Estimation of Own- and Cross-Price Elasticities of Disaggregated

Imported and Domestic Goods in Russia

Nadezhda Ivanova

Graduate Institute of International Studies, Geneva

Abstract

The paper employs panel data analysis to estimate price and income elasticities for disaggregated domestic and imported goods using the Budget Survey of Russian households and prices of imported and domestic goods in Russia. The project is implemented using two types of data: the national level data for average households and households differentiated by income, and the data for the average regional households. Three different specifications of the demand equations: the double-logarithmic, the Linear Approximation to the AIDS, and the specification derived from the maximization of the CES utility function, are estimated for eight categories of traded non-food goods. The application of the instrumental-variable estimators to the regional data enables the endogeneity biases of the elasticity coefficients to be substantially corrected. The results of estimations of elasticities of demand for domestic and imported goods obtained for households differentiated by income indicate certain differences between estimated elasticities. This fact may be important for evaluating the impact of implementation of price and tariff policies on consumers with different levels of income.

September 2005

The author would like to thank Manuel Arellano, Michael Beanstock, David Brown, Mark Schaffer, David Tarr, Charles Wyplosz and Ksenia Yudaeva for useful suggestions and comments. All remaining errors are my own. The author is also very grateful to Konstantin Glushenko and David Brown for sharing some regional data with her.

1

Non-technical summary The price elasticity of demand is a measure of the sensitivity of demand to price changes: it

determines percentage change in demand caused by 1 percentage change in the price. This paper

presents estimates of price and income elasticities of demand for disaggregated traded goods for the

average Russian households and for households, differentiated by income levels. A statistical

analysis is applied to expenditure data obtained from the Household Budget Survey, which

distinguishes between spending on imports and import-competing domestic products within the key

eight categories of the non-food consumer goods: textiles, clothing, footwear, furniture, electric

household appliances, vehicles, TV sets and construction materials.

Up until now, there has been no study estimating elasticities of the tradable goods in Russia at the

disaggregated level. Internationally the most systematic and complete information on the

disaggregated trade elasticities is available for the USA. According to both the estimates obtained

for the USA and the results of this research, demand for imported clothing and footwear appears to

be extremely sensitive to changes in prices, while electric household appliances and textiles prove

to be in the group of moderately price-elastic goods. On the other hand, while the estimates reported

for the USA suggest that furniture and vehicles can be classified as highly import-sensitive, these

categories of goods turn out to be only moderately price-elastic according to the estimates obtained

for the Russian economy.

As far as the absolute values of the estimated elasticities are concerned, the own-price elasticities

range from -0.95 to -8.38 for the imported goods and from -0.68 to -13.87 for the domestic ones.

The evidence on the substitutability between imported and domestic goods is mixed: while for the

imported goods the values of the cross-price elasticities are mostly positive and vary from 0.36 to

5.66, for the domestic goods the substitution effect with the exception of clothing and textile

appears to be negligible. At the same time, the values of the Armington elasticities, which are

estimated in the range of –1.29 to –6.29, imply the presence of some nontrivial degree of

substitutability between the imported and domestic products. All imported goods with the exception

of furniture and vehicles can be viewed as a kind of luxury goods since their estimates of income

elasticities are slightly above unity. On the other hand, all domestic goods (again with the exception

of furniture and vehicles) with their income elasticities below unity represent necessities for the

households.

2

Income inequality is a striking feature of the Russian economy. Hence, to correctly evaluate the

effect of economic policies on the poor, it is appropriate to use the estimates of elasticities that

reflect the behavioural responses of the poor rather than the entire population. It can be claimed that

this paper is the first study internationally, which estimates trade elasticities by income level. There

are certain differences in the estimates of elasticities of demand for domestic and imported goods

obtained for households broken down into income groups. In particular, poorer households appear

to be more sensitive to changes in prices and income than richer households as regards demand for

domestic goods, while regarding the consumption of imported goods, richer households tend to be

more affected by changes in prices and income than the poorer households.

The knowledge of price elasticities of traded goods at the disaggregated level is essential to a

comprehensive analysis of a wide range of controversial policy issues vital to the Russian economy.

The results of the current paper will allow economists to incorporate in their studies the estimates of

elasticities, obtained on the basis of econometric estimations using data on consumer expenditures

and prices in Russia. This may increase the relevance of policy recommendations and conclusions

drawn for the Russian economy.

3

1. Introduction

This project estimates prices and income elasticities of imported and domestic goods for Russia at

the disaggregated level. Estimates of the responses of consumer expenditures to changes in prices

are widely applied to address a variety of economic policy issues.

The truth is that no one seems to have undertaken systematic estimations of trade elasticities for the

Russian economy, especially at the disaggregated level.1 Obtaining estimates of elasticities of

imports and import-competing domestically produced goods will allow researchers to tackle a

number of issues which are hotly debated and have important policy implications in Russia. One of

such issues, whose thorough investigation requires the knowledge of trade elasticities at the

disaggregated level, is an assessment of welfare, output and employment effects of the reduction of

tariff and non-tariff barriers as part of Russia’s WTO accession.

Estimation of price elasticities of imported and import-competing domestically produced goods in

Western economies mostly relies on the use of time-series data at the level of industries.

Unfortunately, this approach appears to have certain limitations in the case of Russia. First, the

construction of domestic demand for domestic goods turns out to be problematic for the Russian

data – the application of this procedure in the case of many Russian industries is reported to have

resulted in negative values of domestic demand. Second, no less importantly, the trade data

collected by the Customs in Russia is commonly viewed as very unreliable.

The paper approaches the problem of the estimation of elasticities for traded goods in an innovative

manner by applying cross-section and time series analysis to expenditure data obtained from the

Household Budget Survey conducted by the Russian Statistical Agency, Goskomstat, on a quarterly

basis. The Survey makes a distinction between spending on imports and import-competing domestic

products within the following key eight categories of aggregated non-food goods: textiles, clothing,

footwear, furniture, electric household appliances, vehicles, TV sets and construction materials.

In addition to obtaining estimates of the parameters of the demand functions for representative

households, this paper also estimates the demand functions for households differentiated by their

1 The results of the estimation of aggregated demand for imports in Russia are presented and discussed in the paper by Дынникова (2001).

4

level of income. In view of high income inequality in Russia, consumers in different income groups

are likely to react differently to changes in prices and incomes. This departure from the concept of

the representative consumer in the estimating the parameters of the demand functions will enable

the effects of economic policies on different income groups, including the poorest ones, to be

evaluated correctly.

The estimations of elasticities are implemented using two types of data – the national level data for

the average households and households differentiated by income, available for the eight quarters of

1999– 2000 and data for the average regional households available for the four quarters of 2000.

The general strategy of estimations is to apply the standard two-stage budgeting approach and

cross-sectional time-series regression analysis employing variations in expenditures and prices

across time, products inside categories of aggregated goods and, when possible, geographical

regions, under the assumption that elasticities are the same inside each category of goods. In terms

of functional specification of demand equations, I estimate the regression equations in the form of

the double-logarithmic demand functions and in the form of the linear approximation to the Almost

Ideal Demand System (LA-AIDS). In addition, I obtain estimates of the Armington elasticities of

substitution between imported and domestic goods for the demand equation derived from the

Constant Elasticity of Substitution (CES) utility function.

The paper is organised as follows. Section 2 presents a brief review of literature relevant to the

objectives of the study and outlines the general estimation strategy. Section 3 examines the data

used in the analysis and the key problems inherent in the data and suggests approaches to address

these problems. The results of the estimations are discussed in Section 4. Section 5 concludes.

2. Literature Review and Estimation Strategy.

The review of three strands of literature is relevant to the purpose of estimating elasticities of

domestic and imported goods from the data of household budget surveys. First, there is a host of

literature on estimation of income and price elasticities for import and export demand and supply

functions in developed countries at both aggregated and disaggregated levels. Stern et al. (1976)

provided an extensive annotated bibliography of own- and cross- price elasticity studies in

international trade as well as summary tables of “median” price elasticities broken down by the

commodity group and country. The survey of Goldstein and Khan (1985), encompassing both

methodological issues and empirical evidence of the estimation of import and export demand

relationships, is the most comprehensive and widely quoted one.

Due to the pioneering contribution by Armington (1969), a systematic scheme of estimation or

derivation of the own- and cross-price elasticities was adapted to imports disaggregated by types of

commodities (and/or by the country of origin). The building blocks of the Armington model and its

application to the problem can be summarized as follows. First, all commodities are distinguished

by the kind and place of production. Types of commodities (or “goods”) correspond to a rather

broad commodity classification. In our case, these are eight aggregated non-food goods (see Section

3). Although the original Armington model makes a distinction between goods produced in

different countries, I assume that consumers differentiate only between domestic and imported

goods. It follows that similar domestic and imported goods are to be imperfect substitutes for one

another. Second, demand for imported and domestic goods is determined by two-stage budgeting,

which incorporates the fact that domestic buyers allocate expenditures to imports in a variety of

categories of goods. At the first stage, total expenditures are allocated to different categories of

goods on the basis of consumer income and prices of goods. At the second stage, expenditures

within each category of goods are allocated between imports ( ) and domestic goods ( ) on the

basis of prices of imports ( ) and domestic goods ( ) and overall consumer demand for that

category of goods (determined at stage one):

miq d

iq

mip d

ip

),,( idi

mi

mi

mi Yppqq = (2.1)

),,( idi

mi

di

di Yppqq = , (2.2)

where - the total expenditures on good i, determined at the first stage of the

consumer’s decision:

idi

di

mi

mi Yqpqp =+

)),,(),...,(( 111 YppPppPYY dr

mrr

dmii = (2.3)

The formal representation of the two-stage budgeting process and the required assumptions of the

separability of preferences and homotheticity of subutility functions can be found in Armington

(1969), and in many empirical papers employing his approach, e.g. in Shiells et al (1986).

5

As regards the estimation strategy, several comments can be made with respect to certain theoretical

limitations of the conceptual framework outlined above. In particular, the application of the two-

stage budgeting, which greatly simplifies econometric analysis, requires homothetic subutility

functions.

Empirical researchers are in fact divided into those who make that assumption only for theoretical

exposition and do not apply it in econometric work, since data do not necessary support that

assumption, and those who stress that failure to impose the restriction would violate the very

principles of two-stage budgeting (e.g. Panagariya et al (1996)). As regards the incorporation of the

elasticities obtained from the econometric analysis into the Applied General Equilibrium (AGE)

models, it may be appropriate to apply the assumption of homothetic subutility into econometric

estimations.2

On the other hand, if this assumption is rejected by the data, econometric estimation carried out

under this restriction may produce biased estimates of other coefficients of the demand functions.

This certainly represents one of the limitations of the standard two-stage budgeting approach.

Another one is the assumption of weak separability of preferences, which the actual data may not

necessary satisfy either. At best we can view these limitations of the two-stage budgeting approach

as an approximation to some more complicated real process of making consumption decisions. The

nonparametric test developed by Varian (1982, 1983) can be used for checking the validity of the

assumption of weak separability.3

However, checking the assumption of homothetic subutilities does not involve the application of the

special nonparametric procedure and can be performed as part of a standard econometric regression

analysis along with estimations of other coefficients of the demand functions. Therefore, to get an

6

2 For instance, AGE model uses price indices of goods that are composed of the domestic and imported variety. In order to ensure that such price indices of the composite goods are defined appropriately, a theoretical assumption that the income elasticities of demand functions for domestic and imported goods equal unity ( myη =1, dyη =1) may need to be explicitly made in the regressions. 3 Applying this test to the data on tradable demand systems, several authors (Winters (1984), Shiells et al (1993)) demonstrate that for some subsamples of the data the assumption of weak separability is indeed too restrictive. However, in most cases, including ours, even if weak separability is rejected by a part of the data, given the shortness of the samples, very little or nothing can be done to relax this assumption, because estimation of more complex tradable demand systems obviously requires much greater degrees of freedom than researches can usually afford.

idea of how well the data and model support the assumption of homothetic subutilities, I undertake

estimations of the demand equations without restriction on the coefficient of the “income” or total

expenditure variables and test if the income elasticities are statistically different from unity.

In some empirical studies, demand equations are assumed to be free from money illusion for

estimation purposes, but I did not make such an assumption and preferred to test it as a hypothesis

instead. Since both quantity and prices are determined simultaneously, price variables may be

endogenous. However, if the assumption of a small country is made and the import supply function

is infinitely price elastic, the price of imported goods can be viewed as given on the world market.

In addition, given the first stage budgeting and expression (2.3), group expenditure, Yi, may be

potentially endogenous as well. To solve the endogeneity problem, some sort of instrumental-

variable estimation (2SLS, 3SLS, GMM) can be applied to prevent the biases in the OLS estimates.

Regarding the functional form of equations (2.1) and (2.2), which must be specified in order to

perform the estimation, the second big strand of literature - Applied Analysis of Domestic Demand

- is also relevant for the purposes of the project. Studies in this field provide a theoretical

framework for econometric estimations of demand functions in general, and in particular, in

international trade. In principle, empirical demand analysis distinguishes three kinds of approaches

to the specification of demand equations (see Theil and Clements (1987), Deaton and Muellbauer

(1980b)). The first approach is the so-called ad-hoc specification of the demand equation, with no

reference to the utility maximization problem of the representative consumer. The double-

logarithmic specification is one of the most widely used examples of such an ad-hoc specification.

Many studies of trade price elasticities (Stern et al. (1976), Goldstein and Khan (1985), Shiells and

Reinert (1993), use the double-logarithmic specification of import demand functions:

mmymmmdmdmm Yppq εηηηα ++++= lnlnlnln (2.4)

where ηmm and ηmd are own- and cross-price elasticities, and ηm is income elasticity of demand for

imported good. While quite attractive in terms of direct interpretation of the estimated coefficients,

the double-logarithmic model (DLM) has serious limitations. Since it is not utility based but ad-hoc

7

specified, there is no reason to expect the double logarithmic demand function to satisfy the adding-

up restrictions, implied from the linear budgeting assumption.4

The second approach of the empirical demand analysis is the specification of the demand equation

on the basis of a specific algebraic form of the utility function. Here one of the most popular forms

of demand equations, broadly used for estimations of elasticities in international trade at the

disaggregated level, is that derived from the Constant Elasticity of Substitution (CES) utility

function:

)/ln()/ln( dmdm ppqq σα += (2.5)

where σ is the elasticity of substitution, e.g. between imported and domestic goods. The CES utility

function satisfies all the requirements of the Armington’s two-stage budgeting and is very popular

with empirical studies, where goods are differentiated by types of commodities and by countries of

origin (see Shiells and Reinert (1993), McDaniel and Balisteri (2002)).

And finally the third, also utility based, approach suggests a more general or flexible form of

specification of demand equations allowing restrictions on particular coefficients of the demand

functions to be tested statistically. In the analysis, I incorporate the linear version of the flexible

form of a demand system, 5 which was proposed by Deaton and Muellbauer (1980a) and called the

Almost Ideal Demand System (AIDS).6 In its general specification, the AIDS is a non-linear system

of the following form:

)/ln(ln PYpw ijj

ijii βγα ++= ∑ , (2.6)

4From Deaton and Muellbauer (1980b, p.17) we know that double-logarithmic model, as a model of constant-elasticities by its definition, “will only satisfy adding-up generally” if demand functions take the trivial form of constant budget share demand functions with income elasticity equalling one ( 1=iyη ), zero cross price elasticities ( 0=ijη ) and

own-price elasticity equalling minus one ( 1−=iiη ) - the ones that come from maximization of classical Cobb-Douglas utility functions (the point also stressed by Marquez (1994)). On the other hand, the bulk of empirical literature using the double-logarithmic specification of the demand functions produces estimates of elasticities, and in particular, trade elasticities, which are significantly different from the predicted trivial benchmark case. In an attempt to reconcile such a contradiction between the theory and empirical studies, it can be suggested that the constant-elasticities obtained from the estimation of the double-logarithmic specification of the demand functions are some approximations to the “true” and not constant elasticities of a wide variety of utility based demand functions, when the “true” elasticities, as functions of level of prices and income, are evaluated at the sample average values of those parameters. 5 It provides a second-order local approximation to an arbitrary expenditure function, i.e. it can exactly match all derivatives of an arbitrary expenditure function up to the second order at any specific point.

