Размещено на http://www.allbest.ru/

INTRODUCTION

As a result of tense geopolitical situation in Russia, the years 2014 and 2015 witnessed a confluence of economic challenges: declining real disposable income, rising consumer price inflation, expensive and hardly accessible external financing, etc. Given retail is the third leading contributor to GDP, the issue of maintaining its efficiency has been generating considerable interest in the last two years.

However, the way economic headwinds affected SMEs and market leaders is different: whereas the first mostly suffer from financial constraints, the second face intense competition for customers and high pressure on prices. Since major players are more capable of executing turnaround strategies to maintain competitiveness, these act as a better target audience for the purpose of current research. Thus, the paper is for large-sized Russian retailers to get an answer on which business strategies towards enhancing sales and getting closer to customers can help create most margins.

Despite this interest, no one to the best of our knowledge has conducted such a research for the Russian retail market yet. As for the foreign scientific literature, most of the studies estimate efficiency of a single retail chain or examine the impact of a sole efficiency driver, and almost never adhere to a complex approach. In order to fill both gaps, our research is to consider 5 business strategies, best reflecting the specifics of retail and most relevant to the current economic situation in Russia. The aim of the research is thus to reveal these drivers of efficiency and determine the one of highest impact. The object of the study is represented by large-sized Russian retailers, while the subject - by their efficiency.

For the purpose of achieving the goal, several consistent stages needed to be taken. First, in order to confirm the relevance of the issues under investigation, the current state of the Russian retail market was investigated. Then, it was of a great use to study the existing scientific literature on the similar topic to determine the gap and compare different approaches. Once the preparation stage was complete, the process of selecting variables and collecting the data began. Having then identified the best-fitting regression model, the Shapley Value Decomposition was employed to identify the strategy of highest contribution to the explanation of efficiency. Finally, we interpreted the results and proposed corresponding recommendations for Russian retailers on how they can better adapt to the changes in the marketplace.

The methodical base of the current research is predominantly constructed from foreign case studies, which attempted to assess the drivers of retail efficiency both from micro- and macro-perspectives for a variety of different retail segments. The most useful were the ones of Daur V., Fisher M. & Raman A. (1999); Sellers-Rubio R. & Mas-Ruiz F. (2007); Zhao Y., Trejo-Pech C.O. & Weldon R.N. (2013). From the theoretical point of view, a deep insight into the specifics of retail activities has been obtained from the book `Retail Strategy: the view from the bridge' by Reynolds J. and Cuthbertson C. (2004), while the `Microeconometrics Using Stata' by Cameron A. and Trivedi P. was of extreme use when constructing a panel regression model.

The ultimate result of this particular research is a formulated list of retail efficiency drivers (i.e. business strategies) ranged according to their contribution to efficiency. The results obtained might have important managerial implications for Russian retailers aspiring to enhance their margin after the recession of 2015. The paper is organized as follows: introduction, literature review, methodology, results, and conclusion.

1. LITERATURE REVIEW ON ESTIMATING RETAILERS' EFFICIENCY

This part of the project is meant to examine theoretical studies in the field of firms' performance measurement in relation to retail. The subject is provided by a large body of literature, however in this paper light is going to be shed on the major ones. Two conceptually different perspectives - productivity and efficiency - have been so far proposed by scientific communities, thus ensuring a variety of methodological frameworks of addressing the issue: ratio analysis (RA), data envelopment analysis (DEA), and stochastic frontier analysis (SFA). Though major work on the topic has been done at the initial stages of performance theory development in the early 1990s, productivity and efficiency terms have first been differentiated only at the turn of the third millennium.

1.1 Productivity VS Efficiency

Despite having a similar nature, the productivity and efficiency terms differ in both perspective and scope. While productivity is related to quantitative achievements (growth in production or sales volume, profits, etc.), efficiency approach deals with qualitative improvements by taking into account optimal utilization of resources. Thus, a business is performing effectively if it managed to increase its output, and efficiently - if it either created more output with the same amount of input or created the same amount of output with less input (Sellers-Rubio & Mas-Ruiz, 2006). This terminology specification resulted in two approaches to performance measurement: conventional and modern ones.

