Brazil: Selected Issues and Statistical Appendix

This paper analyzes several issues regarding fiscal sustainability and fiscal adjustment in Brazil during 1990 and searches for econometric evidence of a monetary dominant regime during some subperiods. The following statistical data are also presented in detail: macroeconomic flows and balances, industrial production, consumer price index, relative public sector prices and tariffs, minimum wage statistics, financial system loans, monetary aggregates, exports by principal commodity groups, direction of trade, detailed balance of payments, total external debt, central government operations, and so on.

Abstract

This paper analyzes several issues regarding fiscal sustainability and fiscal adjustment in Brazil during 1990 and searches for econometric evidence of a monetary dominant regime during some subperiods. The following statistical data are also presented in detail: macroeconomic flows and balances, industrial production, consumer price index, relative public sector prices and tariffs, minimum wage statistics, financial system loans, monetary aggregates, exports by principal commodity groups, direction of trade, detailed balance of payments, total external debt, central government operations, and so on.

V. Exchange Rate Changes and Consumer Price Inflation: 20 Months After the Floating of the Real1

A. Introduction

1. This section reviews the impact of exchange rate changes on domestic inflation following the floating of the real in January 1999. It updates findings by Schwartz (1999), drawing on more than one year of additional data and experience under inflation targeting, and using a more refined methodology.

2. In the first six months following the floating of the real in January 1999, Brazil experienced surprisingly low inflation. With the real losing over 30 percent of its value against the U.S. dollar from early January 1999 to June 1999, cumulative consumer price inflation, as measured by the broadest available index (IPCA) that is also used for inflation targeting, amounted to less than 4 percent over the same period. At the time, Schwartz (1999) attributed the low passthrough2 to consumer prices to three main factors: first, the relatively low percentage share of the overall cost of production affected by exchange rate movements, reflecting the fairly small share of imported inputs in industry; second, the apparent evidence that industries “sacrificed” profit margins by not changing their prices fully in response to the exchange rate shock; and third, some exceptional and seasonal factors, particularly in agriculture, that may have helped to mitigate inflation in the first six months after the real was floated.

3. However, already the third quarter of 1999 witnessed an “exchange rate/inflation scare,” and the general picture of a low and well-contained passthrough appeared threatened by expectations of significantly higher inflation rates down the road. What in June 1999 had seemed to be an exchange rate that was fairly stable in the range of R$1.70–R$1.80 per U.S. dollar, was suddenly moving toward R$2.00 per U.S. dollar. At the same time, what had been a low average monthly inflation rate of 0.35 percent in the second quarter of 1999, became a rather worrisome average monthly inflation of 0.91 percent in the fourth quarter of 1999. Could it be that the first six months after the floating of the real had been misleading?

4. Luckily for Brazil, things began to turn around already in the fourth quarter of 1999, with the exchange rate strengthening again and settling back to about R$1.70–R$1.80 per U.S. dollar, and the 12-month rate of consumer price inflation remaining below 9 percent at end-1999, thereby allowing the BCB to meet its inflation target in the first year of operating within the new monetary policy framework.

5. Notwithstanding occasional shocks that moved both exchange rates and inflation in one or the other direction, the passthrough from exchange rate developments to inflation has indeed remained fairly low so far in 2000: twenty months after the floating of the real, the measured passthrough to consumer prices is less than 28 percent. It is not clear, though, whether this continues to reflect exceptional factors, e.g., the decline in food prices experienced in the first half of 2000, or economic “fundamentals.” In addition, given the renewed strength of demand, with real output expected to grow on average by about 4 percent in 2000, it is also unclear to what extent the passthrough can be expected to remain low in the future, particularly since it has relied on the compression of profit margins, which, no doubt, was helped by the fairly weak domestic demand that prevailed in 1999. Clearly, an answer to this puzzle would hold important information for the monetary policy stance under the inflation targeting regime.

