Estimating the U.S. Consumer Gains from Chinese Import Penetration

Transcription

1 Estimating the U.S. Consumer Gains from Chinese Import Penetration Liang Bai University of Edinburgh Sebastian Stumpner Université de Montréal February 15, 2017 Abstract This paper investigates the size of U.S. consumer gains from the recent growth in Chinese imports, focusing on the years We use the theoretical insight from Arkolakis et al. (2012) that in a large class of trade models, the domestic welfare consequences of any foreign shock are summarized by the change in the country s share of expenditure on domestically produced goods. We compute this domestic share of expenditure (DSE) at the level of roughly 230 consumer good categories, and estimate its effect on domestic prices by using Chinese import penetration into the EU as an instrument. Using barcode-level data to compute category-level price indices, we find large effects of Chinese import penetration on U.S. prices. Comparing the median category in terms of import penetration to a category with no exposure, we find that prices in the median category declined by roughly 0.4% more per year. The effect is driven by both the intensive and extensive margins, suggesting the presence of pro-competitive effects for old goods as well as variety gains from new goods. In contrast to the gains arising from direct imports of consumer goods, results for price reductions as a result of imports of Chinese intermediate goods are inconclusive. We also find no evidence for heterogeneous effects across consumer groups by income or region. A back-of-the envelope calculation suggests that the U.S. consumer price index fell by roughly 0.9% per year due to Chinese import penetration. We are grateful to Jason Garred, Andres Rodriguez-Clare, Thomas Sampson, Catherine Thomas and seminar participants at the LSE, Université de Montréal, DIW Berlin, RAND, the Montréal Micro-Macro workshop, and the China Economics Summer Institute for their helpful comments. Liucija Latanauskaite, Matus Luptak and Tomaz Norbutas provided excellent research assistance. Bai: University of Edinburgh, School of Economics (liang.bai@ed.ac.uk). Stumpner: Université de Montréal, Department of Economics (sebastian.stumpner@umontreal.ca). 1

2 1 Introduction Recent years have seen a surge in U.S. imports from China. What are the welfare consequences of this Chinese import growth for the U.S.? Previous research has shown that it led to a sizeable decline in U.S. manufacturing employment and lifetime earnings of manufacturing workers. 1 How do these welfare losses contrast with the potential gains for U.S. consumers? In this paper, we estimate the size of these gains, and investigate the channels through which they are realized. Focusing on the years , we first estimate the effect of Chinese import growth on consumer prices across different product categories, using barcode-level price data. Our results show that there are sizeable gains for U.S. consumers. Comparing a product category with median China trade shock to one with no change, prices in the median category grew by 0.4% less per year. Such an effect in turn implies that the consumer price index grew by roughly 0.9% per year less due to Chinese import growth. We next investigate the channels underlying this effect. We find that it is driven by both existing goods having lower inflation and the introduction of new goods, suggesting the presence of pro-competitive effects as well as variety gains (Broda and Weinstein 2006). Moreover, Chinese import penetration leads to increased product turnover, more exit of previously consumed products, and a small increase in the number of consumed varieties. Variety growth is 1% higher in a category with median China trade shock compared to one experiencing no shock. In contrast to the gains arising from direct imports of consumer goods, results for price reductions as a result of imports of Chinese intermediate goods are inconclusive. We also find no evidence for heterogeneous effects across consumer groups by income or region. Our main estimation equation is consistent with a large class of theoretical trade models. In particular, we use the insight from Arkolakis et al. (2012) that for a wide variety of trade models, the domestic welfare effects of a foreign shock are summarized by the change in the share of expenditure on domestically produced goods. We identify the variation in the change of domestic share that comes from Chinese supply shocks by using Chinese import penetration into the five largest European economies as an instrument. We construct CES price indices at the level of around 230 consumer product categories using data from AC Nielsen Homescan, and the methodology from Sato (1976), Vartia (1976) and Feenstra (1994). The data supply us with information about consumer goods purchases 1 See Autor et al. (2013) and Autor et al. (2014). 2

3 (prices and expenditure) from a sample of roughly 60,000 U.S. households at the barcodelevel. The data have the advantage that we can define a product at the finest possible level. This ensures that we do not confound individual product price changes with changes in the composition of varieties if products were more coarsely defined. While the data do not cover all consumer expenditure on tradable goods (we only observe barcoded products), we show that Chinese import penetration was likely even larger in product groups that we do not observe in the Nielsen data. This suggests that our results provide a lower bound of the true effect of Chinese import penetration on U.S. prices. We complement the consumer data with information on trade flows at the HS 6-digit level from UN Comtrade and construct a concordance between the Nielsen product modules and HS 6-digit categories. Finally, we make use of U.S. and European industry-level production data to compute the change in the domestic share of U.S. expenditure and Chinese import penetration into Europe. The main threat to identification is that both prices and imports from China may be driven by demand or supply shocks in the U.S. instead. For instance, a positive U.S. demand shock (negative U.S. supply shock) could lead to higher U.S. prices and an increase in imports from China. The resulting bias would therefore make it harder to find any effect. In order to address this concern, we follow Autor et al. (2013) and instrument for the change in the domestic share of U.S. expenditure using Chinese import penetration into the five largest European economies. This allows us to use only the variation in the change of the domestic share that comes from supply changes in China and not from shocks originating in the U.S. 2 A second threat to identification is that supply shocks in the rest of the world (ROW) may be correlated with Chinese supply shocks. There are two reasons why this is unlikely to be a major concern. First, the rise of China is much more prominent than that of any other large country exporting to the U.S. during the time period under study. Second, we observe that across categories, our instrument has a positive effect on the China share of U.S. expenditure, but a negative effect on the ROW share. That is, Chinese imports are displacing both U.S. products and also products from the ROW. If correlated supply shocks among China and the ROW were a concern, we would instead expect a positive relationship between our instrument and the U.S. expenditure share on ROW products. 2 If demand shocks across the U.S. and the EU were correlated, this would lead to a bias in our estimates. However, we would only overestimate the true effect, if demand shocks across the U.S. and the EU were negatively correlated, which we regard as very unlikely. 3