The conventional approach suggests measuring firms' productivity with single output to input ratios. Given a labor-intensive character of retail industry, the vast majority of initial and following studies estimated productivity in retail mainly by the labor factor (Waldorf, 1966; Nooteboom, 1982; Ratchford & Stoops, 1988; Goldman, 1992). However, this idea was later subjected to much criticism on the grounds that labor is not the only meaningful resource used in retail (Good, 1984) as well as that not all the firms' employees are directly involved in the sales process (Lusch & Moon, 1984). The first remark represents the major limitation of the whole conventional approach: when measuring such a complex phenomenon as firms' performance it allows considering no more than one output or input. Profit ratios (ROE, ROA, ROI, etc.), probably the most commonly employed productivity metrics in the recent years, represent the same problem (Sellers-Rubio & Mas-Ruiz, 2007). Another serious drawback of conventional approach is that the productivity value of a separately taken firm cannot be fairly compared to that of its competitors, if only to the sample average (Shockley & Turner, 2014). The modern approach based on the efficiency term is considered to overcome both aforementioned limitations.

The modern approach suggests estimating how rationally a firm uses its resources, as compared to its full potential. This way the efficiency term is based on the concept of production possibility frontier (PPF) and is closely tied to the economic feature of opportunity cost (De Jorge, 2008). In order to determine whether a firm lies on the PPF or beneath it, the modern approach requires application of non-parametric (DEA) and parametric (SFA) techniques (Sarantopoulos, Kioses, & Doukidis, 2007). Having entered the information on a set of outputs and inputs for different firms into special software programs, one can obtain a vector of values ranging from 0 to 1 indicating the corresponding relative efficiency degrees. By thus fully addressing the problems of the traditional approach (multiple factors involved and possible pairwise comparison), the modern one has all the prerequisites for becoming the most demanded technique among practitioners. However, Sellers-Rubio and Mas-Ruiz (2007) note that even large publicly quoted retailers tend to avoid technically complex approaches to performance measurement in favor of regular financial or operating ratios that are more meaningful to investors and stakeholders.

Overall, the literature review of existing approaches to performance measurement indicated there is no full agreement on superiority of one over another from practical and theoretical perspectives. While in the foreign scientific literature there are a surprising number of studies using both productivity and efficiency concepts towards retail performance assessment, for Russia, to the best of our knowledge, this is still not the case. In order to gather foreign experience, some major research papers and their results are going to be discussed later in this section.

1.2 Retail Productivity Assessment

One of the earliest systematic studies on assessing retailers' productivity was conducted in 2000 by Gleason and Mathur, research associates from Illinois, USA. As the main purpose of the paper is to establish interrelation between capital structure and firms' performance with respect to cultural distinctions, the data included information on 198 retailers representing 14 European countries. The cross-cultural specifics of the study demanded the productivity measures to be both financial (return on assets and pretax profit margin) and operational (sales per employee and sales growth). Having first built four corresponding regression models with only capital structure (total debt to total assets ratio) as an independent variable, the authors obtain a surprising result of it positively effecting ROA values. Given its small magnitude, it is fair to assume that European companies fall under the logic of Jensen (1989), who claimed that in some rare cases high leverage might improve firms' performance by forcing management to either pursue value maximizing strategies or behave carefully in order to avoid debt pressure. Then, in order to address cultural distinctions, the authors split retailers into 4 cultural clusters. The interaction terms between cultural clusters and capital structure being included into the model indicate that the effect of the latter one on retailers' performance differs across cultures. From all the control variables used in the model, only firm's size (logarithm of sales) appeared to be significantly positive. Overall, the article provides an exceptional case of leverage positively effecting firms' performance and confirms the importance of conducting a separate research for Russian retail due to the cultural discrepancies' bias. On the other hand, this study may be considered too narrow in the context of this particular research due to only two explanatory variables being used.

A more recent study carried out by Kaya in 2007 aims at comparing the drivers of American retailers' and wholesalers' economic performances over the period from 2000 to 2005. With respect to retail, the author obtains the results consistent with the main findings of Gleason and Mathur: both size and leverage variables appeared to be significant at 1% significance level with both positively effecting the firms' performance (ROA, ROE). As a significant contribution to the topic under consideration Kaya suggests including into the model the share of tangible assets as a proxy of investment decisions adequacy. In other words, since retailers within their primary activity deal mostly with non-tangible assets, and even the trade rooms are usually rented, considerable amount of material possessions may be associated with poor corporate governance. The statistical results of the research confirm the idea of significant inverse relationship between the newly made tangible variable and all versions of performance indicators.

