Department of Economics Master dissertation
Spring 2009
The Exchange Rate Pass-Through Into Domestic Manufacturing Prices During
Two Inflation Regimes
Supervisor: Prof. Nils Gottfries
Author: Roujman Shahbazian
In the beginning of 1990s Sweden implemented several measures in order to maintain price stability. These measures have resulted in an environment in which inflation is lower and more stable. The same development could be seen in other OECD countries. At the same time a decrease in exchange rate pass-through was noticed in many countries. This has led researchers to believe that there may be a connection, between these two phenomena. This dissertation analyzes whether there has been any change in exchange rate pass-through for manufacturing products in Sweden between the high inflation period (1977-1993) and the low inflation period (1994-2006). The result shows that there is a difference in the exchange rate pass-through between the two periods. During the low inflation period the degree of pass-through was lower than during the high inflation period.
Keywords: Exchange rate pass-through, import prices, Sweden, producer prices, inflation regimes, manufacturing products
- SAMMANFATTNING -
I början av 1990-talet vidtog Sverige flera åtgärder för att upprätthålla ett fast penningvärde.
Dessa åtgärder har bidragit till att hålla inflationen på en låg och stabil nivå. Likartad utveckling har också påträffats i andra OECD länder. Samtidigt som inflationen gick ner, observerades en minskning av växelkursförändringens genomslagskraft på inhemska priser i en rad länder. Detta bidrog till att forskare började fundera på om det finns ett samband mellan dessa två fenomen. Denna uppsats analyserar om det har skett en ändring i växelkursförändringens genomslagskraft för tillverkningsindustrin i Sverige under en hög inflationsregim (1977-1993) samt en låg inflationsregim (1994-2006). Studien visar på att det existerar en skillnad mellan tidsperioderna. Under den låga inflationsregimen var växelkursens genomslagskraft mindre jämfört med den höga inflationsregimen.
- TABLE OF CONTENTS -
- INTRODUCTION - ... 5
- PREVIOUS RESEARCH -... 9
- THEORETICAL FRAMWORK - ... 12
3.1 Pricing Model ... 12
3.2 Theoretical Specification ... 13
3.3 Model Specification ... 15
- DATA & ESTIMATION METHOD - ... 17
4.1 Data Description ... 17
4.2 The Model ... 19
4.3 Stationarity ... 19
4.4 Co-integration ... 20
4.5 Error Correction Model ... 20
4.6 Structural Change ... 21
4.7 Residual Analysis ... 22
- RESULTS - ... 23
5.1 Co-integration ... 23
5.2 Error Correction Model ... 24
5.3 Structural Change ... 24
5.4 Estimation Results ... 25
- CONCLUSION - ... 28
- FINAL COMMENTS - ... 30
- APPENDIX - ... 31
- REFERENCES - ... 33
- INTRODUCTION -
During the last two decade we have seen a decrease in inflation all across the OECD countries.
Figure 1 illustrates the development of the inflation rate in five countries in OECD. After the high inflation rate during the 1970s and 1980s the inflation rate has been decreasing in the 1990s. The decrease in volatility in the beginning of 1990s, compared to the time before, is also apparent in Figure 1. Therefore one could easily speak about two different regimes, a low and high inflation regime.
Figure 1
There could be several reasons that contributed to lowering the inflation rate and stabilizing it in the OECD countries. There have been several factors contributing towards this low inflation regime, such as an independent Central Bank, credible monetary policy, implementation of inflation targeting and the fact that many Central Banks around the world have taken into account features like monopolistic competition, nominal rigidities and the short run non-neutrality of monetary policy into their models (Gali, 2008:4-9 and Chari & Kehoe (2006)).
0 5 10 15 20 25
1975 1980 1985 1990 1995 2000 2005
Canada France Sweden UK US
Almost parallel to the stabilization of inflation in the OECD countries, there was large exchange rate deprecation in some of these countries. But the depreciations did not have the same effect on consumer prices or inflation as expected. Cunningham & Haldane (2000) have done an analysis of the depreciation in three countries and the effect that it had on import prices and the consumer index. They analysed among other the large depreciations of pound sterling in year 1992 respectively 1997 and the almost 70 percent depreciation of the Brazilian currency in 1999. They expected that the exchange rate pass-through1 would take some time due to cost in changing prices immediately. But even after some time the exchange rate pass-through to import prices was almost non existing in Brazil and in United Kingdom it was much lower than expected. The consumer index did not experience a sudden increase but had a slow trend as if nothing had happened to the exchange rate.
This common experience has led researchers to think that the exchange rate pass-through has declined. If the pass-through in many countries has been declining, the interesting question is why? What are the causes that can explain this decrease? Many researchers have tried to link the decrease of pass-through with the low inflation regime, but it was Taylor (2000) who first formulated this hypothesis together with a theoretical framework.
What Taylor (2000) sets out to do is try to explain how the expected future persistence of inflation can have an impact on the pass-through. He assumes that there exists an adjustment lag for the pass-through. If actors think that the future path of inflation is going to be low then they would not react to, say an exchange rate shock. Therefore they would not change their prices. But if they think that the high inflation is going to be persistent then they would change their prices accordingly. That is to say that the inflation environment has an impact on the degree of pass- through. In low inflation regimes the pass-through is going to be low and in high regimes one would expect a high pass-through. From now on this is referred to as the Taylor hypothesis.
