Journal Issue
Share
Article

Brazil: Selected Issues and Statistical Appendix

Author(s):
International Monetary Fund
Published Date:
September 1999
Share
  • ShareShare
Show Summary Details

I. Price Developments after the Floating of the Real: the First Six Months1

A. Overview

1. This chapter looks at price developments following the floating of the Real in mid-January 1999. In general, the “passthrough” from the exchange rate depreciation to consumer prices2 has been much lower than initially expected, and only amounted to 10 percent at end-March, and 13 percent at end-June. Although it is far too early to reach definite conclusions only six months after floating the currency, the chapter points toward four main elements that may help to explain why inflation has remained surprisingly low: (i) an economy that does not rely on imported inputs to the extent of other economies; (ii) relatively tight monetary and fiscal policies; (iii) sluggish consumption that has resulted in profit margin reductions at different stages of the production process; and (iv) a number of exceptional and favorable factors, like a good harvest and the slow adjustment in some administered prices (such as electricity tariffs), that have dampened upward pressure on prices in different subsectors of the economy (e.g., food and energy). To the extent that the latter two elements may be expected to be temporary, underlying inflation may be expected to be higher in the future.

B. Devaluation and Inflation: The Experience of Some Other Countries

2. Experiences with inflation following a substantial devaluation have varied significantly across countries, ranging from virtually no effect to hyperinflation. Clearly, hyperinflation outcomes were not due to the devaluation itself but were more likely attributable to continued macroeconomic disequilibria. Still for any given price change, it is difficult to extract only the devaluation-induced part. Therefore, studies have usually looked at passthrough in terms of the overall price changes following devaluation.

3. Borensztein and de Gregorio3 (B&G) who looked at 49 episodes of large devaluations (of which eight were followed by hyperinflation and consequently excluded), show that, in general, the passthrough from devaluation to inflation is not fall. On average, 30 percent of the devaluation passed through to inflation in the first three months, and 55 percent in two years. Hence, changes in the real exchange rate are fairly long lasting. However, individual country experiences varied widely. There were countries with virtually no passthrough after three months (“short run”) and only 15 percent after two years (“long run”), countries with a higher short-run than long-run passthrough, and countries where the short-run passthrough was over 50 percent and the long-run passthrough over 100 percent. Of the three Brazilian devaluation episodes included in the original “49-episodes” B&G data set—February 1983, February 1987, and March 1990—the latter two were excluded as they resulted in hyperinflation.

4. For the more recent experiences that are included in the B&G data set, Latin American countries usually experienced a much higher passthrough than European economies (Figure 1.1). Mexico in December 1994, Colombia in August 1995, Venezuela in May 1994, all experienced a reasonably low short-run passthrough, that ranged from 14 percent in Mexico to 40 percent in Venezuela, but had a fairly high long-run passthrough, that ranged from 69 percent in Mexico to 173 percent in Colombia. In contrast, Finland, Italy, Spain, Sweden, and the United Kingdom—all in September 1992—experienced a short-run passthrough of no more than 7 percent, and long-run passthrough of no more than 24 percent.

Figure 1.1.Pass-Through Experiences in the 1990s 1/

Source: Borensztein and de Gregorio (1999), and Fund staff estimates for Brazil.

1/ The “Europe ‘92” PTIs are calculated as the arithmetic mean of the respective PTIs for Finland, Italy, Spain, Sweden and the U.K.; the “Latin America 90s” PTIs are calculated as the arithmetic mean of the respective PTIs for Colombia, Mexico, and Venezuela; the Brazil PTIs are calculated on the basis of the consumer price index (IPCA); the data for Brazil’s “PTI-2Y” only reflects data to June 1999 (latest available data).

