left(n-1\right)/2\right)}^2/{\sigma}_i^2\right)/4, $$

where:

H _i,j :

denotes the value of the i ^th Hofstede dimension for the j ^th country;

i :

1… 4 denotes the i ^th original Hofstede dimension;

k, j :

1… n indexes the country pairs in which the firm operates

n :

denotes the number of countries in the firm’s portfolio for that particular year.

An economic dissimilarity index was developed using a similar approach to that employed for the cultural dissimilarity index, but using factors assessing the economic climate and technological development of the countries rather than Hofstede’s culture scores. The factor scores for the four dimensions (development status, infrastructure technological advancement and efficiency, international orientation, and market potential) were based on an exploratory factor analysis of 62 economic indicators from the World Bank report (2007).

Calculation of Google Trends brand equity measure

Google Trends has been utilized in the academic literature to analyze product trends by Du and Kamakura (2012). Google Trends provides an index of search volume over time for a given search term. It also allows for comparisons of multiple search terms, giving relative search volume (e.g., 54 vs. 65). However, in this previous work, Google Trends was used to gather longitudinal trends for a single product. Google Trends does not provide absolute value (e.g., number of actual searches) comparable across multiple Google Trends searches. In addition, Google Trends has low granularity, only reporting integer values between 1 and 100, so a comparison between a company with ultra-high global visibility, such as Amazon.com against a small domestic retailer would always give 100 for Amazon and 1 for the retailer. To get around these problems, the following procedure was used.

• i) Google search terms were gathered. For over 95% of the companies, Google had a search code specific to the company that would filter out non-relevant results. Where this was not available (a few small, defunct companies), the company search name was used in conjunction with the type of store.

• ii) Given that Google throttles its Trends service, it would have been infeasible to compare each retailer with every other retailer. A set of 10 comparison retailers was selected from the overall list of retailers. The retailers were selected to be evenly spaced in terms of search volume, from Walmart at the top, to a small Wisconsin retailer (Roundy’s at the bottom), so that each retailer would have at least one non-dominating (e.g., not 100 to 1) comparison.

• iii) A JavaScript program was written to grab the Google Trends data from 2004 to 2015 for sets of two retailers. The program was run for each combination of retailer and comparison retailer. The difference between the retailer and the comparison retailer was recorded for each time point.⁵

• iv) The difference data were aggregated by year for each combination of retailer and comparison. An overall index was calculated for each retailer by aggregating across the 10 comparisons. The retailers were then placed in rank order.

The overall index had strong face validity. Dell had the highest brand equity value from 2000 to 2009. In 2010 it was Walmart and then subsequently from 2011 to 2015, it was Amazon.com.