8

6 Another widely used flexible specification of the demand system is the Rotterdam model suggested by Theil (1965, 1976) and based on his and Barton’s (1966) differential approach.

where jkk j

kjkk

k pppP lnln21lnln 0 ∑ ∑∑ ++= γαα

Deaton and Muellbauer (1980) themselves suggested the linear approximation to (2.6), replacing

the price index P proportionally by the known Stone’s price index :

. ktk

kt pwPP lnlnln * ∑=≈ φφ

The estimates of demand elasticities can be computed from the estimated AIDS’s parameters on the

basis of the following expressions:7

iji

jijiijij w

wPYδ

βββγη −

−+=

)/ln( (2.7)

1)/( += iiiy wβη (2.8),

It should be stressed that most of the studies on trade elasticities are only focused on specification

and estimation of demand equations for imports or exports, ignoring the equations of demand for

domestic substitutes. The third strand of literature that is important for the purpose of the study

seeks to estimate the tradable demand systems, making a distinction between disaggregated

domestic tradables and their imported substitutes. Existing research on the estimation of tradable

demand systems has largely arisen to support the AGE analysis. For instance, the AIDS Model was

applied by Shiells et al (1993) for modelling the North American Free Trade Area and by Shuguang

et al (1999) for estimating parameters of demand for imported and domestic goods in China.

Marquez (1994) applied the Rotterdam Model for the estimation of U.S import demand.

In most studies, disaggregated trade elasticities are estimated using time series data at the industrial

level, i.e. industrial classification is applied to goods. Following this approach, researches find

correspondence between trade statistics and the data on domestic industrial production. Import

statistics are used for import demand variables, while domestic sales of domestically produced

goods are calculated as domestic industrial production less exports of goods in question. The same

approach is used for the price data. This strategy, however, has serious limitations for Russia, as a

transition country. First, the construction of domestic demand for domestic goods by subtracting

exports from industrial production appears to be problematic for the Russian data – the application 7 The demand elasticities computed from the estimated AIDS’s parameters are functions of prices and income, and, consequently, are not constant, but vary over the sample. Inserting the sample average values of budget shares, incomes and prices into expressions (2.7)-(2.8), we can obtain "constant" estimates of the elasticities.

9

10

of this procedure in the case of many Russian industries is reported to have resulted in negative

values (and quantities) of domestic demand. Second, no less importantly, the trade data collected by

the Customs is commonly viewed as very unreliable because of the widely acknowledged and

persistent underreporting of the true values of imports at the customs.

The above caused me to look at the household expenditure data split between spending on imported

and locally produced goods as the source of information on domestic demand. Deaton (1987, 1988,

1990) introduced a methodology allowing the estimation of own- and cross-price elasticities of

domestic demands using household budget surveys. The central idea in Deaton’s analysis is to

assume spatial variation in prices in developing countries and divide households geographically into

clusters. Deaton applied his procedure to budget surveys containing data on expenditures and

purchased quantities but not market prices.8 Applying his procedure to the household budget

surveys in Cote d’Ivoire (1987, 1988) and Indonesia (1990), Deaton estimated own- and cross-

price elasticities of demand for several aggregated consumer goods in those countries. More

recently, Deaton’s approach was applied by Stavrev and Kambourov (1999a, 1999b) for the

estimation of price elasticities using the data of Bulgarian household budget surveys.

As can be seen from Section 3, contrary to the studies by Deaton (1987, 1988, 1990) and Stavrev

and Kambourov (1999), I am unable to use the household level raw data on expenditures and

demographic characteristics from the Goskomstat-conducted Russian Household Budget Survey.

What I can use is some sort of expenditure aggregates constructed on the basis of the household

data. In addition, the Goskomstat Surveys contain only nominal expenditure records that fail to

indicate quantities purchased. On the other hand, detailed data on market prices for both domestic

and imported goods were obtained from another data set from Goskomstat.

The data of the Goskomstat Budgetary Survey and the price data are discussed in detail in Section

3. The next section will also show that given the availability of the data the general strategy of

estimations is to apply the standard two-stage budgeting approach and cross-sectional time-series

regression analysis employing variations in expenditures and prices across time, products inside

8 Deaton built a model in which market prices are treated as unobservable variables, which directly determine quantities purchased and are only “indicated” by unit vales - the ratios of expenditures to quantity. Because unit values reflect the quality choice and because the quality choice will be generally affected by prices, all prices are assumed to affect the unit values of all goods in the model. In addition, Deaton develops a procedure to deal with measurement errors in expenditures and quantities, which will also induce the measurement error in unit values that are likely to be negatively correlated with quantities.

11

categories of aggregated goods and, when possible, geographical regions. In terms of functional

specification of demand equations, I estimate the regression equations in the form of the double-

logarithmic demand functions and in the form of the linear approximation to the Almost Ideal

Demand System (LA-AIDS). In addition, I estimate the Armington elasticities of substitution

between imported and domestic goods for the demand equation derived from the Constant Elasticity

of Substitution (CES) utility function. Section 3 also discusses some problems of estimations

arising from the use of different data sets and possible solutions.

3. Data Description.

The study is implemented using two types of data – the national level data for the average

households and households differentiated by the income level, and the data for the average regional

households. The information on consumer demand comes from the Household Expenditure Survey

of 49,175 households conducted by Goskomstat in all Russian regions (89 regions minus Chechnya)

on a quarterly basis. Based on the results of the Survey, Goskomstat produces aggregated figures

for the 88 regions as well as for the entire economy. In general, aggregated information on

expenditure contains data on average nominal spending in terms of roubles by 100 members of

households on various food and non-food goods and services. In particular, the Survey makes a

distinction between spending on imports and close domestic substitutes for eight categories of

aggregated non-food goods and the following numbers of products within each category of goods:

(1) four textile products, (2) 30 clothing products, (3) nine footwear products, six products in each

of the following categories: (4) furniture, (5) electric household appliances and (8) construction

materials; as well as (7) one TV product, and (6) three types of vehicles.9

In addition to the aggregated results on overall spending of all types of households broken down by

region and for Russia as a whole, Goskomstat produces the same kind of aggregates for different

types of households, in particular, for ten income deciles. The expenditure data for ten income

deciles obtained for the current research are at the national level of aggregation and are not broken

down by the region.

Prices are obtained from the Goskomstat’s data set on Average Prices of Domestic and Imported

Representative Items. These prices are measured for 196 Russian cities and towns of 88 regions

9 For the complete lists of the names of products in each category of goods, see Table 1.4 in Appendix 1.

12

each month, beginning January 1999. The monthly prices of imported and domestic items for the

entire Russian Federation are obtained by Goskomstat by aggregation of this kind of price

information collected in regional cities and towns.

In obtaining quarterly prices as simple three-month averages, I use raw monthly prices in order to

have the same frequency of the price data as that of expenditure data. However, in general, the price

data is more disaggregated than that of household expenditures, therefore some of the products

inside the categories of goods also include sub-products.10 Since the composition of household

expenditures by sub-products is unknown, a product price is calculated as a simple average of the

sub-product prices and, consequently, product expenditures divided by the calculated product prices

are used as proxies for product quantities (see Section 4 for the definition of the variables used in

the regressions). This approach is justified by the application of the Hicksian composite commodity

theorem, since the evidence from the data is that the prices of items belonging to the same products

move in parallel.

The plan of the rest of the section is as follows. Descriptive statistics of overall household

expenditure data and the prices of imported and domestic products for Russia as a whole are

presented in subsection 3.1. Subsection 3.2 displays what new information the incorporation of the

regional dimension provides to the analysis of overall household spending. Subsection 3.3 looks at

the expenditure data of ten income deciles.

3.1. Data on overall spending of all households at the national level

Descriptive statistics of expenditure and price data for Russia as a whole for the common time

series sample of those data (Q1 1999 - Q4 2000) are presented in Tables 1.1 (Appendix 1). The first

column of Table 1.1 gives the average shares of eight aggregated goods in total household

expenditures on non-food goods: the sum of household spending on eight goods accounts for almost

70% of total non-food expenditures. Expenditures on clothing and footwear as well as basic non-

food necessities represent the largest and most stable shares of total spending on non-food goods (as

10 (1) Two of four textile products include two sub-products, (2) 17 of 30 clothing products include two to five sub-products, (3) four of nine footwear products include two to three sub-products, (4) four of six furniture products include two to four sub-products, (6) one of three types of vehicles includes two sub-products, (8) two of six construction material products include two sub-products.

13

can be seen from the coefficients of variations presented in the second column of Table 1.1). The

third column of Table 1.1 presents shares of imports in expenditures on eight categories of goods.

We can distinguish between two different groups of goods. The first group, including footwear,

clothing, televisions, textiles and electric household appliances, contains very high (more than 67%)

and relatively stable shares of imports.11 In the second group, including mostly durables such as

furniture, construction materials and vehicles, the shares of imports are lower (less than 35%) and at

the same time much more unstable12.

The last two columns of Table 1.1 in Appendix 1 show descriptive statistics of consumer prices for

imported and domestic goods for the entire Russian Federation. The striking feature of the data is

the high coefficients of correlation between prices of domestic and imported products for all

categories of goods. The Law of One Price in the relative form13 appears to hold in Russia, at least

for the goods and periods under consideration. It can be seen that goods with very high shares of

imports in expenditures tend to have almost perfectly correlated prices for domestic and imported

goods, while for durable goods with lower shares of imports, relative prices seem to show wider

fluctuations.

The fact that prices for domestic and imported goods are highly correlated means a multicollinearity

problem, and, as a result, some coefficients of price elasticities may appear to become insignificant

in the regression equations. To mitigate this problem and to eliminate such form of heterogeneity as

the fixed-effects, the regression equations (see Section 4) were mostly estimated in the first

differences rather than in levels.

11 To some extent the share of imports for those goods can be overestimated: according to Goskomstat there is evidence that, for example, apparel produced by the small Russian firms often carries foreign labels, possibly to attract more consumers. 12 On one hand, this fact can be observed because of the durable nature of these goods and imperfect quality of the data: we have quarterly figures only for two years, and for such a short period the number of respondents in the surveys might not be sufficient to represent the entire markets of these goods. On the other hand, demand for durables, as expensive goods, may be more sensitive to changes in relative prices. 13 While the Law of One Price in the absolute form says that prices of the same traded goods denominated in the same currency should be equal all across the world, the Law of One Price in the relative form allows the level of the prices of the same traded goods to differ across countries but states that those prices are subject to the equal rates of inflation.

14

3.2. Regional expenditure and price data.

In order to obtain regional prices of imported and domestic products, I have to use prices for the

representative items measured, in general, in 196 Russian cities and towns of 88 regions. Although

measurement of such prices started in January 1999, I am able to obtain monthly data on prices for

regional cities and towns only for the year 2000.14 Appendix 1 provides the details of the algorithm

and the procedure applied for calculation of the monthly and quarterly regional prices for imported

and domestic products from data on the monthly prices of representative items recorded in the cities

and towns.

The number of regions for which prices for both imported and domestic products can be calculated

is presented in Table 1.3 (Appendix 1). It can be seen that even inside each category of goods, the

number of regions varies substantially across the products. Moreover, the list of regions is not the

same for different products. Applying cross section and time-series analysis with variations across

time, products and regions, I have to deal with quite an unbalanced panel, where some products

inside the category of goods are over-represented compared with others in terms of regions, and

some regions are over-represented in terms of products. However, even given this variability in the

number and lists of regions, we can see that the size of the samples for all categories of goods

increased substantially after the incorporation of the regional data.

Table 1.2 of Appendix 1 presents the average shares of imports for the regional household

expenditure data for the same eight categories of goods15. The left column describes the average

shares of imports and their coefficients of variation for all 88 regions, while the next one shows the

same indicators only for regions where the respective regional price data are available. Despite the

selection procedure described above both columns report quite similar results, except for

construction materials.16

Next, the figures in the columns presenting the average shares of imports for the regional data can

be compared with those for the entire economy (see Table 1.1, Appendix 1). Since the data for the

whole of Russia is constructed using some weights for the size of population of different regions,

14 Goskomstat decided to remove price data registered in various towns, keeping only aggregated prices for the whole of Russia for the year 1999, since the collected information in electronic form allegedly took a lot of space. 15 In addition to averaging across regions the data in Table 1.2 is also averaged across time (four quarters of 2000) and the products inside categories of goods. 16 The latter is probably explained by the domination of one particular product in the lists of regions for that good.

15

the consumption pattern of the biggest and riches regions may dominate the aggregated results.

Comparing two tables, we can be see that when the regions are not weighted by their population

(Table 1.2) the average shares of imports are a bit lower for all categories of goods and at the same

time much more unstable than for the whole of Russia. The last two columns of Table 1.2 show the

average regional relative prices for imported and domestic products. Once again comparing those

with relevant figures for the entire economy (see Table 1.1, Appendix 1), we can see that the

regional relative prices show much higher volatility than the same indicators calculated from the

aggregated data.

Incorporation of the regional data into econometric analysis will allow me to work with standard

types of panels, where the size of cross-section significantly exceeds the number of time periods. At

the same time, the types of estimations implemented for the expenditure data at the national level

(for average households and households decomposed by income groups) lie mostly in the domain of

cross-sectional time series analysis when the number of time periods is relatively large compared

with the number of cross-sections. It should, however, be stressed that most variation in the regional

price data comes from variation across regions and products, while prices vary very little over time,

since the regional quarterly price data are available for only one year, 2000, which saw relatively

low inflation. Given the low variation of prices over time, along with the unbalanced nature of the

regional panel, I have to employ the assumption of the random effects model, which prove to be

correct for most of the demand equations of the eight categories of goods at the regional level (see

Section 4.1).

3.3. Expenditure data of ten income deciles.

First of all, some overall measures of inequality in spending among ten income deciles can be

provided 17. Table 1.4.a. (Appendix 1) shows the ratios of total consumer expenditures and

expenditures on non-food goods between each adjacent deciles and the same ratios for the 10th and

1st deciles. It can be seen that total consumer expenditures of the 10th group are almost 10 times

those of the 1st group. The difference becomes even more striking when expenditures on services

and food are excluded: the 10th group spends 21 times more than the 1st group does on non-food

17 Although in the case of expenditure data it would be more appropriate to call them "spending" or "expenditure" deciles, because these groups are made up by ranging expenditure records of each 10% of respondents of the Survey, which are then aggregated to represent the entire population

16

goods alone. The same fact is demonstrated by the last line in Table 1.4.b: less than 20% of the

poorest group’s total consumer expenditures account for non-food goods, while the richest group

spends more than 40% of the total on these items. Looking at the ratios for the adjacent deciles, we

can see that the poorest group and the richest group indeed stand out against their neighbours: the

differences in spending between the 1st and 2nd groups and between the 10th and the 9th are

remarkably larger than those between all other adjacent groups. This fact is noteworthy, since very

rich consumers are notorious for eluding the questionnaires of such surveys and are clearly

underrepresented by them.18

Table 1.4.b. (Appendix 1) shows the average shares of eight aggregate categories of goods in

household expenditures on non-food goods as well as total spending of eight aggregated goods as a

percentage of non-food expenditure by deciles. A number of observations followed. The ratios of

spending on the main non-food necessity – clothing, to total non-food expenditures do not differ

substantially among the deciles, and that is not much less true of another non-food necessity –

footwear. As regards the durables, there are obvious observations that the share of expenditures on

those increases with the rise in total spending. Looking at the share of spending on eight categories

of aggregated goods in total non-food spending by deciles we see that these eight goods are the

most representative for the richest group and the least – for the poorest one.