Taylor presents evidence which shows inflation to be positively correlated with persistence of inflation. Suggesting the low pass-through is caused by low inflation.
1An extent to which exchange rate differences are reflected in prices on the destination market of traded products is referred to as exchange rate pass-through.
There are also other explanations that are offered for this current low exchange pass-through. An explanation that sometimes is offered along with the Taylor hypothesis has to do with the composition of imported goods. This idea has been formulated by Campa & Goldberg (2002).
They suggest that the decrease in pass-through has microeconomic explanations rather than macroeconomic. Their reasoning is that during the last 40 years there has been a change in the composition of imports, this shift has been towards sectors that have lower degree of exchange rate pass-through. They mean that the composition of imports have shifted from sectors that had high (or complete) pass-through effects (like the energy sector or the raw material imports) towards sectors that have relative low pass-through.
Therefore the Taylor hypothesis should not be considered as the only explanation. But one cannot disregard the intuitive as well as the empirical evidence2 that connects this recent phenomenon in exchange rate pass-through with the low inflation regime.
If this hypothesis is true, the same phenomena should also be observed for Swedish data, meaning that the exchange rate pass-through should be lower after the implementation of the inflation-targeting framework in the beginning of the 1990s compared to the time before. The purpose of this dissertation is to see how the exchange rate pass-through has evolved during the period 1977 to 2006, for Swedish data. If the Taylor hypothesis is correct then exchange rate pass-through should be lower during the low inflation regime (sub-sample 1994-2006) compared to the high (sub-sample 1977-1993).
Why is it interesting or important to figure out the degree of exchange rate pass-through or if it has changed? The interesting aspect has to do with the policy implications of pass-through. There are at least three important implications for policy decisions connected to the degree of pass- through. The first policy implication is the so called expenditure switching effects. Consumer would not change their expenditure habits as often if the pass-through is low. In other words the price incentive for consumers to switch their expenditure would be low.3 The second policy
2 See section previous research for more evidence.
3 For example, if a depreciation of the Swedish currency would take place, the price of foreign products would increase relative to Swedish product. Everything else being equal, there would be an increase demand for Swedish products. But if the exchange rate pass-through to import prices is low, then the increased demand for Swedish products would not be as big as it would if there would be a complete exchange rate pass-through.
implication has to do with forecasting. The key function for monetary policy is for the Central Bank to be able to forecast the future path of inflation and a decline in the pass-through to consumer prices could play an important role. If there has been a decline or increase in the pass- through which the inflation forecasters have not adjusted their models to, the forecast could overestimate or underestimate the effects of changes in the exchange rate pass-through on inflation.4 The third policy implication has to do with international transmission of shocks. If the exchange rate pass-through is low then the international transmission of shocks would be dampened, thereby the business cycle would become more synchronized.5
This dissertation is disposed as following: the next section presents a short summary of the previous research. Section 3 develops the theoretical framework that would be used to interpret the data and derives the model specification that is going to be estimated. Section 4 presents the data, the econometrical method and accounts for the tests that is going to be performed in this dissertation. The explicit tests for the data are presented in section 5 which also includes the estimation results for the model specification. The conclusion is discussed in section 6 and the dissertation ends by summarizes the findings of this study along with some final comments.
4 To illustrate the importance of the inflation forecasting, one could have in mind what happened in the fall of 2008 when Statistic Sweden wrongly overestimated the inflation. This had enormous impact on the policy decisions for both Sveriges Riksbank and the government and in the end the Swedish population.
5 To illustrate this lets assume that the U.S. dollar depreciates due to an unforeseeable shock. If the pass-through is complete then the effect of the shock would be positive for the U.S. aggregate output, but negative for U.S. trading partners, everything else being equal. If now the pass-through is low then the shock that happened in U.S. would have less negative effects on U.S. trading partners and therefore the international transmission of shocks would be dampened.
- PREVIOUS RESEARCH -
Table 1 provides a short summary of five articles which have dealt with exchange rate pass- through. The choice of articles is determined by the desire of presenting articles that have studied this phenomenon from different approaches. The table contains the name of the authors, their purpose, the time length of the study, the countries investigated and what conclusions they have reached.
The purposes of all articles are somewhat similar. All articles want to investigate the degree of pass-through during the time period in focus. They have all the same underlying idea, that it is the change in macroeconomic conditions that have lead to a change in the exchange rate pass- through. Meaning the change in inflation regime has contributed to lowering the pass-though effect.
The articles have analysed almost the same time period, which includes the two inflation regimes, i.e. from 1970s up to the 2000s. The number of countries in the sample differs. Choudhri &
Hakura (2001) and Devereux & Yetman (2002) analyses a large sample of countries which are in different stages in their economic development and have different economic structure. While Bailliu & Fujii (2004) and Gagnon and Ihrig (2004) concentrate on the more industrialized countries that have similar economic conditions, thereby the number of countries decreases significantly. Bouakez & Rebei (2008) have only one country in focus and that is Canada.
The first point that all articles make is that the degree of pass-through has decreased over the time period in question. None of the studies have been able to prove a complete exchange rate pass- through during any period. But from a low level in the 1970s and 80s, the pass-through has decreased even further at the end of the time period.
The second point that the articles have in common is the fact that they all associate the decrease in pass-through with the low inflation regime. The low inflation regime has been linked with the
Table 1
Name of authors Purpose Time period Countries Results
Choudhri & Hakura (2001)
Has the exchange rate pass- through been reduced, due to a low inflation regime?