B. General Passthrough Determinants and Some Observations for Brazil

6. A large body of theoretical and empirical literature has reviewed the passthrough from exchange rates to prices. In general, the literature contends that the parameters determining the passthrough can broadly be separated into two categories: initial conditions, and policies implemented following a change in the exchange rate regime. McCarthy’s (1999) research, for example, emphasizes the importance of initial conditions. In analyzing the experiences in several OECD economies, he finds that, in general, the passthrough tends to be low and highly correlated with the degree of openness of the economy. In contrast, Borensztein and de Gregorio (1999), who look at 49 episodes of large devaluations find that passthrough experiences differ significantly across countries, and that the policies which were implemented following the change in the exchange rate regime significantly affected the ultimate outcome: 16 percent of all observations had to be excluded from the initial sample because they resulted in hyperinflation, which reflected, among others, an inadequate policy stance. In comparing passthrough experiences in a panel data study of 71 countries, Goldfajn and Werlang (2000) find that the passthrough is higher the longer the time horizon considered, and that it is correlated with the business cycle, the size of the initial real exchange rate misalignment, and the degree of openness of the economy.

7. To explore these issues for Brazil, as an initial exercise, a graphical examination on the extent to which movements in the exchange rate affect other prices in the economy seems useful. Figure 5.1 plots, for the floating exchange rate period, monthly changes in the exchange rate against changes in import unit values (expressed in reais);3 wholesale prices and consumer prices.4 Accordingly, after the floating of the real in January 1999, we detect some correlation between wholesale prices and the nominal exchange rate with a one-month lag, while consumer price inflation seems to react to significant movements in the exchange rate with a lag of three months. Small movements in the exchange rate do not seem to affect any price indicator. Note that, as shown in Figure 5.1, in all cases the fluctuations in the price indices are much smaller than the fluctuations in the exchange rate, and, overall, the passthrough is fairly small, particularly for consumer prices.

Figure 5.1.
Figure 5.1.

Brazil: Monthly Change in Prices

(in Percent)

Citation: IMF Staff Country Reports 2001, 010; 10.5089/9781451805901.002.A005

Source: IBGE, BCB, FUNCEX, and Getulio Vargas Foundation.

8. Figures 5.2 and 5.3 repeat the same exercise with quarterly and annual changes, respectively. Wholesale prices appear to adjust to exchange rate developments with a one-quarter lag; consumer prices appear to follow a similar pattern, although the correlation is less strong. In Figure 5.3, the correlation between wholesale prices and exchange rates seems to occur at longer lags and be less strong than in the previous figure; as in Figure 5.2, there continues to be a positive lagged correlation between exchange rates and consumer prices.

Figure 5.2.
Figure 5.2.

Brazil: Three-Month Change in Prices

(in Percent)

Citation: IMF Staff Country Reports 2001, 010; 10.5089/9781451805901.002.A005

Source: IBGE, BCB, FUNCEX, and Getulio Vargas Foundation.
Figure 5.3.
Figure 5.3.

Brazil: 12-Month Change in Prices

(in Percent)

Citation: IMF Staff Country Reports 2001, 010; 10.5089/9781451805901.002.A005

Sources: IBGE, BCB, FUNCEX, and Getulio Vargas Foundation.

9. These graphical examinations seem to suggest that there is a clear correlation between exchange rate developments, wholesale prices, and consumer prices. Large changes in the exchange rate seem to have a lagged effect on prices.

10. How strong is the correlation between the exchange rate and prices? Table 5.1 shows cross correlations between changes in the exchange rate, and changes in import unit values (in reals) (IMP), wholesale prices (WPI), and consumer prices (CPI) at different time horizons and frequencies, using a sample period from January 1996 to September 2000.5 Accordingly, there is an immediate, almost full impact of exchange rate changes on import prices, suggesting that Brazil is a price taker in international markets. In general, however, correlations between exchange rate changes and other prices are higher for longer periods, i.e., the effects of exchange rate changes on prices show up more strongly in cumulative changes over longer periods of time than in month-to-month changes. Table 5.1 also seems to suggest a logical chronological sequence of price developments being passed on through the supply chain, where the effect of the exchange rate change on prices peaks first for import prices (without a lag), then for wholesale prices, and finally for consumer prices. As expected, in the short run, wholesale price changes show a stronger correlation with exchange rate changes than consumer prices, suggesting that, to some extent, exchange rate shocks were absorbed through lower profit margins of wholesalers.

Table 5.1.

Brazil. Cross Correlations Between Changes in the Exchange Rate in Month t and Changes in Different Price Indicators in Month t+k1/

article image
Source: Authors’ estimates.