4 This paper fits into a large and diverse literature on quantifying the effects of trade integration (see Costinot and Rodriguez-Clare (2013) for a recent survey). Within the micro approaches of this literature, there has been work examining the implications of international trade for productivity, inequality and labor markets. There has been surprisingly little work studying the effects on prices and consumer surplus. The most closely related papers to ours are Broda and Romalis (2008) and Broda and Weinstein (2006). In particular, Broda and Romalis (2008) showed that Chinese import penetration was associated with lower consumer prices in the U.S. Our work differs, however, in several important respects. First, we go beyond previous work by investigating the channels through which Chinese import penetration affects domestic prices, such as intensive margin price growth, variety effects, and the role of imported intermediate goods. Second, our main empirical equation is consistent with a large class of trade models, which allows us to make more easily the link between Chinese import penetration and U.S. welfare changes. This seems particularly relevant in light of the work by Autor et al. (2013) that focused on the negative employment consequences of Chinese import penetration. Finally, we adopt a different empirical strategy, using Chinese import penetration to Europe as an instrument for the change in the share of expenditure on domestic goods. Our paper is also closely related to the work by Amiti et al. (2016) who estimate the effect of China s WTO entry on the U.S. manufacturing price index. While their work concentrates more on the origins of the Chinese export surge due to productivity improvements and trade policy, our paper focuses more on the channels of adjustment in the U.S. by using barcode-level price data. This paper is also related to a much more recent and smaller literature on the implications of China s rise for the global economy (Autor et al. (2013), Hsieh and Ossa (2011), Costa et al. (2014)). Most closely related to our paper, Autor et al. (2013) and Autor et al. (2014) focus on the effect of rising Chinese imports on U.S. manufacturing employment and the long-run effects on workers labor market outcomes. We build on their work and, in particular, on their identification strategy, but consider instead the effects of Chinese imports on U.S. product markets and consumer prices. The rest of the paper is organized as follows. Section 2 will describe in detail our data sources and the construction of key variables. Section 3 will outline our empirical strategy and discuss summary statistics. Section 4 will present our results on inflation, product variety and turnover, as well as robustness checks. Section 5 offers some concluding remarks. 4

5 2 Data Sources and Methodology 2.1 Data Sources We use data from several sources in the empirical analysis. The two main datasets we employ are data on international trade flows at the HS 6-digit level, and data on household purchases and product prices at the barcode (Universal Product Code, or UPC) level. The international trade data used in this paper (i.e. bilateral imports and exports) come from BACI (Base pour l Analyse du Commerce International), which takes as input the UN s Commodities Trade Statistics database (COMTRADE), at the HS 6-digit level (2002). This is a widely used dataset, covering over 5,200 commodities and 240 countries/regions. The data on the prices of consumer goods and volume of purchases come from a proprietary dataset from AC Nielsen. This longitudinal dataset contains a sample of around 60,000 US households who continually provide information to Nielsen about their household demographics, what products they buy, as well as when and where the products were bought. The panelists were provided with in-home scanners to record all of their purchases. A key advantage of the Nielsen data is the availability of price and expenditure information at the barcode level. For the period we study, over 1.5 million UPC codes are present. These are grouped into roughly 1,100 different product modules by Nielsen. Some examples include olive oil, pasta-spaghetti, dental accessories, cameras, batteries and printers. In order to merge the two aforementioned datasets, we constructed a concordance between the digit HS codes and 1200 Nielsen "product modules". A large part of the HS codes are accounted for by intermediate goods (e.g. industrial chemicals, minerals, copper wires, semiconductor devices, etc.), and therefore do not match directly to Nielsen product modules. Often times, complex merges are required in which several HS codes and several Nielsen product modules are combined to form one product category. 3 The merge was carried out with the help of online tools such as the US Census Bureau s Schedule B Search Engine 4 and the Canadian Importers Database 5, which can be used to identify the relevant HS codes for a given product. We aimed to produce the largest number of merged categories possible, while ensuring all relevant Nielsen modules and HS commodities are included. The resulting concordance contains 3 For example, the Nielsen products "Milk chocolate" and "Dark chocolate" were combined with the HS 6-digit codes "Chocolate in blocks weighing less than 2kg" and "Chocolate in blocks weighing more than 2kg" to form one category named "Chocolate"