Despite the outstanding quality of Kaya's findings, several researchers (Ahmad, Abdullah, & Roslan, 2012) have proposed a way to specify them by addressing the impact of short- and long-term debt separately from each other. In their own paper authors estimate the performance of 58 Malaysian retail companies over a 5-year period (from 2005 to 2010). Being measured by ROA, productivity is regressed on the short- and long-term debt, size, assets growth, and sales growth. Opposite to what was expected, the short- and long-term debts have a unidirectional impact on retailers' performance. The significantly negative coefficients on corresponding variables reflect relatively high cost of capital for Malaysia as a developing country as well as managers' unreadiness to perform efficiently under debt pressure. With regard to the size variable, it turned out to be insignificant, which has never occurred in the previous studies. The reason behind this is the size variable being incorrectly measured by the logarithm of assets instead of sales. This phenomenon was fully addressed by a group of researches from UK and USA back in 1999.

In their seminal paper, Gaur, Fisher, and Raman (1999) analyze the performance of 293 public-listed retailers in USA with the long-term stock returns as a measure of this performance. Though this view of productivity is not of our focus, the authors provide a valuable insight into the intra-sectoral differences with respect to retail. The first important conclusion concerns the absence of any variation between retail segments in leverage and sales growth values, which is explained by them being entirely under the management control. The second crucial merit of this work is an attempt to assess the Gross Margin Return on Invested Inventories (GMRoII), which is of a great importance for inventory-intensive retail companies. The authors conclude on its systematical variation between retail segments explained by the following logic: the slower inventory turns, the higher gross margin. For instance, while jewelry or fashion apparel retail can afford slow inventory turns, the grocery and computer retail are forced to quickly realize the products due to their perishability and obsolescence, respectively. This different relationship between performance and inventory turns for different retail segments requires the inclusion of corresponding interaction terms.

Zhao, Trejo-Pech, Weldon (2013) have extended the previous study by simultaneously taking into account both cross-sectional and time-series effects. This study examines the performance of more than 250 companies across 13 different sectors of economy for the 1986 - 2008 time period. The major contribution of the paper is a thorough explanation of the selection criteria when choosing independent variables. The authors propose applying the DuPont extended decomposition to derive financial indicators affecting profitability estimated by ROI. As a result, a set of indicators, specifically gross margin, interest expenses, SG&A expenses, R&D expenses, accounts receivable, inventory turn, PP&E expenses, and short- and long-term debt, is considered as the best proxy for ROI estimation. Having conducted a two-way panel regression analysis for the industry that consists of three sectors (food wholesalers, retailers, and food service), the researchers revealed no intra-industrial variation in the results. Thus, the findings for the whole industry hold for each separately taken segment, including retail which is of our main interest. According to the results of the regression analysis, only two variables - interest expenses to sales and long-term debt to assets - manage to explain variation in ROI. Consistent with Ahmad, Abdullah, and Roslan, interest expenses and the amount of debt significantly negatively influence the retailers' performance.

While most of the existing studies on the assessment of retailers' effectiveness prefer standard ROA and ROE as productivity measures, with respect to explanatory variables such a full coherence has not yet been established. Nevertheless, the most common ideas covered in the literature are short- and long-term debt burden, assets tangibility, inventory turns, sales and assets growth, different types of expenses, and firm's size. Although the effect of these metrics on retail performance differed for some studies, it can be justified by the founding of American scientists on the bias arising from cultural distinctions. This contributes a lot to the necessity of carrying a separate research for the Russian retail market. The most recent works provided statistically significant evidence for both intra-sectoral and intertemporal variation in financial performance of retail market participants, which can be also checked within the current paper.

1.3 Retail Efficiency Assessment

The efficiency perspective on retailers' performance has gained a lot of attention from academicians in the last decade. The methodological base of this modern approach employs both parametric (SFA) and non-parametric (DEA) techniques. The major difference between them is that, unlike DEA, stochastic frontier analysis requires specification of a functional relationship between the inputs used and the outputs obtained. However, Gong and Sickles (1992) assure that neither technique uniformly dominates the other. Though both methods have been criticized for making inferences about individual observations rather than central tendency, they allow considering multiple inputs and outputs thus providing better efficiency estimates compared to traditional productivity ratios. Having been applied to a variety of industries, the aforementioned methodologies have overwhelmed current scientific literature. The following paragraphs are supposed to present best practices of SFA and DEA application in relation to retail.