1979 to 2000 71 There is a strong positive and significant relationship between pass-through and average inflation rate across countries. Their conclusion is that high inflation rate is associated with large pass-through compared to the opposite.
Devereux & Yetman (2002)
Is the degree of pass-through related to the inflation environment (i.e. monetary policy regime)
1970 to 2001 122 Macroeconomic factors at least partly play a role in determining the rate of pass-through of exchange rates to prices. Their second point is that monetary policy regime does have a role for the rate of pass-through.
Bailliu & Fujii (2004)
Has the transition to a low- inflation environment, resulted in a decline in the degree of pass- through?
1977 to 2001 11 Exchange rate pass-through declines with a shift to a low- inflation environment brought about by a change in the monetary policy regime.
Gagnon & Ihrig (2004)
Test whether pass-through has declined in each country, following a change in the inflation environment.
1971 to 2003 20 Estimates of the exchange rate pass-through coefficients were, in most cases, smaller in the low inflation environment.
Bouakez & Rebei (2008)
Investigates whether the exchange rate pass-through really has declined in Canada.
1973 to 2006 1 The exchange rate pass-through has declined after the implementation of inflation targeting in Canada.
changes in macroeconomic behaviour, such as change in monetary policy and implementation of inflation targeting. That is to say that their finding is in support of the Taylor hypothesis, that in low inflation regimes there is lover pass-through rates compared to high inflation regimes.
All of the above mentioned studies, excluding Bouakez & Rebei (2008), include Sweden in their dataset however their analysis have a comparative nature. The comparative approach has the advantage to produce unambiguous results using sophisticated econometric techniques. The disadvantage is that it oversimplifies behaviours and presumes a static view of cultural and institutional changes (Clark (2002)).
Therefore this dissertation is similar to the approach used by Bouakez & Rebei (2008), i.e.
concentrating on one single country and evaluate whether or not the pass-through has changed.
Thereby cultural and institutional aspects, such as Sweden’s most important trade partners, would be taken into account.
- THEORETICAL FRAMEWORK -
This section begins by presenting the different models that could have been used in the dissertation. Thereafter advancing to constructing a theoretical model for the typical firm and discussing the underlying assumptions for it. The section would end by deriving the model specification that is going to be estimated.
3.1 Pricing Model
The pricing model that is used in this dissertation is influenced by the staggered pricing model that Taylor (2000) has proposed and has been used in other studies of pass-through effects (Choudhri & Hakura (2001)). But there are certain differences in the way Taylor’s model is presented in this dissertation. This is done in order to differentiate and drive equations for the price process of the domestic vs. the importing firm (or the home vs. the foreign firm). But also in order to present the pricing model in a more synoptically fashion and thereby taking into account for nominal neutrality6.
The Taylor model is a time-dependent pricing model which assumes that wages and prices are set by multi-period contracts and in the end of each period a fraction of the contracts expires and are renewed. The prices and wages are predetermined and also fixed, meaning that the contract specifies the same price (or wage) for each period it is in effect. These multi-period contracts lead to a gradual adjustment and not to a direct adjustment (Romer, 2006;319-26).
Two other models, the Fischer model and Caplin-Spulber model, could have also been used in this dissertation. The Fischer model has the same assumptions as the Taylor model with the exception of fixed prices. Meaning that the multi-period contracts are predetermined but the price can change each period according to the contract. In order to simplify, the Fischer model was
6 Nominal neutrality refers to the fact that a percental increase of all nominal explanatory variables (for example say the price of a firm, the competitor’s price or a price index for other products) would not affect the quantity that is demanded.
excluded. The Caplin-Spulber model has also been excluded, because of the state-dependent pricing assumption. Under state-dependent pricing firms decide to change a price (or wages) in response to a change in economic conditions. This differs from the time-dependent pricing which at a certain point of time the prices and wages are changed. (Romer, 2006;309-33).
3.2 Theoretical Specification
How would one go about when constructing a model for the typical firm? One thing is clear, firms are not price-takers, at least not in the short run.7 Therefore the relative price of a firm must depend to some degree on the costs and exchange rates (Gottfries 2002:587-8). The price decision process of a firm is quite complex, but usually a firm has some market power and thereby the price setting becomes an endogenous decision variable for the firm (Taylor 2000;1394-5). The argument above explicitly implies that the assumption of the idea of the perfect competitive market, were the firm is assumed to be a price-taker, can not apply is this context. Another assumption of the perfect competitive market is that the products in the market are perfect substitutes, in other words one cannot differentiate between the goods. This assumption would not hold if the price is to be a decision variable of a firm. Because the fact that the firm perceives itself to have some market power, must imply that the product the firm is selling is to some degree differentiated from other products in the market. Therefore relaxing this assumption would allow different firms to charge different prices. The constructed model has to incorporate at least this two assumptions meaning: firms are not price-takers and there is to some degree imperfect substitutes between the products. Below, these assumptions would be used to drive equations for the price of the domestic and the importing firm.
Assume that the demand for production of a firm is determined by the relative price, the price elasticity and demand.