6 322 distinct categories, spanning 1138 Nielsen product modules and 846 HS 6-digit commodities. Examples of such categories include coffee, computer software, and creams and cosmetics. As mentioned in the introduction, our identification strategy requires the computation of Chinese import penetration in Europe. This, in turn, requires category-level expenditure and output data in Europe. We focus on the five largest European economies (Germany, France, United Kingdom, Italy and Spain), both because of their relative size and their availability of such output data for our study period. In particular, our output data for Europe come from the UNIDO Industrial Statistics Database at the 4-digit level of ISIC (INDSTAT 4). It contains information on more than 150 manufacturing sectors and sub-sectors for the period 1990 onwards. Here we follow a twostep procedure to construct expenditure data at the category level. First, we allocate output from each ISIC code to its corresponding HS codes, using a concordance obtained from WITS 6, and making a proportionality assumption where the output-to-export ratio is assumed to be constant within each ISIC code. Second, using expenditure data at the HS level, we then bring it to the category level using our own concordance between HS and Nielsen. A similar procedure is used to compute category-level expenditure data for the US, using output data at the NAICS 6-digit level from the Bureau of Economic Analysis (BEA). Finally, in order to study the indirect effects of Chinese imports on US consumer prices via intermediate inputs, we use the detailed direct requirements table from the BEA to measure input-output linkages across the roughly 430 BEA commodity codes. 2.2 Constructing Category-Specific Inflation Rates To construct category-specific inflation rates, we start with the following non-symmetric CES consumption function. Consumption in product category i at time t is given by an aggregate over different varieties k: C it = ( k a k i ) σ 1 σ 1 σ 1 σ c k it σ The terms a k i denote unobserved product quality, which is assumed to be constant for a barcodelevel product over time. The ideal price index for this consumption bundle is given by: 6 P it = ( k ) 1 1 σ a k i p k 1 σ it 6

7 When the set of goods remains constant over time, it is straightforward to compute the inflation rate, defined as π it = P it P it 1 From Sato (1976) and Vartia (1976), we can write inflation as the weighted geometric sum of individual price changes: ( p k it ) ω k it π it = k p k it 1 The weights ω k it sum to one and can be expressed as a function of expediture shares sk it = pk it ck it k pk it ck it : ω k it = s k it sk it 1 log(s k it ) log(sk it 1 ) k s k it sk it 1 log(s k it ) log(sk it 1 ) For very small changes of the market share, this weight equals the market share s k it. With a changing set of goods, Feenstra (1994) shows that inflation can be written as: π it = ( k I E t p k it p k it 1 ) ω k it } {{ } intensive margin ( λit λ it 1 ) 1 σ 1 }{{} extensive margin where I E it denotes the set of goods that are present in both time periods. The full inflation rate is now the product of the inflation rate at the intensive margin, multiplied by a correction factor that measures the contribution of the extensive margin (new and disappearing varieties). The term λ it of the correction factor is defined as follows: λ it = ExpenditureE it Expenditure it, which is the fraction of expenditure at time t that is going towards previously available varieties. Intuitively, when the share of expenditure going towards new varieties in period t exceeds the share of expenditure going towards old varieties in period t 1, the adjustment term reduces overall inflation. The extent to which the introduction of new goods affects the inflation rate depends on the elasticity of substitution σ. In particular, if the new goods are highly substitutable, the effect on inflation will be more muted. Given our sample, there are two alternative ways of computing the intensive margin inflation 7

8 rate. First, we can restrict the sample to include only those goods that appear in every year of the Nielsen data between 2004 and We call these goods long stayers. Second, we can restrict the sample to include all goods that appear in any consecutive pair of years. We call these goods short stayers. In other words, the second way of computing the intensive margin involves updating the definition of old and existing goods each year. It is worth noting that the price of each good p k it is the average unit price for that UPC code among all Nielsen purchases in a given year, while the expenditure on each good is the total among all Nielsen purchases that year. The expenditure shares of each good in a given year s k it is, in turn, computed as the ratio of total expenditure on a given UPC code that year over total expenditure on the product category it belongs to in the same year. Finally, to compute inflation rates, we need a value for the elasticity of substitution between product varieties. We assume a value of 5 in our baseline specification, which is roughly the mean of the elasticities of substitution estimated by Broda and Weinstein (2006) for product groups at a similar level of aggregation. In robustness exercises, we also consider different values, and find that the results are qualitatively unchanged. 2.3 Inflation and Changes in the Domestic Share of Expenditure (DSE) How does a positive supply shock in China affect prices in the U.S.? Arkolakis et al. (2012) show that in a large class of trade models, the domestic welfare effects of any foreign shock can be summarized by the change in the share of expenditure on domestically produced goods. In the simplest example, preferences in country j are Cobb-Douglas across categories of goods and CES within categories, trade follows an Armington model with perfect competition, labor is the only factor of production and labor supply is fixed. Utility in country j is: U j = k (C k j ) αk j with C k j = ( i a k i 1 σ c k ij ) σ σ 1 σ 1 σ The corresponding price index in category k is: P k j = ( i ) 1 1 σ a k i p k 1 σ ij 8