1.3.1 Stochastic Frontier Analysis

Stochastic frontier analysis as a non-parametric method of economic modeling has been simultaneously introduced by two groups of researches - Aigner, Lovell, and Schmidt; Meeusen and Van den Broeck - back in 1977. The stochastic element of the proposed model helps to assess either maximum output that can be obtained under fixed inputs (profit efficiency) or minimum costs required to produce fixed output (cost efficiency). Nevertheless, the realization of this tempting idea requires making assumptions on the type of interrelationship between the inputs and outputs, which is almost practically impossible (Sellers-Rubio & Mas-Ruiz, 2007). Therefore, the SFA technique substantially concedes the DEA one on the amount of studies, where it is employed.

One of the first and most successful studies on assessing retailers' efficiency using SFA considers 47 retail outlets of Portugal's leading hypermarket chain (Barros, 2005). This research estimates a stochastic generalized Cobb-Douglas cost function with two outputs (sales and earnings), two inputs (labor and capital prices), contextual variables outside (population density within 5 minutes from the outlet, surface area of other competing outlets within 10 minutes and the index of purchasing power per capita) and under (share of temporary workers in the total work-force and stuff absenteeism) management control. Overall, the results of the regression model built on the vector of efficiency indicate that the chosen Cobb-Douglas function fits the data well: right model specification, high value of R² and satisfactory F-statistic, significant coefficients on all the included into the model variables. However, the authors consider such a successfully determined functional relationship as a lucky accidence that is almost impossible to achieve otherwise. As a result, the population density, competing outlets, purchasing power, and the share of temporary workers turned out to be the best drivers of operational costs' minimization. The only limitation of this early study is its too narrow research perspective, as it considers only one retail distribution chain and not the whole retail market.

A more recent study of a bigger scale analyses profit efficiency of 24 retailers in South Africa over the period 2005-2006 (Akinboade, Mokwena, & Kinfack, 2008), thus proving the applicability of SFA on the macro level. Unlike the previous study, this one estimates the production frontier function expressed by taxable profit (EBT). Having made an assumption on functional relationship between inputs and output, the authors employ a log-linear regression model on EBT with the following set of explanatory variables: sales revenue, gross dividend, gross interest, other income, total expenses, interest expenses, current assets, and gearing level. As a result, all the variables appeared to be significant at any reasonable significance level (all the expenses have a negative sign, while others - positive), except for the leverage. Following this, the vectors of efficiency calculated for 2 years in question indicated that these are stable, which speaks in favor of the quality of chosen explanatory variables. However, one can argue that this stability might owe to the market specifics: South Africa is slightly subjected to crisis and monopolistic phenomenon. In addition, the results may not be representative due to relatively small sample size and short time horizon.

Having analyzed the abovementioned studies, one can recapitulate on the applicability of stochastic frontier approach for assessing the efficiency of retailers' performance. The technique can estimate both cost and production frontiers with no diminution in the explanatory power of the corresponding regression models. However, the results' accuracy relies strongly on the quality of assumed functional form of inputs and outputs, which is challenging to select in the real life. Thus, the complexity of SFA methodology can be confirmed by relatively small amount of literature compared to that for the DEA approach, which enables to achieve comparable results with less efforts employed.

1.3.2 Data Envelopment Analysis

Data Envelopment Analysis is a non-parametric method of the DMUs' relative performance estimation, characterized by multiple inputs and outputs. Since it has first been proposed by Charnes, Cooper, and Rhodes in 1978, the DEA technique has been successfully applied as a performance evaluation tool in a variety of fields including manufacturing, education, health care, banking, insurance etc. Such popularity owes a lot to the relative simplicity of DEA application, which does not require the knowledge of the functional relationship between inputs and outputs. Nonetheless, its main disadvantage hides in the interpretation of obtained efficiency estimators. The DEA only allows comparing each object with its analogs: a DMU is considered efficient if it produces the highest output among firms with the same amount of input. Thus, it is only possible to estimate the relative efficiency of each particular firm, but never its potential output.

One of the most comprehensive works on DEA application with relation to retail was conducted in 2010 by Indian researchers Gupta and Mittal. The general data set includes 43 grocery retailers in Deli. However, the study represents not as practical as theoretical value. The authors agree with the founders of DEA methodology (Charnes, Cooper, & Rhodes, 1978) on the importance of precisely chosen sets of inputs and outputs. According to them, the selection criteria should be as follows: literature review, managers' subjective opinions and data availability. Thus, the authors recommend using sales, gross profit, number of customers, customers' satisfaction, customer conversation ratio, etc. as outputs and number of employee, information systems, distribution centers, inventory volume, etc. as inputs. These clearly reflect the specifics of retail industry by taking into account its inventory- and labor-oriented nature. However, the data on the majority of proposed inputs-outputs is hardly obtainable for it requires access to corporate information (customer conversation ratio or customers' satisfaction).