Table 1.4.c. shows the average shares of import for each category of goods by deciles. Here again

the most differences are observed in the durables, while the shares of imports for clothing and

footwear are almost equally high (no less than almost 77%) for all the deciles. The latter may

indicate a variety of facts. It may prove that the Russian light industry does indeed offer very little

even to the least affluent consumers, but, on the other hand, given the doubts on too high shares of

imports in clothing and footwear for the data on overall households, those figures on income deciles

may provide additional evidence to view them as probably overestimated. Another obvious

observation is that although according to the data, both the 10th and the 1st groups spend more than

75% on imported clothing, and the rest on the clothing produced domestically, quality of both

imported and domestic goods purchased by the richest and the poorest can in most cases be

expected to be very different. The same argument is of course valid for all categories of goods. And,

18 On the subject of the underrepresention of the richest part of population in particular in Goskomstat Expenditure Survey, and the suggested way to address the problem, see Aivazian and Kolenikov (2001). On the issue of measuring poverty in Russia, see also Ovcharova et al (1998).

17

in fact, as prices paid by consumers of different income groups are not observed, there is nothing I

can do to differentiate between the products and goods consumed by different groups. Estimating

the demand equations for different income deciles I have to assume explicitly that consumers in all

income groups pay the same set of prices for imported and domestic goods,19 the difference comes

only from purchased quantities or from the shares of spending on imported and/or domestic

substitutes. However, the increase in the volatility of import shares for goods such as clothing and

footwear for each deciles compared with those for the average households may imply that

consumers with different levels of income respond differently to changes in prices for those goods.

On the other hand, as may have been expected, the share of imported durable goods, while still

quite high, is, nevertheless, more stable in the case of each income group than for all types of

households in total.

The remarkable feature of the expenditure data broken down into income deciles is the presence of

a significant number of zeros in expenditure records, particularly for the lowest income groups.20

This may mean the problem of measurement errors with zeros in expenditure records of some

income groups – the problem of purchase infrequency. Following Deaton (1990) and Stavrev and

Kambourov (1999b), in estimating the demand functions of different income groups I estimate not

only double-logarithmic demand functions (which force me to exclude the zero observations), but

also demand functions in the form of the linear approximation to the AIDS (where expenditure

shares of goods are used instead of quantities of goods as dependent variables). That type of

demand equations enables zero expenditure on some goods to be modelled. The use of

instrumental-variables for estimation of share equations of demand functions may mitigate the

problem of infrequent purchases even further.

4. Results of Econometric Estimation

As mentioned above, the estimations of elasticities are implemented using two types of data – the

national level data for the average households and households differentiated by the income level

and data for the average regional households. In terms of functional specification of demand

19 This assumption can be valid in the situation when prices paid by consumers from different income groups for products of different quality are subject to the rate of the inflation that is common to the entire category of goods. 20 As Table 1.4.c. shows, for the most expensive durables – cars – constant purchases are almost exclusively reported by the two highest deciles. The situation is quite similar for motorcycles, which, in principle, means that coefficients of demands for vehicles can be estimated and compared for only a few highest income groups. This is still relevant but less dramatic in the case of other durables – furniture, electrical household appliances and construction materials.

equations, I estimate the regression equations in the form of the double-logarithmic demand

functions, like (2.4), in the form of the linear approximation to the Almost Ideal Demand System

(LA-AIDS) (see 2.6) and, in addition, I obtain the estimates of the Armington elasticities of

substitution between imported and domestic goods for the demand equation derived from the

Constant Elasticity of Substitution (CES) utility function (see 2.5).

Subsection 4.1 discusses the results of the estimation of the demand equations for the regional data.

The results and comments on the results of the estimation of the double-logarithmic and LA-AIDS

demand functions for the average households and for the ten income deciles are presented in

Subsections 4.2.

4.1. Estimations using data at the regional level of aggregation.

For the expenditure and price data at the regional level of aggregation with variation across regions,

products within categories of goods, and across four quarters I apply panel data analysis under the

assumption that elasticities are the same inside each category of goods. Every product inside every

region constitutes a cross-section dimension of the panel for each category of goods, i.e. the number

of cross-section units for each category of goods is the number of all products in all regions. In

order to take into account the effects of particular products and geographical locations of the

regions, the product dummies and dummies for the six federal districts (seven exiting districts

minus one) are included in the panel regressions, whenever some of them appear to be significant.

First I estimate the following two regression equations in the standard double-logarithmic form like

(2.4) for each of the eight categories of goods:

mititmy

mitmm

ditmd

mi

mit uYppq ++++= lnlnlnln ηηηα (4.1)

dititdy

ditdm

ditdd

di

dit uYppq ++++= lnlnlnln ηηηα (4.2)

where i stands for a product, t stands for time, mi

mim

i pV

q = is the quantity of an imported product i

and di

did

i pV

q = is the quantity of a domestic substitute purchased by consumers in particular category

of goods, and are expenditures on imported and domestic products, and are prices miV d

iV mip d

ip

18

of imported and domestic products, is total expenditure, or money “income”

allocated for spending on a product composed of domestic and imported products. The subscript for

the category of goods is skipped for the sake of space.

di

mii VVY +=

To see how the inclusion of the records of zero purchases affect the results, I use the Linear

Approximation to the AIDS (LA-AIDS) as a specification of the demand equations. Since under the

two-stage budgeting approach the system of demand functions consists of only two equations (for

imported and domestic goods), and the coefficient of those two equations satisfy the adding-up

restrictions imposed by the linear budgeting assumption, I can estimate only one of the two

equations of our LA-AIDS and then use the adding-up restrictions and several modifications of

formulas (2.7) and (2.8) applied for the linear version of AIDS, to calculate the elasticities of the

demand system. In particular, the demand equation for imported goods is estimated:

mititm

mitmm

ditmd

mi

mit uPYppw ++++= *)/ln(lnln βγγα (4.3)

where YV

wm

imi = is the budget share of imported products and lnP* =wm*lnpm+ wd*lnpd is the

Stone Price Index. It has turned out that after the inclusion of the zero purchases the LA-AIDS

specification of the demand equations show worse statistical performance than the DL does,

although in general the price and income elasticities obtained from estimation of both specifications

are fairly close to each other. For the sake of space the results of the estimations of the LA-AIDS

obtained using the data at the regional level are not reported.

Table 2.1 (Appendix 2) presents the results of the estimation of DL equations (4.1)-(4.2) for all

categories of goods, except for clothing, under the random effects (RE) assumption about the

unobserved heterogeneity, while for clothing and domestic footwear the fixed effects (FE)

assumption proves to be valid. As a standard approach to that RE sort of unobserved heterogeneity,

when the variance of the random effects is estimated first (see Green (2000), Wooldridge (2002),

Baltagi (1995), Hsiao (2003), Arellano (2003)), the Generalized Least Square (GLS) estimator is

applied. The results of the test for the statistical significance of the variance of the RE, which is

reported at the bottom of Table 2.1, show that for the nine DL demand equations the unobserved

heterogeneity is indeed present, in seven cases it can be rejected at any standard level of 19

20

significance, while in one case (demand for imported electric appliances) evidence is not so explicit.

21 The results of the Hausman specification test, which are also reported at the bottom of Table 2.1,

prove that the assumptions of the RE panel model, namely that unobserved effects are uncorrelated

with the explanatory variables, appear to be correct for all equations, with the exception of the

above mentioned function of demand for imported and domestic clothing and domestic footwear.

Since the potential endogeneity of all explanatory variables in equations (4.1)-(4.2) is likely to be a

serious problem in estimations, the instrumental variable (IV) techniques are applied. For the cases

where the unobserved heterogeneity turns out to be significant I apply either the Generalized 2SLS

estimator (G2SLS) or FE-IV and First Difference (FD) estimators, depending on whether the RE or

FE type of panel model prove to be appropriate.22 Where unobserved heterogeneity does not appear

to be the problem, the 2SLS estimator with robust standard errors is applied.23

Looking for instruments for regional prices of imported and domestic goods, I follow the approach

of Glushenko (2002), who studied the degree of integration of the Russian consumer goods market.

Glushenko stressed that the main source of variation in prices across regions is the difference in per

capita demand, which is in turn approximated by the difference in per capita income. Following this

finding, I have tried other variables, which also indicate the difference in the living standards of

Russian regions in general. These variables are the average wage per worker in regions, the value of

the main food basket in regions, and per capita gross regional product. Among other reasons

determining the market fragmentation of regions, Glushenko found factors such as the quality of

regional transport infrastructure, costs of retailers in regions, intra-regional transport costs,

subsidies, “shuttle trade”, and organized crime to be significant. All those factors are tested as

candidates for the IVs for prices of imported and domestic goods in regions, and some of them have

indeed been found to be significant. In addition to the variables just mentioned I have also tested

some variables, which reflect the production side in the regions. In principle, there may be two

sources of supply of consumer goods in the region: the primary one - industrial enterprises located

in the region and producing one or another consumer good, while the secondary source is retail

businesses and it’s present in all regions regardless of whether the goods are produced in the

21 The presence of the unobserved heterogeneity can be rejected at the 10% of significance level, but not at 5%. 22 Both types of the IV estimators, except for the FD-IV one, are covered in Baltagi (1995). The FD-IV estimator is discussed in Hsiao (2003) and Wooldridge (2002). 23 For the latter I use the version of the extended IV estimator programmed for Stata by Baum, Schaffer and Stillman (see Baum, Schaffer, and Stillman (2003)).

21

particular region. For example, I have found that producer prices in the regions significantly affect

the prices of domestic textile products, while the physical volume of fabrics produced in the regions

affects the prices of both domestic and imported textiles. Moreover, I have found that prices are

influenced by the amount of taxes paid by producers and retailers, such as the value added tax and

the tax on imputed income. The data on most of the variables discussed above are published by

Goskomstat. Data on taxes are available on the website of Russia’s Ministry of Finance. The author

is very grateful to K. Glushenko and D. Brown for providing the data on variables such as “shuttle

trade”, subsidies, index of the economic power of organized crime and index of infrastructure

development in the regions (see Glushenko (2002) and Brown and Earle (2001)).

In search of the IVs for the total expenditure or “income” allocated for spending on composite

(domestic and imported) good, Yi, I look for factors, which reflect consumers’ decisions at the first

stage of the budgeting process, namely total consumer expenditures, consumer expenditures on food

and non-food goods, and consumer expenditures on services. Information on these variables was

obtained from the Household Budget Survey conducted by Goskomstat, which is also the main

source of information on demand (see Section 3).

Table 2.2 presents the results of the estimation of the first-stage equations of the IV estimators for

the demand equations of the six goods – textiles, clothing, footwear, furniture, electric household

appliances and construction materials. For these goods, all or some of the explanatory variables in

equations (4.1)-(4.2) appear to be endogenous according to the results of the endogeneity test (see

below). The results of the estimations of the first-stage equations of the IV estimators and in

particular the test of joint significance of the instrumental variables are viewed as the test checking

the relevance of the suggested instruments. According to this test, all variables reported in Table

2.2. appear to be relevant instruments for the explanatory variables suspected to be endogenous in

equations (4.1)-(4.2). Most of the suggested IVs vary across regions and quarters, and several of

them vary only across regions.

Table 2.3 reports the results of the estimation of the second-stage equations of the IV estimators for

the demand equations of the six goods. For vehicles and TV-sets, all explanatory variables in

equations (4.1)-(4.2) appear to be exogenous: for vehicles Table 2.3 just repeats the results of Table

2.1 and for TV sets Table 2.3. presents the results of the application of the FGLS estimator with the

22

adjustment for panel heteroskedasticity. The bottom line of Table 2.3 shows the results of the

overidentification test, which can be used to check the validity of the additional instruments, in the

sense that they are uncorrelated with the error term of the equation, in the situation when we have

more instruments than we need to identify the equation. Since I have failed to reject the null

hypothesis in all cases, I can be confident in the overall set of the suggested instruments for all

demand equations.

Comparing the estimates in Tables 2.1 and 2.3 we can see that IV estimates of the elasticity

coefficients differ significantly from those presented in Table 2.1. This fact may imply that the

respective explanatory variables are endogenous and the estimates of the coefficient for those

variables in Table 2.1 are seriously biased. This conjecture is indeed confirmed by the results of the

Hausman tests on endogeneity24 reported at the bottom of Table 2.2. Given these findings, I discuss

below the estimates of elasticities obtained with the use of the IV technique and summarized in

Table 2.4.

Income elasticities of the imported goods range from 0.49 to 1.24 and those of the domestic goods

vary from 0.21 to 0.98. All domestic goods, with the exception of furniture and vehicles, can be

viewed as necessities with their income elasticities below unity25 (domestic textiles have the lowest

income elasticity of 0.21, although this income elasticity is the only one which is not statistically

significant), while all imported goods, again with the exception of furniture and vehicles, have

elasticities slightly above unity and represent, rather, a kind of luxury goods (imported electric

household appliances have the highest elasticity of 1.24).

All own-price elasticities for both imported and domestic goods are negative and statistically

significant. For all goods, except for imported textiles and domestic vehicles, the own-price

elasticities are greater than one in the absolute terms. Clothing appears to be the most price-elastic

among the eight categories of goods, with the own-price elasticity of –8.38 for imported and –13.87

for domestic clothing. Domestically produced clothing represents only a small share of household

expenditures versus imported clothing, so the obtained high price elasticities for domestic clothing

24 The Hausman test on endogeneity and Sargan test on overidentifying restrictions are programmed in Stata by Baum, Schaffer and Stillman (see Baum, Schaffer, and Stillman (2003)) 25 For domestic furniture and vehicles we cannot reject the hypothesis that their income elasticities statistically differ from one.

23

are in line with the findings of Reidel (1988, 1994), who showed that demand for products of

developing countries is very price-elastic on the world market. Another category of domestic goods,

which turns out to have a fairly high absolute value of own-price elasticity is TV sets (-4.48).

Most of the cross-price elasticities of imported goods are significant and positive (domestic goods

are expected to be substitutes for imports and vise-versa), while demand for domestic goods, with

exception of textiles and clothing, turned out to be insensitive to changes in prices of imported

substitutes. The highest and statistically significant estimates of cross-price elasticities are again

obtained for clothing: 5.66 for imported and 17.36 for domestic products.

In addition to the tests discussed above, Table 2.3 also presents the results of the test of

homotheticity and test of the “no money illusion” hypothesis. It has turned out that the hypothesis

that income elasticity of demand equals one cannot be rejected with 5% of significance level only in

four out of 16 cases (these four cases are imported clothing and construction materials and domestic

furniture and vehicles). The hypothesis of “no money illusion” or that demand functions are

homogeneous of degree zero can be rejected at the 5% significance level in all cases but two

(imported clothing and construction materials). Many other studies (e.g. see Deaton, A. and

Muellbauer J. (1980a)) confirm and comment on this departure from the classical assumptions of

the consumer demand theory.