9 and the aggregate price index is: P j = ( P k j α k j ) α k j Taking the wage in country j as the numeraire, welfare in country j is given by the negative of the price index: W j = 1 P j Welfare changes are then related to category-level prices changes: d log(w j ) = k α k j d log(p k j ) A foreign shock (e.g. a productivity shock in China) affects the price index in the U.S. (country j) to the extent that there are changes in the U.S. bilateral terms of trade: d log(p k j ) = i θ k ij(d log(p k ij) d log(p k jj)) where θ k ij = Xk ij X k j is the share of expenditure on goods from country i. The changes in the bilateral terms of trade can be inferred from expenditure changes by using the import demand equations. We then obtain: d log(p k j ) = 1 1 σ i θ k ij(d log(θ k jj) d log(θ k ij)) = 1 σ 1 d log(θk jj) That is, the U.S. benefits from lower prices if the Chinese supply shock results in a lower share of U.S. expenditure on domestically produced goods. The change in the U.S. domestic share includes all relevant general equilibrium effects, i.e. indirect effects of the Chinese supply shocks on the U.S. bilateral terms of trade with third countries. In the simple case of the Armington model, the coefficient on the domestic expenditure share is related to the elasticity of substitution between different varieties. In a more general class of models, Arkolakis et al. (2012) show that it equals one over the trade elasticity. 2.4 Measuring Changes in DSE and Import Penetration Given the equation derived in Section 2.3, our main explanatory variable is the log change in domestic share of expenditure (DSE) for a product category i between 2004 and This is 9

10 computed as follows: DSE it = Expenditurei t Imports i t Expenditure i t We follow Autor et al. (2013) in measuring Chinese import penetration as the change in U.S. imports from China from , normalized by total U.S. expenditure in that product category for 2004: ImportP enetration i = Importsi CHN,2012 Importsi CHN,2004 Expenditure i 2004 Total U.S. expenditure on product category i are computed as total production plus imports minus exports. 7 We similarly compute a measure for Chinese import penetration of intermediate inputs (IPII). To do so, we collect input shares from the BEA input-output tables, and compute IPII as the weighted average of input-level import penetration ratios, with weights γ ij given by input shares: S IP II i = γ ij ImportP enetration j j=1 3 Empirical Strategy and Summary Statistics Our main empirical estimation exploits cross-product variation in the import penetration measures to identify the effect of trade shocks on consumer prices and product varieties in the US between 2004 and Following the equation derived in section 2.3, we relate the log change of the price index to the log change in the domestic share of expenditure in a given category: d log(p k ) = α + βdse k + ɛ k (1) In subsequent specifications we will decompose the effect on inflation into intensive and extensive margins. Specifically, we will have two measures of inflation at the intensive margin: first using the same set of goods in consecutive years, second using the same set of goods in all years. For the extensive margin, we use as the dependent variable the Feenstra correction factor outlined in Section 3, which captures the contribution to inflation from new product varieties. We then turn our attention to the effect of import penetration on the growth in consumer 7 We compute production at the category-level using production data for 6-digit NAICS industries, and using a crosswalk from NAICS to HS. 10

11 expenditure for category k, the change in the number of goods purchased as well as a measure of product turnover. 3.1 Identifying Trade Shocks The main threat to identification is that both prices and imports from China may be driven by demand or supply shocks in the U.S. rather than Chinese supply shocks. For instance, a positive U.S. demand shock should lead to higher U.S. prices and may also affect the U.S. domestic share of expenditure. Likewise, a positive U.S. supply shock would tend to lower U.S. prices and reduce the domestic share of expenditure, as U.S. products become more competitive. This bias would therefore go against us finding any effect. To address this concern, we follow Autor et al. (2013) and use Chinese import penetration into EU countries as an instrument. Intuitively, if the growth in US imports from China during is driven either by productivity growth in China or a reduction in trade barriers as a result of China s accession to the WTO, we should observe an increase in Chinese exports to other developed countries, such as the EU. In practice, we restrict our EU sample to its five largest economies: Germany, France, United Kingdom, Italy and Spain, due to insufficient data on industrial output at the ISIC - Rev 3 sector level for other EU countries. 8 A potential concern with the identification strategy is that demand or supply shocks may be correlated between the U.S. and EU countries. However, this would only be an important concern if shocks across the U.S. and the EU were negatively correlated. If, for instance, demand shocks between the U.S. and the EU were positively correlated, one would observe an increase in Chinese import penetration into the EU and also an increase in U.S. prices, which is contrary to what we find. A second threat to identification is that supply shocks in the rest of the world (ROW) may be correlated with Chinese supply shocks. We believe that this is unlikely to be a major concern. First, the rise of China in world trade and among U.S. imports is much more prominent than that of any other large country exporting to the U.S. during the time period under study. Figure?? shows the evolution of U.S. import shares by origin region. While the shares of EU countries, Canada and Mexico, and other emerging countries has either remained stable or declined over time, China s share in U.S. imports has increased by almost 10 percentage points from Data on output at the sectoral-level is needed in order to compute total expenditure at the product level. This, in turn, is needed to construct our import penetration measures. 11

