Master Thesis Double Degree Program:
M.Sc. International Economics and Business – University of Groningen
M.A. International Economics – University of Göttingen
Global Sourcing and Import Price Pass-Through:
Relevance of the Import Sourcing Bias
Student Name: Esther Reichelt
Student Number: S2417839 (Groningen) 21157204 (Göttingen) Email: estherreichelt@t-online.de
Affiliation: University of Groningen, Faculty of Economics and Business University of Göttingen, Faculty of Economic Sciences
Global Sourcing and Import Price Pass-Through:
Relevance of the Import Sourcing Bias
Esther Reichelt
Supervisor: Dr. Robert Inklaar Co-Assessor: Dr. Laura Birg
July 5, 2013
Abstract
Academic research has recently paid increasing attention to possible overstatements in productivity growth due to mismeasurement in import price indices. This paper adopts a broader approach by investigating whether the bias in import price changes resulting from mismeasurement induced through domestic producers’ import sourcing shifts is systematically linked to domestic price dynamics. Executing a standard pass-through analysis, it cannot be confirmed that the results are significantly influenced by the bias. The findings are robust to various sample changes but stress the importance of a careful methodological basis of price index data.
i
Table of Content
Introduction ... 1
Theoretical Background ... 4
Input Sourcing Strategies ... 4
Pass-Through ... 7
Import Sourcing Bias ... 12
Methodology ... 15
The Pass-Through Model ... 15
Data ... 19
Econometric Analysis ... 22
Results ... 27
Baseline Specification ... 27
Disaggregation of Total Intermediate Inputs ... 31
Import Sourcing Bias ... 33
Robustness Checks ... 33
Discussion ... 38
1
Introduction
During the last decades, improvements in infrastructure, institutional reforms, structural changes and increasing economic and productive capabilities in emerging and developing economies have given rise to a new phenomenon of globalization. While the first period of globalization was characterized by domestic producers shifting the sourcing of intermediate inputs1 from domestic to foreign suppliers, a process commonly referred to as offshoring, further reductions in communication and transportation costs have led more recently to substantial changes in the optimal sourcing strategy. As wage demands in emerging economies rise, suppliers in developing countries become an increasingly valid and cost efficient alternative to those in emerging economies. Hence, domestic firms in industrialized countries increasingly shift the sourcing of intermediate inputs away from the economies initially targeted for offshoring and source from countries that are able to provide the required inputs at an even lower cost. The motivation for these shifts in intermediate input sourcing remains the same as for offshoring: to profit from factor price arbitrage made possible by wage differentials across countries (Houseman, 2008).
The increasing importance of offshoring and import sourcing shifts has motivated numerous studies to evaluate the economic consequences of these developments for the industrialized countries. Academic literature has been particularly interested in the implications for productivity, the labor market and overall welfare. However, some scholars have recently pointed out that productivity is overstated as cost savings from offshoring are not captured appropriately in import price statistics (Houseman, 2008; Houseman, Kurz, Lengermann & Mandel, 2012). Therefore, the standardly used import price data are commonly found to be unsuitable to evaluate productivity gains in a globalized world. In consequence, general economic statistics are likely to be distorted which can have a substantial impact on monetary and fiscal policies.
In particular, Mandel (2007) stresses how flawed import price data generate a “phantom GDP” (p.1). This indicates that the import price index biases affect domestic statistics not necessarily directly but rather indirectly, as distorted import prices are included in the calculation of domestic statistical measures. Hence, it is justified to take the pass-through
2
analysis as an empirical instrument to assess the transmission of the import price index measurement error.
Pass-through analysis typically investigates how the transfer of price changes to higher levels of the value chain is influenced by industrial organization and other economic and geographic factors. It is therefore used to investigate the transmission of shocks, as these often run through the price channel. The results of a pass-through analysis allow the determination of implications for producers, consumers and monetary policy in response to the shocks (Campa & Goldberg, 2005). The focus of this study is to determine how changes in import prices, including all potential measurement errors, are transferred to domestic output prices. This paper therefore takes a broader approach than initial studies dealing with import price index biases that focused mainly on productivity management to evaluate to what extent biases in import price indices can influence domestic statistics.
In this paper, I will draw on research of Inklaar (2013) who develops a measure for the import sourcing bias that approximates the extent of mismeasurement in the standard import price index due to domestic producers shifting sourcing from an expensive to a cheaper foreign supplier. The study is in line with the remarkable research of Houseman et al. (2012) who rely on confidential import value and intermediate input price data at a commodity level. Their highly disaggregate data allows them to capture the import price index bias resulting from offshoring in addition to the import sourcing bias calculated by Inklaar (2013). Houseman et al. (2012) therefore capture measurement errors resulting from both dynamics, a shift in sourcing of producers from domestic to cheaper foreign suppliers as well as from foreign suppliers to even cheaper foreign suppliers. Although the study of Inklaar (2013) captures only one specific aspect of the mismeasurement, his bias measure can be more easily implemented, since the study works with publicly available data. As the bias under consideration, i.e. the import sourcing bias, is still sizable, I consider the measure to be appropriate to evaluate the extent to which biases might affect economic evaluations.
3
assumed to overestimate import price changes the findings of this study can give an additional explanation for incomplete pass-through.
The analysis is based on a simple pass-through model that is derived from the growth accounting equation expressed in prices (dual growth accounting). Using predominantly data from the World Input-Output Database (WIOD) and import sourcing bias information from Inklaar (2013), an ordinary least squares (OLS) regression analysis is carried out. Contrary to the literature, results show a very high pass-through for intermediate input prices in general and domestic and imported intermediate input prices in particular. While substantial differences are found between different import price measures, for the most reliable, WIOD-based import price data the hypothesis of complete pass-through cannot be rejected. The proxy for the import sourcing bias is found to be insignificant in all implementations. The results prove to be surprisingly robust to a large variety of robustness checks.
While these unexpected results might be related to structural and data limitations in the research framework, the lack of statistical significance of the import sourcing bias indicates that the phenomenon of import sourcing substitution might not yet be prevalent enough in global trade patterns to have a sizeable economic and econometric impact. While further refinements to the research frame are required, the problem of import sourcing substitution and import price index biases in general is likely to grow in the near future and thus has to remain an important part of the future research agenda.