The fact that the level of aggregation of data used for the estimation of the import-demand

elasticities for the US and other economies differs substantially from that used in the current study

precludes direct comparisons between the absolute values of elasticities obtained for other

economies and those of the current paper. However, some relative comparisons between estimates

of elasticities for different categories of goods can be drawn. For instance, according to both the

estimates obtained for the USA26 and the results of this research, demand for imported clothing and

footwear appears to be extremely sensitive to changes in prices, while electric household appliances

and textiles prove to be in the group of moderately price-elastic goods. On the other hand, while the

estimates reported for the USA suggest that furniture and vehicles can be classified as highly

import-sensitive, these categories of goods turn out to be only moderately price-elastic according to

the estimates obtained for the Russian economy.

26 See Stern et al (1976) and Shiells et al (1986).

24

In addition to the won- price, cross-price and income elasticities of the demand functions for

imported and domestic goods, I also estimate the Armington elasticity of substitution between

imported and domestic goods by running the regressions of form (2.5) for seven categories of goods

using the regional data. There are not enough observations for vehicles to estimate that sort of

equation. It should be mentioned that the CES utility function is homothetic by definition, while our

results of the estimations for the DL specification show that in most cases this assumption is not

really supported by the data. On the other hand, the CES utility functions are so popular, especially

in the field of the AEG, where most of demand for our estimates probably lies, that it makes sense

to estimate the Armington elasticities directly. Table 2.5 reports the results of estimation of

Armington elasticities under the RE assumption for all goods except for clothing, since that

assumption has turned out to be invalid for this good (see the Hausman specification test), so

equation (2.5) for clothing is estimated with FE.

Since the application of the IV estimators to the demand equations in DL form (4.1)-(4.2) implies

that prices of many imported and domestic goods are indeed endogenous variables for those

equations, it is appropriate to suggest that the relative prices of imported and domestic goods may

be endogenous variables of equation (2.5) as well. To test this hypothesis I have searched for the

IVs for relative prices. Table 2.6 reports the results of the estimations of the first-stage equations of

the IV estimators for equation (2.5) for the seven goods. Since the effects of some IVs, which I use

for the price variables in the DL equations, cancel each other, while the effects of others remain

important for the relative prices, I have looked for additional IVs. The additional IVs for the relative

prices are the per capita gross regional product (textiles), the index of real industrial production in

regions (clothing and furniture), the CPI of non-food goods, average January temperature, and

output of small enterprises of the transportation sector (footwear), the share of fully depreciated

assets in total fixed assets of transportation and communications enterprises (furniture and TV sets)

in 1999, regional budgets’ expenditures on industry, regional production and retails sales of

refrigerators (electric household appliances), the size of the regional territory (TV sets and

construction materials), the size of deposits with Sberbank at the end of 1999, and total foreign

investment in the regions (TV sets).

25

Table 2.7 presents the results of the estimation of the second-stage equations of the IV estimators

for the Armington elasticities of the seven goods. The results of the overidentification test at the

bottom line of Table 2.6 suggest that I may be confident in the used IVs for all goods except for

clothing and footwear. Although there are certain difference in the estimates of the Armington

elasticities reported in Table 2.5 and Table 2.7, the results of the endogeneity test reported in Table

2.6 imply that the relative prices are indeed endogenous only for one category of goods - electrical

household appliances. The IV estimates of the Armington elasticities reported in Table 2.7 are

statistically significant at the conventional levels of significance and vary from -1.29 for clothing to

-6.29 for TV sets. It should, however, be stressed again that the results of the estimation of the

Armington elasticities should be considered with some degree of caution, since the relevant demand

equation (2.5) is derived from the maximization of the homothetic CES utility function, while that

assumption generally does not appear to be supported by the data, as follows from our analysis of

the income elasticities using the DL specification of the demand functions.

4.2. Estimations using the data at the national level of aggregation: average households and

households broken down into income groups.

The strategy for the estimation of the demand equations using the national level of aggregation (for

households broken down by income and for average households) is to apply that data for panel (or

cross-sectional time series) regression analysis employing variation across time and products within

categories of seven goods27, under the assumption that elasticities are again the same inside each

category of goods. As in the case of the regional data, I estimate the regression equations in the

standard double-logarithmic form (4.1) - (4.2), LA-AIDS specification of the demand equations

(4.3) and equation (2.5) with the Armington elasticities for each of the seven categories of goods.

As it can be seen below, the estimations of the double-logarithmic and LA-AIDS specification of

the demand equations obtained using the data at the national level of aggregation result in quite

similar coefficients of elasticities.

While the maintained hypothesis is that demand elasticities are the same for all products inside each

category of goods for both the DL and LA-AIDS specification of the demand equations, I allow for

the product specific fixed effects (FE). The fixed effects appear to be significant for most of the

27 Dealing with the panels not enlarged by the regional expenditure and price data I exclude the 7th category of goods – televisions, because there are only eight observations available.

deciles for all goods, except for furniture, where it never shows up. In general, there are two widely

used approaches to get rid of the significant FEs, which represent the unobserved heterogeneity and

are correlated with the explanatory variables. One approach is to obtain the so-called FE-estimator,

which is equivalent to running regressions in the time-demeaned variables. 28 The other one is to

estimate the equations in the first differences (FD). The choice between the FE and FD estimators

depends on the assumptions and nature of the idiosyncratic error terms, . As pointed out by

Wooldridge (2002, p.284), the FE estimator is more efficient when are serially uncorrelated,

while the FD is more efficient when follows a random walk. For the five out of seven categories

of goods, namely for textiles, furniture, electric household appliances, vehicles, and construction

materials, the panels of the data at the national level of aggregation have the higher time dimension

(T=8) than the number of cross-section units (N). For one category of goods – footwear, N and T

are almost the same size (N=9), and only for clothing (N=30) the panel may to a certain extent be

considered as satisfying the standard classical assumptions, when N is large and T is fixed. In other

words, the problem of so-called time series persistence in the panels at the national level may exist,

when N is fixed while T can go to infinity and the time dependence of the data may take the form of

unit roots. On the other hand, the available number of time periods - eight quarters, is not large

enough to test for the presence of the panel unit roots, while for the larger T, such tests can, in

principle, be implemented (see Hsiao (2003)). Having said that, I regard the estimation of the

equations in the FD as a superior strategy for the panel data at the national level compared with the

estimation of the time-demeaned equations, and, accordingly, run the regressions in FD whenever

there is evidence of the FEs

itu

itu

itu

29. At the same time, when the FEs appear to be insignificant for some

goods (such as furniture) and income deciles, the demand equations are estimated in levels,

although this approach may be vulnerable to the potential presence of the unit roots.

The types of panel or cross-section time series data I work on are likely to have a more complicated

structure of the variance-covariance matrices of the error terms than assumed for the correct

26

28Time-demeaning means removing the panel level means from each variable. 29 In some cases, the constant term appeared to be significant in the regression equations estimated in FD, which is a somewhat contentious issue, since there is no well-established theoretical guidance that would justify the presence of the time trend in the original demand functions, but, on the other hand, statistical significance of the time trend may imply the importance of some omitted and time-varying variables, or changes in consumer preferences over time. Actually, the inclusion of statistically significant constant term in the equations in FD appears to substantially change the value of estimated price elasticities only for furniture and electric household appliances (the results are not reported here).

27

standard errors of the classical OLS estimates. In particular, the model of the panel data may be

subject to cross-section heteroskedasticity and in general to contemporaneous cross-section

correlations. In addition, the models for panel data may be characterized by either common serially

correlated idiosyncratic error terms or by unit-specific serially correlated errors. The econometric

technique which is normally used to deal with such types of covariance structures is Feasible

Generalized Least Squares estimation (FGLS), proposed by Parks(1967), or calculation of Panel-

Corrected Standard Errors (PCSE) for OLS estimates, suggested by Beck and Katz (1995). 30

To test for the panel (or group-wise) heteroskedasticity and contemporaneous cross-section

correlations, I have applied the LR tests discussed, for example, in Greene (2000), since the iterated

FGLS without serial correlation assumptions produces the Maximum Likelihood estimates of the

parameters. To test against the presence of AR(1) serial correlation in the error terms of equations

(4.1)-(4.3), in the case of the common constant term (no significant FE), I estimate these

regressions by the Pooled OLS, save the residuals and then re-estimate these equations by Pooled

OLS, adding the saved lagged residuals and testing the significance - of the coefficients of the

lagged residuals (Wooldridge (2002), p.176). In the case of the significant FE the same procedure

is used for the equations in the FD (Wooldridge (2002), p.282).

For most of the goods and income groups, the application of the LR tests indicates at least the

presence of panel heteroskedasticity - the variation in disturbances variance across products. 31 In

cases when there is evidence of its presence, the correction for cross-section heteroskedasticity or

cross-sectional correlation causes standard errors of the estimated coefficients to decrease as

expected. The adjustment is made by applying either the relevant version of FGLS or the

appropriate form of the PCSE to the Polled OLS estimates, suggested by Beck and Katz (1995).

Where the models in either levels or FD are tested for the presence of the serial correlated error

terms, I allow for such structure of the covariance matrices by running the relevant version of the

FGLS or again by applying the appropriate form of the PCSE .

30 Beck and Katz (1995) have argued that in the case of contemporaneous cross-section correlations and/ or unit-specific serially correlated errors, the standard error obtained from FGLS may be too optimistic for panels with a relatively small difference in the number of cross sections (N) and number of time periods (T). 31 The application of the FGLS for the panel with N>T precludes the correction (and accordingly test) for cross-sectional correlation. In addition, the implementation of FGLS in Stata, which is used to perform the estimations, does not allow for cross-sectional correlation in the case of unbalanced panels.

28

The estimated elasticities of DL regression equations (4.1)-(4.2) for the seven categories of goods

for average households as well as for separate ten income groups are reported in Tables 3.1-3.3 of

Appendix 3. Table 3.1 and Table 3.2 present respectively the estimates of the own- and cross- price

elasticities of the DL model, while income elasticities of the DL model are shown in Table 3.3. The

elasticities calculated from the estimated coefficients of the LA-AIDS are in general quite close to

those estimated from the DL equations, although the estimated LA-AIDS equations show poorer

performance than the DL equations. Given that cross-price elasticities obtained from the estimation

of the DL specification are not quite significant for many deciles and goods (see Table 3.2), only

the own-price elasticities calculated from the estimates of the LA-AIDS specification are presented

(see Table 3.4). Comparing results reported in Table 3.1 and Table 3.4, we can see that elasticities

estimated from both specifications of the demand equations are very close for textiles, clothing and

footwear, which are goods showing high and stable shares of imports, while the elasticities obtained

from the estimation of two specifications are more divergent for furniture and construction

materials – goods with much less stable shares of imports. Estimates of the Armington elasticities

of substitution for the ten income groups are presented in Table 3.5 (Appendix 3).

The results of the estimation of elasticities of demand for domestic and imported goods obtained for

households differentiated by income indicate a certain difference between estimated elasticities.

Looking at the estimates of own-price elasticities obtained from the estimation of the DL equation

for domestic goods, where there are more statistically significant coefficients than in the estimation

of LA-AIDS specification, we can see that for all goods, except for clothing and footwear, the

average estimates of own-price elasticities for the five poorest groups are higher than those for the

five richest groups (see the bottom of Table 3.1). On the other hand, as regards demand for

imported goods in all categories of goods except for textiles and construction materials, the own-

price elasticities appear to be higher on average for the poorer deciles. In other words, poorer

households appear to be more sensitive to changes in prices and income than richer households as

regards demand for domestic goods, while regarding the consumption of imported goods, richer

households tend to be more affected by changes in prices and income than the poorer households.

The same observation can apply to the estimates of income elasticities (income elasticities within

income deciles) of demand for domestic and imported goods (see Table 3.3). Although the income

elasticities are approximately about unity in most of the cases, average income elasticities of

29

demand for domestic goods are higher in all goods for the five poorest groups of households than

for the five richest groups. The situation is the opposite as regards demand for imported goods –

poor households seem to have lower income elasticities for imported goods.

The results of the estimations using the regional data (subsection 4.1) show that all of the right-hand

side variables of the DL specifications of demand equations (4.1)-(4.2) can be endogenous. The

same observation may prove to be correct for the equations estimated using the national data.

However, the incorporation of available relevant instruments such as the production costs (wages,

producer prices) of the Russian industries concerned, and different tariffs on imported goods, into

the estimation of the demand equations at the national level of aggregation for average households

and households differentiated by income has produced almost no changes to the estimates of

elasticities. Moreover, the results of the endogeneity test have failed to confirm the endogeneity of

the suspected variables.32 For that reason as well as for lack of space, the IV estimators of

elasticities for households differentiated by income are not reported.

5. Conclusions

The paper employs cross-sectional time series analysis to estimate price and income elasticities for

disaggregated domestic and imported goods in Russia using the Budget Survey of Russian

households and data on prices of imported and domestic goods. Three different specifications of the

demand equations: the double-logarithmic (DL), the Linear Approximation to the AIDS (LA-AIDS)

(Deaton and Muellbauer (1980a)), and the specification derived from the maximization of the CES

utility function, are estimated for eight categories of traded non-food goods using the regional level

of aggregation for the average households and national level data for average households and

households broken down into ten income groups. The DL specification of demand equations appear

to show better statistical performance than the LA-AIDS specification does, although in general

price and income elasticities obtained from estimating both specifications are fairly close.

32 Such failure may to a certain extent be due to the absence of truly reliable instruments. On the other hand, since households in all income deciles face the same vector of prices in the estimations, the prices are expected to be exogenous rather than endogenous variables in the demand equations of households broken down by income group.

30

Since the results of the analysis of the regional data show that price and “income” variables are

indeed endogenous for demand equations of most of goods under consideration, the instrumental-

variable estimation is applied, which enables the endogeneity biases of the elasticity coefficients to

be substantially corrected. Income elasticities of the imported goods range from 0.49 to 1.24 and

those of the domestic goods vary from 0.21 to 0.98. All imported and domestic goods appeared to

be fairly elastic relative to the changes in their own prices: the respective elasticities range from -

0.95 to -8.38 for the imported goods and from -0.68 to -13.87 for domestic ones. On the other hand,

with the exception of clothing, the substitution effect turned out to be quite modest for the imported

goods and mostly negligible for domestic products. At the same time, the values of the Armington

elasticities, which are estimated in the range of –1.29 to –6.29, imply the presence of some

nontrivial degree of substitutability between the imported and domestic products.

The results of estimations of elasticities of demand for domestic and imported goods obtained for

households differentiated by income indicate a certain difference between estimated elasticities.

Poorer households appear to be more sensitive to changes in prices and income than richer

households as regards demand for domestic goods, while regarding the consumption of imported

goods, richer households tend to be more affected by changes in prices and income than the poorer

households. The incorporation of the instrumental variables into the estimations of elasticities using

the national level of aggregation produces coefficients very similar to those obtained without the

use of IVs.

31

Appendices

A1. Construction of the regional prices In order to obtain regional prices of imported and domestic products, I used the price of the

representative items measured, in general, in 196 Russian cities and towns of 88 regions. The prices

for imported goods and domestic substitutes are recorded in up to five regional cities and towns,

depending on the region. At the same time the price of each particular imported or domestic item,

is, as a rule, recorded not in all 196 towns, and respectively, not in all 88 regions.

The data provide evidence that there is very high degree of variability of prices of supposedly the

same item measured in different cities and towns even within the same region. In order to construct

the series of the most representative regional prices, which will not be biased to prices prevailing in

some particular, even the biggest towns in the regions, it was decided to maintain the maximum

number of towns representing each particular region where prices of each particular item are

measured relatively regularly. For this purpose sometimes the average regional inflation figure for

each particular item (based on prices measured in other available towns) was used to determine the

price of the item for a town, which was missed by Goskomstat in a specific month.