The overview of some other relevant studies for various retail segments - apparel, grocery, pharmacy, and e-retail (Xavier, Moutinho, & Moreira, 2015; Barros & Alves, 2003; Patel & Pande, 2013; Lu & Hung, 2011) - indicated that sales is the most commonly used indicator of output; while operational costs, capital, the number of sales employees, floor space, and the number of POS appeared to dominate among inputs. Thus, a deeper analysis of existing literature enables us to conclude that there is no actual need in corporate information when collecting data on the most commonly applied metrics. In addition, the relative homogeneity of inputs preferences by different retail segments is so far the most promising finding, as it means we can avoid segmentation when conducting this particular research. Overall, these specified outputs and inputs turn into the efficiency values after having applied the DEA technique.

Some academicians have recently moved even further and provided studies, where they regress the efficiency terms obtained on some meaningful for retail factors. One of these researches first employs DEA to estimate the efficiency of 40 outlets of a Portuguese apparel retailer (Xavier, Moutinho, & Moreira, 2015). Once the technical efficiency values for each outlet are obtained, they are regressed on the following independent variables: the purchasing parity index per capita, the population density, the number of employees, the seniority of the staff, and location (a shopping mall or a street one). The results indicate that the number of employees and the location type appeared to affect the store efficiency most. These findings are consistent with the ones of Patel and Pande (2013), who applied the analogous procedure to 46 outlets of an Indian pharmacy retailer. Out of all the range of explanatory variables, again, only two appeared to be significant: the number of employees and the location type. Although both studies consider a retail chain instead of the whole market (which is more of our interest), they at least make an attempt to reveal possible efficiency drivers.

The literature review on the DEA technique application for estimating efficiency in retail indicated its high popularity due to the significant advantages it has over the stochastic frontier analysis. However, some academicians argue that no functional relationship between inputs and outputs is more of DEA's limitation, rather than an advantage. Thus, it was followed by a new wave of works, in which researches made their attempts to compare the results of DEA and SFA. In 1992, Gong and Sickles managed to provide evidence that these can be used interchangeably without lowering the quality of the research. Their Spanish colleagues (Sellers, & Mas, 2007) developed their idea further by simultaneously applying traditional productivity measures (profit ratios) and modern efficiency techniques (DEA and SFA) to estimate efficiency of 491 retailers operating in Spain in 2004. Once the corresponding productivity and efficiency values have been derived, Pearson and Spearman correlation indexes were calculated to test the degree of agreement between the results provided by different methods. Overall, as a main conclusion the authors claim that none of the methodologies is better than the others; hence, the most appropriate methodology depends on the characteristics of the production process and the aim and scope of the analysis. Therefore, taking into account statistical interchangeability of the three techniques discussed in this section (ratio analysis, stochastic frontier analysis, and data envelopment analysis) as well as the technical complexity associated with the two latter ones, in this particular research we decided to adhere to the traditional approach.

2. METHODOLOGY OF ASSESSING EFFICIENCY IN RETAIL

Consistent with all the previously mentioned studies, this particular research employs a multiple regression analysis in order to identify major drivers of retail efficiency in Russia. Though the idea is not groundbreaking, the way it is addressed within this work distinguishes it from many others. The key common weakness with all the existing studies is that they, to the best of our knowledge, examine the impact of a sole efficiency indicator (usually leverage) supported by a number of control variables. Thus, this particular research proposes a more complex and retail-oriented approach: the independent variables best reflecting the specifics of retail industry represent major and most relevant strategies of improving retail efficiency (organic growth, inorganic growth (M&A), operational performance improvement, private labels, multichannel selling, and franchising). In addition, such an approach has never been applied to the Russian retail market yet. Both data management and regression analyses were performed via statistical software STATA.

2.1 Data and Sample

The initial sample was obtained using the Interfax SPARK database and included financial information on 703 large-sized Russian retailers over the period 2010-2015. The sample was then reduced to 485 to include only those firms that are operating in one of the major retail sectors: FMCG, electronics and home appliances, DIY, household, drogery, fashion, children's goods, and art. For our empirical analysis we excluded the pharmacy sector due to its clear seasonal sales pattern. Finally, the firms with missing observations for more than three years have been discarded, which resulted in 316 entities within the sample.