4
Theoretical Background
Offshoring as a process to relocate business activities to a foreign country (Olsen, 2006)2 is an important strategic option in a firm’s reaction to increasing competitive pressures in a more and more globalized and open world economy. Industrial upgrading and structural changes in traditional low-labor-cost countries could, however, increase wages and lead to shifts in international sourcing of firms to even cheaper suppliers (Gereffi, 2009). An analysis of price pass-through provides a way to determine how domestic consumers in the sourcing economies are affected by these developments. However, cost savings from shifts in sourcing from domestic to foreign or between foreign suppliers are not completely reflected in import price indices based on the standard methodology applied by statistical agencies to construct the respective measures (Diewert & Nakamura, 2010). This could conceal important effects of changes in import prices on the domestic economy.
Input Sourcing Strategies
The global fragmentation of the production process is a distinct characteristic of globalization (Baldwin, 2006). It was made possible by considerable reductions in transportation costs and, even more importantly, by advances in information and digital technologies that transformed work processes and allowed a rapid transmission of information as well as fast, easy and direct communication across large distances. From the very beginning cost-saving behavior of firms to establish or maintain a competitive advantage has been seen as the main driver to offshoring and outsourcing (Olsen, 2006). A firm’s global sourcing strategy thus combines the locational and organizational/contractual choice for the production of intermediate inputs in order to exploit competitive advantages at the firm level and the comparative locational advantages of various sourcing countries in light of global competition (Kotabe & Murray, 2004). This development is reflected in an increased vertical specialization across countries as reflected in Hummels, Ishii and Yi (2001). The main idea of vertical specialization of countries on different stages of the production process is captured in the global commodity chain literature, which developed a comprehensive framework to analyze the geography and organization of internationally fragmented production stages using a network-based methodology (Bair & Dussel Peters, 2006).
5
The potential for cost-savings from international production fragmentation is based mainly on the vast wage differentials between industrial and emerging as well as developing countries. However, low costs of other production factors such as land, energy and natural resources are also important (Gereffi, 2009). While there is a large literature on the effects of a shift from domestic to foreign sourcing3, i.e. offshoring, the dynamics of international sourcing, in the sense of import sourcing decisions over time, are less well studied. This is, however, an important field of research as firms choose the location with the lowest production cost, i.e. wages, transport, and contracting cost (Marin, 2006). Production cost in each location can change over time inducing shifts in foreign sourcing from one country to another, which becomes more and more important as pressures from the international competitive environment increase. Shifts in international sourcing are already identified as the prevalent business strategy for lowering production cost (Diewert & Nakamura, 2010, p. 239).
Figure 1 shows the origin of intermediate inputs used in the EU-15 countries. It is evident that intermediate input sourcing outside of the EU-15 had increased substantially between 1995 and 2008. In particular the new EU member states, other emerging economies, and the rest of the world increased their share in EU-15 total intermediate inputs. This is particularly impressive as intermediate input sourcing in the EU-15 more than doubled between 1995 and 2008, when looking at absolute numbers. While the represented data can already be taken as first evidence that sourcing of intermediate inputs shifts to poorer countries, it has to be stressed that the graph does not reveal whether changes between the groups are due to offshoring or import sourcing substitution. Furthermore, the largest share is still produced within the EU-15 and includes the domestically produced intermediate inputs of each country.
3 See for example Antràs and Helpman (2004) as well as Grossman and Helpman (2004; 2005) for some theory on offshoring and outsourcing and Hummels, Ishii and Yi (2001) and Johnson and Noguera (2012) for empirical evidence
80% 85% 90% 95% 100%
2008 1995
EU-15
Other Industrial Economies New EU Members Emerging Markets Rest of the World
0%
Source: Own calculations based on data from the World Input Output Database
Other Industrial Economies include Australia, Canada, Japan, Korea and the United States; the new EU member s include EU-27 without EU-15, Emerging Economies include Brazil, China, Indonesia, India, Mexico and Russia
6 AUS AUT BEL BGR BRA CAN CHN CYP CZE DEU DNK ESP EST FIN FRA GBR GRC HUN IDN IND IRL ITA JPN KOR LTU LUX LVA MEX MLT NLD POL PRTROU RUS SVK SVN SWE TUR TWN USA ROW .4 .6 .8 1 2008 .4 .6 .8 1 1995
Figure 2: Domestic Share of Total Intermediate Inputs
Source: Own calculations based on data from the World Input Output Database Typically, between 60 and
90% of all intermediate inputs are produced domestically. However, this share also decreased in recent years, as can be seen from figure 2, which gives a direct evidence for an increasing importance of offshoring. In general the descriptive evidence shows that intermediate input
sourcing became
substantially more important and globalized in recent years.
The descriptive evidence presented here is supported in the empirical literature. The firm-level study of Marin (2006), for example, shows the importance of the Eastern European countries for German and Austrian firm’s offshoring strategies, but it does not reflect where the offshored production processes were taking place before. However, as will be shown later, this could be of crucial importance for the reliability of officially reported price data and therefore all analytical and empirical conclusions drawn from them.
7
the World Trade Organization (WTO) in 2001 which indicates that factors other than production costs can influence sourcing costs, and that institutional factors are important. It can be assumed that these sourcing shifts are important more generally in the framework of manufacturing production, implying that firms source their raw materials and intermediate inputs from the cheapest provider, taking institutional uncertainty, transport and trade costs as well as quality and/or technological content into account. The gains from input sourcing substitution occur at firm level and can generate higher profits that are either invested, redistributed to shareholders, or passed on to consumers through a decrease in prices (Farrell, 2005). The interest here is on the latter channel as it reflects a potentially immediate impact on the domestic economy. In the following section, I intend to summarize the main findings of a large strand of literature that has investigated the pass-through of changes in price components into domestic prices.
Pass-Through
Pass-through analysis at the country level provides a structured approach to analyze how the domestic economy is influenced by changes at the international level that are transmitted via the price channel. In general, pass-through describes how prices change in response to variation in underlying cost components and mark-up (Goldberg & Knetter, 1997). The process is typically divided into two stages in the through literature: First, the pass-through of exchange rate changes into import prices and second, the pass-pass-through of import price changes into domestic producer prices or directly into consumer prices. This paper abstracts from the determination of import prices and focuses directly on how import price changes are translated to domestic output prices. The approach can be justified by Feenstra (1989) who finds a symmetry in pass-through of tariffs and exchange rates. His results therefore imply that the exact nature of import price changes is irrelevant for the pass-through evidence. Figure 3 gives a simplified systematic overview of the pass-through process as it is typically reflected in the literature. The approach taken by the different studies can vary, some investigate for example directly how import prices pass-through into consumer prices (Berner & Birg, 2012), or how exchange rate changes affect consumer prices (Taylor, 2000). The bold arrow indicates the focus of this paper.
