APPENDIX I.1 The VAR Model
The methodology employed in the chapter was first proposed by McCarthy (1999) and is based on a vector autoregressive (VAR) model that incorporates a recursive distribution chain of pricing. A five–variable VAR, that includes the price of oil in domestic currency (ptoil), output (yt), the nominal exchange rate (et), the wholesale price index (wpit), or, if available, the producer price index (ppit) and the consumer price index (cpit) is estimated. All variables are introduced in logs and first differences to render them stationary. Formally, the system is:
The rationale for this model is that oil shocks identify supply shocks, while output identifies demand shocks. The inclusion of output also measures the cyclical position of the economy. The exchange rate is therefore allowed to respond to supply and demand shocks, and to its own shock. Wholesale prices respond to these shocks, and to their own shock. The CPI responds to all the shocks of the system. The ordering of the variables denotes the “price chain” structure of the model, as it seeks to identify the effects of exchange rate innovations at each stage of the distribution process. The shocks are identified (orthogonalized) using the Cholesky decomposition of the variance–covariance matrix of the reduced form residuals.
In all cases, the sample covers the period 1995–2004, and uses monthly data. In all cases, six lags of each variable are introduced in the VAR. Due to data limitations, the starting date for Poland is November 1996; the starting date for Brazil is January 1996 in order to exclude observations from that country’s last hyperinflationary period.
Hassan, M, 2004, “Driving Forces behind the Real Effective Exchange Rate: The Case of Egypt,” unpublished manuscript, Central Bank of Egypt and Cairo University.
Kara, H., Küçük Tuğer, H., Özlale, Ü., Tuğer, B., Yavuz, D., and E. Yücel, 2005, “Exchange Rate Pass–Through in Turkey: Has It Changed and to What Extent?,” Central Bank of the Republic of Turkey Working Paper 05/04.
McCarthy, J., 1999, “Pass–Through of Exchange Rate and Import Prices to Domestic Inflation in Some Industrialized Economies,” BIS Working Paper No. 79.
Rabanal, P., and G. Schwartz, 2001, “Exchange Rate Changes and Consumer Price Inflation” 20 Months after the Floating of the Real,” in Brazil: Selected Issues and Statistical Appendix, IMF Country Report 01/10.
Prepared by Pau Rabanal. The author would like to thank the Monetary Policy Unit of the Central Bank of Egypt (CBE) for useful comments on an earlier version of this paper.
The new series was released in January 2004, with the initial observation going back to July 2003. A further backward revision with the new weights has not been produced.
The main advantage of this methodology over single–equation regressions or cumulative pass–through calculations is that it takes into account the influence of other macroeconomic variables (e.g., supply shocks, the cyclical position of the economy, the effects of commodity prices) on the price level. One potential shortcoming is that the model is linear: it assumes that large and small exchange–rate shocks (in either direction) have the same proportional effect on prices. The methodology is also not well equipped to deal with parameter instability, a likely consequence of changes in the exchange rate regime, or with very short sample periods.
See, for instance, Rabanal and Schwartz (2001) and Belaisch (2003) for the case of Brazil; Billmeier and Bonato (2002) for Croatia; Leigh and Rossi (2002) for Turkey; and Bhundia (2002) for South Africa.
See Appendix I.1 for a brief description of the VAR methodology and other details of the estimation.
The estimations do not take into account possible changes in the monetary policy and exchange–rate regime. Belaisch (2003) and Kara, et. al (2005) investigate this possibility for the cases of Brazil and Turkey, and find evidence of a decline in the pass–through coefficients in both countries in recent years.
The pass–through at horizon j is defined as PTt,t+j = PTt,t+j/Et,t+j, where Pt,t+j is the cumulative response of the price level j periods after the shock, and Et,t+j is the cumulative response of the nominal exchange rate. For example, if six months after the shock, the nominal exchange rate has depreciated by 3 percent and the price level has increased by 2 percent, the pass–through level would be 66.6 percent.
The results for Brazil, South Africa, and Turkey are similar to those obtained in the studies cited in footnote 4. The VAR included the producer price index (PPI) in the four countries where it is available (the Czech Republic, Mexico, Poland, and South Africa). In the other three countries (Brazil, Israel, and Turkey), the VAR was estimated with the WPI instead.
The data used in the estimation is described in Appendix I.2. The estimations used the parallel market rate from January 2001 to September 2004 (i.e., the exchange rate series plotted in Figure 1) under the assumption that the parallel market rate was the one that influenced the pricing behavior of retail importers. The results were broadly similar when the estimations were done using the official exchange rate.
The behavior of CPI and WPI inflation during 2003–04 presented an additional complication for the choice of the sample period. Because these series exhibited low and stable values between 1999–2003, and started rising afterwards, a VAR estimated with a starting date in 1999 or 2000 delivered unstable roots. Hence, more observations from the past, when both inflation and the exchange rate were stable, had to be added to make the system stationary.
Roughly one third to one half of the items in the CPI series that was used until July 2003 is believed to have consisted of goods with administered prices, including food items, utilities, transportation, and rent.
The two series of commodity prices used in those estimations were obtained from the IMF Research Department.
Additional robustness checks (not reported in Table 3) included: starting the sample period in 1997, introducing M2 in domestic currency (M2D) in the system, or changing the ordering of the variables. The results did not change significantly with any of these changes.