5. The more recent experiences in East Asia—Indonesia, Korea, Malaysia, Philippines, and Thailand—all show that the passthrough from devaluation to inflation has been lower than expected, with the exception of Indonesia. An analysis by Stone4 shows that during the 20-month period June 1997 to February 1999, passthrough ratios (on the basis of nominal effective exchange rates) ranged from 23 percent in Malaysia to 49 percent in the Philippines, and only in Indonesia exceeded 100 percent. Hence, even after 20 months, real exchange rates remained substantially below their pre-crisis levels, particularly in those economies that have achieved some degree of stabilization following the crisis.

C. Brazil: Passthrough Experience Since the Floating of the Real

6. With only six months passed since the Real was floated, it is still too early to reach final conclusions about Brazil’s inflation experience as a result of the float. What is commonly agreed though, is that, so far, the passthrough to inflation was quite low (Figure 1.2), and certainly much lower than expected. The passthrough from a “big” exchange rate movement to consumers depends principally on an economy’s degree of openness, the specific goods that are imported, the behavior along the supply chain, and the economic policies that accompany the exchange rate movement.

Figure 1.2.Brazil: Exchange Rate and Price Level 1/

(12-month rate of change)

Source: Central Bank of Brazil

1/ Based on the consumer price index (IPCA).

7. Why has the passthrough been so low in Brazil? There are four elements that may help to explain the phenomenon: an economy that does not rely on imported inputs to the extent of other economies; sluggish domestic demand that has resulted in profit margin squeezes at different stages of the production process; relatively tight fiscal and monetary policies; and a number of exceptional and favorable factors, like a good harvest and the slow adjustment in some administered prices (such as electricity tariffs), that have dampened upward pressure on prices in different subsectors of the economy (e.g., food and energy). All four are interrelated.

Openness and the impact of devaluation on the cost structure

8. Given Brazil’s degree of openness, how did devaluation affect the cost structure of enterprises operating in different sectors of the Brazilian economy?

9. Using the 1995 input-output (I-O) matrix, a recent study by Pereira and Carvalho5 suggests that a devaluation of the Real vis-à-vis the U.S. dollar by about one-third (say, from R$L21 per U.S. dollar to R$1.82 per U.S. dollar) would increase production prices by 8.24 percent after the devaluation has worked itself through the system.6 This calculation assumes that there is no change in profit margins (that is, all cost increases from the devaluation are fully passed through along the supply and production chain).

10. Using the 1995 I-O matrix suggests that (variable) costs subject to exchange rate devaluations range from below 1 percent to close to 50 percent, depending on the industry (Table 1.1). In 11 out of 42 industries variable costs subject to exchange rate fluctuations exceed 20 percent of total variable cost. This total is the sum of direct and indirect effects. Direct effects are experienced by industries that use imported inputs. Of course, even industries that do not use imported inputs would be affected by the devaluation to the extent that their domestic suppliers use imported inputs and pass on these higher costs to them.

Table 1.1.Brazil: Impact of a Devaluation of the Real on Industrial Costs in Different Sectors of the Economy 1/
Percent of Costs Sensitive to a Devaluation (percent)Impact on Costs of a Devaluation (in percent) of the Real against the U.S. Dollar
SectorTotalDirect (1st

round)

