EC910 Econometrics B. Exchange Rate Pass-Through and Inflation Dynamics in. the United Kingdom: VAR analysis of Exchange Rate.

Transcription

1 EC910 Econometrics B Exchange Rate Pass-Through and Inflation Dynamics in the United Kingdom: VAR analysis of Exchange Rate Pass-Through Department of Economics The University of Warwick Abstract This aim of this study is to analyse exchange rate pass-through and domestic price with a VAR which allows for investigation on ERPT effects along the distribution chain. 1

2 1 Introduction Exchange rate pass-through (ERPT) is the percentage change in local currency import prices following a percentage change in exchange rate between importing and exporting countries. A one-to-one response of import prices to exchange rate changes is known as a complete pass-through. The rate of ERPT has important implications for the effect of monetary policy on domestic inflation as well as for the transmission of macroeconomic shocks and the volatility of the real exchange rate. Although there are a vast amount of literature investigated on the relationships between exchange-rate pass-through and import price and the relationship between import price and inflation dynamics in the UK, this paper contribute to the literature by analysing exchange rate pass through with import price and then consumer price with VAR which allows us to account for the fact that distribution costs along the price chain may cause a decline in the degree of pass-through from upstream to downstream prices (Campa and Goldberg, 2008). From a policy perspective, the pass-through into consumer prices has important implication for the effects of monetary policy on domestic imflation. Quarterly data from 1986 Q2 to 2009 Q3 is used in this study. 2

3 2 Studies on ERPT Campa and Goldberg (2005) looked at cross-country time series and then industry specific data to analyse exchange rate pass-through (ERPT) into import prices of 23 OECD countries by OLS regression on variables in log differences. They have also analysed differences among countries in aggregate pass-through where they concluded that pass-through into import price is lower for countries with low inflation and low exchange rate volatility. They have also looked at data in a disaggregated level where they have divided import prices into five categories and concluded that ERPT for manufacturing and food products are incomplete where energy and raw material imports has a result closer to unity. Mumtaz et al. (2006) looked at cross-sectional variation in the estimated industry-specific pass-through rates and they have found evidence for short-run and long-run partial pass-through into food and manufacturing sectors. Furthermore, they tested on variation on estimated short-run ERPT rates and explore the possibility of structural breaks over their sample. They considered a simple time-varying parameter model to capture any variation in the estimated industry-specific short-run ERPT rates over time. There are various literatures which looked at ERPT with a vector autoregressive model. Ito and Sato (2006) did a VAR analysis on five variables to examine 3

4 pass-through effects of exchange rate changes on domestic prices in the East Asian economies. They have stated that VAR analysis has advantages over single equation analysis on ERPT. This technique allow us to identify shocks through Choleski decomposition and it enables us to analyse ERPT into set of domestic prices along the distribution chain from import level, producer level to the consumer level. An and Wang (2001) estimated ERPT for nine OECD countries by a VAR with sign restrictions which tends to emphasise on the distribution chain where they then analysed the extent of ERPT to price along the distribution chain by examining the impulse responses of prices with respect to exchange rate shock. They have found that the degree of ERPT declines along distribution chain. 4

5 3 Methodology 3.1 Data This study aims to investigate the transmission of fluctuation in exchange rate to import price and domestic prices. Variables being used in this study are listed in table 1 below. Table 1 Data Source Sample Size Consumer Price Index of All-Items in the UK Organisation for Economic 1973 Q Q4 Co-operation and Development Industrial Production Index Datastream 1985 Q Q4 Import Price Index Datastream 1973 Q Q4 UK Money Supply M0: notes & Coins in circulation outside Bank of England Datastream Nominal Effective Trade-weighted Exchange Rate Datastream 1985 Q Q Q Q3 oil price Datastream 1973 Q Q4 Producer Price Index Datastream 1980 Q Q3 Output gap Datastream 1980 Q Q4 5

6 Table 2 Variables Consumer Price Index industrial Production Index Import Price Index Money Supply (M0) Nominal Effective Exchange Rate(Trade weighted) oil price Producer Price Index Output gap cpi ipi imp money_supply neer oilprice ppi outputgap To test for unit roots, a Dickey-Fuller with GLS (DF-GLS) is used to test on all variables (with 4 lags). The results reflect that most of the variables are non-stationary where output gap is stationary which matches to other studies. The following table (table 3) presents the test statistics of each variables used in this study. All variables (except output gap) are in natural log. 6