12 12. Second, we observe that our instrument has a positive effect on the China share of U.S. expenditure, but a negative effect on the ROW share. That is, Chinese imports are displacing both U.S. products and also products from the ROW. If correlated supply shocks among China and the ROW were a concern, we would instead expect a positive relationship between our instrument and the U.S. expenditure share on ROW products. With these points in mind, we run the following first-stage regression: DSE k = γ + δm k EU,Chn + ɛ k (2) where m k EU,Chn is the China import penetration measure for the EU in category k. 3.2 The Role of Intermediate Inputs Besides the direct effect of import penetration on final consumer goods, there could also be an indirect effect via intermediate inputs. For example, the price of PCs in the US may be affected by imports of PCs, as well as by imports of semi-conductors. To capture this indirect effect, we construct a measure called import penetration in intermediate inputs (IPII), for each of our 230 product categories. In our regressions, we include this as an additional explanatory variable, instrumented similarly by an analogous measure for the five EU economies. 3.3 Summary Statistics Table 1 shows summary statistics for some of our key outcomes and regressors. Assuming an elasticity of substitution of 5, the average cumulative inflation rate during was 2.4%, or 0.3% per annum. Using only goods that are present in consecutive years, and therefore disregarding the contribution of new varieties through the Feenstra correction factor, the average cumulative inflation rate becomes 14.1% during , or 1.6% per annum. There is a great deal of variation across product categories for each of these measures, with standard deviations of 26 and 20 percentage points respectively. At the same time, the U.S. domestic share of expenditure declined in the majority of consumer goods categories. The average decline equals 0.15 log points, driven mostly by some categories with very large import penetration. The median category saw a decline of the domestic share by roughly 4.3%. There is also a large amount of variation in the change of the domestic share of expenditure across categories, with a standard deviation of 0.38 log points. 12

13 The distribution of the log change in the domestic share of expenditure is illustrated in figure??. Finally, Chinese imports into the EU also grew rapidly. For the average category, the increase in EU imports from China from equaled roughly 5% of total expenditure in Results In this section we discuss the estimated effects of import growth from China on inflation and product variety in the United States between 2004 and We will examine both the intensive and extensive margins of inflation, separately identifying the contribution of new varieties. We will test whether any effect on inflation differs across income groups as well as regions. We will also split our sample period into three sub-periods ( , and ), corresponding to pre-, during- and post-recession respectively. Finally, we will check for robustness of our results to alternative values of the elasticity of substitution. 4.1 Effect on Inflation Rates Before discussing the results on inflation, Table 2 (Panel B) presents the first-stage of the IV regression. Columns 3 and 4 are unweighted and weighted specifications respectively. The main coefficient on China s import penetration in the five-largest European economies is strongly negative and highly statistically significant across both specifications. Intuitively, the product categories that saw a larger increase in Chinese import growth to Europe also saw a larger decrease in the share of US domestic expenditure. Table 3 decomposes this first-stage relationship further by looking separately at the relative China share in US expenditure, as well as the rest of the world share. Here we observe that a rise in Chinese import growth to Europe is positively correlated with the change in relative China share of US expenditure, and negatively correlated with the change in the rest of the world share. Taken together, this implies Chinese imports are displacing both US products and also ROW products. Table 2 (Panel A) presents our estimated effects of changes in the share of domestic expenditure on US inflation rates. Columns 1 and 2 are simple OLS estimates, while columns 3 and 4 are results when the US domestic share of expenditure is instrumented by Chinese import penetration in Europe. 13

14 As discussed in the previous section, if demand shocks in the U.S. affect both prices and imports from China, then the least squares specifications would underestimate the true effect. We see some suggestive evidence of this in our results. While the OLS and IV coefficient estimates are both positive and highly statistically significant, the IV estimates are larger in magnitude. For instance, in the unweighted IV specification (Column 3), comparing a product category with median change in the share of domestic expenditure to one with no change, prices in the median category grew by 3.5% less cumulatively over our study period, or 0.4% less per year. Using estimates from the weighted regression yields a slightly larger effect. 4.2 Intensive vs. Extensive Margins of Inflation Rates Given the formulation of our price index, inflation rates could be affected either by (i) changes in the prices of existing goods (the intensive margin), or (ii) the introduction of new goods (the extensive margin). As discussed in the previous section, Feenstra (1994) develops a convenient decomposition of the inflation rate into these two margins, which allows us to estimate the effect on them separately. The results are presented in Table 4 (Panel A). There are two different ways of computing the intensive margin, depending on the definition of "existing" goods. First, we can compute the year-on-year inflation rate, using only the set of goods that are consumed in both periods. We then update the set of "existing" goods for each consecutive pairs of future years, before compounding these annual inflation rates into a cumulative value. This is the procedure followed in constructing our short-stay inflation measure (columns 1 and 2). Alternatively, we can focus exclusively on the set of goods which are consumed in all of the years in our sample. This is the approach followed in constructing our long-stay inflation measure (columns 3 and 4). Finally, columns 5 and 6 present results on the extensive margin, which captures the contribution of new goods. Both the unweighted and weighted IV coefficient estimates suggest that there is a sizeable effect on the intensive, as well as the extensive margin. Results are smaller in magnitude, but not in statistical significance, if we use the long-stay, rather than the short-stay inflation measure. 4.3 Expenditure, Varieties and Product Turnover Given the richness of the Nielsen data, there are a number of additional outcomes we can examine. Columns 1 and 2 of Table 4 (Panel B) report results on expenditure growth. Here we 14