Figure 3: Stylized Pass-Through Process
= exchange rate; = import price; = domestic output price / producer price; = consumer price
8
Originally, the motivation in the pass-through literature was to assess the implications of exchange rate movements for an economy’s external balance and inflation (Goldberg & Knetter, 1997). In mostdeveloped economies the effect on consumer prices is relevant due to inflation targeting monetary policy; low and/or lagged rates of pass-through could lead to monetary policy overshooting as inflationary pressures are misjudged, therefore, pass-through evidence has to be taken into account appropriately. Furthermore, incomplete pass-through4 has implications for global trade (Marazzi & Sheets, 2007) as it implies that relative prices adjust only slowly which will in turn result in a slow adjustment of trade volumes thus influencing external imbalances (Krugman, Obstfeld, & Melitz, 2012). Soon, the focus therefore shifted to explaining the finding of incomplete pass-through. Specific interest was paid to the role of imperfect competition for explaining incomplete pass-through which increasingly led to a pass-through analysis at the industry level (Goldberg & Knetter, 1997). Next to incomplete pass-through, a consistent finding in literature is that pass-through decreased during the last decades, a development which has been assigned to globalization dynamics (Gust, Leduc & Vigfusson, 2010; Marazzi & Sheets, 2007). Based on the previous section’s argument, that input sourcing changes are indeed an economic reality, shifts in import sourcing can therefore nevertheless be assumed to increase the variability in production cost of firms which practice global sourcing. The pass-through analysis is consequently considered to be an appropriate method to investigate in more detail how import price changes due to import sourcing substitution translate into domestic prices.
The literature has identified three factors in particular to explain the consistent finding of incomplete pass-through throughout empirical studies: imperfect competition, barriers to price adjustments, and local cost components (Nakamura & Zerom, 2010). The mechanisms seem to work in a similar way at each stage of the pass-through process.
One of the earliest and most fundamental approaches to explain incomplete pass-through was to relax the assumption of purchasing power parity which assumes “that the exchange rate between two countries’ currencies equals the ratio of the countries’ price levels” (Krugman, Obstfeld, & Melitz, 2012, p. 479). The concept is based on product homogeneity as well as perfect information and competition. Dornbusch (1987) stresses the importance of imperfect competition by dismissing the assumption of a constant mark-up. Instead, profit maximizing
9
firms are likely to adjust their mark-up and thus the import price in the light of exchange rate changes in the same way that they vary the domestic price due to cost changes (Knetter, 1995). The observation of incomplete pass-through is therefore in line with international pricing-to-market and strategic interaction. This finding was recently confirmed in a multi-sector analysis by Auer and Schoenle (2012) who stress that the entire market structure of an industry influences the overall level of exchange rate pass-through into import prices even if only a small number of firms is actually affected. When importing firms adjust their prices due to changes in the exchange rate, the overall price level in an industry changes which affects the pricing decision of all firms in the respective industry (Auer & Schoenle, 2012). The pass-through phenomenon is thus related to state dependent pricing (Dotsey & King, 2005) and price complementarities (Gust, Leduc, & Vigfusson, 2010) which see firms’ profit maximizing price setting as a strategic decision in interaction with the pricing decision of competitors.
Imperfect competition also substantially influences the later stages of pass-through (Francois, Manchin & Norberg, 2010; Hellerstein, 2008; Nakamura & Zerom, 2010). In particular, the concept helps to explain another consistent finding in the literature: the fact that pass-through to import prices or producer prices is much higher than to consumer prices (Clark, 1995). Bacchetta and van Wincoop (2003) develop a theoretical model incorporating imported intermediate inputs. Competition faced by domestic firms leads to an optimal pricing strategy of firms that reduces the pass-through to consumer prices.
In the remainder of this paper imperfect competition is considered to be particularly relevant. Firms’ global sourcing strategy is assumed to be a reaction to increased (global) competition. The firms’ goal is to reduce costs in order to achieve a competitive advantage. Differences in the ability of firms to use their international options in the most efficient way as well as economy-, and industry-specific factors are contemplated as the foundation of competitive distortions.
pass-10
through. In highly volatile environments, inflationary pressures force more frequent price changes that could then incorporate price adjustments from import price changes without additional cost. However, the Campa and Goldberg (2005) reject the hypothesis that macroeconomic conditions are responsible for different levels of pass-through across countries. Their findings are strongly related to Gopinath and Itskhoki (2010) who find a substantially higher rate of pass-through of exchange rates into “prices at the dock” for goods with a high frequency of price adjustment than for those with a low frequency of adjustment. Implementing a menu cost model, their results are in line with evidence of nearly complete pass-through in countries with high inflation (Taylor, 2000). However, as the inflation environment is rather similar across the sample of mostly European countries chosen below for the empirical analysis, this factor is considered to be less relevant for this study.
As this paper takes import prices as given, it is more related to the discussion on evidence indicating that pass-through is more hampered within the economy than across borders. Here, the role of local cost components is particularly pronounced in explaining the observation that pass-through of exchange rate changes to import prices is much higher than the pass-through to domestic (consumer) prices. Specifically, two main explanatory approaches can be distinguished with respect to how local cost components influence pass-through: (1) the role of the distribution sector in cushioning price changes at the import level and (2) the role of intermediate imports and their complex interaction with domestic costs in the production process.
Burstein, Neves and Rebelo (2003) provide evidence for the first channel. In their study the authors try to explain the movements in real exchange rates during exchange-rate based stabilizations by focusing on the influence of distribution costs for tradable goods. These costs include in particular transporting, wholesaling and retailing (Burstein, Neves, & Rebelo, 2003, p. 1190). Burstein, Neves and Rebelo (2003) show that distribution costs add significantly to the average consumer good price; ignoring this component can therefore bias results on the price development since it is assumed that all goods require distribution services to be traded. As distribution services are provided at a local level, the associated costs can explain differences in price levels across countries as distribution costs are determined mainly by local production factors, in particular labor, which are country-specific.
11
a setting with vertical interactions between monopolistic producers and retailers. Corsetti and Dedola (2005) find robust theoretical and empirical evidence that the presence of distribution costs is responsible for the high degree of local currency price stability enjoyed in most developed economies as it further reduces the extent of exchange rate variability pass-through into consumer prices. This can be due both to mark-up adjustments or the existence of local cost components, which reduce the relative share of the final price that is subject to the exchange rate shock (Corsetti & Dedola, 2005, p. 148).