P D
Y Pi
d
d
(1)
7 But one can not exclude the possibility that firms might be price-takers in the long run.
where d stands for domestic firm, i for importing firm, Y is the production, P is the price of the good, D is capturing the demand shock, α is a positive constant and λ is the price elasticity. The domestic production,Y , would increase either by a positive demand shock or if the relative priced
) /
(Pd Pi , between the domestic and the importing firm, would decrease. Assume constant return to scale and that a firm’s marginal cost is equal to the wage level. Then the profit function of the domestic firm consists of two components the demand function and the mark-up8:
P D W P
P i
d d
d
d ( )*
(2)
Where π stands for the profit function and W is the wage. We multiply the two components with each other and get the profit function:
P D W P P W
P P
P i
d d d
i d d d
d *
(3)
Differentiate the profit function with respect to domestic price, set it to zero, divide with D and thereafter simplifying gives us the domestic price function:
) (
2 *
1 i d
d P W
P
(4)
Repeating the same steps as above for the foreign exporter gives the following price function:
) (
2 *
1 d i
i P QW
P
(5)
8 Mark-up is the amount by which price exceeds marginal cost.
The above function is the equivalent price function for the importing firm with the exception of QWi which stands for the marginal cost in the domestic currency. Now there are two equations and two unknowns and we can solve for Pdby substituting equation (4) in (5) and simplifying:
d i
d QW W
P *
2 2
1
(6)
Repeating the same procedure yields the price function of the importing firm:
i d
i W QW
P *
2 2
1
(7)
The pricing models that are expressed in equation (6) and (7) satisfy nominal neutrality. That is if the cost for both the importing and domestic firm increase with 10 %, i.e. QWiandWd, then the price would also increase 10 %.
3.3 Model Specification
We now take a log linear approximation to optimal price function:
d t t
f t do
t
p q w
p
0 1( )
2(8)
Where o denotes the optimal level, ptf is foreign producer prices which is assumed to be the approximation of Wiand the lower-case letters denote logs. Equation (8) implies that if the prices are flexible then the optimal price level of the domestic firm is determined by the domestic wage costs and the importing prices in the domestic currency. Nominal neutrality implies that the hypothesis 1 2should equal to one. Assume that firms set its prices to last for a period of time and suppose for simplicity that it is for two periods. Under these assumptions the firms expected optimal price would then be:
1
0
1) ( 2
/ 1
i
do t
t E p
x
(9)
Where xtapplies in period t and t + 1 and the term E denotes the expectation. For the sake of simplicity we assume a specific process forw, which is the assumption of random walk.
Meaning:
w t d t
t
w
w
1 1(10)
Where the error term is a noise process and therefore is independent and identically distributed.
The process implies that the error term has a long-lasting effect onw. This gives us that the expected optimal price for period t + 1 is equal to the optimal price of period t.9 Hence:
do t
t p
x . Combining equation (8) and (9) yields:
1
0
1
0 2 1
0
( ) 2
2
j jd j t j
t f
j t do
t
p q w
p
(11)
Where the j has substitute the i, in order to emphasize that we are dealing with the price in the previous period. Taking away the summation sign and simplifying yields10:
) ( )
(
2 3 4 51
0
p q w p q w f t
p
tdo tf t td tf t td(12)
Where f(t) is polynomial in t and is capturing a linear, quadratic and cubic year trend form.
9 E(ptdo1) (ptdo)
10 For the derivation, see appendix section: 8.2 Derivation of equation (12).
- DATA & ESTIMATION METHOD -
This section begins by presenting the data used in the dissertation. Thereafter discusses the different steps that are associated with the dissertations econometrical method and accounts for the tests that are going to be performed. Also discussed is how spurious correlation can be avoided and how autocorrelation and heteroskedasticity can be adjusted for.
4.1 Data Description
The data used in this dissertation are quarterly and covers Swedish manufactured products between 1977-2006. This choice is first and foremost due to availability and accessibility of data.
In order to evaluate how the exchange rate pass-through has evolved during the two inflation regimes, it is appropriate to divide the sample into two parts. As a breaking point year 1994 has been chosen. One could wonder why year 1993 was not chosen, it was nevertheless from the beginning of that year that Sweden adopted the inflation targeting. But in monetary policy it is usually assumed that it takes between one to two years before any decision made by the central bank has an effect on the economic activity. The reasoning behind this assumption is usually based on the argument about nominal rigidities, sticky prices, etc., (Gali (2008) and Clarida (1999)).
Producer price index (PPI) and labour cost index for manufacture products are used, where labour cost is the total amount of cost for each hour (SCB 2008;AM0301). In the beginning of the 1990s Statistic Sweden changed its nomenclature SNI69 to SPIN2002, which are two different product classifications. Therefore both time index series had to be sorted in order to match (MIS 1992/6;113-132). Precise definitions of the variables used in the data are given in the appendix.
It is quite important to be aware of the fact that transport costs, cultural links, etc., plays a important role for Sweden’s importing partners (Gottfries 2002:585-6). Therefore weighted
indexes of foreign PPI and exchange rates have been constructed, with weights determined by the share of Swedish imports from different countries. The share of Swedish imports from different countries changes year to year and in a 30 year period the changes could be significant. Therefore the best thing would have been to create weighted indexes for each year. But this would have been too time consuming and therefore we assume that the share of Swedish imports for each year have been the same as year 2005.
Figure 2 illustrates the domestic price and labour cost relative to foreign export price in Swedish currency during both inflation regimes. Until the end of 1992 Sweden had a fixed exchange rate regime and thereafter the krona has been floating. Figure 2 illustrates several important features.
In the beginning of the 1980s Sweden decided to increase the competitiveness of its industry.