Refers to data from January 1996 to September 2000, using end-of-month exchange rates. Exchange rate changes are measured on the basis of the R$/US$ rate. The highest correlation in each column is shown in bold. The results did not change significantly when using another starting date for the cross-correlations.

11. Table 5.1 also suggests that, with the correlation of the 12-month rate of change between the exchange rate and prices peaking after about nine to twelve months, the full passthrough of exchange rate shocks seems to occur relatively more rapidly in Brazil than in most industrialized countries that operate with an inflation targeting framework.

12. Following these preliminary assessments we proceeded to evaluate the exchange rate passthrough in a more comprehensive model framework, using a VAR model similar to the one used by McCarthy (1999), and as adapted for Brazil. This allows us to measure passthrough from exchange rate changes to several price indicators at different time horizons, based on impulse response functions that trace the dynamic effects of the system.

13. McCarthy (1999) estimated a six-variable VAR with the following ordering for the endogenous variables: the price of oil in domestic currency, output,6 the nominal exchange rate (relative to the U.S. Dollar), import price inflation, producer price inflation and consumer price inflation. The idea behind this ordering is that supply and demand shocks are identified by the first two variables. Nominal exchange rate shocks are identified by the third variable, while the other three variables contain sequential shocks that can be attributed to the various stages of the supply chain; this allows us to trace the dynamic effects of a nominal exchange rate shock on all price indicators. The VAR is estimated recursively in a nonstructural way. This estimation procedure allows us to recover the structural shocks, by making use of the Cholesky decomposition of the variance-covariance matrix of the reduced form residuals.

14. To apply McCarthy’s VAR methodology for the case of Brazil, two issues needed to be addressed. First, the price of oil (and oil-based products) is an administered price in Brazil. Hence, international oil price developments will not translate immediately to domestic prices. Still, changes in international oil prices will eventually translate into domestic price adjustments, and therefore reflect a relevant supply shock that has to be taken into account Second, there is no producer price index in Brazil, and a wholesale price index (WPI) has to be used instead. This series is a satisfactory indicator of passthrough at intermediate stages of production, also since it contains a higher percentage of tradable goods than the CPI.7

C. Some Statistical Preliminaries

15. To determine the stationarity of the variables in the system, unit root tests were run on the (natural logarithms) of the price of oil in domestic currency, output, the nominal exchange rate, import unit values (in reais), wholesale prices, and consumer prices. The null hypothesis of a unit root (i.e., nonstationarity) could not be rejected at the 5 percent level for the price of oil, the nominal exchange rate, import unit values (inreais), wholesale prices, consumer prices and output; the null hypothesis of a second unit root was rejected for all series, although it could not be rejected for consumer prices at the 1 percent level. In general, these results suggests that taking the first difference of the various series would induce stationarity.

16. Next, we ran Granger causality tests for both model specifications (levels and first differences). The exercises focused on Granger causality between the nominal exchange rate and the three price indicators. These tests helped further to determine the dynamic behavior of the variables in the system. We used three lags of each variable on the basis of the results from pairwise regressions, starting from six lags, and eliminating the highest nonsignificant lag.8 To check the stability of the results over time, separate tests were performed for the period January 1995-September 2000, and for the sub-periods starting in January 1997 and January 1999. The results are shown in Table 5.2.

Table 5.2.

Brazil: Granger Causality Tests Between the Nominal Exchange Rate and Three Price Indicators.

article image
Source: Authors’ estimates.Note: The table shows Granger causalities between the nominal exchange rate and import unit values (in reais)(IMP), wholesale prices (WPI) and consumer prices (CPI), respectively. Yes means rejection at the usual 5 percent level of the null hypothesis of no Granger causality running from exchange rates to prices; Yes* means rejection at a 10 percent level of the same null hypothesis. Granger causality tests for the quarterly changes yielded results that were similar to the monthly changes and are omitted here.