Focusing on the apparel industry, Berner and Birg (2012) confirm the role of more general local cost components in reducing pass-through. Their finding is of importance for the line of reasoning in this paper, as the import price index bias is assumed to be particularly relevant for domestic products using imported intermediate inputs. By definition the use of imported intermediate inputs implies a large share of domestic value added.
Next to distribution services, intermediate inputs are identified to explain how local cost components reduce pass-through. Goldberg and Campa (2010) confirm the basic results of Corsetti and Dedola (2005): exchange rate fluctuations are absorbed to a substantial extent by firms in the imperfect competitive distribution sector that keep their prices stable. However, their study is interesting as it also incorporates the second explanatory approach for differences in pass-through into import and consumer prices; they include in their model the use of imported inputs in the production of both tradable and non-tradable goods, through this they identify the use of imported intermediates in the domestic production process as the main transmission channel of international shocks in the OECD5. The direct consumption of imported final goods is therefore found to be relatively less relevant (Goldberg & Campa, 2010, p. 393). Assuming full pass-through into imported inputs, the response of domestic prices to changes in exchange rate are assumed to be limited by producers’ flexibility in the production process, i.e., their ability to substitute the more expensive foreign intermediate inputs by domestic products.
To summarize the previous findings, it is assumed in this study that manufacturers in the countries under consideration face severe competition both internationally and domestically. Changes in production costs can thus be expected to be only partially passed-through to the next level. Pass-through is further muted as (imported) intermediate inputs are combined with
12
additional local costs, in particular for distribution services. Incomplete pass-through is therefore expected.
A further point that is stressed in the review of the pass-through literature is that imported components in domestic production are of increasing importance in the empirical analysis as the fragmentation of the production process increases (Goldberg & Campa, 2010). However, several researchers recently have doubted the appropriateness of standard aggregate price indices to capture price changes resulting from the substitution of goods from high-cost to low-cost suppliers (Houseman et al., 2012; Inklaar, 2013) which would substantially limit their use in identifying effects from offshoring.
Initial evidence on the importance of the methodology of price indices for pass-through evidence is ambiguous. While Campa and Goldberg (2005) find that shifts in the composition of countries’ import bundles account for a large share of pass-through changes, Marazzi and Sheets (2007) cannot confirm the hypothesis that changes in the methodological construction of import prices are responsible for the evidence of decreasing pass-through. The authors run their analysis twice, each time using data based on different methodologies (import price surveys or census unit value indices). As the results are sufficiently similar, it is assumed that their findings are robust to the methodology used when constructing the underlying price indices.
The following section will deal with the challenges globalization poses on statistical agencies to capture the price changes from cost savings due to sourcing shifts in their index numbers.
Import Sourcing Bias
There is evidence for a serious flaw in the way official statistics treat offshoring and shifts in import sourcing. A well known problem in the construction of price indices is that price changes and product replacements are not captured properly by statistic agencies that calculate monthly published import price data (Nakamura & Steinsson, 2012). These flawed data are nevertheless used by governments to calculate important economic key measures such as output and productivity, which has potentially large economic and political implications (Mandel, 2007).
13
product (Inklaar, 2013). Thus, a bias occurs even when no lag exists between the time of entry of a new supplier into the market and the time that his product is picked up in the import price sample (Houseman, 2011). The idea is quite straightforward, in particular when assuming that neither the price of the domestic (high-cost) supplier nor the price of the foreign (low-cost) supplier change6. As a consequence, neither the corresponding domestic price index nor the import price index will document a price change. However, when domestic producers start sourcing from a foreign supplier, its market share will increase and producers will enjoy price decreases. If the two products were regarded as substitutes, a weighted average of the prices would capture the realized cost savings, with the product’s share in total expenditure as weight. The same intuition holds if domestic producers are already sourcing from a foreign country but switch to a lower priced product of a new foreign producer.
The bias problem is well documented for the frequently used data of the Bureau of Labor Statistics (BLS) but the definition of the import price index (MPI) according to Eurostat (2010, p. 7) indicates that the methodological approach of the European statistical agency is prone to similar problems:
Import price indices aim to measure the transaction price development of imported goods purchased from non-domestic areas by domestic services. All the related services are initially excluded from the scope. The price indices should track the price movements of comparable items over time. Like output price indices, MPI are calculated as a weighted average of price changes for a selection of representative products.
Diewert and Nakamura (2010) give an overview of the biases in input price indices due to changes in sourcing once firms decide to purchase, rather than produce, an input. The offshoring bias occurs when a producer switches from a domestic to foreign sourcing: The actual price for the intermediate good will then drop out of the producer price index, which captures price changes for all domestic outputs, and become part of the import price index. The price change experienced by the domestic producer, however, will fall into the “price collection gap” between the two indices (Diewert & Nakamura, 2010, p. 242). Houseman et al. (2012) have studied this bias extensively, relying on confidential data on domestic and imported intermediate inputs at the commodity level. Their very detailed dataset allows them to capture both international sourcing dimensions discussed: offshoring as well as import sourcing shifts. Inklaar (2013) focuses on the import sourcing bias occurring from a domestic
14
firm switching among foreign suppliers. This enables an analysis based exclusively on import price indices which allows for a broader application.
To give an example of how different methodological approaches in underlying data might influence economic analysis, this study will follow the methodology of Inklaar (2013), i.e. assume that price changes based on official statistical price indices are well proxied by the use of Törnqvist indices. A Törnqvist index is the weighted geometric averages of relative prices of two time periods using (unweighted) arithmetic averages of the value shares in the two time periods as weights (OECD, 2005; Reinsdorf, Diewert, & Ehemann, 2002):
1 ∆ ö ∏ / / / ! " # #/#!# With % ∑ ' ()*0,1-
Lower case letters indicate prices and quantities ' of single products while upper case letters indicate the corresponding aggregates. In this example, two substitution goods with different price levels, .)*/, 0-, are compared. ∆ indicates a change from one time period ( to another.