15 do not find any effect of a change in the share of domestic expenditure on overall expenditure growth in a given product category. Columns 3 and 4 report results for the number of different product varieties (as proxied by UPC/bar-codes) consumed by households. Here we find that variety growth is 1% higher in a category with median change in the share of domestic expenditure compared to one experiencing no change. Beyond variety growth, we can test whether import growth from China led to more product turnover. Here we construct a measure of product turnover inspired from the labor literature on worker turnover as follows: Turnover ct = N Entry ct + Nct 1 Exit N ct 0.5(N ct + N ct 1 ) That is, we measure the degree of product turnover between periods t and t 1 as the ratio of turnover in excess of the absolute change in the number of products, divided by the average number of products in t and t 1. We compute this turnover variable for each category and year, and then take an average within a category over time. Results are in columns 5-6 and show a higher degree of turnover in categories where the share of domestic expenditure fell. 4.4 Heterogeneous Effects by Income Group and Region Another advantage of using the Nielsen Homescan data is that we can estimate heterogeneous treatment effects for different income groups and regions within the US. Here we first divide individuals into five similarly sized groups according to their reported annual household income, before carrying out our analysis separately for each of the four US census regions. The results are shown in Table 5. The first finding is that our earlier result on inflation holds across income groups and regions. While it is true that the point estimate decreases monotonically with income, suggesting that poorer households benefited more from the growth in Chinese imports, these differences are not statistically significant at conventional levels. Similarly, we cannot reject the hypothesis of equal treatment effect across the four census regions. 4.5 Effect of Intermediate Inputs Theoretically the effect of Chinese import growth can affect US inflation both via final goods and intermediate inputs. To investigate this additional channel, we construct a measure for the 15

16 share of domestic expenditure in intermediate inputs in the US using input-output tables from the BEA, before instrumenting for it using an analogous measure for Europe. Table 6 presents the IV results, where we include this measure as an additional regressor. Here we first note that the effect of changes in the share of domestic expenditure in final goods remains unchanged, while an increase in the share of domestic expenditure in intermediate inputs during our study period does not have a statistically significant effect on US inflation. 4.6 Sub-Period Analysis and Robustness Table 8 presents our inflation results by sub-period. We split up our sample into the period prior to the Great Recession ( ), during- ( ), and post-recession ( ). The estimated coefficients are positive and economically meaningful for the pre- and post-recession periods. Compared to Table 2 (Panel A), the point estimates are naturally lower. For the time during the recession, however, we find that the instrument does not have sufficient predictive power. The F-test value of 2.97 is substantially below acceptable levels. The irrelevance of the instrument during the recession period suggests that supply shocks in China were not the main driver of changes in trade flows to the U.S. and Europe during that time period. This is consistent with evidence by, for instance, Eaton et al. (2013). Table 9 presents results from a robustness check, using alternative values of the elasticity of substitution in computing cumulative inflation rates. Specifically, we experimented with values of 2, 3 and 10 (while the main regressions used a value of 5). In all cases our results are essentially unchanged, both qualitatively and quantitatively (although the point estimates are necessarily larger for smaller values, as the contribution of new goods is more important in these cases). 4.7 Aggregation Using our cross-sectional estimates, we can carry out a simple aggregation exercise, where we compute the effect of changes in the share of domestic expenditure, due to Chinese import growth, on an overall consumer price index. To do so, a number of additional assumptions are needed. First, we assume that the effect on inflation is zero for the category in which the share of domestic expenditure has remained constant. We make a similar assumption for the effect of intermediate inputs. Next, we need to make an assumption regarding utility across different categories. Since we observe no effect on category-level expenditure, we postulate a Cobb-Douglas formulation across different categories. 16

17 We denote the elasticity with respect to category i by ω i. We then compute the yearly aggregate change in inflation due to Chinese import growth as follows: log ( P2012 ) China = ω i ˆβ log(θi ) = 6.9% P 2004 i This suggests that Chinese import penetration reduced U.S. inflation for the consumer goods we observe by 0.9% per year. 5 Concluding Remarks In this paper we have tried to estimate the size of U.S. consumer gains from the recent growth in Chinese imports by adopting a micro-econometric approach. Using barcode-level price data from AC Nielsen and exploiting cross-product variation in import penetration between 2004 and 2012, we find that prices declined by more in product categories with higher Chinese import penetration: comparing a category with median China trade shock to one with no change, prices in the median category grew by 0.4% less per year. We addressed the potential endogeneity of imports by using Chinese exports to European countries as an instrument. We find that the effect of import penetration on prices is driven by both the intensive and extensive margins, suggesting the presence of pro-competitive effects for old goods as well as variety gains from new goods. In addition, we find evidence for net gains in the number of consumed varieties, and newly introduced goods appear to strongly displace old varieties. Finally, we find no significant evidence for reductions in final goods prices induced by imported intermediate inputs from China. Taken together, these results suggest substantial gains to U.S. consumers from the recent growth in trade with China. These have the potential to offset at least some of the negative labor market consequences due to import competition. 17