Only after implementation of that kind of adjustment, the monthly prices registered in towns were

converted into monthly regional prices, applying the standard procedure of weighting the towns'

prices by the relative size of population of those towns. Since for each particular item, the lists of

towns, and respective regions where prices for imported and domestic substitute are measured, do

not coincide, but just intersect at some nonempty set of regions, for each particular item only

regions for which observations available on both the imported and domestic prices were selected.

After applying that selection procedure on the level of items the monthly regional prices of products

and then the quarterly regional prices of products were calculated.

32

Table 1.1. Expenditure and import shares and price data for the national level of aggregation, 1999:q1–2000:q4

% of total expenditures Share of imports (%) Relative prices Correlation between domestic and import

prices*

Average Coefficient of variation Average Coefficient

of variation Average Coefficient of variation Levels Differences

Footwear 11.3 17.6 87.2 6.1 1.4 4.8 1.00 0.92

TV sets 4.2 19.1 86.0 7.2 1.2 0.6 1.00 0.99

Clothing 30.6 14.9 81.9 5.9 1.3 5.1 0.99 0.80

Electric household appliances 4.3 35.7 69.8 15.5 1.9 8.3 0.78 0.54

Textiles 0.8 18.9 68.0 10.1 1.6 8.0 0.97 0.64

Furniture 5.3 44.3 25.8 74.9 2.0 8.5 0.91 0.69

Construction materials 4.7 32.2 19.7 48.1 1.5 11.7 0.93 0.53

Vehicles 7.0 34.8 20.3 107.7 2.5 36.7 0.52 0.10

Sum of 8 categories 68.2 4.8 69.9 5.7

Note: * — Correlation coefficients between domestic and import prices are average correlation coefficients of prices for domestic and imported products. Correlation coefficients are calculated for logarithms of prices.

Table 1.2. Regional household expenditure and price data, 2000:q1–2000:q4

Share of imports across regions, (%)

All 88 regions Regions merged with price data

Relative prices: imports/domestic

Average Coefficient of variation Average Coefficient

of variation Average Coefficient of variation

Textiles 61.4 61.7 64.9 54.5 1.62 46.0

Clothing 80.7 28.9 80.3 28.1 1.37 38.2

Footwear 84.8 25.5 83.5 26.5 1.32 27.9

Furniture 16.5 196.3 22.7 165.2 1.65 64.9

Electric household appliances 47.6 86.0 51.3 77.8 2.26 45.0

Vehicles 24.9 167.9 27.1 159.5 2.03 67.4

TV sets 76.9 42.7 76.0 42.0 1.23 8.0

Construction materials 16.6 163.4 42.6 66.7 1.95 45.1

33

Table 1.3. Products and regions inside categories of goods (T — number of quarters)

Number of product Name of product Number of regions, R Number of observations per product, R×T

1. Textiles (356 observations)

1 Cotton textiles 6 24 2 Wool textiles 9 36 3 Silk and synthetic textiles 35 140 4 Other textiles 39 156

2. Clothing (5524 observations)

1 Men's winter overcoat, coats 5 20 2 Men's in-between-season

overcoats, coats 18 72 3 Men's suits 48 192 4 Men's trousers 77 308 5 Men's shirts 79 316 6 Women's winter overcoats, coats 7 28 7 Women's in-between-season

overcoats, coats 21 84 8 Women's suits, dresses 28 112 9 Women's skirts, trousers 71 284

10 Women's blouses, shirts 72 288 11 Women's underwear 21 84 12 Boys' winter overcoats, coats 33 132 13 Boys' in-between-season

overcoats, coats 37 148 14 Boys' trousers 40 160 15 Boy's shirts 70 280 16 Girls' suits, dresses 1 4 17 Girls' blouses, shirts 5 20 18 Boys' underwear 50 200 19 Men's headwear 57 228 20 Women's headwear 61 244 21 Boys' headwear 32 128 22 Men's jackets, sweaters 72 288 23 Women's jackets, sweaters 23 92 24 Boys' jackets, sweaters 58 232 25 Men's sportswear 63 252 26 Boys' sportswear 68 272 27 Men's underwear 72 288 28 Men's socks 80 320 29 Boys' socks 41 164 30 Women's stockings, socks 71 284

34

Number of product Name of product Number of regions, R Number of observations per product, R×T

3. Footwear (1884 observations)

1 Men's winter boots 46 184 2 Men's shoes 57 228 3 Women's winter boots 37 148 4 Women's in-between-season boots,

shoes 57 228 5 Women's shoes 37 148 6 Boys' winter boots 45 180 7 Boys' shoes 50 200 8 Women's sandals 71 284 9 Boys' sandals 71 284

4. Furniture (368 observations)

1 Tables and chairs 4 16 2 Cupboards 7 28 3 Beds and sofas 12 48 4 Other furniture 18 72 5 Kitchen furniture 25 100 6 Living and bedroom furniture 26 104

5. Electric household appliances (1188 observations)

1 Fridges 11 44 2 Washing machines 29 116 3 Vacuum cleaners 30 120 4 Small house equipment 66 264 5 Sewing machines 78 312 6 Lighting equipment 83 332

6. Vehicles (488 observations)

1 Bicycles 16 64 2 Motorcycles 47 188 3 Cars* 59 236

7. TV sets (196 observations) 1 Televisions 49 196

8. Construction materials (348 observations)

1 Plywood 2 8 2 Other construction materials 3 12 3 Slate, roofing paper 2 8 4 Windowpane 1 4 5 Bricks 2 8 6 Wall-paper 77 308

Note: * — New domestic cars are considered as imperfect substitutes for second-hand imported cars.

35

Table 1.4a. Comparative measures of spending between income deciles (annual data for 1999): ratios of expenditures between deciles: Di/Dj

D10/D1 D2/D1 D3/D2 D4/D3 D5/D4 D6/D5 D7/D6 D8/D7 D9/D8 D10/D9

Total consumer expenditures 9.75 1.52 1.26 1.22 1.21 1.21 1.20 1.24 1.23 1.54

Expenditures on non-food goods 21.76 1.63 1.32 1.31 1.35 1.37 1.30 1.36 1.2 1.95

Table 1.4b. Expenditure shares by income deciles (Russian Federation, no regional decomposition), 1999:q1–2000:q4, %

Deciles

1 2 3 4 5 6 7 8 9 10

0.7* 0.8 0.9 0.9 0.9 1.0 0.9 0.9 0.9 0.6 Textiles

33.2** 21.8 24.8 5.4 22.1 40.7 28.1 24.3 15.7 29.5

28.3 30.5 31.4 32.4 32.3 32.5 33.2 32.2 31.7 25.0 Clothing

9.3 8.8 9.5 10.1 7.6 14.5 14.7 13.8 12.8 27.3

12.1 13.0 13.3 13.3 12.8 12.4 11.5 10.1 9.6 6.6 Footwear

26.3 18.1 18.4 18.1 15.0 27.9 24.8 20.3 15.1 17.7

0.2 0.4 0.7 1.2 3.3 4.6 5.7 3.9 5.5 7.5 Furniture

40.1 30.4 61.7 49.7 181.2 80.3 60.6 33.4 46.5 62.1

0.6 0.7 0.8 1.1 1.5 3.3 3.2 3.2 3.4 5.2 Electric household appliances

30.2 31.5 32.4 19.8 47.8 150.4 59.3 34.9 48.8 77.0

0.1 0.1 0.1 0.1 0.2 0.2 0.5 3.0 1.6 18.5 Vehicles

100.3 71.6 60.2 73.4 85.9 81.8 126.6 171.9 69.0 40.6

0.1 0.1 0.3 0.5 0.6 0.8 1.3 2.4 2.2 2.8 TV sets

79.3 125.9 82.6 51.6 64.9 41.8 85.4 64.3 46.2 65.4

1.2 1.5 1.8 1.8 2.2 2.6 2.9 3.6 4.5 3.5 Construction materials

20.3 21.4 22.6 26.7 19.6 40.2 41.3 44.3 58.8 31.6

43.2 47.2 49.3 51.3 53.6 57.4 59.2 59.3 59.4 69.6 Sum of 8 categories

11.9 9.6 9.3 10.3 11.7 7.3 7.7 11.7 8.7 4.7

18.7 20.5 21.7 23.4 25.3 28.6 32.1 33.7 34.8 42.5 All non-food goods as % total consumers' expenditures 11.7 10.9 11.2 12.5 14.9 9.9 10.8 13.0 12.6 11.0

Note: * — Average; ** — Coefficient of variation.

36

Table 1.4c. Import shares by income deciles (Russian Federation, no regional decomposition), 1999:q1–2000:q4, %

Deciles

1 2 3 4 5 6 7 8 9 10

58.0* 62.9 62.9 62.3 62.8 67.7 64.3 65.4 70.7 72.7 Textiles

18.6** 16.5 15.7 14.3 17.0 15.0 16.5 17.7 17.2 15.5

76.9 78.5 78.5 80.2 80.1 81.8 82.7 84.5 83.3 82.9 Clothing

12.8 11.7 11.9 11.6 13.5 12.9 13.6 11.5 12.6 12.8

80.2 82.5 81.7 84.6 84.9 86.4 88.3 89.6 87.4 89.4 Footwear

13.6 9.0 10.6 10.3 10.0 9.6 8.7 6.2 9.8 8.2

2.7 11.8 7.7 13.5 13.0 16.2 24.1 23.4 24.5 27.0 Furniture

5.4 15.8 13.9 15.8 13.0 16.3 23.4 22.2 21.2 20.4

34.3 31.5 39.0 51.3 42.0 51.6 60.2 64.3 61.1 69.7 Electric household appliances

34.6 30.5 32.9 31.5 27.9 24.4 27.1 22.4 23.9 21.2

Vehicles, of which

Bicycles 22.9 13.3 3.5 16.7 10.8 17.9 21.9 17.0 33.3 23.6

35.7 19.3 5.9 22.6 12.5 27.2 24.7 22.3 42.2 27.1

N of records 5 7 8 8 8 8 8 8 7 8

Motorbikes 0.0 10.5 4.0 0.0 10.3 5.2 1.9 2.7 11.8 5.7

– 21.1 10.5 0.0 16.0 11.5 4.6 6.6 20.6 15.2

N of records 1 2 7 3 6 5 6 6 5 7

Cars – – – 100.0 100.0 33.3 100.0 100.0 61.9 29.7

– – – – 0.0 57.7 0.0 0.0 45.2 28.2

N of records 0 0 0 1 2 2 5 3 8 8

31.0 31.4 59.8 81.0 70.4 88.3 81.7 89.2 84.1 86.0 TV sets

31.0 34.1 25.2 11.8 18.1 8.1 8.0 10.1 9.3 8.7

10.4 11.2 14.5 14.0 16.6 22.1 24.9 28.1 27.4 27.8 Construction materials

13.4 14.0 17.5 14.8 20.0 18.8 21.6 23.7 23.4 22.9

75.8 77.2 77.6 78.5 76.5 75.7 77.9 78.2 73.0 57.4 Sum of 8 categories

1.7 2.3 1.8 1.3 9.4 9.7 9.5 3.9 7.9 14.9 Note: * — Average; ** — Coefficient of variation.

37

A2. Estimations using the data at the regional level of aggregation Table 2.1a. Own-price, cross-price and income elasticities, double-logarithmic specification of the demand functions (Textile and Clothing)

Textile Clothing

Imported, lnqm Domestic, lnqd Imported, lnqm Domestic, lnqd

RE RE FE FE

lnpm –1.130*** [9.50] –0.131 [0.61] –0.846*** [12.37] –0.229 [1.07]

lnpd –0.028 [0.27] –0.810*** [4.10] –0.192*** [3.79] –0.014 [0.09]

lny 0.997*** [32.13] 0.600*** [11.97] 1.016*** [168.94] 0.711*** [33.61]

p1 –1.294*** [5.49]

Central –0.418*** [3.44]

North_West –0.513*** [3.81]

Volga –0.488*** [3.57]

Ural –0.879*** [4.57] 1.147*** [3.38]

Oil producing region 0.352* [1.73] –0.995*** [2.67]

Far_East –1.676*** [3.28]

Siberia –0.734*** [2.78]

South –1.165*** [4.92]

Constant 0.663 [1.35] 0.644 [0.91] –0.215 [0.72] –4.044*** [4.41]

Observations 294 264 5331 4593

Number of cross-sections 85 84 1379 1337

R-squared 0.89 0.57 0.88 0.27

Test Var(u) = 0: Prob > chi2 0.1625 0.0000 0.0000 0.0000

Hausman specification test: Prob>chi2 0.0623 0.5084 0.0000 0.0000

Absolute value of z statistics in brackets, * — significant at 10%; ** — significant at 5%; *** — significant at 1%; p2, p3, p6 — product dummies. Estimates of the elasticities shown in bold are of the expected sign and significant at 10%, 5% or 1%.

38

Table 2.1b. Own-price, cross-price and income elasticities, double-logarithmic specification of the demand functions (Footwear and Furniture)

Footwear Furniture

Imported, lnqm Domestic, lnqd Imported, lnqm Domestic, lnqd

RE FE RE RE

lnpm –1.012*** [28.32] –0.291 [0.68] –1.243*** [4.01] –0.066 [0.59]

lnpd 0.028 [0.72] –0.294 [0.96] 0.367 [1.09] –0.737*** [5.00]

lny 1.015*** [213.33] 0.751*** [33.94] 0.820*** [9.32] 0.911*** [29.42]

p3 –0.557* [1.77] 0.14 [1.11]

p5 0.081** [2.32]

p8 0.103*** [3.21]

Central –0.148*** [5.54]

North_West –0.291*** [9.49]

Volga –0.171*** [6.26]

Ural –0.168*** [4.50]

sea_border 0.060** [2.06]

dq2 0.063*** [3.81]

Constant –0.299*** [3.12] –2.916 [1.45] –0.449 [0.30] –1.418** [2.18]

Observations 1661 1304 85 216

Number of cross-sections 471 458 47 80

R-squared 0.96 0.59 0.56 0.81

Test Var(u) = 0: Prob > chi2 0.0000 0.0000 0.3038 0.0319

Hausman specification test: Prob>chi2 0.6968 0.0000 0.3963

Absolute value of z statistics in brackets, * — significant at 10%; ** — significant at 5%; *** — significant at 1%; p2, p3, p6 — product dummies. Estimates of the elasticities shown in bold are of the expected sign and significant at 10%, 5% or 1%.

39

Table 2.1c. Own-price, cross-price and income elasticities, double-logarithmic specification of the demand functions (Electric household appliances and Vehicles)

Electric household appliances Vehicles

Imported, lnqm Domestic, lnqd Imported, lnqm Domestic, lnqd

RE RE RE RE

lnpm –1.213*** [9.69] 0.335* [1.69] –1.121*** [9.86] –0.290** [2.52] lnpd 0.176* [1.91] –1.024*** [7.12] 0.239* [1.88] –0.681*** [5.15] lny 1.012*** [40.43] 0.683*** [27.70] 0.915*** [31.57] 0.983*** [33.30] p2 0.302** [2.05] –0.705*** [3.66] p3 0.361*** [4.00] –0.411*** [3.01] p6 –0.445** [2.36] 0.570* [1.90] Constant –0.257 [0.53] –1.553** [2.16] –0.739* [1.95] –0.149 [0.54] Observations 716 857 49 120 Number of cross-sections 262 279 39 70 R-squared 0.76 0.68 0.96 0.91 Test Var(u) = 0: Prob > chi2 0.068 0.0000 0.61 0.675 Hausman specification test: Prob>chi2 0.3241 0.1595 0.5121

Table 2.1d. Own-price, cross-price and income elasticities, double-logarithmic specification of the demand functions TV sets and Construction materials)

TV sets Construction materials

Imported, lnqm Domestic, lnqd Imported, lnqm Domestic, lnqd

RE RE RE RE

lnpm –1.857*** [3.51] 0.239 [0.13] –1.024*** [7.77] 0.136 [1.05] lnpd 1.04 [1.24] –4.195 [1.33] 0.193 [1.29] –1.203*** [8.62] lny 1.051*** [30.66] 0.500*** [4.78] 1.092*** [21.46] 0.715*** [18.93] p3 –0.451*** [3.00] Metropol 0.595** [2.11] –1.006*** [3.50] South 0.388** [2.23] –0.504*** [2.95] Constant –1.982 [0.37] 26.423 [1.27] –2.272*** [3.94] 1.334*** [2.77] Observations 157 130 307 338 Number of cross-sections 48 46 82 87 R-squared 0.86 0.16 0.70 0.71 Test Var(u) = 0: Prob > chi2 0.2542 0.9116 0.0122 0.0000 Hausman specification test: Prob>chi2 0.9345 0.1871 0.1871 0.0526

Absolute value of z statistics in brackets, * — significant at 10%; ** — significant at 5%; *** — significant at 1%; p2, p3, p6 — product dummies. Estimates of the elasticities shown in bold are of the expected sign and significant at 10%, 5% or 1%. Metropol — dummy for Moscow, St. Petersburg and Moscow oblast.