15
In sum, the standard approach used by statistical agencies to determine price changes is to calculate the weighted average price changes (1) while the real price change must be calculated as change in the weighted average prices (2). Inklaar (2013) uses these two concepts to calculate the import sourcing bias ∆:
3 ∆ ∆ ö4 ∆ 23
As it is assumed that import price do not capture all cost savings from import sourcing substitution, I expect the true import price index to be lower than the reported one. Price changes are hence assumed to be overestimated which would imply that real import growth has been understated, leading to a potentially severe overestimation of value added and, therefore, productivity. A further consequence from the overestimation of import price changes is that pass-through is underestimated, as cost-savings from import sourcing shifts to cheaper producers are not acknowledged adequately in the data. Consequently, incomplete pass-through is assumed to be partially an artifact; controlling for the import sourcing bias in the pass-through analysis is expected to result in a pass-through value that is higher than conventional estimates.
Methodology
The goal of this paper is to determine whether differences in the methodology of constructing price indices have a significant impact on economic results. Starting from standard national accounting identities that relate output prices to its labor, capital and intermediate inputs components, a disaggregation into domestic and imported intermediate inputs is carried out. To implement this approach I develop a pass-through model from the growth accounting theory. I then explain how the different components are measured and which data sources are used before I go into more detail on the econometrical model and the estimation techniques employed.
The Pass-Through Model
16
input @ and multifactor productivity B, I take the total differentiate with respect to time ( and divide by output % to express the production function in the form of growth rates7:
4 !E! BF!FGGE!; BF!FH!HE;IEI
Defining the relative shares of factor inputs J as KL F! FL L ! B FM FL L
! we can rewrite 4 into the standard primal growth accounting equation with continuous growth rates:
5 !E! KG GE
G; KH HEH; IE I
Thus, the growth rate of output is the productivity growth plus the weighted average of factor input growth with the weights being the output elasticities. These are identical to the factor income shares KL OL with nominal output P % (with being the output price level) when inputs are paid the value of their marginal product (Hulten, 2009), implying perfect competitive factor markets.
This straightforward result can be easily generalized for discrete time periods and additional input factors (Timmer, O'Mahony, & van Ark, 2007), in this case intermediate inputs Q: 6 ∆ ln % KU∆ ln Q; ;KUG∆ ln ?; KUH∆ ln @; ∆ ln B
Under the given assumptions the weights averaged across two time periods KU add up to 1. Assuming translog technologies for input factors, the growth rate of total input of each factor is the weighted average of the growth rate of the factor sub-types. This allows an easy inclusion of different types of capital (human, physical) and labor (high-/low-skilled). And more specifically in this application, it allows to disaggregate intermediate inputs into those imported V and domestically produced W:
7 ∆ ln Q Y∆ ln V ; 1 4 Y ∆ ln W
The weight Y being the import share in total intermediate inputs: Y Z[ Z["Z\ .
To use the evidence on primal growth accounting for a simple pass-through equation I exploit the equivalence of this result with the dual growth accounting equation. The national (income)
7 Continuous growth rate: ]E
17
accounting identity states that the sum of the value of all output products ` is equal to the total production cost, determined as the sum of the price of each input times its quantity (Jorgenson & Griliches, 1967). In this application the dual growth accounting equation is:
8 ∑ b b%b V ; cW ; d? ; e@
With b being the price of output `, being the price of imported inputs, c the price of domestic intermediate inputs, d being the rental price of capital and e being the wage rate: L) *, c, d, e-.
Following the procedure of Solow (1957), I differentiate totally with respect to time E and divide both sides with the corresponding total value:
9 ∑ Kb bg hiE ;!!hiE j Kg
[E
[;Ej ; Kcg \E
\;cEcj ; KGkE;GEGl ; KHkmEm;HEHl
As before the weights K] ∑ n]
i
i !i are the relative shares of output ` / input J in total nominal
output (with _ refering to type of output or input respectively)8.
Aggregating from the product level, an index of the quantity of total output and input can be defined in terms of the weighted average of the individual growth rates, which follows the Divisia index approach:
10 !E! ∑ Kb b!!hEi
The corresponding Divisia price indices are:
11 E ∑ Kb b iEh and for factor prices L: 12
OE
O ∑ KL o OE
O
In terms of Divisia indices, according to Jorgenson & Griliches (1967), a natural definition of the total factor productivity is the ratio of total output to total factor input: B !L
13 IEI !E!4LEL !E!4 K E
4 Kc cEc4 KG GEG4 KH HEH Which can equivalently be expressed in prices:
18
14 IEI E 4 K [E
[4 Kc \E
\4 KG E4 KH mEm
Solving for the growth rate of the domestic output price index and exploiting that ∆ ln _ is a good approximation of the percentage growth rate of a variable _, I obtain a simple pass-through model:
15 ∆ ln KU∆ ln ; KUc∆ ln c; KUG∆ ln d ; KUH∆ ln e ; ∆ ln B
Hulten (2009) shows how the continuous time model can be operationalized by discrete approximations (e.g. Törnqvist-Index) in this way. The continuous income shares are then replaced by the corresponding average income shares KUL (Hulten, 2009, p. 13).
Limitations
The model is based on some restricting assumptions. However, it is still assumed to be suitable to provide insights into pass-through processes. The model’s foundation in the growth accounting framework makes it prone to all weaknesses of this approach which are comprehensively discussed in Hulten (2009). Relaxing the assumption of constant returns to scale, for example, he shows, that with a redefinition of cost shares, the primal growth equation still holds, but the accounting identity which is the foundation of the equality of primal and dual growth accounting is no longer valid.
19
Data
The construction of the dataset as well as the empirical analysis will be carried out mainly in the data analysis and statistic software Stata, except where direct calculations with data of the input-output tables from the WIOD9 require the use of the Scilab software for numerical computations. The estimation of the pass-through model is based on a panel data set of 14 manufacturing industries, defined according to the ISIC rev. 3 classification, of 39 countries including all 27 European Union (EU) members, and 12 other major advanced and emerging economies10. The dataset covers the years 1995-2008 and is thus identical to the countries and time period covered in Inklaar (2013). I use two main data sources: Information on the input-output structure and socio-economic variables are taken from the WIOD and additional import price data based on the United Nations (UN) Comtrade database are provided by Inklaar (2013)11. The latter contains import price changes for the period 1995-2008 according to the two methodologies reflected in the paper. Additionally, the Eurostat databases12 and the World Bank’s world development indicators13 are used for data on alternative capital measures at the country level, which unfortunately reduces both the sample size and length. Domestic prices are taken from the WIOD socio-economic accounts (SEA) which provides data on price levels of gross output. This measure has the advantage that it most closely resembles the dependent variable according to the model and is directly available at the required level of aggregation.