18 References Amiti, M., Dai, M., Feenstra, R., and Romalis, J. (2016). How did china s wto entry benefit u.s. consumers? mimeo, FRBNY. Arkolakis, C., Costinot, A., and Rodríguez-Clare, A. (2012). New trade models, same old gains? The American Economic Review, 102(1): Autor, D., Dorn, D., and Hanson, G. H. (2013). The china syndrome: Local labor market effects of import competition in the united states. The American Economic Review, 103(6): Autor, D., Dorn, D., Hanson, G. H., and Song, J. (2014). Trade adjustment: Worker-level evidence. The Quarterly Journal of Economics, 129(4): Broda, C. and Romalis, J. (2008). Inequality and prices: Does china benefit the poor in america? mimeo, University of Chicago. Broda, C. and Weinstein, D. (2006). Globalization and the gains from variety. Quarterly Journal of Economics, 121. Costa, F., Garred, J., and Pessoa, J. P. (2014). Winners and losers from a commodity-formanufactures trade boom. mimeo, FGV. Costinot, A. and Rodriguez-Clare, A. (2013). Trade theory with numbers: Quantifying the consequences of globalization. Handbook of International Economics, 4. Eaton, J., Kortum, S., Neiman, B., and Romalis, J. (2013). Trade and the global recession. mimeo, University of Chicago. Feenstra, R. C. (1994). New product varieties and the measurement of international prices. American Economic Review, 84(1): Hsieh, C.-T. and Ossa, R. (2011). A global view of productivity growth in china. Technical report, National Bureau of Economic Research. Sato, K. (1976). The ideal log-change index number. Review of Economics and Statistics, 58: Vartia, Y. (1976). Ideal log-change index numbers. Scandinavian Journal of Statistics, 3:

19 Figure 1: Changing Composition of US Expenditure Figure 2: Changing Composition of US Imports 19

20 Figure 3: Product Heterogeneity Figure 4: Change in Domestic Share of US Expenditure 20

21 Figure 5: First Stage Figure 6: Effect on Inflation 21

22 Table 1: Summary statistics Panel A: Inflation (category level, ) Variable Mean Std. Dev. Min Max Obs Inflation (Elasticity of Substitution = 3) Inflation (Elasticity of Substitution = 5) Inflation (Elasticity of Substitution = 10) Panel B: Inflation (intensive vs. extensive margin, ) Inflation (using only goods present in consecutive years) Inflation (using only goods present in all years) Inflation (Feenstra Correction Factor, Elasticity of Substitution = 5) Panel C: China Import Penetration, Log Change in Domestic Share of Expenditure (US) Import Penetration (Europe)

23 Table 2: Import Penetration and Inflation ( ) (1) (2) (3) (4) Panel A: Cumulative Inflation OLS IV Share of Domestic Expenditure (US) 0.166** 0.221** 0.873*** 1.083*** (0.074) (0.107) (0.201) (0.253) Weights No Yes No Yes R n/a n/a N Panel B: First Stage China Import Penetration (Europe) *** *** (0.412) (0.277) Weights No Yes R N F-stat Notes: Robust standard errors in parentheses. *** p<0.01, ** p<0.05,* p<0.10. Share of domestic expenditure is the log change in the same variable for the US during The instrument used is China import penetration in the five largest European economies (Germany, France, United Kingdom, Italy and Spain). In constructing the inflation rate reported here, we have used a within-category elasticity of substitution equal to 5. The weights used in columns 2 and 4 are total US expenditure in a given category in The sample includes all product categories that fall within the 1st and 99th percentiles of China import penetration, cumulative inflation and share of domestic expenditure in the US. All import penetration measures are defined as (Imports 2012 Imports 2004 )/Expenditure

24 Table 3: Decomposing the First Stage (1) (2) (3) (4) Relative China Share Relative ROW Share Chn Import Penetration in EU, *** 1.486*** *** *** (0.590) (0.229) (0.449) (0.150) Observations R-squared Weights No Yes No Yes Notes: Robust standard errors in parentheses. *** p<0.01, ** p<0.05,* p<0.10. Share of Chinese (rest of the world) expenditure is the simple change in share of U.S. expenditure towards goods from China (the rest of the world). The instrument used is China import penetration in the five largest European economies (Germany, France, United Kingdom, Italy and Spain). The weights used in columns 2 and 4 are total US expenditure in a given category in The sample includes all product categories that fall within the 1st and 99th percentiles of China import penetration, cumulative inflation and share of domestic expenditure in the US. All import penetration measures are defined as (Imports 2012 Imports 2004 )/Expenditure