Therefore two offensive devaluations in September 1981 and October 1982 were conducted. The economic condition in Sweden during the latter part of 1980s, lead up to the floating and the large depreciation of the krona in the end of 1992. This event had a significant effect on the relative price and labour cost, as it can be observed. Around 1994 the confidence for Swedish economical policy was restored which lead to a return to a more normal situation.
Figure 2
Note: Domestic prices, pd, and labour cost, wd, relative to foreign import prices in Swedish currency, (pf q). The variables have been seasonally adjusted.
Figure 2 also illustrates the relationship between costs and prices on one hand, and pass-through on the other. It is quite obvious that relative prices must depend on costs relative to foreign prices.
-.12 -.08 -.04 .00 .04 .08 .12
78 80 82 84 86 88 90 92 94 96 98 00 02 04 06
Relative price Relative cost
Looking at figure 2 we see that variation in relative cost is much bigger than relative price. This simple bivariate analysis indicates that the pass-through is not complete, according to the data.
The fact that we have converted the foreign prices to Swedish currency together with the fact that the pass-through is not complete implies the following: prices are set in krona and are not immediately adjusted when exchange rates change.
4.2 The Model
When creating a pricing model it is important to take into account both the short-run and long-run relationship between the variables in the model, therefore an error correction model (ECM) would be used in this dissertation. ECM originates from a structural economic relationship and a long-run equilibrium condition. With the help of algebra the structural relationship is manipulated in order to formulate a short-run term which expresses the deviation from the long-run equilibrium, this term is referred to as the error correction term (Wooldridge 2006;652-3).
Therefore in the estimating process both the short and long-run relationship is been considered.
A condition for the ECM to be used is that there exist a long-run relationship between the variables in the model, or putting it on other words they should be co-integrated. The reason why it is appropriate to use ECM is because the problem of “spurious correlation” is avoided. If there exist a spurious correlation in the regression, meaning highR , it indicates that there is a strong 2 relationship between the dependent and independent variables even due such relationship does not exist.
4.3 Stationarity
The first step to deal with spurious correlation is to test whether or not the time series used in the model are stationary. A time series is assumed to be stationary if the expected value and the variance are constant over time and the covariance only depend on the distance between the two time periods and not the time period themselves (Wooldridge 2006;380-4). If these three assumptions are fulfilled then the series could be called stationary. One could test it by a unit root test. The most used unit root test is the Augmented Dickey-Fuller (ADF) test. The null hypothesis is that the series has a unit root and therefore cannot be regarded as stationary.
If the time series of the model cannot reject the null hypothesis of ADF test, differences can be used in order to transform the time series into a stationary one. A consequence of differentiating the time series is that much of the information in the series gets lost and therefore the economic interpretation becomes less meaningful. A solution to this dilemma is to test if the including variables, in levels, in the model are co-integrated.
4.4 Co-integration
Time series that are co-integrated would not need differentiating and at the same time we would be assured that spurious correlation is avoided. Co-integration is when time series that are not stationary in themselves become stationary through a linear combination. A condition for a co- integration relationship between variables is that the variables are integrated off the same order (Murray 2006;789-93). Co-integration is therefore about relationships between time series, that there exist a long-run relationship and that they follow each other. The time series may diverge from each other during a short period but they have to follow the same path in the long-run.
The most used and popular method for testing for co-integration has been proposed by Engle and Granger (Murray 2006;791-801 & Wooldridge 2006;647-54). The test is performed in two steps.
The first step is to estimate the long-run relationship between the variables in the model and save the residuals. The variables have to be non-stationary and integrated of the same order. The second step is to perform an ADF test on the residuals that has been saved. If the test is able to reject the null hypothesis then that is an indication of a co-integration relationship.
4.5 Error Correction Model
When a co-integrated relationship between the variables has been confirmed, the model can be formulated as an ECM. Assume that the simple long-run equilibrium relationship between the variables can be denoted as in Eq (8). The general dynamic relationship between our dependent variable and explanatory variable is as the model specification, i.e. Eq (12).
The error correction term, that is the difference between the long run equilibrium relationship and the dynamic model, can then be written as:
) ( )
(
1 1 2 1 3 1 4 1 5 11 0 1
1
p p q w p q w f t
s
t tdo tf t td tf t td
(13)
Taking differences of our long-run equilibrium and inserting the error correction term yields the ECM:
1 2
1
0
(
tf t)
td tdo
t
p q w s
p
(14)
We should expect 011. If p in the previous period has overshot the equilibrium the error do correction term would work to push p back towards the equilibrium. And in the opposite way, do if ptdo1 has fallen short of the equilibrium then error correction term would correct p back to the do equilibrium (Wooldridge 2006;652-54).
4.6 Structural Change
The dissertations point of departure is that there may be a structural change in the time period 1977-2006, which has to do with the change of inflation regime. This is what the theory states but can this be seen in the data? A formal test that tests if there has been a structural change is the Chow breakpoint test.
The idea of the breakpoint Chow test is to fit the equation separately for each subsample and to see whether there is a significant difference in the estimated equations. If there is a difference then we can proceed by dividing the sample into two different periods to test our hypothesis that the exchange rate pass-through has changed during the two different inflation regimes. The null hypothesis of the Chow breakpoint test is that there is no structural change.