40

Table 2.2a. First-stage IV estimates, double-logarithmic specification of the demand functions (Textile)

Imported (OLS regression with robust standard errors)

lny lnpm lnpd

Coef. t-statistic Coef. t-statistic Coef. t-statistic

North_West –0.026 –0.10 0.126 2.76 0.152 2.46

Volga 0.858 3.85 –0.050 -0.51 0.227 1.82

Ural 0.759 2.68 0.276 2.25 0.163 0.97

Oil prod.region 0.349 1.09 –0.302 -1.89 -0.824 -4.63

Far_East –0.168 –0.41 0.569 3.41 0.983 5.73

Siberia 0.730 2.62 0.027 0.18 0.135 0.77

South –0.038 –0.13 0.434 3.92 0.397 2.7

Producer prices 0.535 1.25 0.085 0.58 0.758 4.32

Costs of retail trade in 1999 –0.298 –3.58 0.139 4.28 0.161 3.96

p1 –0.619 –1.97 –1.647 -15.14 -1.631 -24.36

Index of organized crime –0.030 –0.46 –0.087 -3.13 -0.099 -3.7

Production of fabric –0.007 –1.81 –0.002 -3.21 -0.004 -4.32

Imputed income tax 0.040 0.44 0.065 1.53 -0.079 -1.74

Expenditure on food goods –1.870 –4.13 0.038 0.21 0.188 0.74

Expenditure on non-food goods –1.362 –2.64 1.090 3.94 0.958 3.22

Expenditure on services 1.612 3.39 –0.774 -3.21 -0.803 -3.28

Expenditure on alcohol 0.810 2.21 –0.039 -0.29 -0.291 -2.12

Constant 20.019 5.48 –1.857 -1.12 -3.241 -1.53

Observations 294 294 294

Partial R-squared of excluded instruments

Shea Partial R2 0.31 0.23 0.19

Partial R2 0.31 0.48 0.40

F(13, 276) 9.79 28.13 93.89

P-value 0.0000 0.0000 0.0000

Wu-Hausman test: P-val. 0.0035 0.0020 0.0014

Durbin-Wu-Hausman test: P-val. 0.0032 0.0019 0.0013

41

Domestic (First-stage G2SLS regression)

lny lnpd

Coef. t-statistic Coef. t-statistic

lnpm -0.446 -2.29 0.780 14.63

North_West 0.090 0.26 0.027 0.28

Volga 1.095 3.18 0.245 2.60

Ural 1.224 2.69 0.043 0.35

Oil prod.region 0.284 0.59 -0.237 -1.81

Far_East 0.838 1.31 0.397 2.26

Siberia 1.387 3.46 0.091 0.83

South 1.262 3.88 0.089 1.00

Producer prices

Costs of retail trade in 1999

p1 -1.393 -2.91 -0.336 -2.57

Index of organized crime

Production of fabric

Imputed income tax

Expenditure on food goods -1.572 -2.60 0.309 1.87

Expenditure on non-food goods -0.967 -1.35 -0.089 -0.45

Expenditure on services 0.811 1.16 -0.162 -0.84

Expenditure on alcohol 1.168 2.47 -0.035 -0.27

Constant 17.655 4.16 -0.075 -0.06

Observations 264 264

Prob > chi2 0.0000 0.0000

Wu-Hausman test: P-val. 0.0086 0.0000

Durbin-Wu-Hausman test: P-val. 0.0078 0.0000

Estimates of the coefficients of the excluded instruments shown in bold are significant at 10%, 5% or 1%.

42

Table 2.2b. First-stage IV estimates, double-logarithmic specification of the demand functions (Clothing, First-stage within regression)

Imported Domestic lnpm lnpd lnpm lnpd

Coef. t-statistic Coef. t-statistic Coef. t-statistic Coef. t-statisticlny 0.004 3.21 0.004 2.05 0.007 3.99 0.004 1.98 Price of gasoline 0.247 12.85 0.261 9.75 0.253 12.13 0.275 9.65 Wage in retail trading 0.148 8.09 0.226 8.82 0.139 6.85 0.204 7.33 Turnover of retail trading 0.038 3.04 0.047 2.68 0.044 3.17 0.042 2.25 VAT 0.009 2.52 0.012 2.61 0.006 1.59 0.013 2.52 Dummy for Q2 0.023 7.5 0.030 7.01 0.025 7.45 0.034 7.4 Dummy for Q3 0.015 5.93 0.021 5.88 0.016 5.66 0.022 5.89 Constant 3.521 42.68 2.593 22.5 3.395 36.84 2.620 20.75 Observations 5267 5267 4536 4536 R-sq within 0.48 0.43 0.48 0.42 Wu-Hausman test: P-val. 0.0000 0.0000 0.0000 0.0000 Durbin-Wu-Hausman test: P-val. 0.0000 0.0000 0.0000 0.0000

Estimates of the coefficients of the excluded instruments shown in bold are significant at 10%, 5% or 1%.

Table 2.2c. First-stage IV estimates, double-logarithmic specification of the demand functions (Furniture)

Imported (OLS regression with robust s.e.)

Domestic (First-stage G2SLS regression)

lny lnpm lnpd

Coef. t-statistic Coef. t-statistic Coef. t-statistic lnpd 0.223 0.64 lnpm 0.243 0.77 p3 0.914 3.1 Average wage 1.376 2.03 lny 0.131 4.58 0.117 5.31 Value of food basket 0.011 2.56 0.001 0.31 Gross regional product per capita 0.143 0.9 0.260 2.14 Regional territory 0.014 0.14 –0.219 –2.91 Subsidy to enterprises in 1999 –2.452 –5.28 –1.231 –3.45 Consumer expenditure 25.141 2.2 Expenditure on non-food goods –7.023 –1.43 Expenditure on food goods –12.897 –1.92 Expenditure on services –6.034 –4.89 Constant –26.672 –3.15 7.380 5.99 7.584 8.01 Observations 85 189 189 Partial R2 0.24 F(13, 276) 10.19 P-value 0.0000 Prob > chi2 0.0000 0.0000 Wu-Hausman test: P-val. 0.0564 0.0127 0.0316 Durbin-Wu-Hausman test: P-val. 0.0511 0.0122 0.0302

Estimates of the coefficients of the excluded instruments shown in bold are significant at 10%, 5% or 1%.

43

Table 2.2d. First-stage IV estimates, double-logarithmic specification of the demand functions (Electric household appliances, First-stage G2SLS regression)

Imported (lny) Domestic (lny)

Coef. t-statistic Coef. t-statistic

lnpm –0.060 –0.32 –0.208 –0.78

lnpd 0.210 1.52 0.117 0.61

p2 0.028 0.11

p3 –0.483 –3.64 –0.807 –4.56

p6 –0.450 –1.57 –0.739 –1.84

Consumer expenditure 3.769 4.88 12.031 3.04

Expenditure on non-food goods –2.998 –2.18

Expenditure on food goods –2.283 –4.29 –7.350 –3.14

Expenditure on services –0.804 –2.61 –0.650 –1.3

p2 0.755 3.44

Constant –6.585 –3.13 –15.942 –3.87

Observations 716 857

Prob > chi2 0.0000 0.0000

Wu-Hausman test: P-val. 0.0022 0.0004

Durbin-Wu-Hausman test: P-val. 0.0022 0.0004 Estimates of the coefficients of the excluded instruments shown in bold are significant at 10%, 5% or 1%.

Table 2.2e. First-stage IV estimates, double-logarithmic specification of the demand functions (Construction materials, First-stage G2SLS regression)

Imported (lnpd) Domestic (lnpd)

Coef. t–statistic Coef. t-statistic

lnpm 0.590 16.55 0.672 20.04

lny –0.052 –2.53 -0.062 -4.41

Metropol –0.225 –1.93 -0.263 -2.15

South –0.163 –2.55 -0.197 -3.03

Expenditure on services 0.163 3.28 0.152 3.31

Value of food basket 0.002 2.95 0.002 2.8

Imputed income tax –0.077 –2.65 -0.085 -2.86

Constant –0.080 –0.18 -0.223 -0.51

Observations 307 338

Prob > chi2 0.0000 0.0000

Wu–Hausman test: P–val. 0.0279 0.0635

Durbin–Wu–Hausman test: P–val. 0.0267 0.0612 Estimates of the coefficients of the excluded instruments shown in bold are significant at 10%, 5% or 1%.

44

Table 2.2f. First-stage IV estimates, double-logarithmic specification of the demand functions (Footwear)

Imported (OLS regression with robust s.e.)

D.lnpd D.lny

Coef. t-statistic Coef. t-statistic

D.lnpm 0.448 5.12 8.872 4.58 D. CPI of non-food goods 1.600 5.75 -9.530 -2.15 D. Producer prices –0.237 –2.16 3.494 2.17 D. Price of gasoline –0.060 –1.3 -2.353 -2.44

Observations 1101 1101 Partial R-squared of excluded instruments Shea Partial R2 0.05 0.015 Partial R2 0.11 0.032 F(13, 276) 22.95 6.34 P-value 0.0000 0.0003 Wu–Hausman test: P–val. 0.0030 0.0007 Durbin–Wu–Hausman test: P–val. 0.0030 0.0007

Estimates of the coefficients of the excluded instruments shown in bold are significant at 10%, 5% or 1%.

Table 2.3a. IV estimates, double-logarithmic specification of the demand functions (Textile)

Imported Domestic 2SLS, robust (lny, lnpm, lnpd) G2SLS RE-IV (lny, lnpd)

lnpm –0.947*** [4.05] 1.369* [1.71] lnpd 0.362** [2.06] –2.681*** [2.95] lny 1.237*** [20.93] 0.214 [1.50] North_West –0.375** [2.40] 0.616** [2.23] Volga –0.537*** [4.45] 1.280*** [3.67] Ural –1.009*** [5.08] 1.674*** [3.70] Oil producing region 0.428** [2.54] –1.270** [2.44] Constant –3.356*** [3.73] 2.966*** [2.68] Observations 294 264 Number of cross-sections 85 84 R-squared 0.84 0.33

Test lny = 1: Prob > F 0.0001 0.0000

lny + lnpd + lnpm = 0: Prob > F 0.0004 0.0000 Overidentification test: P-val. 0.1329 0.2462

t statistics in brackets, * — significant at 10%; ** — significant at 5%; *** — significant at 1%. Estimates of the elasticities shown in bold are of the expected sign and significant at 10%, 5% or 1%.

45

Table 2.3b. IV estimates, double-logarithmic specification of the demand functions (Clothing)

Imported (FE-IV, lnpm, lnpd) Domestic (FE-IV, lnpm, lnpd)

lnpm –8.378*** [2.91] 17.359** [2.35]

lnpd 5.663** [2.48] –13.871** [2.32]

lny 1.026*** [72.64] 0.655*** [14.22]

Constant 10.377*** [2.65] –27.146*** [2.92]

Observations 5267 4536

Number of cross-sections 1373 1329

R-squared 0.54 0.10

Test

lny = 1: Prob > F 0.0641 0.0000

lny + lnpd + lnpm = 0: Prob > F 0.0061 0.005

Overidentification test: P-val. 0.8889 0.3483 t statistics in brackets, * — significant at 10%; ** — significant at 5%; *** — significant at 1%. Estimates of the elasticities shown in bold are of the expected sign and significant at 10%, 5% or 1%.

Table 2.3c. IV estimates, double-logarithmic specification of the demand functions (Footwear)

Imported (2SLS, FD, robust, cluster; d.lnpd d.lny)

lnpm –2.814*** [3.08]

lnpd 1.243* [1.83]

lny 1.147*** [18.19]

Observations 1101

Number of cross-sections 461

R-squared 0.96

Test

lny = 1: Prob > F 0.0199

lny + lnpd + lnpm = 0: Prob > F 0.1099

Overidentification test: P-val. 0.5031

t statistics in brackets, * — significant at 10%; ** — significant at 5%; *** — significant at 1%. Estimates of the elasticities shown in bold are of the expected sign and significant at 10%, 5% or 1%.

46

Table 2.3d. IV estimates, double-logarithmic specification of the demand functions (Furniture)

Imported Domestic

2SLS, robust, cluster, lny G2SLS RE-IV, lnpm, lnpd

lnpm –1.150*** [3.24] –0.173 [0.55]

lnpd 0.484 [1.30] –1.019* [1.95]

lny 0.488*** [2.98] 0.962*** [18.80]

Constant –0.144 [0.09] 1.548 [0.70]

Observations 85 189

Number of cross-sections 47 72

R-squared 0.48 0.75

Test

lny = 1: Prob > F 0.0031 0.4602

lny + lnpd + lnpm = 0: Prob > F 0.0000 0.0000

Overidentification test: P-val. 0.1618 0.5854 t statistics in brackets, * — significant at 10%; ** — significant at 5%; *** — significant at 1%. Estimates of the elasticities shown in bold are of the expected sign and significant at 10%, 5% or 1%.

Table 2.3e. IV estimates, double-logarithmic specification of the demand functions (Electric household appliances)

Imported (G2SLS RE-IV, lny) Domestic (G2SLS RE-IV, lny)

lnpm –1.200*** [9.21] 0.350* [1.70]

lnpd 0.143 [1.41] –1.024*** [6.85]

lny 1.238*** [13.83] 0.479*** [6.27]

p3 0.431*** [4.07] –0.578*** [3.77]

p6 –0.356* [1.83] 0.525* [1.68]

p2 –0.741*** [3.70]

Constant –1.483*** [3.31] –0.496 [0.59]

Observations 716 857

Number of cross-sections 262 279

R-squared 0.76 0.66

Test

lny = 1: Prob > F 0.0080 0.0000

lny + lnpd + lnpm = 0: Prob > F 0.0000 0.0000

Overidentification test: P-val. 0.3966 0.9146 t statistics in brackets, * — significant at 10%; ** — significant at 5%; *** — significant at 1%. Estimates of the elasticities shown in bold are of the expected sign and significant at 10%, 5% or 1%.

47

Table 2.3f. IV estimates, double-logarithmic specification of the demand functions (Vehicles)

Imported (RE, GLS) Domestic (RE, GLS)

lnpm –1.121*** [9.86] –0.290** [2.52]

lnpd 0.239* [1.88] –0.681*** [5.15]

lny 0.915*** [31.57] 0.983*** [33.30]

Constant –0.739* [1.95] –0.149 [0.54]

Observations 49 120

Number of cross-sections 39 70

R-squared 0.96 0.91

Test

lny = 1: Prob > F 0.0034 0.5577

lny + lnpd + lnpm = 0: Prob > F 0.0000 0.0000

t statistics in brackets, * — significant at 10%; ** — significant at 5%; *** — significant at 1%. Estimates of the elasticities shown in bold are of the expected sign and significant at 10%, 5% or 1%.