The prices of intermediate inputs are included in the SEA data. Prices for domestic intermediate inputs, however, are not directly available. While there are input price indices issued by some statistical agencies, their composition is often not sufficiently transparent and the level of aggregation and country coverage does not fit the sample chosen here. It therefore proved easier to construct a domestic intermediate input price index based on information on domestic prices and the input-output structure at industry level as provided in the WIOD. Price changes faced by each industry . are calculated by summing the price changes experienced by all domestic industries `, each weighted by pqb, the two year average share of industry ` in the total domestic inputs of industry .:
9http://www.wiod.org/database/index.htm
10 Namely Australia, Brazil, Canada, China, Indonesia, India, Japan, Korea, Mexico, Russia, Turkey and the United Sates
11 My thanks are due to Dr. Robert Inklaar for providing me with the readily processed data set. 12http://epp.eurostat.ec.europa.eu/portal/page/portal/statistics/search_database
20
16 ∆ ln c ∑ pq
b∆ ln bc r
bs
The data on total intermediate and domestic intermediate input prices can be used to generate a first measure of import price changes tY capturing the official statistics perspective, based on the following decomposition:
17 ∆ ln Yu∆ ln ; 1 4 Yu ∆ ln c ∆ ln ∆ vw xy yz ∆ vw \ z
With Yu being the two year average of the import share in total intermediate inputs Y.
The construction of income shares is straightforward but requires some calculation effort in the case of intermediate inputs. While the total value of intermediate inputs is reported in the WIOD SEA, the share of both domestic and imported intermediate inputs in nominal output must be calculated using information on the input-output structure provided in the WIOD; the numerical computation software Scilab is used to aggregate domestic and imported intermediate inputs that are used in the manufacturing sectors of each country. Gross output is taken directly from the WIOD SEA. The weights are then constructed by dividing for total, domestic, and imported intermediate inputs per industry through the total output of each industry; this is done for each country. In a last step, the two year averages are taken
Alternative import price data (tY1 for the corrected, tY2 for the standard Törnqvist import price changes) are provided by Inklaar (2013) based on unit values in US $. In order to clear them from changes in nominal exchange rate, any appreciation or depreciation over time is eliminated based on data on exchange rates provided at the WIOD homepage.
A proxy for the wage rate was constructed in accordance with the procedure of Hall (1988) by dividing labor compensation through the number of persons engaged. The number of persons engaged captures both employed and self-employed participants of the labor force. Data on both are available from the WIOD SEA. A more detailed approach is taken by separating labor input according to skill type into high-skilled, medium-skilled and low-skilled which allows abstracting from changes in average wage due to changes in the skill structure of the labor force. The necessary data on labor compensation and hours worked by skill type is available in the WIOD SEA. Different types of labor input are taken into account according to following disaggregation which is basically a more general expression of equation 7 :
18 ∆ ln e ∑ {H∆ ln e
21
The weights { measure the average share of each skill type | in the value of total labor compensation (Timmer, O'Mahony, & van Ark, 2007). The overall income share of labor is calculated by dividing total labor compensation through nominal output.
An appropriate measure of capital inputs is still a highly controversial issue14. The WIOD provides data on total capital compensation and the real fixed capital stock as well as a deflator of gross fixed capital formation which allows the calculation of a nominal capital stock. However, conclusions drawn from these data have to be handled with care as the WIOD treats capital measures as residuals, which results in frequent negative values for capital compensation. Following the strategy for the determination of wages, for an initial measure the total capital compensation is divided through the capital stock in current prices to determine the (ex-post) return on capital d1. In addition to problems with data quality, this procedure has some serious limitations as discussed in Hulten (2009) and Inklaar (2010). As this measure simply equates capital compensation with residual income it only imperfectly reflects economic reality. A better approach would be to calculate the (ex-ante) cost of capital d2 more directly, based on information from capital markets. It is commonly found that this reduces the contribution of capital to output growth in standard growth accounting (Inklaar, 2010, p. 407). Oliviera Martins, Scarpetta and Pilat (1996) derived a simplified equation for the rental price of capital motivated by the commonly used implementation of Hall and Jorgenson (1967):
19 d2 C 4 3 ;
C are government bond yields with 10 years' maturity for which Eurostat provides annual data for Japan, Turkey and the US as well as for the EU member states except Luxemburg and Romania. 3 is the expected inflation rate which is proxied by the actual inflation rate as reported by Eurostat. This assumes perfect foresight which is of course a very restricting assumption. The real cost of capital faced by firms is then represented by C 4 3 . The rate of depreciation is provided at industry level in Erumban et al. (2013), assuming homogeneity across countries, an assumption that further reduces the level of detail. is the deflator for fixed capital formation and is taken from the WIOD SEA as described above. It is important to stress that this capital cost measure is very likely to underestimate the real share
22
of capital in total costs as taxes, typically of crucial importance, are not taken into account (Inklaar, 2010, p. 410).
For a third alternative (also ex-ante) capital measure d3, the real cost of capital C 4 3 is directly captured by using real interest rate data from the Worldbank which allows a better country coverage.
20 d3 .3Z< ;
Under perfect conditions, ex-ante and ex-post capital measures should be identical. Surprisingly, the correlation between the two ex-ante measures and the ex-post measure is slightly negative15. The correlation between the two ex-ante measures is reassuringly high at 0.85. However, first evidence on the capital measures indicates problems with the data quality. Despite these data shortcomings, the income share of capital is calculated by dividing total factor compensation through nominal output.
MFP cannot be measured directly but is determined as a residual in the standard growth accounting model and will therefore become part of the error term in the econometric model which will be described in more detail in the following section.
Econometric Analysis
Before beginning the econometric analysis it is necessary to check and clean the data to ensure that incorrectly coded data are appropriately controlled for. Maximum and minimum values are examined for each variable, in order to evaluate whether data are hampered by outliers: All growth rates are in a sensible range of approximately 5% and are in the mean slightly increasing. An important exception is the growth of d2 that is on average negative, this is very likely due to the weak data basis and the strong assumptions attached to calculating this measure. A further exception is the growth of import prices tY, calculated as the residual from total and domestic intermediate input price changes, and for which growth rates are substantially higher. However, when eliminating observations that contained a division through zero from the dataset, only five observations lie outside the sensible range of 5% for this measure. They indicate that particularly industry 23 (coke, refined petroleum and nuclear fuel) seems to be prone to outliers. Therefore, a robustness test is included to
23
check whether the results determined in the econometric analysis are sensitive to the inclusion of individual industries.