25 Table 4: Import Penetration, Inflation, Expenditure Growth, and Product Scope ( ) (1) (2) (3) (4) (5) (6) Panel A: Inflation (Various Margins) Short-Stay Long-Stay New Varieties Share of Domestic Expenditure (US) 0.533*** 0.621*** 0.266*** 0.320*** 0.340*** 0.462*** (0.122) (0.162) (0.060) (0.112) (0.084) (0.111) Panel B: Additional Outcomes Expdt Growth No. of Products Excess Turnover Share of Domestic Expenditure (US) ** *** *** *** (0.125) (0.229) (0.121) (0.272) (0.093) (0.133) 25 Weights No Yes No Yes No Yes N st-stage F-stat Notes: Robust standard errors in parentheses. *** p<0.01, ** p<0.05,* p<0.10. In Panel A, columns 1-4 present results on the intensive margin of the inflation effect, while columns 5-6 the extensive margin. Specifically, columns 1-2 use an inflation rate constructed using the same set of goods throughout the sample period, while columns 3-4 study the effect on an inflation rate constructed using the same set of goods in consecutive years. In other words, we have excluded the effect of new goods on inflation in columns 1-4. The instrument used is China import penetration in the five largest European economies (Germany, France, United Kingdom, Italy and Spain). The weights used are total US expenditure in a given category in The sample includes all product categories that fall within the 1st and 99th percentiles of China import penetration, cumulative inflation and share of domestic expenditure in the US. All import penetration measures are defined as (Imports 2012 Imports 2004 )/Expenditure 2004.

26 Table 5: Heterogeneous Effects of Import Penetration on Inflation ( ) (1) (2) (3) (4) (5) Panel A: By Income Group < 30k (30k 50k) (50k 70k) (70k 100k) > 100k Share of Domestic Expenditure (US) 0.617*** 0.575*** 0.549*** 0.495*** 0.488*** (0.122) (0.103) (0.096) (0.086) (0.077) Weights Yes Yes Yes Yes Yes N st-stage F-stat (1) (2) (3) (4) Panel B: By Region Northeast Midwest South West 26 Share of Domestic Expenditure (US) 0.538*** 0.545*** 0.622*** 0.498*** (0.100) (0.101) (0.111) (0.085) Weights Yes Yes Yes Yes N st-stage F-stat Notes: Robust standard errors in parentheses. *** p<0.01, ** p<0.05,* p<0.10. In Panel A,. The instrument used is China import penetration in the five largest European economies (Germany, France, United Kingdom, Italy and Spain). The weights used are total US expenditure in a given category in The sample includes all product categories that fall within the 1st and 99th percentiles of China import penetration, cumulative inflation and share of domestic expenditure in the US. All import penetration measures are defined as (Imports 2012 Imports 2004 )/Expenditure 2004.

27 Table 6: Import Growth in Intermediate Inputs and Inflation ( ) Dependent Variable: Cumulative Inflation (1) (2) IV Share of Domestic Expenditure in Final Goods (US) 0.821*** 1.047*** (0.182) (0.277) Share of Domestic Expenditure in Intermediate Inputs (US) (2.020) (3.834) Weights No Yes N st-Stage F-stat Notes: Robust standard errors in parentheses. *** p<0.01, ** p<0.05,* p<0.10. Share of domestic expenditure is the log change in the same variable for the US during The instrument used is China import penetration in the five largest European economies (Germany, France, United Kingdom, Italy and Spain). In constructing the inflation rate reported here, we have used a within-category elasticity of substitution equal to 5. The weights used in columns 2 and 4 are total US expenditure in a given category in The sample includes all product categories that fall within the 1st and 99th percentiles of China import penetration, cumulative inflation and share of domestic expenditure in the US. All import penetration measures are defined as (Imports 2012 Imports 2004 )/Expenditure

28 Table 7: Barcode level: Average effect on products (1) (2) (3) (4) (5) (6) (7) Price Expenditure Market Share Exit Price Expenditure Market Share Share of Domestic Expenditure (US) 0.786*** 1.190* *** 1.357*** 1.244*** (0.276) (0.718) (0.751) (2.090) (0.102) (0.378) (0.323) Observations 496, , , , , , ,002 Weights No No No No No No No Sample All All All All Stayer Stayer Stayer Price Expenditure Market Share Exit Price Expenditure Market Share 28 Share of Domestic Expenditure (US) 0.618*** 2.066** 1.957** *** 0.298*** 1.315* 1.177* (0.162) (0.975) (0.962) (1.343) (0.076) (0.733) (0.651) Observations 496, , , , , , ,002 Weights Yes Yes Yes Yes Yes Yes Yes Sample All All All All Stayer Stayer Stayer Notes: Robust standard errors in parentheses. *** p<0.01, ** p<0.05,* p<0.10. The instrument used is China import penetration in the five largest EU economies (Germany, France, United Kingdom, Italy and Spain). The sample includes all products observed in 2004.

29 Table 8: Import Penetration and Inflation (sub-period analysis) (1) (2) (3) (4) (5) (6) Dependent Variable: Inflation Inflation Inflation Share of Domestic Expenditure ( ) 0.640** (0.255) (0.344) Share of Domestic Expenditure ( ) (0.193) (0.055) Share of Domestic Expenditure ( ) (0.187) (0.259) Weights No Yes No Yes No Yes N Notes: Robust standard errors in parentheses. *** p<0.01, ** p<0.05,* p<0.10. The instrument used is China import penetration in the five largest European economies (Germany, France, United Kingdom, Italy and Spain). The weights used are total US expenditure in a given category in The sample includes all product categories that fall within the 1st and 99th percentiles of China import penetration, cumulative inflation and share of domestic expenditure in the US.