7 Table 3 variables (in natural log) cpi imp money_supply neert oilprice ppi outputgap test statistics of test statistics of test statistics first difference 2nd difference (critical value) (critical value) (critical value) (-2.900) (-2.925) (-2.956) (-2.932) (-2.892) (-3.448) -2.4 (-1.943) order of Integration I(2) I(1) I(2) I(1) I(1) I(2) I(0) *All critical values are at 5% significance. For simplicity, all variables, excluding output gap, are treated as I(1) variables. The DF-GLS test was proposed by Elliott, Rothenberg and Stock (1996). Time series is transformed via a generalised least square (GLS) regression before carrying out the test. Elliott et al. have shown that the test has significantly greater power than the previous version of augmented Dickey-Fuller (ADF) test. 7

8 3.2 The model A vector autoregressive model (VAR) is used to analyse exchange rate pass-through in the United Kingdom as well as the effect of exchange rate pass-through on domestic inflation in the UK. The benchmark model is a seven-variable VAR which consist of oil price, money supply, nominal effective trade-weighted exchange rate (neert), output gap, import price index (IMP), producer price index (PPI) and Consumer Price Index (CPI). First, oil price is included to identify supply shocks and output gap is added to capture demand side effect. Second, money supply is included to capture the effects of monetary policy on inflation. Third, nominal effective trade-weighted exchange rate is used in the VAR where currencies of trading partners of the United Kingdom are weighted by their volume of trade with the UK. Fourth, the consumer price index (CPI) is used as a measure of domestic inflation. Import price index and producer Price index (PPI) are added into the model in order to analysis the effects of shocks on prices at different stages of the pricing chain from imports to retail. 8

9 The reduced form representation of the model can be written as t ALy t where ' yt loilpricet, outputgapt, lmoney_supply t, lneert t, limp t, lppit, lcpi t AL I AL AL and 1 p p, is a vector of constants and t is a vector of residuals, l indicates natural logarithms and denotes first differences. Impulse response functions are used to analyse the dynamic responses implied by estimated VAR. It enables us to look at the sign, the size and the persistence of the effects of shocks to one variable on another variable. To do this analysis we need to identify the structural form of the contemporaneous interactions between the variables. The structural model can be identified when at least k 2 k /2 restrictions where k is the number of endogenous variables. This is done through a recursive ordering (i.e. a Choleski decomposition). 9

10 The matrix below describes the relationship between the reduced form VAR residuals and the structural shocks. oilprice t outputgap t money_supply t neert t imp t S S21 S S31 S32 S S41 S42 S43 S S51 S52 S53 S54 S S S S S S S 0 ppi t cpi S71 S72 S73 S74 S75 S76 S77 t 1 t 2 t 3 t 4 t 5 t 6 t 7 t In other words, the following ordering is : oilprice outputgap money_supply neert imp ppi cpi The ordering of the last four variables reflects how prices are passed along the distribution chain. This specification allows exchange rate shocks to affect all prices in the distribution chain immediately by pass-through effects. Money follows output gap, similar to Peersman and Smets (2001). Economics activity only reacts to monetary shocks with a lag, while as an asset price, the exchange rate response immediately to both monetary and real shocks. Oil prices are ordered first because it is unlikely to be affected by the other variables contemporaneously while it is likely to affect other variables in the system immediately (as supply shocks). This choice of ordering largely corresponds to previous studies (McCarthy (2006) and Hahn (2003)). With a relatively large system like this, parameter parsimony must be considered. Following Ivanov and Killian (2005), AIC is used to choose the optimal lag length which comes up to be four lags. 10

11 4 Results The IRFs used are the accumulated responses, since the variables are in log-differenced format to get the approximate percentage change in a price index. Since the innovation is one standard deviation (in this case, the standard deviation of log NEER changes is ), the IRFs have to renormalized to get the standard interpretation of pass-through as the percentage change, for a 1% exchange rate shock. The confidence intervals are ± 2 standard errors, as calculated by Monte Carlo integration of 1000 draws. This makes the implicit assumption that the residuals are approximated well by a multivariate normal distribution. Bootstrapping, as described in Kilian (1998), is more robust In the presence of serial correlation, outliers and heteroscedasticity. However, this procedure is not implemented in Eviews yet. Ideally, this should be supplemented with Sims and Zha (1999) style error bands, which characterizes the shape of likelihood uncertainty. Both are left for future work. Figure 1 indicates the pass-through to import prices. As expected, this is the largest response to exchange rate shocks (as it has the most direct link). Normalising by the 1 SD innovation of NEER, the first quarter has an ERPT of 35%, rising to 72% by the fifth quarter. These numbers are roughly in line with existing literature - very close in the short-run, while higher in the longer-run, but neither would be outside the 11