All weights KUL have to be in the range between zero and one. Values outside this range exist for labor and capital income shares and are due to negative values for capital compensation and missing gross output that leads to dividing through zero when determining the income shares. All observation below zero or above one are therefore dropped from the dataset to avoid potential problems and distortions in the later estimation process.
The analysis starts with an econometric adaption of the baseline theoretical model specified in 6 to determine the most appropriate estimation method and model specification:
21 ∆ ln ; KU ∆ ln ; KUG ∆ ln d ; KUH ∆ ln e ; The model is based on national accounting data and describes the effect of growth rates of the main price components in industry . in country in year ( on the overall domestic price changes. As mentioned before, MFP will be captured as the time varying component of the error term and can, by construction as residual in a growth accounting equation, be considered to be independent of the exogenous variables. Consequently, its exclusion from the explicit formulation of the model should not result in an omitted variable bias. Whether this assumption is economically sensible will be discussed in more detail later. The variables in log differences are discrete approximations of continuous growth rates used to operationalize the theoretical model (Hulten, 2009). The logarithm is therefore not applied to the data for import prices that are directly available as changes from one year to the other.
24
The variables are checked for stationarity in various ways. Scatter plots indicate a weak trend for the import price variables tY1 and tY2, and strong trends for total, imported and domestic intermediate input price weights. All other variables seem stationary upon visual inspection. This first impression is confirmed in time series plots for randomly chosen countries. The formal Fisher-type augmented Dickey-Fuller unit root test is applied to all variables and non-stationarity is rejected for all variables except for the two year averages of factor input income shares. However, a scatter plot analysis of the factor price variables weighted with their respective income share indicates that the problem of non-stationarity is considerably reduced or even eliminated when the income share and factor price variables are interacted. Thus, it is assumed that the standard OLS procedure will not lead to spurious results. This conclusion is made despite the potential low power of the unit root test employed, as it tests against the null hypothesis of all panels having a unit root. As the dataset contains a large number of panels the probability of rejecting the null hypothesis is by construction quite high.
A correlation matrix is used (see appendix II) to check for multicollinearity among the variables, which could lead to spurious inference. A correlation below 0.8 is commonly considered to be unproblematic. In the following analysis I do not expect multicollinearity to be a problem, as a correlation above the critical threshold is only observed between alternative measures which are not included simultaneously in the regression and the income share weights. No critical correlation is found for the factor prices weighted with their respective two-year average income share. Surprisingly, the correlation between the import price changes derived from the WIOD national accounting data tY and the two alternative import price change measures as determined by Inklaar (2013) tY1/tY2 is low. This already indicates that results are very likely to be sensitive to the methodology used in data construction.
Estimation Procedure
25
regression is re-estimated with White country-industry cluster-robust standard errors if the null hypothesis of homoskedasticity is rejected, which would also correct for autocorrelation. The pooled OLS estimation gives some insights in potential problems that need to be accounted for in more detailed estimations. However, pooling across countries and industries without an adequate treatment of the panel dimension of the dataset constitutes an inappropriate generalization and reduction of the available information. Fixed Effects (FE) are employed in order to allow for heterogeneity across countries, industries, and time. The resulting model is therefore:
22 ∆ ln , ; KU ∆ ln ; KU G ∆ ln d
; KU H ∆ ln e ; ; }
The index . indicates a unique country-industry identifier. , captures all effects that are common to each specific country-industry pair but do not vary over time (country-industry-FE). This, for example, captures price effects due to a country’s industrial structure based on the concept of comparative advantage. A basic F-test is used to test for the significance of country-industry specific heterogeneity. represents the time FEs that capture all changes over time that do not vary across country-industry pairs, such as changes in the global economic activity or global shocks in more general. To justify the inclusion of time dummies, their joint significance is determined with an F-test. } is the remaining error component varying across time and country-industry pairs when controlling for the time and country-industry invariant components of the original error term .
The FE model is appropriate, in the sense that it yields unbiased and efficient estimates, if the country-industry specific heterogeneity is not distributed randomly; when heterogeneity is random, the random effects (RE) model yields more efficient estimates. Therefore, a RE model is estimated additionally and the Hausman test is applied to identify whether the estimates of FE and RE model are significantly different from each other. If this is the case, the RE model is assumed to result in biased estimates implying that heterogeneity is not random and ignoring this leads to omitted variable biases. According to the results of the Hausman test, the FE model is chosen.
26
comparison of goodness of fit measures, information content and plausibility of estimation results.
In a next step, total intermediate inputs will be disaggregated into imported and domestically produced intermediate inputs:
23 ∆ ln , ; KU ∆ ln ; KU c ∆ ln c
;KU G ∆ ln d ; KU H ∆ ln e ; ; } This specification allows the comparison of the differences across various import price measures. According to the findings in the pass-through literature, incomplete pass-through, i.e. an estimated coefficient smaller than one, is expected. In a last step, the difference between the two methodological approaches for constructing import price indices according to equation (3) is incorporated into the model to proxy the input sourcing substitution bias and to determine its significance in this specific example of economic analysis:
24 ∆ ln , ; KU ∆ ln ; KU t. ; KU c ∆ ln c ;KU G ∆ ln d
; KU H ∆ ln e ; ; }
According to the findings discussed for the pass-through and import sourcing literature, the coefficient for imported intermediate input price pass-through, , is expected to increase when controlling for the import sourcing bias. As measurement issues in the import price indices are found to overestimate import price changes, the bias according to the definition in (3) commonly assumed to be positive. Hence, for a negative sign is expected.
The analysis is followed by a critical comparison of the final results and the findings and extensive robustness checks.
Robustness Checks
27
Additionally, import sourcing substitution is assumed to be carried out mostly by advanced countries. Inklaar (2013) determines a consistent structure in the input substitution bias across advanced and emerging economies. I therefore split the sample and re-run the analysis on advanced and emerging economies separately.
Furthermore, Houseman, Bartik and Sturgeon (2013) stress the importance of the exceptional growth of the computer and electronic products industry. Together with the findings of the initial analysis of the dataset that indicate a prevalent role of single industries regarding extreme outliers and low data quality, regressions are run excluding one industry at a time. The same is done excluding one year at a time.
Last but not least, incorporating MFP in the error term might prove to be a strict assumption after all, as shocks on multifactor productivity can actually be assumed to influence the model parameters systematically and induce simultaneity. To control for this and to allow for an alternative estimation procedure, the analysis is also carried out based on difference and system General Method of Moments (GMM).