11 If we had specified the ECT the other way round, we would then have expected a positive value for .
4.7 Residual Analysis
When estimating the model specification it is important to check if there exist autocorrelation and heteroskedasticity. If that is the case then it has consequences for the estimations. Firstly the ordinary least squares (OLS) are no longer the optimal method for estimations. Secondly the standard errors that have been estimated through OLS are no longer correct. It is imperative to check for the existence of autocorrelation and thereafter performing the test for heteroskedasticity. Because the heteroskedasticity test is no longer valid if the residuals are correlated with each other.
With the Durbin Watson statistica one can get an indication whether or not there is an autocorrelation. But the best thing is to use the Eviews Serial Correlation LM test. We are dealing with quarterly data, the test should then be performed for the last four lags. The null hypothesis is that there is no autocorrelation. If the null hypothesis is rejected then there exist autocorrelation and we have to adjust for it. This can be done with the Newey-West estimator, which gives us correct standard errors even due the residuals are correlated with each other and hypothesis testing can then be performed as usual (Woodridge 2006;416-423).
As stated earlier it is crucial to perform the heteroskedasticity test after correcting for autocorrelation. The Whites Heteroskedasticity test is a good test to determine if there is heteroskedasticity. The null hypothesis of the test is that there is no heteroskedasticity and correcting for it can be done with the Newey-West HAC.
- RESULTS -
The first part of this section presents the tests that have been performed in order to ensure that spurious correlation is not present. The second part provides the regression results from the model specification estimation.
5.1 Co-integration
As stated earlier co-integration means that the time series in question are not stationary by themselves, but together they become stationary because of the fact that they have a long-run relationship. The results in table 2 has been compiled through the procedure described in section 4.1.2 Co-integration. The null hypothesis of the test is that there exists a unit root, i.e. no co- integration relationship. The t-statistic for the sample in column (1) results in the fact that the null hypothesis is rejected on the one percent significant level, due to -3.906 < -3.90. Performing the same test for the subsamples in column (2) and (3) would yield the same result, i.e. the null hypothesis is rejected, but on the five percent significant level instead. Thus we can arrive to the conclusion that a co-integration relationship exists between the time series.
Table 2
Null hypothesis: The estimated equation (…) has unit root
(1) 1977-2006
(2) 1977-1993
(3) 1994-2006
Test critical
values
1 % level -3.90 -3.90 -3.90
5 % level -3.34 -3.34 -3.34
10 % level -3.04 -3.04 -3.04
t-Statistic -3.906150 -3.613942 -3.724478
Note: The test critical values for the test is taken from Wooldridge (2006: table 18.4) and not from the ADF test. The lag length of the test was automatic based on Schwarz Info Criterion.
5.2 Error Correction Model
Using Engle-Grangers two-step procedure gives the estimation outputs for the ECM, which is presented in table 3. The ECT, st 1, is the residuals from our dynamic model specification, see Eq (13), which is used to estimate the ECM. The ECM has been estimated for i, where i = 2 and 4.
All equations have been estimated for the whole period: 1977-2006.
Table 3
Equation i i = 2 & 4
(1)
Equation with( 2)
(2)
Equation with ( 4)
Constant 0.002346***
(0.000705)
0.002368***
(0.000739)
)
log(pf q 0.266002***
(0.031338)
0.278897***
(0.033936)
w
log 0.266002***
(0.031338)
0.143343***
(0.026880)
1
st -0.151134**
(0.070226)
-0.184764**
(0.079632)
2
Radj 0.414411 0.427195
Note: *** p < 0.01, ** p < 0.05, * p < 0.10. Numbers in parentheses are standard errors.
All coefficients have the right sign and are significant on the five percent level. The R of the adj2 estimated models are all just above 40 percent. We see that there is an “error correction” in effect. The point estimate for the error correction term is -0.185 point, see column (2), implying that 18.5 percent of last year’s disequilibrium is made up in the current year. A 95 percent confidence interval for this parameter would be between -0.0255 and -0.343912. For the equation in column (1) the point estimate is -0.1511, meaning that a little bit more that 15 percent of last year’s disequilibrium is made up in the current year.
5.3 Structural Change
Before proceeding and estimating the model specification for the time periods 1977-1993 and 1994-2006, it is appropriate to perform a formal test in order to determine if there is a structural
12 A 95 percent confidence interval is calculated as following: -0.1847 2 (0.0796)
change at year 1994. Table 4 show the result of a Chow breakpoint test. The null hypothesis of the test is that there is no break at the specified breakpoint, meaning year 1994. From the p-values we can see that the null hypothesis is rejected. Therefore the test indicates that there has been a structural change in the sample at year 1994.
Table 4
Null Hypothesis: No breaks at specified breakpoints Chow Breakpoint Test: 1994Q1
Value p-value
2
F – statistic 7.774491 0.0000
Log likelihood ratio 81.34734 0.0000
4
F – statistic 8.928984 0.0000
Log likelihood ratio 89.58201 0.0000
Note: the equation sample had the time period of 1977-2006 and all of the equation variables were included in the test.
5.4 Estimation Results
Table 5 contains estimated equations for the periods 1977-2006, 1977-1993 and 1994-2006 furthermore two specifications have been estimated depending on i, where i = 2 and 4. The estimated models explain over 90% of the variation in the depending variables.
We can see that all of the coefficients for log (pf q) and log whave the right sign and are significant on the one percent level, with the exception of logw in column (3). The economic theory tells us that the two coefficients should sum up to one, meaning that the optimal price level of the domestic firm should be determined by the domestic wage costs and the importing prices in the domestic currency, if we have total flexible prices. Adding the coefficients together we see that the sum is fairly close to one in most cases. The estimates can be considered reasonably stable across periods.