12 confidence bands. For example, the seminal paper on ERPT, Campa and Goldberg (2005) estimates a short-run and long-run ERPT of 0.36 and 0.46, respectively. Nevertheless, their story of incomplete pass-through still holds. A caveat is that the confidence bands become quite large (since these are accumulated responses), such that it becomes not statistically significant from zero at 8 quarters away. The ERPT to PPI is much lower, since imported goods only form a proportion of the intermediate goods and by the effect of distribution costs (Campa and Goldberg, 2008). In addition, the competition factor in the pricing-to-market cause for incomplete pass-through is also enhanced. Only 1.7% of the NEER shock is passed-through on impact (not statistically significant from zero), while rising to 15% later on. From the large error bands, the point estimates are only statistically significant from zero at only 2 to 4 quarter horizons (about 12% ERPT at these horizons). Similar to PPI, the ERPT to CPI is even lower, as it compounds the incomplete pass-through reasons once more. Domestic factors tend to influence CPI more for an economy which is not as exposed to foreign trade (more evidence on this from the variance decompositions). At most, the pass-through is at 5% at long horizons, and it does not become statistically significant from zero at any point. Also note the scale on the IRFs the error bands are still very narrow, thus enabling some policy 12

13 conclusions to be made. Figure 1 IRF for Import Price Figure 2 IRF for PPI 13

14 Figure 3 IRF for CPI 14

15 Variance Decomposition The Forecast Error Variance Decompositions (FEVD) also reveal interesting points. A significant proportion of import price variance is explained by fluctuations in NEER (approximately 40% for the point estimates), reflecting the results from the IRF analysis that the ERPT is quite high. Both of this could be caused by the fact that pricing-to-market in UK imports is apparently not extensive, perhaps due to high overseas competition. Interestingly, oil price variation holds a lot of sway in explaining the variance of import prices (around 20%), despite oil accounting only for a small part of UK imports. This may be explained from Campa and Goldberg s (2005) finding that energy imports have much higher pass-through than other imports, as well as the possibility of oil prices affecting the production cost (and hence, selling price) of other imports. The decomposition of PPI variance has similar results to the IRF analysis. The elasticity of PPI with respect to import prices is quite high (approximately peaking at 35% at 3-quarters, in the IRFs). This is mirrored in why import price seems to be an important factor in explaining PPI fluctuations, at around 20%. Oil prices appear to have an even bigger role, at explaining around 30% of the variance. Similarly, CPI is also less well-explained by external factors and more by domestic factors. The vast 15

16 majority is explained by innovations in CPI itself, and the rest is by PPI and output gap (at longer horizons), which is what is generally expected. Variance Decomposition Percent DLIMP v ariance due to DLCPI Percent DLIMP variance due to DLIMP Percent DLIMP variance due to DLMONEY_SUPPLY Percent DLIMP variance due to DLOILPRICE Percent DLIMP variance due to DLNEERT 80 Percent DLIMP variance due to DLPPI Percent DLIMP v ariance due to OUTPUTGAP Figure 4 FEVDs for Import Price 16

17 Figure 5 FEVDs of PPI Figure 6 FEVDs of CPI 17

18 5 Robustness Alternative Model To check for robustness, industrial production index (IPI) is used as an alternative measure of demand size effect instead of output gap where results are similar to those from the benchmark model. Since a a larger number of observations of output gap is available therefore it is chosen in the benchmark model. Limitations An and Wang (2011) has stated that a stationary VAR model has a standard recursive identification assumption which some of the variables cannot interact with other variables and can lead to a significant difference in results. Also Bache (2005) finds that impulse response functions in first differences are biased when estimating ERPT, even when the VAR model is specified with a large number of lags. Therefore they have adopted a VAR model in level of sign restriction developed by Uhlig (2005). This method enables restrictions to be made more explicitly by being directly imposed on impulse responses. Also, sign restrictions are weak which they do not lead to exact identification of the reduced form of VAR. Furthermore, external effects may affect the estimation where the economic information (GDP,CPI..etc) of all trading partners of the UK can be added into the model by first principle components (to retain parsimony in such a high-dimensional 18

19 system) to control for it. However, due to limited time and length, this is not carried out in this study. 19