Since the next step was to collect specific operational information (number of outlets, sales square, private labels share in total sales, etc.), as a final selection criterion we needed the firm's general information to be available on the Internet (whether on official website or in media). After eliminating entities for which all the necessary information was not available, the dataset narrowed to 90 large-sized Russian retailers or 540 firm-year observations over a 6-year period.

2.2 Indicator of Retailers' Efficiency

The literature review indicated that the difference in the results obtained by traditional and modern approaches to evaluating firms' performance is not statistically significant. Thus, the decision was made in favor of the first one due to its overall simplicity and universality.

Out of all return ratios we opted for ROS. The first and foremost reason is that retail is a mostly cash-generative industry. Given the relative scarcity of other sources to enhance financial returns, expanding sales is the singular most important driver of financial performance in retail. A manufacturing company, for example, may use its product R&D base to augment sales and profits via a range of new products. There is no product per se to develop in retail, and gains to be potentially reaped from innovation come mainly from more skilful changes in the composition of stores, formats, marketing, etc. Thus, ROS is one of the most crucial figures to track in retail. In addition, we could not use ROIC or EVA as dependent variables for only 8% of the retailers in the sample is public.

The general formula for calculating return on sales is net income to sales ratio. However, since it is a common practice that retailers manipulate both tax and depreciation charges, we opted for EBITDA as a proxy for net earnings. EDITDA margin is thus considered to reflect those factors that are dependent of operational efficiency only and eliminate any bias from managerial fraud. Thus, the dependent variable is a result of dividing EBITDA by the sales revenue for each retail company. According to the descriptive statistics, the average EBITDA margin within the sample equals 1.6% with standard deviation of only 7%.

The following paragraphs are to explain the rationale behind choosing both control and explanatory variables.

2.3 Control Variables

Prior to considering the explanatory variables essential for estimating retailers' efficiency, it is necessary to determine the control ones. Though control variables are not of our primary interest, these are crucial for estimators' consistency. According to the methodological base of the previous studies, the control variables for retail as probably for any other sector of economy should reflect firms' size, experience, liquidity, and leverage (Tab. 1).

It is commonly recommended to include either logarithm of sales or assets as a proxy for firm's size. In this particular research, we opted for the logarithm of assets, as sales is already used in the calculation of EBITDA margin. To control for the firm's seniority and experience we use the age of the retailer, measured by the time span between 2015 and the year of foundation of the firm. Both size and age variables are expected to have a positive effect on the firm's efficiency due to greater capacities and more experienced management.

By analogy with previous studies, we also include current ratio and leverage as these have an apparent effect on the dependent variable, which needs to be removed from the equation. The decision was made to compute leverage as total debt divided by total assets instead of well-known debt-to-equity ratio. The reason is that we want to avoid equity in the denominator, which in some cases can be negative or small positive due to drastic net income fluctuations. Whereas current ratio is expected to have a positive impact on retailer's efficiency, leverage is not.

Finally, we introduced a control variable indicating, whether a retailer is engaged in manufacturing processes or not. The rationale behind is that producer-retailers are thought to be more efficient, since they are able to directly control for COGS and set higher trade margins.

As soon as the regression model is fully specified, one can examine the signs of the coefficients in front of the control variables to quickly check its adequacy.

Tab. 1. Control variables

2.4 Determinants of Retailers' Efficiency: Hypotheses Development

The set of explanatory variables was determined by either already common or only gaining popularity strategic practices that retailers can potentially employ in order to accelerate their margins: organic and inorganic growth, multichannel selling, private labels, and franchising. Their choice will be consistently explained in the following paragraphs. Overall, five corresponding hypotheses were made within this paperwork.

2.4.1 Organic and Inorganic Growth

Businesses constantly face the trade-off between the two expansion strategies - organic or inorganic growth. While organic growth refers to business expansion achieved by the firm's internal resources, inorganic growth is attained via mergers, acquisitions, or takeovers. Companies can thus choose to grow organically developing their own in-house competencies, investing in product line innovations or inorganically - leveraging upon the market, products, and revenues of other firms.

Both strategies are characterized by their own advantages and disadvantages, benefits and risks. For instance, unlike extensive growth strategy, the organic one allows expanding while maintaining the overall control over the direction the business is headed. However, when market competition is fierce, it is the inorganic strategy that seems more preferable with the aim to enlarge customer base and get access to new geographical locations.