Results
Baseline Specification
28
useful when the underlying distribution is unknown. The bootstrapped standard errors are largely in line with the robust standard errors. In particular, they do not change in a way that would indicate spurious significance of inference based on standard country-industry cluster robust standard errors. For simplicity, cluster-robust standard errors are in the following assumed to sufficiently approximate standard errors robust to non-normality.
Table 1 reports initial regression results when domestic prices are regressed on the prices of labor and capital inputs as well as total intermediate inputs. In the pooled OLS regression without year dummies (column 1) all estimates show the expected positive sign and are highly significant. An additional (not reported) t-test shows that all estimates are also significantly different from one, rejecting the null hypothesis of complete pass-through. However, the magnitude of through for each of the factor inputs is surprising. In particular, a pass-through of intermediate input prices into domestic prices larger than one is not sensible from an economic perspective and clearly contradicts the coherent finding of incomplete pass- through in the literature. Surprisingly, empirical evidence on the pass-through of intermediate
Table 1: Econometric Model Specification
(1) (2) (3) (4)
Pooled OLS Pooled OLS FE RE
a_I*dlpi 1.223*** 1.221*** 1.199*** 1.208*** [0.022] {0.022} [0.022] [0.022] [0.022] a_K*dlr1 0.164*** 0.177*** 0.189*** 0.184*** [0.029] {0.032} [0.029] [0.026] [0.027] a_L*dlw 0.435*** 0.426*** 0.303*** 0.351*** [0.059] {0.055 } [0.059] [0.049] [0.052] Constant 0.002* 0.010*** 0.014*** 0.012*** [0.001] {0.001} [0.002] [0.002] [0.002] Observations 6612 6612 6612 6612 R-squared 0.925 0.926 0.925 0.925 F-test 2308 647.8 511.0 8743.19a) Prob > F 0 0 0 0
Year dummies No Yes Yes Yes
Country-industry effects No No Yes Yes
Number of icb) - - 542 542
Country-industry cluster robust standard errors in brackets; bootstrapped standard errors in braces *** p<0.01, ** p<0.05, * p<0.1
a)
Wald Test Statistic; b)
29
inputs specifically is scarce. Dhyne et al. (2006) find that goods with large labor inputs are subject to less, goods with a high share of intermediate inputs are subject to more frequent price adjustments. Together with the finding of Gopinath and Itskhoki (2010) that a high frequency of price adjustment is connected to higher pass-through, the general finding of a higher pass-through of intermediate input prices seems broadly in line with the theoretical literature. On the contrary, the theoretical and empirical analysis of Clark (1995) indicates that price changes only weakly travel along the production chain. Changes in intermediate input prices translate into producer price changes, but these are not found to pass-through into consumer prices. The dependent variable here is the domestic output price, which is the most closely related to the producer price. A higher pass-through can therefore be expected. The rather low pass-through values for changes in wage rate and capital cost indicate the expected adjustment costs attached to the literature’s general finding of incomplete pass-through. The magnitude of wage pass-through is in line with the findings of Hellerstein (2008) on the price composition of beer. In a preliminary estimation, Hellerstein (2008) finds pass-through of both capital and labor cost of around 0.35.
The inclusion of year dummies, as reported in column 2, is justified from a theoretical as well as an econometrical perspective. As year dummies capture all global shocks that have a homogenous effect on pass-through across countries it helps to eliminate the biases these overall shocks might induce. Furthermore, the time fixed effects prove to be jointly significant.
30
model is identified as the appropriate way to capture country-industry specific heterogeneity. An F-test performed on the results confirms the joint significance of the country-industry fixed effects.
Although the parameter estimates do not vary substantially depending on the estimation method, the fixed effects model is chosen as the most appropriate model as it allows for most unobserved effects and helps to reduce the risk of an omitted variable bias. The remaining error term contains only factors that vary across countries, industries and time.
Table 2 reports estimation results of the total intermediate input model (22) estimated including country-industry and time FEs and compares different capital and wage measures. Clearly, including the two ex-ante measures of capital cost further reduces the pass-through of changes in the price of capital into domestic output prices while at the same time influencing the estimated coefficient for wage pass-through. When including the capital measure based on Eurostat data (d2, column 2 and 5) the sample size is substantially reduced and capital cost pass-through becomes negative which is not sensible from an economic perspective. For this specification, the hypothesis of complete pass-through of intermediate input price changes
Table 2: Variable Model Specification
(1) (2) (3) (4) (5) (6) a_I*dlpi 1.199*** 1.068*** 1.126*** 1.199*** 1.068*** 1.126*** [0.022] [0.039] [0.034] [0.022] [0.039] [0.034] a_K*dlr1 0.189*** - - 0.189*** - - [0.026] - - [0.026] - - a_K*dlr2 - -0.026** - - -0.026** - - [0.012] - - [0.012] - a_K*dlr3 - - 0.045** - - 0.045** - - [0.017] - - [0.017] a_L*dlw 0.303*** -0.018 0.105*** - - - [0.049] [0.062] [0.037] - - - a_L*dlwS - - - 0.300*** -0.022 0.105*** - - - [0.048] [0.062] [0.037] Constant 0.014*** 0.009*** 0.018*** 0.015*** 0.009*** 0.018*** [0.002] [0.003] [0.002] [0.002] [0.003] [0.002] Observations 6612 2907 5360 6612 2907 5360 R-squared 0.925 0.899 0.859 0.925 0.899 0.859 F-test 511.0 304.8 276.9 510.7 304.9 276.7 Prob > F 0 0 0 0 0 0 Number of ic 542 361 516 542 361 516
Country-industry cluster robust standard errors in brackets *** p<0.01, ** p<0.05, * p<0.1
31
cannot be rejected at the 5% significance level. Including the capital measure d3 (column 3 and 6), which is calculated based on real interest rate data, substantially reduces the goodness of fit. Therefore the ex-post capital measure d1 is identified as yielding the most reasonable pass-through results and the best model specification. In contrast, the choice of a specific wage measure does not substantially influence goodness of fit or parameter estimates. Thus, the wage rate taking into account the skill level composition of labor is chosen as it is expected to describe actual wage changes more closely. However, the high correlation between the two wage measures indicates that both measures are appropriate and will yield nearly identical results.
Disaggregation of Total Intermediate Inputs
Table 3 reports the regression results according to equation (23) when disaggregating total intermediate inputs into imported and domestically produced intermediate inputs. The estimation is carried out according to the most favorable specification determined in the previous section.