Table 2.3g. IV estimates, double-logarithmic specification of the demand functions (TV sets)

Imported (FGSLS) Domestic (FGSLS)

lnpm –1.521*** [8.28] –0.213 [0.14]

lnpd 0.862*** [2.63] –4.480* [1.92]

lny 1.040*** [61.74] 0.457*** [5.32]

Constant –3.254 [1.29] 33.280** [2.01]

Observations 157 130

Number of cross–sections 48 46

Test

lny = 1: Prob > F 0.0180 0.0000

lny + lnpd + lnpm = 0: Prob > F 0.0386 0.0075

t statistics in brackets, * — significant at 10%; ** — significant at 5%; *** — significant at 1%. Estimates of the elasticities shown in bold are of the expected sign and significant at 10%, 5% or 1%.

48

Table 2.3h. IV estimates, double-logarithmic specification of the demand functions (Construction materials)

Imported (G2SLS RE-IV, lnpd) Domestic (G2SLS RE-IV, lnpd)

lnpm –1.545*** [4.77] 0.572 [1.59]

lnpd 1.032** [2.07] –1.828*** [3.65]

lny 1.105*** [20.29] 0.682*** [15.02]

Metropol 0.630** [2.08] –1.043*** [3.35]

South 0.578*** [2.68] –0.673*** [3.00]

Constant –3.281*** [3.87] 2.045*** [2.77]

Observations 307 338

Number of cross-sections 82 87

R-squared 0.66 0.69

Test

lny = 1: Prob > F 0.0539 0.0000

lny + lnpd + lnpm = 0: Prob > F 0.0784 0.0000

Overidentification test: P-val. 0.6382 0.6868 t statistics in brackets, * — significant at 10%; ** — significant at 5%; *** — significant at 1%. Estimates of the elasticities shown in bold are of the expected sign and significant at 10%, 5% or 1%.

Table 2.4. Estimates of Price and Income elasticities obtained with the use of the regional data (summary of Table 2.3)

Imported goods Domestic goods

Own-price elasticity

Cross-price elasticity

Income elasticity

Own-price elasticity

Cross-price elasticity

Income elasticity

ηmm ηmd ηmy ηdd ηdm ηdy

Textile –0.95 0.36 1.24 –2.68 1.37 0.21

Clothing –8.38 5.66 1.03 –13.87 17.36 0.66

Footwear –2.81 1.24 1.15

Furniture –1.15 0.48 0.49 –1.02 –0.17 0.96

Electric household appliances –1.2 0.14 1.24 –1.02 0.35 0.48

Vehicles –1.12 0.24 0.92 –0.68 –0.29 0.98

TV sets –1.52 0.86 1.04 –4.48 –0.21 0.46

Construction materials –1.55 1.03 1.11 –1.83 0.57 0.68 Estimates of the elasticities shown in bold are of the expected sign and significant at 10%, 5% or 1%.

49

Table 2.5. Armington elasticities

Text

ile

Clo

thin

g

Foot

wea

r

Furn

iture

Elec

tric

hous

ehol

d ap

plia

nces

TV se

ts

Con

stru

ctio

n m

ater

ials

RE RE RE RE RE RE RE

lnpr –0.976*** –0.838*** –0.675*** –1.367** –1.343*** –4.555* –1.095*** [3.29] [9.28] [3.68] [2.08] [5.19] [1.74] [4.32] Central –1.683*** –2.073*** –1.425*** [4.56] [15.36] [8.64] North_West –1.867*** –2.554*** –2.181*** 0.903* [4.66] [18.18] [12.53] [1.84] South –0.523*** –1.483* [3.65] [1.91] Volga –1.564*** –2.211*** –1.567*** –2.592 –0.516* [4.05] [16.42] [9.29] [1.50] [1.80] Ural –2.894*** –2.041*** –1.758*** [5.42] [11.50] [7.73] Siberia –0.798*** –0.471*** [5.62] [2.59] Moscow –0.95 0.547** [1.30] [2.09] Metropol 0.528*** 1.656*** [3.37] [2.91] Oil producing region 1.451*** 0.417** 0.624*** [2.59] [2.44] [2.66] Constant 2.076*** 3.853*** 3.003*** 0.471 1.024*** 2.567*** –0.055 [7.57] [31.42] [18.80] –0.93 [3.42] [4.14] [0.29] Observations 230 4542 1261 59 590 116 303 Number of cross–sections 76 1333 455 36 224 45 82 R–squared 0.37 0.36 0.24 0.13 0.18 0.10 0.15 Test Var(u) = 0: Prob > chi2 0.0003 0.0000 0.0000 0.5152 0.0870 0.6277 0.0000 Hausman specification test: Prob>chi2 0.1054 0.0000 0.277 0.8041 0.6005 0.8365

Test lnpr = –1: Prob > F 0.935 0.0919 0.0765 0.5755 0.1848 0.1748 0.7084 Absolute value of z-statistics in parentheses, * — significant at 10%; ** — significant at 5%; *** — significant at 1%. Product dummies are not reported. Estimates of the elasticities shown in bold are of the expected sign and significant at 10%, 5% or 1%.

50

Table 2.6a. First-stage IV estimates, Armington elasticities (Textile, OLS regression with robust standard errors)

Coef. t-statistic Central 0.125 1.48 North_West 0.045 0.53 Volga –0.231 -2.74 Ural 0.108 1.12 Moscow –0.691 -5.23 Oil producing region 0.401 3.02 Producer prices –0.624 -4.3 Gross regional product per capita 0.315 4.41 Imputed income tax 0.104 3.11 Constant –0.003 0.000 Observations 221 Partial R2 0.15 F(3, 209) 14.30 P-value 0.0000 Wu-Hausman test: P-val. 0.5414 Durbin-Wu-Hausman test: P-val. 0.5307

Estimates of the coefficients of the excluded instruments shown in bold are significant at 10%, 5% or 1%.

Table 2.6b. First-stage IV estimates, Armington elasticities (Clothing, First-stage G2SLS regression)

Coef. t-statistic Central –0.216 –7.59 North_West –0.172 –7.05 South 0.026 0.86 Volga –0.136 –4.52 Ural –0.123 –3.82 Siberia –0.186 –7.43 Moscow 0.040 0.75 Metropol 0.054 1.95 Central 0.073 2.23 Real industrial production –0.055 –9.41 Turnover of retailers 0.138 6.59 Average wage –0.052 –2.15 VAT tax 0.019 2.71 Border with non–CIS –0.126 –6.34 Border 0.036 2.71 Constant 0.734 4.03 Observations 4515 Prob > chi2 0.0000 Wu-Hausman test: P-val. 0.6731 Durbin-Wu-Hausman test: P-val. 0.6722

Estimates of the coefficients of the excluded instruments shown in bold are significant at 10%, 5% or 1%.

51

Table 2.6c. First-stage IV estimates, Armington elasticities (Footwear, First-stage G2SLS regression)

Coef. t-statistic

Central –0.144 –5.25 North_West –0.198 –6.88 Volga 0.008 0.25 Ural –0.071 –1.81 Siberia –0.043 –1.22 Oil producing region 0.088 2.13 Border 0.022 1.23 Sea_border –0.096 –3.29 Dummy for Q2 –0.028 –1.85 Dummy for Q3 –0.009 –0.68 CPI for non-food goods –0.376 –3.32 Average wage 0.093 2.38 January temperature –0.071 –4.13 Output of small transport business –0.037 –4.44 Budgetary revenue 0.097 5.06 Producer prices –0.179 –3.94 Constant 1.773 3.31 Observations 1164 Prob > chi2 0.0000 Wu-Hausman test: P-val. 0.2369 Durbin-Wu-Hausman test: P-val. 0.2337

Estimates of the coefficients of the excluded instruments shown in bold are significant at 10%, 5% or 1%.

Table 2.6d. First-stage IV estimates, Armington elasticities (Furniture, OLS regression with robust standard errors)

Coef. t–statistic

p3 –0.131 –0.73 Volga –0.046 –0.1 F25R 0.008 2 Real industrial production 0.447 3.22 Gross regional product per capita –0.368 –1.42 % of fully depreciated assets in transport 0.036 2.05 Constant 0.845 0.47 Observations 51 Partial R2 0.35 F(4, 26) 6.41 P-value 0.0010 Wu-Hausman test: P-val. 0.1401 Durbin-Wu-Hausman test: P-val. 0.1227

Estimates of the coefficients of the excluded instruments shown in bold are significant at 10%, 5% or 1%.

52

Table 2.6e. First-stage IV estimates, Armington elasticities (Electric household appliances, First-stage G2SLS regression)

Coef. t-statistic p6 –1.121 –28.98 p2 –0.503 –13.41 p3 –0.454 –12.29 Budgetary expenditure on industry –0.021 –2.68 Production of refrigerators –0.001 –2.16 Retail sales of refrigerators 0.000 –4.76 Volga 0.087 2.63 Constant 1.172 42.36 Observations 554 Prob > chi2 0.0000 Wu-Hausman test: P-val. 0.0000 Durbin-Wu-Hausman test: P-val. 0.0000

Estimates of the coefficients of the excluded instruments shown in bold are significant at 10%, 5% or 1%.

Table 2.6f. First-stage IV estimates, Armington elasticities (TV sets, OLS regression with robust standard errors)

Coef. t-statistic North_West 0.011 0.71 Metropol 2.445 4.57 Regional territory –0.061 –12.1 Border non_CIS 0.126 7.02 Border –0.071 –6.16 Foreign Investment 0.008 2.52 Deposit with Sberbank 0.115 4.73 Imputed income tax –0.019 –2.54 Exports 0.015 3.22 Output of small trade business 0.064 4.71 % of fully depreciated assets in transport –0.003 –2.95 Constant –0.763 –3.66 Observations 102 Partial R2 0.73 F(11, 89) 25.81 P-value 0.0000 Wu-Hausman test: P-val. 0.4193 Durbin-Wu-Hausman test: P-val. 0.4096

Estimates of the coefficients of the excluded instruments shown in bold are significant at 10%, 5% or 1%.

Table 2.6g. First-stage IV estimates, Armington elasticities (Construction materials, First-stage G2SLS regression)

Coef. t-statistic Metropol 0.248 1.87 Value of food basket –0.002 –3.06 Imputed income tax 0.076 2.3 Regional territory 0.033 1.54 Constant 0.458 3.67 Observations 303 Prob > chi2 0.0028 Wu-Hausman test: P-val. 0.1742 Durbin-Wu-Hausman test: P-val. 0.1716

Estimates of the coefficients of the excluded instruments shown in bold are significant at 10%, 5% or 1%.

53

Table 2.7. IV estimates, Armington elasticities

Text

ile, 2

SLS,

robu

st

Clo

thin

g, G

2SLS

RE-

IV

Foot

wea

r, G

2SLS

RE-

IV

Furn

iture

, 2S

LS, r

obus

t, cl

uste

r

Elec

tric

hous

ehol

d ap

plia

nces

, G

2SLS

RE-

IV

TV se

ts,

2SLS

, rob

ust

Con

stru

ctio

n m

ater

ials

, G

2SLS

RE-

IV

lnpr –1.360* –1.289*** –1.838** –2.646** –3.096*** –6.288** –2.531** [1.95] [2.78] [2.24] [2.45] [7.61] [2.19] [2.10] Central –1.650*** –2.281*** –1.706*** [5.58] [14.28] [7.98] North_West –1.761*** –2.706*** –2.568*** 1.117** [4.54] [16.97] [11.21] [2.43] South –1.686*** –0.557*** [5.37] [3.79] Volga –2.651*** –2.392*** –1.644*** –2.759*** [6.14] [16.10] [9.34] [4.54] Ural –1.650*** –2.166*** –1.935*** [5.58] [11.65] [8.06] Siberia –0.928*** –0.669*** [5.78] [3.30] Moscow –0.992*** 0.669** [2.68] [2.43] Metropol 0.480*** 1.930*** [3.05] [3.16] Oil producing region 1.832*** 0.422** 0.750*** [3.21] [2.47] [3.03] border –2.281*** –0.204* [14.28] [1.88] Constant 2.211*** –2.706*** 3.438*** 1.332** 3.047*** 2.892*** 0.71 [6.48] [16.97] [11.11] [2.42] [7.74] [4.61] [0.97] Observations 221 4515 1164 51 563 102 303 Number of cross-sections 1331 423 33 208 82 R-squared 0.37 0.37 0.23 0.08 0.15 0.1 0.13 Test lnpr = –1: Prob > F 0.6063 0.5330 0.3081 0.1390 0.0000 0.0763 0.2046 Overidentification test: P-val. 0.4287 0.0539 0.0024 0.3015 0.1794 0.8046 0.4658

Absolute value of z-statistics in parentheses, * — significant at 10%; ** — significant at 5%; *** — significant at 1%. Product dummies are not reported.

54

A3. Estimations using the data at the national level of aggregation Table 3.1. Own-price Elasticities, double-logarithmic specification

Deciles Textiles, FGLS (PC)

Clothing, FGLS (PH)

Footwear,FGLS(PH)

Furniture,FGLS (PH)

Electric household

applicances,FGLS (PH)

Vehicles, OLS, PCSE

Construction materials,

FGLS (PH)

Domestic goods Overall households –1.19 –0.88 1.36 –0.42 –0.07 –1.52 –0.95

1 –1.80 –0.50 0.28 –1.08 –0.75 –0.83 2 –1.52 –0.76 –1.07 –0.82 –1.65 –0.86 3 –3.52 –0.11 –0.25 –1.04 0.23 –0.72 4 –0.65 –0.84 –1.54 –1.27 –0.67 –0.33 5 –1.01 –0.89 6.77 –1.21 –1.42 –0.89 6 –1.25 –1.73 –0.12 –1.24 0.22 –0.33 7 –2.47 –0.27 6.17 –1.14 –1.79 –0.47 8 –1.29 –2.14 –5.18 –1.71 –2.62 –0.43 9 0.27 –0.17 1.90 –0.98 –0.89 0.44

10 –0.75 –1.35 0.72 0.67 0.88 –0.99 0.01 Imported goods

Overall households –1.14 –1.16 –0.85 –2.85 –1.05 –2.95 –0.19

1 –1.85 –0.97 –0.79 –1.01 –1.43 2 –1.23 –0.95 –1.17 –0.21 –0.79 –2.82 3 –1.89 –0.87 –0.86 0.13 –0.73 0.15 4 –1.24 –0.99 –1.16 –1.16 –0.77 0.77 5 –0.43 –1.21 1.33 –2.12 0.15 0.55 6 –0.79 –1.50 –0.70 –1.12 –0.98 0.54 7 –1.47 –1.14 –0.90 –3.95 –4.89 0.39 8 –1.26 –1.21 –1.57 –3.42 –2.19 0.43 9 –1.13 –1.07 –0.79 –2.77 0.81 0.01

10 –0.80 –1.03 –0.87 –0.49 –0.95 –0.83 0.36 Domestic goods (average across deciles)

Five poor deciles –1.70 –0.62 0.84 –1.09 –0.85 –0.73 Five rich deciles –1.10 –1.13 0.70 –0.88 –0.84 –0.16

Imported goods (average across deciles) Five poor deciles –1.33 –1.00 –0.53 –0.84 –0.63 –0.56 Five rich deciles –1.09 –1.19 –0.96 –2.35 –1.64 0.35

Estimates shown in bold are of the expected sign and significant at 5% significance level, those shown both in bold and in italic bold are of the expected sign and significant at 10% significance level. FGLS — Feasible Generalized Least Square, PH — Panel (cross-section) heteroskedasticity, PC — Panel cross-section correlation, PCSE — Panel Corrected Standard Errors (Beck and Katz (1995)). In about half of the cases FGLS and PCSE correct also for the presence of the serial correlation in the error terms. Blank cells mean the lack of sufficient number of observations to perform estimations.