20 6 Conclusion The aim of this study is to analyse the effect of exchange rate pass-through on import price, producer price index and consumer price. The VAR analysis enabled us to analyse the effect of ERPT along the distribution chain. From the analysis of impulse response function, it suggested that the degree of pass-through to the exchange rate shock varies across different price indices. We can observe the effect of ERPT is the largest on import price index then PPI and the smallest on CPI which matches to the findings of An and Wang (2011) who shows that ERPT declines along distribution chain. Results from variance decomposition reveal the fact that 40% of import price fluctuation can be explained by nominal effective exchange rate which is a significant proportion. However, when we further investigate along the price chain it shows that ERPT on CPI is minimal where domestic factor actually dominates ERPT on the effect on domestic price. For future work, a VAR can be done on disaggregated data to look at the effect of ERPT on different industry. Policy implications Given the recent strong depreciations of the sterling from rounds of quantitative easing, the Bank of England and consumers may be worried about not only domestically generated inflation, but also imported inflation. The analysis showed here that the economy shows surprising resilience to imported inflation that while 20

21 the import prices, and to a lesser extent, producer prices, respond fairly strongly to exchange rate fluctuations, consumer prices remain fairly muted to exchange rate movements. 21

22 Reference An L and Wang J (2011) Exchange Rate Pass-through: Evidence Based on Vector Autoregression with Sign Restrictions Federal Reserve Bank of Dallas Globalisation and Monetary Policy Institute Working Paper No.70 Bache.I.W (2005) Assessing the Structural VAR Appproach to Exchange Rate Pass-through Norge Bank Working Paper No.32 Campa J.M and Goldberg L.s (2005) Exchange Rate Pass-through into import prices Campa J.M and Goldberg L.S (2008) The Sensitivity of CPI to Exchange Rates: Distribution Margins, Imported Inputs, and Trade Exposure, Federal Reserve bank of New York and NBER Elliott, G., Rothenberg, T.J. and Stock, J.H. (1996), Efficient Tests for an Autoregressive Unit Root, Econometrica, Vol. 64, No. 4, Hahn E (2003) Pass-through of External Shocks to Euro Area Inflation, Working paper 243, European Central Bank Herzberg V, Kapetanios G and Price S (2003) Import Prices and Exchange Rate Pass-through: theory and evidence from the United Kingdom, working paper no. 182 Bank of England Ito T and Sato K (2006) Exchange Rate Changes and Inflation in Post-Crisis Asian 22

23 Economies : VAR Analysis of The Exchange Rate Pass-Through, National Bureau of Economic Research Working Paper Kilian L and Ivanov V(2005) A Practitioner s Guide to Lag Order Selection for VAR Impulse Response Analysis, Studies in Nonlinear Dynamics and Econometrics, 9(1), March 2005, Article 2. Killian (1998) Small-sample Confidence Intervals For Impulse Response Functions McCarthy J (2000) Pass-through of Exchange Rates and Import Prices to Domestics Inflation in Some Industrialised Economies, Staff reports 111, Federal Reserve Bank of New York Mumtaz H, Oomen O and Wang J (2006) Exchange Rate Pass-through into UK Import Prices, working paper no. 312 Bank of England Peersman G anf Smets F (2001) The monetary transmission mechanism in the Euro area: more evidence from VAR analysis, European Central Bank working paper 91 Sims C and Zha T (1999) Error Bands for Impulse Response, Econometrica, Vol.67, No.5 (September,1999), Stulz J (2007) Exchange Rate Pass-Through in Switzerland: Evidence from vector autoregressions, Swiss National Bank Economic Studies No

24 Appendix 1 : IRF tables of Benchmark model Table A IRF of Import Price DLMONEY_S Period DLIMP UPPLY DLNEERT DLOILPRICE DLPPI OUTPUTGAP DLCPI E ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Cholesky Ordering: DLOILPRICE OUTPUTGAP DLMONEY_SUPPLY DLNEERT DLIMP DLPPI DLCPI Standard Errors: Monte Carlo (1000 repetitions) 24

25 Table B IRF of PPI DLMONEY_S Period DLIMP UPPLY DLNEERT DLOILPRICE DLPPI OUTPUTGAP DLCPI ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Cholesky Ordering: DLOILPRICE OUTPUTGAP DLMONEY_SUPPLY DLNEERT DLIMP DLPPI DLCPI Standard Errors: Monte Carlo (1000 repetitions) 25

26 Table C IRF of CPI DLMONEY_S Period DLIMP UPPLY DLNEERT DLOILPRICE DLPPI OUTPUTGAP DLCPI E ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Cholesky Ordering: DLOILPRICE OUTPUTGAP DLMONEY_SUPPLY DLNEERT DLIMP DLPPI DLCPI Standard Errors: Monte Carlo (1000 repetitions) 26

27 Appendix 2 : IRF graphs of alternative model Accumulated Response to Cholesky One S.D. Innovations ± 2 S.E. Accumulated Response of DLCPI to DLNEERT Accumulated Response of DLIMP to DLNEERT Accumulated Response of DLPPI to DLNEERT