17. The results in Table 5.2 suggest that there is Granger causality running from the nominal exchange rate to wholesale and consumer prices, but no Granger causality running from the nominal exchange rate to import prices. Since these results reflect lagged relationships, they do not contradict the findings of the simple causality exercise, which suggested that there is a strong contemporaneous relationship between import prices (in reais) and the nominal exchange rate. The exchange rate and import prices (in reais) show an almost full contemporaneous correlation, and there is no lagged causal relationship: hence, exchange rate movements provide no guidance for forecasting future import prices, which supports the suggestion that Brazil is a price taker in international markets.

18. Granger causality between the exchange rate and wholesale prices and consumer prices is more pronounced when running the test on variable levels (“a more depreciated exchange rate causes a higher price level”) instead of changes in variables (“exchange rate changes cause price changes”). The causal links between the exchange rate, the WPI, and the CPI are likely to strengthen with a longer data series for the floating exchange rate regime.

D. Estimating the VAR for Brazil

19. The VAR model that was estimated consisted of six variables with the following ordering: the price of oil in reais, output, the nominal exchange rate (reais per U.S. dollar), import unit values (in reais), the WPI (IPA-DI) and the CPI (IPCA). The model was estimated using three lags of the endogenous variables, as suggested by the Granger causality tests; this lag length was sufficient to induce white noise residuals. To account for the floating of the real in January 1999, and reduce a possible estimation bias, the model included a dummy variable that takes the value of one for January 1999 and is zero otherwise. The sample period for the model estimates was January 1995 to September 2000. While this covers a relatively long period of fixed exchange rates, discarding pre-January 1999 data would not yield a sufficient number of observations.

20. The same model was estimated in levels (“Model 1”) and first differences (“Model 2”).9 This allows evaluating the impact of the level of the exchange rate on the level of wholesale prices and consumer prices, and the impact of exchange rate changes on changes in wholesale and consumer prices.

21. The columns of Table 5.3 show the actual passthrough from December 1998 onward,10 and the estimated cumulative passthrough to the price level of a shock in the level of the nominal exchange rate under the two model specifications (levels and first differences).11 Since the main focus here is on the dynamic effects of a shock to the nominal exchange rate to wholesale prices and consumer prices, we omit the results for all other variables.

Table 5.3.

Brazil: Actual Passthrough and Passthrough Estimates

(in Percent)

article image
Source: Authors’ estimates.

Cumulative percentage deviations of prices from their long-run values in response to a 1 percent standard deviation shock in the nominal exchange rate.

22. In general, the two model estimates (levels and first differences) are rather different. Interestingly, the estimates from Model 1 (levels) appear closer to the actual passthrough since the real was floated than the estimates from Model 2 (first differences).

23. For wholesale prices, the results from Model 2 suggest that the passthrough occurs fairly rapidly and in a nonmonotonic fashion, with the cumulative passthrough peaking at 75 percent already after six months and then dropping to 63 percent after 36 months. This is inconsistent with the actual passthrough experience so far, which has occurred more slowly and continued to increase cumulatively. Actually, a 75 percent passthrough to wholesale prices was reached only after about 19 months, not already after 6 months as predicted by Model 2. In general, the estimates of Model 2 appear counterintuitive, also because they seem to imply that wholesalers have a significant degree of market power to pass external shocks through to retailers fairly rapidly. This makes it all the more difficult to understand why then the cumulative passthrough to wholesale prices would decline after six months, as predicted by Model 2.

24. In contrast, the passthrough estimates from Model 1 are by and large more in line with actual observations, although the actual passthrough is overestimated for the first nine months and underestimated thereafter. Both model estimates suggest that roughly two-thirds of the initial exchange rate shock can be expected to have been passed through to wholesale prices after 20 months; this compares to a somewhat higher actual passthrough of just below 80 percent.

25. For consumer prices, both model specifications suggest that the cumulative passthrough increases monotonically over time, reaching, depending upon model specification, 26 percent to 44 percent after 36 months (Table 5.3). Again, the passthrough estimated by Model 2 was stronger and occurred much faster than in Model 1. Accordingly, the estimates from Model 2 suggest that, already after six months, the cumulative passthrough reaches a level that exceeds the cumulative passthrough under Model 1 after 36 months.

26. Again, Model 1 tracked the actual experience in the first 20 months better than Model 2. After 20 months, Model 1 predicts that roughly 23 percent of the initial exchange rate shock would have been passed through to consumer prices, which compares to 28 percent actual passthrough. Also, as suggested in Table 5.3, most of the passthrough to consumer prices occurs during the first year. While the cumulative passthrough to consumer prices continues to increase through 36 months, the additional increases beyond the first 20 months are fairly small.

E. Concluding Remarks

27. This section has analyzed the passthrough from exchange rate developments to prices in Brazil during the first 20 months after the real was floated in January 1999. While the actual passthrough has been moderate since January 1999, it would seem interesting to know whether model-based estimates would generate a similar outcome, or whether the low actual passthrough may be considered an outlier. In modeling the passthrough, this section has applied an approach first suggested by McCarthy (1999). McCarthy’s (1999) model has the advantage that it tries to isolate the effect of exchange rate fluctuations from other factors (e.g. oil price shocks), so that the measured change in prices is considered to be only partially due to exchange rate factors. The model suggests that different prices react rather differently to the same exchange rate shock.

28. The analysis presented here shows that import prices (in reais) have a strong concurrent relationship with exchange rate shocks, reflecting the fact that Brazil is a pricetaker in international markets. At the same time, developments in the exchange rate do not explain future developments in import prices, which depend on conditions in international markets, in which Brazil is a price taker.

29. Both actual observation and model estimates suggest that, as a rough rule of thumb, after 18 months about two-thirds of the initial exchange rate shock has been passed through to wholesale prices, and one third of these two-thirds (or two-ninth of the initial shock) has been passed through to consumer prices. Hence, the passthrough to wholesale prices occurs more rapidly and is more pronounced compared to the passthrough to consumer prices.

30. The model estimates are comforting in the sense that they suggest that, 20 months after the floating of the real in January 1999, the initial shock has worked itself through the system already, given relatively short transmission lags. In addition, the estimates suggest that the relatively small fluctuations in the foreign exchange rate that have been experienced following the initial shock are unlikely to have a sizeable impact on consumer prices, and are more likely to be absorbed along the supply chain.

31. However, the relatively large difference in the passthrough to wholesale prices and to consumer prices suggests that, up to now, much of the absorption of exchange rate shocks has taken place somewhere along the supply chain, and must have led to a significant compression of profit margins.

32. Profit margins may not remain compressed forever, and two issues need to be kept in mind in this regard. First, a significant pickup in domestic demand could render the model estimates unstable, and result in a significantly larger actual passthrough. With much of the period for which data were used relating either to the fixed exchange rate regime or to a trough in the business cycle, it is unlikely that estimates based on historical data can provide good guidance on likely future outcomes. This calls for continued caution, particularly concerning the stance of monetary and fiscal policies.

33. Second, regardless of the stage of the business cycle, another large exchange rate shock, as unlikely as it may seem, may not necessarily have the same quantitative effect as the shock that occurred in January 1999, particularly considering that profit margins along the supply chain have already been compressed. Yet, if it were to happen, it could hit prices more quickly and more strongly than before, which may leave little time for a policy response.

34. These considerations suggest that the model estimates presented in this section may not be stable over time. In particular, as the output gap closes, the passthrough from exchange rate developments to prices, particularly consumer prices, may be significantly stronger than what has been observed in the past.

References

  • Borenzstein, Eduardo, and José de Gregorio, 1999, “Devaluation and Inflation after Currency Crisis” (Washington, D.C.: International Monetary Fund) (draft).

    • Search Google Scholar
    • Export Citation
  • Goldfajn, Ilan, and Sérgio Werlang, 2000, “The Pass-Through from Depreciation to Inflation: A Panel Study.” Working Paper Series (Central Bank of Brazil), No. 5 (July).

    • Search Google Scholar
    • Export Citation
  • Instituto de Pesquisa Económica Aplicada (IPEA), 2000, Boletim Conjuntural, No. 49 (April).

  • McCarthy, Jonathan, 1999, “Pass-Through of Exchange Rates and Import Prices to Domestic Inflation in Some Industrialized Economies,” Bank for International Settlements Working Paper, No. 79.

    • Search Google Scholar
    • Export Citation
  • Nelson, Charles R., and Charles I. Plosser, 1982, “Trends and Random Walks in Macroeconomic Time Series,” Journal of Monetary Economics, Vol. 10, pp. 139162.

    • Search Google Scholar
    • Export Citation
  • Rabanal, Pau, and Gerd Schwartz, 2000, “Testing the Effectiveness of the Overnight Interest Rate as a Monetary Policy Instrument” (Washington, D.C.: International Monetary Fund) (Section VI in this document).

    • Search Google Scholar
    • Export Citation
  • Schwartz, Gerd, 1999, “Price Developments in Brazil After the Floating of the Real: the First Six Months,” In Brazil—Selected Issues and Statistical Appendix, IMF Staff Country Report No. 99/97, September (Washington, D.C.: International Monetary Fund).

    • Search Google Scholar
    • Export Citation
  • Sims, Chris A., James Stock and Mark Watson, 1990, “Inference in Linear Time Series Models with Some Unit Roots”, Econometrica, Vol. 58, pp. 113144.

    • Search Google Scholar
    • Export Citation
1

Prepared by Pau Rabanal and Gerd Schwartz.

2

Passthrough is defined here as the cumulative consumer price inflation relative to the cumulative depreciation of the real vis-à-vis the U.S. dollar.

3

The series reflects unit values for imports of goods and nonfactor services, as published by FUNCEX, where the original index was transformed using the prevailing exchange rate.

4

Wholesale prices reflect the wholesale price index (IPA-DI) of the Getulio Vargas Foundation (FGV); an alternative wholesale price index (IPA-OG), also published by FGV, which has a higher component of imported goods than the IPA-DI, delivered very similar results. For a similar comparison, using only the industry component of the IPA-OG index, see IPEA (2000). Consumer price inflation is measured on the basis of the IPCA, published by the IBGE.

5

While, ideally, the passthrough should be measured using only data for the floating exchange rate regime that started in mid-January 1999, there are only some 20 observations so far. Using some data from the fixed exchange rate period is likely to bias our estimates toward a low passthrough. Therefore, it reduces the measured impact of the large depreciation following the floating of the real in January 1999.

6

In applying Mc Carthy’s (1999) model to Brazil, output was proxied alternatively by industrial production, and by a series of monthly output proxies, prepared by the BCB. Both series yielded rather similar results, and we decided to use the monthly series of output proxies, as it seemed more comprehensive.

7

About 90 percent of the components of the WPI are tradables, while the CPI contains about 50 percent of tradables. Initially we considered including wages in the VAR. However, since the sample suggested that the exchange rage did not Granger-cause wages, and that wages did not Granger-cause price indicators, wages were dropped from the VAR.

8

The Granger causality tests for the subperiod January 1999 to September 2000 only used one lag.

9

The unit root tests generally suggested to estimate the VAR in first differences. Still, by estimating the VAR in levels, we allow for the possibility of cointegration between variables. Using Johansen’s cointegration test in the e-Views econometrics package, we identified three cointegrating relationships between the five variables in the system. While this test is useful in identifying the number of cointegrationg relationships, it does not offer guidance on which variables are actually cointegrated. However, given the evidence on the presence of cointegration, we assume that estimating the VAR in levels is a valid strategy. See Nelson and Plosser (1982), and Sims et al. (1990) for a discussion on estimating VAR models when series are nonstationary and possibly cointegrated.

10

The passthrough is measured here by PTt, t+j=Pt, t+j/Et, t+j, where PTt, t+j denotes the cumulative passthrough after j months, Pt, t+j the cumulative change in the price level after j months, and Et, t+j is the cumulative change in the nominal exchange rate after j months. As presented here, the passthrough is measured based on changes in the R$/US$ rate, i.e., a change in the exchange rate from R$l.2 per U.S. dollar to R$1.8 per U.S. dollar implies a depreciation of 50 percent, i.e., (1.8/1.2-1)* 100. As measured by Schwartz (1999), the same change would have implied a depreciation of 33.3 percent, i.e., (1.2/1.8-1)* 100.

11

For the specification in first differences, monthly responses were accumulated to obtain the cumulative response of the levels of the variables. We estimate the VAR in both cases with three lags of every endogenous variable. Therefore, when we estimate the VAR in first differences, we are effectively using up to four lags of every endogenous variable.

Brazil: Selected Issues and Statistical Appendix
Author: International Monetary Fund