Table 5
Equation period
(1) 1977-2006
(2) 1977-1993
(3) 1994-2006
(4) 1977-2006
(5) 1977-1993
(6) 1994-2006
Constant -1.308545***
(0.271151)
-1.746737***
(0.315968)
-3.023025***
(0.682258)
-0.927009***
(0.260419)
-1.157226***
(0.281634)
-3.705608***
(0.350470)
Log (pf q) 0.554511***
(0.056408)
0.589255***
(0.054345)
0.568648***
(0.095336)
0.528497***
(0.049271)
0.537729***
(0.058495)
0.716498***
(0.050131)
Log w 0.617922***
(0.108763)
0.846834***
(0.112789)
0.269604 (0.300895)
0.394834***
(0.106368)
0.540855***
(0.094731)
0.592861***
(0.117794)
Trend -0.002313
(0.007039)
-0.019572**
(0.008378)
0.287965***
(0.059063)
0.012046*
(0.007089)
0.003138 (0.008378)
0.225644***
(0.033475)
Trend2 -0.000450**
(0.000207)
0.000585 (0.000525)
-0.012179***
(0.002224)
-0.000783***
(0.000194)
-0.000379 (0.000518)
-0.009893***
(0.001312)
Trend3 8.63E-06*
(3.16E-06)
-2.90E-05 (1.74E-05)
0.000167***
(2.92E-05)
1.15E-05***
(2.73E-06)
-3.08E-06 (1.66E-05)
0.000136***
(1.75E-05)
SeasonalD1 0.001662
(0.002962)
0.011755**
(0.004737)
-0.006274 (0.007362)
0.000586 (0.001946)
0.001788 (0.002917)
0.005268***
(0.001729)
SeasonalD2 -0.005741**
(0.002642)
0.003413 (0.004421)
-0.011273*
(0.006130)
-0.001732 (0.001155)
-0.000519 (0.001807)
-0.002295 (0.002089)
SeasonalD3 0.004098**
(0.001764)
0.009146***
(0.002226)
-0.001574 (0.003761)
0.003685***
(0.001310)
0.006766***
(0.001530)
0.003180*
(0.001807)
pf 2log
0.369596**
(0.148125)
0.512827***
(0.162971)
0.433437**
(0.187110) - - -
q
2log -0.211994***
(0.045170)
-0.139098***
(0.047649)
-0.186058***
(0.066617) - - -
w
2log 0.082639
(0.087424)
-0.172336 (0.113672)
0.366485
(0.249434) - - -
pf
4log - - - 0.290988***
(0.077212)
0.252731***
(0.057485)
0.447633***
(0.057985)
q
4log - - - -0.138721***
(0.032911)
-0.110143***
(0.033663)
-0.231174***
(0.043118)
w
4log - - - 0.226750**
(0.089232)
0.108178 (0.108110)
-0.110016 (0.127155)
2
R adj 0.997566 0.998183 0.928136 0.998194 0.998642 0.960569
Note: ***p < 0.01, ** p < 0.05, * p < 0.10. Numbers in parentheses are standard errors. All estimations were done by OLS procedure in Eviews6 allowing for adjusted for autocorrelation and conditional heteroskedasticity.
Almost none of the coefficients for ilogpf and ilogw have the expected sign. But the coefficient in focus off this study ilogq has the right sign and is significant at the one percent level in all estimated equations. In columns (1) and (4) where the estimates are for the whole period, we can note that the ilogq-coefficients are -0.212 and -0.139. This could have two interpretations meaning that prices are set at an earlier point in time or that firms have lagging information when exchange rates are concerned. But since the information about exchange rate is updated and available daily, it must mean that prices do not reflect the most recent information about exchange rates. Therefore one could arrive to the conclusion that prices are set a few quarters in advance and the ilogq-coefficients reflect this. In a questionnaire study done by Assarsson in 1989 the price setting in manufacturing markets in Sweden were analyzed. In the study Assarsson shows that most Swedish industrial firms change their prices once or twice per year (Assarsson 1989:128-9). A similar study was also performed by Blinder (1998) which also confirmed that prices are sticky. Our result that prices are set in advanced is consistent with both Assarsson’s and Blinders findings.
Can the fact that prices are set at an earlier point of time tell us something about the competition in the manufacturing market? Usually when firms have perfect information about the demand curve, there is no difference between choosing the price or the quantity produced.13 But if the assumption about perfect information in not valid, i.e. the firm does not have information about conditions in time t when making its decisions, then setting the price does not yield the quantity produced. That is to say that if prices are set at an earlier point of time, it indicates that the firms are competing in prices. Gottfries (2002) found in his study that prices are predetermined relative to exchange rates when it comes to manufactured goods and therefore indicated that it appears that firms compete in prices rather than quantities. Since our result shows that prices seem to be set at an earlier point of time, we can confirm the result that Gottfries (2002) found.
13 Under the assumption that the typical firm is small, relative to the market.
- CONCLUSION -
The purpose of this dissertation was to evaluate how the exchange rate pass-through had evolved during the two inflation regimes. Therefore it is in order to compare the two sub-periods (1977- 1993) and (1994-2006) in table 5 with each other. At first glance we observe that the exchange rate pass-through, both in two and four quarters, is more sluggish in the later time period.