The major example of the uncertainty about the issue is Russian FMCG market leaders Magnit and X5 Retail Group. Adhering to respectively organic and inorganic growth, these players are in a constant tight competition for leadership permanently overtaking one another. We thus attempted to provide statistical evidence on which strategy is more efficient by including corresponding variables into the regression model. The hypotheses made upon these variables are as follows:

Hypothesis 1: Retailers with higher market presence growth achieve higher efficiency (Tab. 2).

Tab. 2. Hypothesis 1: Independent variables, expected results and argumentation

Hypothesis 2: Retailers, which have been involved in M&A, show lower efficiency in the year of M&A, and higher - in the subsequent years (Tab. 3).

Tab. 3. Hypothesis 2: Independent variables, expected results and argumentation

2.4.2 Private Labels

Private labels (PL) are the brands that are manufactured or packaged for sale under the name of the retailers rather than that of the manufacturer. Though the PL strategy is now significantly more advanced in consolidated retail markets of US and Europe, no countries are immune to the impact and Russia is no exception.

Renewed interest in PL strategy comes with each major recession, as it gives them an opportunity to capitalize on a price-sensitive market (PLs are usually 20-30% cheaper than named brands). A recent survey conducted by Nielsen indicated that consumer demand for PLs has increased in Russia in 2014-2015. Though 60% of respondents consider PLs to be the option for people who cannot afford branded goods, 40% of Russian consumers had to admit that PLs represent better price-quality ratio. Given the recessive economic situation, two-thirds of Russians would not refuse trying new product categories under PL.

Having first been introduced at the beginning of 2000s, PLs represent a relatively new phenomenon in Russia. Combined with the low concentration of the Russian retail (10 largest retailers hold only 24% of the market as for 2015), this helps explain modest (4%) level of penetration. The FMCG sector was a pioneer in the field. While in 2014 the average share of PLs in total sales of FMCG retailers was 10-13%, by the end of 2015 the figure reached 20%. Now it is not only FMCG sector in Russia to see potential in private labels. Inspired by latest European practices, Russian DIY retailers have recently joined the race, as well. In 2015, the average share of private labels in total sales of DIY retailers accounted for 6%.

The reason behind market players starting to launch private labels is not only their higher marginality as opposed to traditional brands, but also:

- lower operational risks (e.g. from supply chain interruptions, changes in conditions for granting trade discounts, etc.);

- better price to quality ratio (hence, higher customer loyalty).

Given the existing demand and financial benefits, PL strategy can potentially enhance margins and build customer loyalty during times of economic instability. This assumption determines the third hypothesis of the paper as follows:

Hypothesis 3: Retailers, which employ private labels more actively, achieve higher efficiency (Tab. 4).

Tab. 4. Hypothesis 3: Independent variables, expected results and argumentation

2.4.3 Multichannel Selling

Multichannel selling is a business strategy of utilizing more than one sales channel or method to offer the goods or services to customers, i.e. physical stores, websites, social media platforms, online marketplaces, and mobile applications. In Russia, though, multichannel retail is just under development. Thus, this paragraph is to explain how Russian retailers can benefit from keeping with the global trend.

According to the Global Retail Survey conducted by PwC in 2015, the most popular reasons why Russians prefer to buy products online instead of in stores are lower prices and better deals (58%), no need to travel to a traditional store (55%), opportunity to shop 24/7 (30%), and wider variety of products (28%). First, the fact that 58% of the respondents shop online because it is cheaper makes recession a perfect time to start moving in the multichannel direction. Second, selling online enables to provide around-the-clock hours of operation, thus raising profit margins. Thirdly, Russians more than citizens of any other country (55%) cited `no need to travel to traditional stores' as one of the reasons to buy product online. We connect it with constant transport problems in big cities and long distances in rural areas of Russia, which makes e-retail even more appealing in terms of market coverage. In addition, developing additional brick-and-mortar stores in remote geographic areas requires significant initial financial investments and typically takes several months to become fully functional. This path to new markets is much faster and safer when switching to multichannel selling. Overall, the above arguments are reflected in the fourth research hypothesis is as follows:

Hypothesis 4: Retailers, which employ multichannel selling more actively, achieve higher efficiency (Tab. 5).

Tab. 5. Hypothesis 4: Independent variables, expected results and argumentation

2.4.5 Franchising

Franchising is defined as a business system under which one company (or franchisor) grants another (or franchisee) the right to operate a business using its trademark. Franchising is primarily a mean of distributing or providing goods or services to the consumer. In retail, however, it is considered more as a relatively low-cost method of expanding an existing business, especially on foreign markets or markets that are geographically or culturally remote from the franchisor. Thus, the vast size of Russia makes franchising mostly attractive for retailers seeking to expand their activities into new geographic market areas.

Not only does franchising allow the rapid expansion of a distribution network, but it also can help capital formation. Traditional ways of raising capital include venture capital lenders, various forms of bank and commercial financing, or private/public placement of securities through investment banking channels. In this sense, franchising might be thought of as a worthy alternative as the franchisee makes a significant capital contribution to the strengthening and expansion of both the scale and the goodwill associated with the franchisor's business enterprise and brand.

Franchising as a business model only came to Russia once formal franchisee legislation was adopted in 1994. Since then, the number of franchise operations has been steadily growing. By the end of 2015, there are over 160 domestic franchising businesses in Russia, with a total of 2900 franchisees. According to data from the Russian Franchising Association (RFA), two-thirds of all franchises are established domestically with retail trade constituting 46% of all active Russian franchisers.

Though franchising in Russia has not yet developed to its fullest, we assume it can have a positive impact on retailers' efficiency due to its both strategic and financial benefits mentioned above. Thus, the final hypothesis of the research is determined as follows:

Hypothesis 5: Retailers, which employ franchising, achieve higher efficiency (Tab. 6).

Tab. 6. Hypothesis 5: Independent variables, expected results and argumentation

2.5 Estimation Methodology

As our dataset has a panel structure, it encompass features of cross-sectional and time-series data, thus allowing one to construct informative regression models. There are two types of panel regressions: fixed- and random-effects models.

With regard to fixed-effects models, these consider each entity as unique and not randomly chosen from the general population. This specification provides most adequate results when applied to countries, regions, industrial sectors, etc. In this type of models, the coefficients on regressors are computed so, as if dummies for each entity were included in the model. Unlike fixed-effects models, random ones are usually applied to randomly selected samples. While the fixed-effects models automatically control for time-invariant characteristics between individual entities, in the random-effects models these should be included manually.

In order to decide on the model best fitting our dataset, several statistical tests can be employed. The F-test results in the left-down corner of the fixed-effects regression output is to indicate whether a pooled model, which assumes no individual variance between entities, is better or not. If the corresponding p-value is close to zero, then the choice should be made in favour of the fixed-effects model. To choose between pooled and random-effects model, one should perform the Breusch-Pagan Lagrange multiplier test. Random and fixed-effects models can be, in turn, compared by applying the Hausman test.

While the overall significance of the fixed-effects model is to be determined by the value of R-squared coefficient, for the random-effects models one should analyze the Wald-chi square statistic.

3. EMPIRICAL FINDINGS

According to the methodology proposed above, the current section is to present and provide interpretation for the major results of data management and regression analysis. First, we will comment on the dataset general characteristics and descriptive statistics of key variables. Then, some experiments with the model specification are to be made to achieve the one of highest explanatory power. Having decided on the model specification, we will consider and choose between its random- and fixed-effects variations. When the model best fitting the dataset is fully established, the regression coefficients will be thoroughly interpreted. Finally, the Shapley Value Decomposition technique will help us determine the business strategies of highest contribution to the explanation of margin.

3.1 Descriptive statistics

The dependent variable is one of highest priority when first approaching the data. Fortunately, our dependent variable (EBITDA margin) appeared to be almost normally distributed (Fig. 1). Though we could have taken its logarithm to achieve smoother distribution, this would not be reasonable for the dependent variable is already a percentage. Have we done that, the coefficients on independent variables would be impossible to interpret. One entity was considered as having influential values of margin and was thus removed from the dataset to ensure higher quality of the future model. Appendix 1 presents the description and summary statistics for the variables, which later appeared significant in any of the constructed regression models. No evident mistakes in the data have been revealed based on the descriptive statistics.

Fig. 1. Density distribution of the dependent variable

Though control variables are those we are not particularly interested in, these are thought to have a substantial influence on the dependent variable and thus should be included to the model. Therefore, the summary statistics for the controls provides general information about the sample:

- 17 is the average age of the firm as of 2015.

- Two-thirds of the sample is represented by retailers from FMCG, electronics and home appliances, fashion, and DIY sectors (Fig. 2)

- 8% of the firms have been listed for the whole period in question or became those (which is why we gave up on EVA and ROIC as dependent variable at the initial stages of the research).