55

Table 3.2. Cross-price Elasticities, double-logarithmic specification

Deciles Textiles, FGLS (PC)

Clothing, FGLS (PH)

Footwear,FGLS(PH)

Furniture,FGLS (PH)

Electric household applicances, FGLS (PH)

Vehicles, OLS, PCSE

Construction materials,

FGLS (PH)

Domestic goods

Overall households 0.29 0.05 –2.35 –0.33 1.36 0.35 0.01

1 0.56 –0.06 –1.27 0.08 –0.19 –0.13

2 0.16 0.03 0.83 –0.12 2.52 –0.10

3 1.03 –0.57 –0.69 0.09 2.68 –0.23

4 0.19 0.09 1.23 0.29 –0.41 –0.71

5 –0.51 0.54 –8.83 0.21 –1.93 –0.19

6 0.76 1.49 –0.39 0.27 –0.94 –0.08

7 1.10 0.53 –8.56 0.41 4.69 0.62

8 0.33 1.68 5.96 0.67 2.40 –0.47

9 –0.95 –0.46 –2.30 0.08 –1.96 –0.75

10 0.05 0.87 –1.56 –1.18 1.16 0.03 0.64

Imported goods

Overall households 0.08 0.09 –0.21 1.74 –0.64 1.34 –1.65

1 0.75 –0.14 –0.22 0.03 –0.35

2 0.39 –0.07 0.05 –1.58 –0.31 0.36

3 1.34 –0.17 –0.15 –1.96 –0.28 –1.49

4 –0.06 –0.03 0.07 0.55 –0.19 –1.99

5 –0.33 0.08 –1.99 1.50 1.41 –1.85

6 –0.36 0.34 –0.30 0.67 –0.13 –2.10

7 0.50 –0.03 –0.09 2.81 0.19 –2.19

8 0.33 0.15 0.41 2.74 –0.42 –1.94

9 –0.03 0.00 –0.25 1.89 –0.09 –1.82

10 –0.33 –0.02 –0.14 –0.58 –0.61 0.15 –2.43 Estimates shown in bold are significant at 5% significance level, those shown both in bold and in italic bold are significant at 10% significance level. FGLS — Feasible Generalized Least Square, PCSE — Panel Corrected Standard Errors (Beck and Katz, 1995). Blank cells mean the lack of sufficient number of observations to perform estimations.

56

Table 3.3. Income Elasticities, double-logarithmic specification

Deciles Textiles, FGLS (PC)

Clothing, FGLS (PH)

Footwear,FGLS(PH)

Furniture,FGLS (PH)

Electric household

applicances, FGLS (PH)

Vehicles, OLS, PCSE

Construction materials,

FGLS (PH)

Domestic goods

Overall households 0.90 1.06 0.91 0.84 0.13 1.12 0.97

1 1.01 0.86 0.90 1.00 0.79 0.98

2 1.10 0.96 0.88 1.00 0.64 0.95

3 1.12 0.95 0.92 0.98 0.91 0.93

4 0.91 0.94 0.94 1.02 0.71 1.07

5 1.02 0.82 0.89 1.03 0.84 1.01

6 0.20 0.93 0.82 1.05 0.28 0.78

7 0.85 0.71 0.86 0.79 0.08 0.59

8 1.37 0.98 0.95 1.15 0.64 0.91

9 0.05 0.89 0.74 1.01 0.66 0.77

10 0.88 0.88 0.93 0.63 0.25 0.94 0.76

Imported goods

Overall households 1.02 0.99 1.01 0.75 1.31 0.28 1.29

1 0.99 1.01 1.00 0.87 1.01

2 1.00 1.01 1.01 –0.38 0.74 1.60

3 0.91 1.01 1.02 0.71 0.96 1.11

4 1.11 1.00 1.01 0.60 1.17 0.88

5 1.03 1.03 1.02 0.38 1.08 1.03

6 1.23 1.00 1.02 0.09 1.20 1.16

7 1.10 1.04 1.02 0.85 1.29 1.14

8 0.83 1.00 1.00 0.34 1.11 1.16

9 1.27 1.01 1.03 0.63 1.14 1.11

10 1.07 1.00 1.00 0.79 1.11 0.67 1.06

Domestic goods (average across deciles)

Five poor deciles 1.03 0.90 0.91 1.01 0.78 0.99

Five rich deciles 0.67 0.88 0.86 0.93 0.38 0.76

Imported goods (average across deciles)

Five poor deciles 1.01 1.01 1.01 0.33 0.96 1.13

Five rich deciles 1.10 1.01 1.01 0.54 1.17 1.12 Estimates shown in bold are significant at 5% significance level, those shown both in bold and in italic bold are significant at 10% significance level. FGLS — Feasible Generalized Least Square, PCSE — Panel Corrected Standard Errors (Beck and Katz, 1995). Blank cells mean the lack of sufficient number of observations to perform estimations.

57

Table 3.4. Own-price Elasticities, Linear Approximation to AIDS

Deciles Textiles, FGLS (PC)

Clothing, FGLS (PH)

Footwear, FGLS(PH)

Furniture,FGLS (PH)

Electric household

applicances,FGLS (PH), OLS, PCSE

Vehicles, OLS,PCSE

Construction materials,

FGLS (PH)

Domestic goods

Overall households –1.25 –1.30 0.04 –1.37 0.03 –1.44 –0.97

1 –1.89 –0.52 0.16 –0.71 –0.53 –0.82

2 –1.78 –0.82 –1.37 –0.89 –0.55 –0.86

3 –3.89 –0.41 –0.36 –1.33 –0.67 –1.78 –0.71

4 –0.82 –0.88 –1.60 –1.21 –0.58 5.56 –0.45

5 –0.87 –1.20 9.84 –1.29 –1.72 –0.93 –0.86

6 –0.81 –2.43 0.81 –1.45 –0.99 –0.53 –0.39

7 –2.55 –0.77 –0.45 –1.73 –0.76 –1.23 –0.69

8 –1.49 –1.89 –4.28 –1.37 –2.27 –0.49 –0.58

9 –1.52 –0.80 0.71 –0.94 –1.36 –6.80 –0.58

10 –0.56 –1.00 0.03 –1.37 –0.14 –1.00 –0.37

Imported goods

Overall households –1.15 –1.13 –0.79 –1.99 –1.05 –2.12 –0.97

1 –1.59 –0.95 –0.74 –0.24 –0.54 –0.21

2 –1.28 –0.97 –1.19 –0.64 –0.45 –0.33

3 –1.81 –0.88 –0.88 –3.22 –3.07 –2.60 0.21

4 –1.14 –0.99 –1.22 –2.15 –0.67 1.46 2.69

5 –0.66 –1.18 1.33 –2.52 0.80 –0.81 0.02

6 –1.11 –1.48 –0.71 –2.80 –1.06 –0.42 0.77

7 –1.78 –1.13 –0.91 –3.07 –2.51 –1.65 –0.27

8 –1.17 –1.22 –1.54 –2.17 –2.28 –3.66 –0.19

9 –1.41 –1.01 –0.78 –0.99 1.11 –9.76 –0.16

10 –0.98 –1.05 –0.88 –1.99 –1.42 –1.06 –1.74 Elasticities shown in bold are of the expected sign and correspond to the estimated coefficients of the LA-AIDS, which are significant at 5% significance level, those shown both in bold and in italic bold are of the expected sign and correspond to the coefficients, which are significant at 10% significance level. FGLS — Feasible Generalized Least Square, PH — Panel (cross-section) heteroskedasticity, PC — Panel cross-section correlation, PCSE — Panel Corrected Standard Errors (Beck and Katz, 1995). In about half of the cases FGLS and PCSE correct also for the presence of the serial correlation in the error terms. Blank cells mean the lack of sufficient number of observations to perform estimations.

58

Table 3.5. Armington Elasticities

Deciles Textiles, FGLS (PC)

Clothing, FGLS (PH)

Footwear,FGLS (PH)

Furniture,FGLS (PH)

Electric household appliances,FGLS (PH)

Vehicles, OLS, PCSE

Construction materials,

FGLS (PH)

Average –1.58 –0.74 2.22 –2.01 –0.93 –0.78 –0.08

(–4.15) (–1.79) (1.28) (–6.61) (–0.39) (–0.92) (–0.94)

1 –2.40 0.12 0.07 –0.98 0.19

(–5.27) (0.20) (0.23) (–3.17) (0.04)

2 –1.16 –0.60 –1.61 1.33 –2.62 –0.69

(–1.58) (–1.28) (–1.02) (0.92) (–1.99) (–0.15)

3 –2.51 0.14 –0.26 –3.78 –8.25 0.77

(–4.54) (0.30) (–0.57) (–5.22) (–3.42) (2.48)

4 –1.75 –0.90 0.35 –2.52 –0.24 0.94

(–2.46) (–1.67) (0.31) (–1.88) (–0.98) (3.04)

5 –0.31 –0.17 0.27 –1.24 –16.29 0.98

(–0.52) (–0.29) (0.53) (–0.98) (–1.58) (1.95)

6 –2.10 –1.23 0.83 0.47 –0.94 1.81

(–2.07) (–1.37) (1.46) (0.36) (–3.31) (2.33)

7 –2.99 1.13 0.95 –4.82 –1.03 1.18

(–3.94) (1.71) (0.33) (–3.66) (–2.51) (0.97)

8 –1.40 –1.98 0.01 –2.18 –1.03 1.49

(–2.68) (–2.79) (0.02) (–1.89) (–0.14) (1.75)

9 –1.41 0.28 1.00 –2.53 –0.57 –0.08

(–1.30) (0.31) (1.05) (–2.22) (–0.19) (–0.04)

10 –0.98 –0.64 0.72 0.95 0.34 –1.13 –0.22

(–1.65) (–0.77) (1.12) (0.90) (0.24) (–1.22) (–0.06)

Average across deciles

Five poor deciles –1.63 –0.28 –0.24 –1.65 –5.67 0.44

Five rich deciles –1.78 –0.49 0.70 –1.62 –0.65 0.84 t-statistics are in brackets. Estimates shown in bold are of the expected sign and significant at 5% significance level, those shown both in bold and in italic bold are of the expected sign and significant at 10% significance level. FGLS — Feasible Generalized Least Square, PH — Panel (cross-section) heteroskedasticity, PC — Panel cross-section correlation, PCSE — Panel Corrected Standard Errors (Beck and Katz, 1995). In about half of the cases FGLS and PCSE correct also for the presence of the serial correlation in the error terms. Blank cells mean the lack of sufficient number of observations to perform estimations.

References Aivazain S, Kolenikov S. (2001) Poverty and Expenditure Differentiation of the Russian Population, Economic

Education and Research Consortium, Working Paper No. 01/01.

Armington P. (1969) A Theory of Demand for Products Distinguished by Place of Productions, IMF Staff Papers 16,

159–178.

Arellano M. (2003) Panel Data Econometrics (Oxford University).

Baltagi B.H. (2001) Econometric Analysis of Panel Data (John Wiley & Sons, Ltd).

Baum, C.F., M.E. Schaffer, and S. Stillman (2003), "Instrumental Variables and GMM: Estimation and Testing",

Boston College, Department of Economics, Working Paper No.545.

Brown, D., and Earle J. (2001), "Competition-Enhancing Policies and Infrastructure: Evidence from Russia", SITE

Working Paper, No 161.

Deaton A., Muellbauer J. (1980a) An Almost Ideal Demand System, The American Economic Review 70 (6), 312–

326.

Deaton A., Muellbauer J. (1980b) Economics and Consumer Behavior (Cambridge University Press).

Deaton A. (1987) Estimation of Own- and Cross-Price Elasticities from Household Budget Data, Journal of

Econometrics 36, 7–30.

Deaton A. (1988) Quantity Quality and Spatial Variation of Price, The American Economic Review 78 (3), 418–430.

Deaton A. (1990) Price Elasticities from Survey Data: Extensions and Indonesian Results, Journal of Econometrics

44, 281–309.

Green W.H. (1997) Econometric Analysis (Upper Saddle River N.J.: Prentice Hall).

Glushenko K. (2002) Common Russian Market: Myth rather than Reality, EERC Working Paper, No 01/11.

Goldstein M., Khan M. (1985) Income and Price Effect in Foreign Trade, in R. Jones and P. Kenen, eds., Handbook

of International Economics (Amsterdam, North-Holland) 1042–1099.

Hsiao C. (2003) Analysis of Panel Data (Cambridge University Press).

McDaniel C.A., Balisteri E.J. (2002) A Discussion on Armington Trade Substitution Elasticities, U.S. Trade

Commission, Working Paper, No.2002-01-A.

Ovcharova L., Turuntsev E., Korchagina I. (1998) Indicators of Poverty in Transitional Russia, Economic Education

and Research Consortium, Working Paper No. 98/04.

Panagariya A., Shah S., Mishra D. (1996) Demand Elasticities in International Trade: Are They Really Low? The

World Bank Policy Research Working Papers, 1712.

Reidel J. (1988) The Demand for LDC Exports of Manufactures: Estimates from Hong Kong, The Economic Journal

98 (March 1988), 138–148.

Reidel J., and Athukorala P. (1994) Demand and Supply Factors in the Determination of NIE: A Simultaneous Error-

Correction Model for Hong Kong: A Comment, The Economic Journal XX (November 1994), 1411–1413.

Shiells C.R., Deadorff A., Stern A. (1986) Estimates of the Elasticities of Substitution between Imports and Home

Goods for the United States, Weltwirtschaftliches Archiv 122 (2), 497–519.

Shiells C.R., Reinert K.A. (1993) Armington models and terms-of-trade effects: some econometric evidence for

North America", Canadian Journal of Economics XXVI (2), 299–316.

Shuguang Z., Yansheng Z., and Zhongxin W. (1999) Measuring the Costs of Protection in China (Institute for

International Economics).

Stavrev E., Kambourov G. (1999a) Estimation of Income, Own- and Cross- Price Elasticities: An Application for

Bulgaria, Institute for Advanced Studies, Vienna, Transition Economic Series, No. 6.

Stavrev E., Kambourov G. (1999b) Share Equations versus Double Logarithmic Functions in the Estimation of

Income, Own- and Cross- Price Elasticities: An Application for Bulgaria, Transition Economic Series, No.7

(Institute for Advanced Studies, Vienna).

Stern R.M., Francis J., Schumacher B. (1976) Price Elasticities in International Trade (Trade Policy Research

Centre, London: The MacMillan Press Ltd).

Theil H., Clements K.W. (1987) Applied Demand Analysis: Results from System-Wide Approaches, Volume 7 of

Series in Econometrics and Management Sciences (Ballinger Publishing Company, Cambridge, Massachusetts).

Varian H.R. (1982) The Nonparametric Approach to Demand Analysis, Econometrica 50, 945–973.

Varian H.R. (1982) Nonparametric Tests of Consumer Behaviour, The Review of Economic Studies 50, 99–110.

Winters L.A. (1984) Separability and Specification of Foreign Trade Functions, Journal of International Economics

17, 239–263.

Wooldridge J. (2002), "Econometric Analysis of Cross Section and Panel Data", MIT Press, Cambridge, MA

Дынникова O. (2001) Плюсы и минусы слабой эластичности российского импорта по отношению к

реальному обменному курсу, Публикации Экономической экспертной группы, www.eeg.ru/publications.html

60