Looking at the estimated equations in column (5) and (6) the difference is apparent, which is more than 12 percent. The same result is observed for estimations in column (2) and (3), but the difference is almost five percent. This result suggests that the exchange rate pass-through has changed.
In the short run, meaning two and four quarters, the exchange rate pass-through has become more sluggish during the later time period. This result is similar to Taylor (2000) and the articles that have been summarized in the section Previous Research. Furthermore the mentioned articles try to link this decrease in pass-through with a shift in the monetary policies responsiveness to inflation. The idea is that if actors believe that the future path of inflation is going to be low, then they would not react to a temporary exchange rate chock. Because they would assume that the shock is temporary.14 If one has this reasoning in mind, then it is not farfetched to assume that the monetary policies responsiveness to inflation has had an impact on the exchange rate pass- through in Sweden. Returning to figure 1 in the introduction, we see that the decrease of pass- through is co-occurring at the same time as the inflation has stabilized. Thus, as the Taylor hypothesis predicted.
One importing aspect of pass-through is its policy implications as discussed in the introduction.
Our result indicates that the degree of pass-through has declined during the low inflation regime, which would also dampen the expenditure shifting effect. The expenditure shifting effect is one
14The reasoning is as following: In a high inflation environment firms are more likely to increase their prices more than necessary, in order to insure themselves. This is because they can not predict whether or not the inflation rate next period is going to be 8, 11 or 14 percent. But in a low inflation environment they can assume the inflation to be low and stable even in the future.
of the central points for the argumentation for a flexible exchange rate. As early as in 1953 Milton Friedman argued that one of the advantages of flexible exchange rate is that they could allow for rapid change in relative prices between countries: “A rise in the exchange rate … makes foreign goods cheaper in terms of domestic currency, even though their prices are unchanged in terms of their own currency, and domestic goods more expensive in terms of foreign currency, even though their prices are unchanged in terms of domestic currency. This tends to increase imports, [and] reduce exports...” (Friedman 1953:162). The underlying assumption in the quotation above is that there is significant pass-through of the exchange rate change to the buyer of the good.
Friedman’s reasoning above is based on an assumption that appears to be at odds with empirical evidence. In the section Previous Research we took part of the empirical evidence which concludes that the exchange rate pass-through is not complete. But in this study we have also indicated that the degree of pass-through have decreased even further during the low inflation regime. Therefore the expenditure shifting argument would lose its merit the more sluggish the exchange rate pass-through becomes.
- FINAL COMMENTS -
The dissertation has been able to show that the degree of exchange rate pass-through has decreased during the low inflation regime when compared to the high inflation regime, for Swedish manufacturing products. It has also been able to show that the typical firm sets its prices in advanced and therefore do not take into account the most recent available information when exchange rates are concerned. The findings indicate therefore that firms are competing in prices rather than quantities.
The most interesting aspect of exchange rate pass-through is its policy implications. There is considerable evidence that the pass-through effect has decreased further, during the low inflation regime, in most advanced economies. An interesting implication is that the effect of expenditure shifting should then theoretically be limited. Because if the effect of exchange rates pass-through on prices has decreased, then the exchange rate does not play as big role as before in adjusting relative prices. This could have an effect when the choice of exchange rate system is concerned.
However the extent of such policy implications is a topic for future research.
- APPENDIX -
7.1 Definitions of variables
Seasonally unadjusted data are used and seasonal dummies are included in all equations. The type of trade-weights used in this dissertation is standard procedure at organisations such as the IMF.
do
p
t : PPI for all Swedish manufacture products between 1977-2006, sorted so that the nomenclature SNI69 and SPIN2002 are matched. Source: Statistic Sweden.d
w
t : Labour cost index for all manufacture products for blue-collar workers between 1977- 2006, sorted so that the nomenclature SNI69 and SPIN2002 are matched. Source: Statistic Sweden.f
p
t : Weighted PPI for eleven OECD countries between 1977-2006. Weights: see below. Source:OECD.
q
t : Weighted exchange rate index for eleven OECD countries between 1977-2006. Weights: see below. Source: OECD & Sveriges Riksbank.)
( p
tfq
t : Is the sum of the weighted PPI and exchange rate index for eleven OECD countries. It is assumed to capturing the price of the importing firm’s products in the domestic currency) (t
f
: Year trends. Have used linear, quadratic and cubic forms.Weights
: Eleven countries have been included in this dissertation which make up approximately 67.8 % of the total imports to Sweden year 2005. These countries are, sorted according by the share of imports: Germany 18%, Denmark 9.8%, Norway 8.1%, UK 6.9%, Netherlands 6.8%, Finland 6.1%, France 5.1%, U.S 3.4%., Japan 2.1%, Spain 1.6% and Canada 0.3%. Source: Kommerskollegium.7.2 Derivation of equation (12)
) 2 (
) )
2 ((
12 1
1 1
0
d t d t t
f t t
f t do
t
p q p q w w
p
) 2 (
) )
( ) 2 ((
1 1
0
d t d
t d t t
f t t
f t t
f t do
t
p q p q p q w w w
p
d t t
t d
t t
f t do
t
p q w p q w
p
0 1( )
22
1 *2
12
2By changing the notation of some of the parameters, such as ptf ptf
2
1
3
t
t q
q 2
1
4 and d
t d
t w
w 2
2
5 the equation (12) is given:
