Exchange rate pass through of China s Exporters Abstract

Transcription

1 Exchange rate pass through of China s Exporters Abstract This paper offers a neat method of estimating how export prices react to exchange rate shocks by precisely controlling exporters marginal costs based on a comprehensive database from China customs. It utilizes the observation that most China s exporters export their product to more than one destination. As the same product of an exporter exported to multiple destinations should have the same marginal cost, we construct a difference in difference estimator at firmproduct level to eliminate the effect of the marginal cost on prices. We find significant evidence of producer currency pricing for China exporters during ; the aggregate exchange rate pass through to export prices denominated in foreign currency is nearly complete (98.6%). With the proposed estimation method, this will be the first paper to my knowledge that is able to estimate and compare the distributions of the exchange rate pass through across different industries, trade modes and firm types. 1

2 1. Introduction How export prices react to exchange rate shocks is one of the classic topics studied in international macroeconomics. The price of the export depends on many factors including the nominal exchange rate, the marginal cost of the product, the degree of the vertical and horizontal competition and the distribution cost. In order to get a precise estimate of the percentage changes of export prices in responds to a one percent exchange rate shock (termed as the exchange rate pass through), all other factors that influence the price should be isolated or controlled. It poses great difficulties to empirical works on estimating the exchange rate pass through as the marginal cost of an exporter is in general unobservable. The project offers a new identification strategy that overcomes the above problems and provides more precise estimates of the exchange rate pass through based on a comprehensive database from the Chinese custom. 1.1 Our Current Progress This project can be divided into three phases: 1) purchasing and processing the data; 2) programing and implementing our identification strategy and; 3) analyzing results and performing robustness checks. To assure that we get the best price, we had a long negotiation with the data provider HuaMei Information. Last week, we have finalized and signed the contract. Although we haven t got the full access of the data, we have been given a short period of limited access of the database. Based on this access, we have designed the key identification strategy of our exchange rate pass through estimation, its associated matching algorithms and programmed the required program in R project to implement our identification strategy. The key results are given below. 2. Key Findings 2.1 The aggregate exchange rate pass through to export prices denominated in foreign currency is nearly complete (98.6%) Empirical studies have emphasized three important observations of ERPT. First, the ERPT is far from complete for import prices. Campa and Goldberg (2005) find that the ERPT for import prices are 0.52 and 0.46 for the Euro area and OECD countries respectively. Second, they also provide the evidence that incomplete ERPT is not only a short run phenomenon but keeps incomplete 64% after one year. This means the price rigidity along is unlikely give a good explanation for the incomplete ERPT. Third, the ERPT is declining over time. 2

3 Marazzi et al. (2005) report the ERPT to import prices in the United States has declined considerably from around 0.5 in the 1980s to roughly 0.2 during the last decade. The recent literature by Gopinath, Itskhoki, and Rigobon (2010) and Gust, Leduc, and Vigfusson (2010b) confirm with this finding. Campa and Goldberg (2005) also show that the ERPT has a decline trend for overall OECD countries in the last two decades. Compared to above estimations using aggregate level price indices, our micro approach finds a substantial high exchange rate pass through of China s exporters. The estimated quarterly exchange rate pass through is around 98.6%, which means after an exchange rate shock, the export price denominated in RMB only move by 1.4% and pass 98.6% of the shock to the foreign prices. If a product was selling to the United Kingdom at a price of 500 RMB or 50 pounds 1 and there is a sudden 10% depreciation of sterling pounds, our estimation would suggest the RMB price of the product would only decrease by 0.14% (1.4*0.1%) and the price denominated in sterling will increase by 9.86% (98.6*0.1%). The product will be sold in the market at the price of [500*(1-0.14%)] RMB or [50*(1+9.86)%] pound. Our estimation suggests that Chinese exporters nearly fully passed the exchange rate shocks to their trade partners during 2000 and Figure 1: Exchange rate pass through estimates of the existing literature While being different from some aggregate pass through estimates from developed countries, our result is consistent with recent papers that were using 1 The implied exchange rate between sterling and RMB was 1:10. 3

4 highly disaggregated micro data to estimate exchange rate pass through. As illustrated in figure 1, the exchange rate pass through of exporters from France and Belgium are relatively low. On the other aspect, Kim et al (2013) estimated an average pass through of the US imports from China during 2005 and 2008 and find an average pass through of 80% (imports)/ 20% (exports). They also find that a large fraction of goods in their dataset never changed their prices. The high pass through of China s exporters during 2000 and 2006 could be explained by various aspects including business mode, market share, import intensity and the degree of homogeity and degree of competition in the field. For instance, Amiti, Itskhoki and Konings (2014) find that Belgium exporter with for low import intensity and low export share tend to do less pricing-to-market and thus will pass more exchange rate shocks in to their prices denominated in the destination currency. Berman, martin and Mayer (2014) also find the exporters with lower productivity will do less pricing to market and pass more exchange rate shocks to their trade partners. These studies suggest China s exporters during 2000 to 2008 may be relatively less productive and owned low market shares compared to their global competitors. We will further explore this part once we get the full access of the database Our identification strategy v.s. the conventional approach From micro pricing theories, the optimal price of a product depends on its marginal cost, its market share, its distribution cost and other macro factors, such as the bilateral exchange rate and the consumer price index. The conventional estimation equation is given in figure 2, where p is the logged export prices, e is the nominal exchange rate, mc represents marginal cost. i, c, t represents product identifier, country identifier and the time period respectively. We take first difference D for our variables as export prices and exchange rates are nonstationary series. The exchange rate pass through estimator is given by b. The difficulty of pass through estimation lies in the fact that the marginal cost is in general unobserved in the data. The existing literature has used various proxies to control for the marginal cost. For estimations of the aggregate pass through, the most used two are the nominal wage index and the PPI of the exporting country. Although they may be good proxies of the average cost, they fail to capture the true marginal cost. 4

5 Figure 2: The conventional estimation equation and various proxies used for the marginal cost In the past few years, a range of highly disaggregated firm level datasets emerged and became available to academic researchers. These datasets have improved the approximation accuracy of the marginal cost by taking in to account the estimated productivity of the exporter and the cost of its total imported inputs. This method gives the most precise proxy of the firm s marginal cost so far. However, the productivity may differ across products within a firm and it is not clear which imported input is used to produce the exported product. This paper provides an innovative and neat identification strategy to control the marginal cost, which enables us to estimate exchange rate pass through at firm and product level. This method is implemented and tested in a Chinese custom database, named Chinese imports and exports of customhouse database. It covers all entries of China s exports and imports at 8-digit harmonized system (HS) code level on monthly basis from 2000 to On top of the conventional information like the price and quantity, this database contains information about trading firms, the location of production and trade classifications (general trade, processing materials supplied by clients, processing with imported materials, etc.) More importantly, this database enables us to trace trade transactions at the firm level, which greatly enlarges the scope of the research methods that can be applied and gives the possibility of estimating the exchange rate pass through in a more precise way. The estimation method can be implemented by two steps. The first step utilizes the observation that most China s exporters export their product to more than one destination. As the same product of an exporter exported to multiple destinations should have the same marginal cost, we construct a difference in 5

6 difference estimator at firm-product level to eliminate the effect of the marginal cost on prices. For example, during July 2004, the exporter with identifier was exporting tomato paste in airtight containers to 7 different destinations 2. If we take differences between the pricing equation of two different destinations, Korean and Vietnam for example, the unobserved marginal cost will be differenced out. Therefore, for each matched pair, we will have two pass through estimation with b KR and b VN being the pass through estimation of Tomato paste in airtight containers from firm for Korean and Vietnam respectively. Following the same procedure, for each exporter that was selling this product during , we will be able to construct a series of price variations that is orthogonal to the change of the marginal cost. Figure 3: An illustration of the method used In the second step, we will run panel regressions with fixed effects at the firmproduct level with destination CPI as controls. Out main estimation equation is 2 Unit value is denoted in US dollars. 6

7 given in figure 4. Figure 4: The main estimation equation Figure 5 gives a summary of our pass through estimates for China s exporters. The pass through estimate is calculated for top 30 trade partners that China was exporting during 2000 and In the first two columns, we run the fixed effect regression with differenced logged exporter prices as the dependent variable. As a result of our matching process, the pass-through estimates are given in pairs. The coefficient in front of e 1 (nex1) and e 2 (nex2) gives the exchange rates pass through estimate for the first group of countries and the second group of counties 3. The exchange rate pass through to prices denominated in RMB is , which means 1 percent change of bilateral exchange rate will only lead to prices denominated in RMB to change percent. The second column adds one lag to our base line estimation. The coefficients in front of these two first lags are not significant 4 and the pass-through estimates do not differ significantly from our base line estimation. After estimating exchange rate pass through of prices, we repeat the analysis using the volume (column 3) and the traded value (column 4) as the dependent variable to estimate the elasticity of volume and traded value with respect to exchange rates. One percent exchange rate appreciation of RMB leads to around 0.065% decrease of quantity exported and % of total value traded. It is worth to note that these two elasticities are still relatively low compared to other countries, e.g. Belgium ( ) and 3 In this preliminary analysis, the destination countries in these two groups are randomly assigned. We will categories the countries into groups according their economic performances and geographic aspects the once we got the full access of data. 4 The clustered heterosedastic robust standard errors has been reported in the brackets and used to test the significant of coefficients. 7

8 France ( ). Figure 5: Aggregate exchange rate pass through 8

9 On the top of our main identification strategy, we have also tried other model specifications given in figure 6. Figure 6: Other Algorithms have been written and applied to the sample data 9

10 2.2 Summary Statistics of China s Trade Activities during Figure 7: Main trade partners 10

11 Figure 8: Trade Classifications Export Import % 1%1% 6% 3%5% % % 44% % 29% 0.50 Assembling Supplied Materials General Trade Goods Stored at Special Tax free Zones Goods Traded at Special Tax free Zones 0.50 Imported Inputs for Foreign invested Enterprises Processing Imported Materials Small Scale Border Trade The Rest Figure 9: Major Customs Export 0.00 Import % 7% 7% 10% 25% 18% % 6% 6% 21% 11% 15% Huangpu Qingdao Shanghai Shenzhen The Rest Tianjin 11

12 Value per Unit Figure 10: Transportation Methods Export 0.00 Import % 1% 16% 2% % 62% % 61% Air Motor Veicale Rail Sea Figure 11: The unit value (in USD) of China s exports is increasing faster than is exports Exports Imports Balance

13 Volume (in billion units) Value in Billion USD Figure 12: Trade Balances Exports Imports Balance Figure 13: The trend of quantity/volume traded Exports Imports Balance

14 3. Revised Project Plans 17 April: Get full access of the database 17 April - 1 May: conducting empirical analysis using already programmed algorithms 1 May 31 May: extensions and robustness checks. 1 June 31 June: writing up. 14