Impact
Indirect

Impac
20

Percent
30

Percent
50

Percent
Agriculture and fisheries8.72.56.21.72.64.4
Mineral extraction10.82.58.32.23.25.4
Oil and gas extraction7.22.44.81.42.23.6
Nonmetallic minerals10.62.87.82.13.25.3
Iron and steel19.47.112.23.95.89.7
Metallurgy (noniron)29.514.914.65.98.914.8
Other metallurgy13.42.411.02.74.06.7
Machines and tractors13.96.67.22.84.26.9
Electric materials19.46.512.93.95.89.7
Electronic equipment47.529.418.19.514.223.7
Cars, trucks, and busses27.515.811.75.58.313.8
Other vehicles and vehicle parts15.54.910.63.14.77.8
Wood and furniture9.21.87.41.82.84.6
Paper and graphics (printing)17.97.010.93.65.49.0
Rubber26.210.216.15.27.913.1
Chemical elements12.05.07.02.43.66.0
Oil refinery30.919.011.96.29.315.5
Chemicals29.514.215.35.98.814.7
Pharmaceuticals and perfumes23.514.19.44.77.011.7
Plastics22.17.614.64.46.611.1
Textiles27.410.716.75.58.213.7
Clothing20.15.015.24.06.010.1
Footwear17.66.011.63.55.38.8
Coffee6.10.25.91.21.83.0
Processing of vegetable products12.54.87.72.53.86.3
Animal slaughter8.10.67.41.62.44.0
Milk products9.81.97.92.02.94.9
Sugar11.21.89.42.23.35.6
Vegetable oils13.23.89.42.64.06.6
Food products13.95.08.92.84.26.9
Other industries12.34.47.92.53.76.1
Public services9.55.04.61.92.94.8
Civil construction8.22.35.91.62.54.1
Commerce7.11.55.61.42.13.6
Transport22.011.310.74.46.611.0
Communications6.43.72.71.31.93.2
Financial sector2.61.01.60.50.81.3
Services to families7.71.76.11.52.33.9
Services to enterprises5.21.43.81.01.62.6
Housing rent0.90.40.50.20.30.4
Public administration5.31.73.51.11.62.6
Private nonmarket services1.00.20.80.20.30.5
Source: Thiago Pereira and Alexandre Carvalho (IPEA/DISET), as published in Boletim de Politica Industrial, No.7 (April 1999)

11. In an economy that mainly imports production inputs, the passthrough of an exchange rate movement to consumers may be occurring more slowly than in an economy that mainly imports finished consumer products, as it takes time for the price effect to work through the supply chain. In Brazil, in most industries, indirect effects (that are experienced by purchasing production inputs from those who import directly) outweigh the direct effects (that are experienced when importing directly).

12. The 1995 I-O matrix suggests that the shares of imported goods in the variable costs of enterprises, i.e., direct effects, range from below 1 percent in some industries (e.g., coffee, private nonmarket services, animal slaughter) to close to 30 percent (electronic equipment) (Table 1.1). Imported goods accounted for more than 10 percent of variable costs in only 9 out of 42 industries. Hence, the direct impact of a devaluation is comparatively small. However, using the technical coefficients of the I-O matrix, the indirect impact of the devaluation on the cost or production ranges again from less than 1 percent in some industries to around 18 percent in the electronic equipment industry. In 16 out of 42 industries, these indirect effects exceeded 10 percent of total variable costs.

13. In 34 out of 42 industries, indirect effects exceeded the direct effects. This implies that most industries are relatively more affected by a devaluation indirectly, that is because they buy their inputs from other domestic suppliers who buy imported inputs, than directly, through the prices of their own direct imports. This in turn may mean that the effects of a devaluation may take some time before they show up in consumer prices. The extent to which this will happen would depend on the profit margin squeezes along the supply chain.

Profit margin developments

14. Margins may be squeezed (for different reasons) at different stages of the supply chain, as illustrated in Figure 1.3, which is adapted from Haldane.7 In the stylized example, the size of the passthrough from a devaluation to retail prices would depend on what happens to the profit margins of foreign exporters, domestic input suppliers, domestic wholesalers, and domestic retailers. Of course, there are several key factors influencing the latter, including the business cycle and, more generally, the type of policies pursued by the government.

Figure 1.3.Passthrough from Exchange Rate Changes to Retail Prices

15. Apart from simply observing that wholesale price developments have outrun consumer price developments this year, there are two other indicators that seem to suggest that profit margins have indeed been cut over the past several months somewhere along the supply and production chain. First, a survey by the National Industry Federation (CNI) suggests that producers did not expect to pass through to prices a large part of the devaluation. Second, calculations by the Institute of Economic and Administrative Research (IPEA) suggests that, in the initial stages of devaluation, profit margins have indeed been squeezed in most industries.

16. In March this year, a survey carried out by CNI, showed that 71 percent of small enterprises and 78 percent of large enterprises believed that price increases due to the devaluation should only be passed on to consumers partially. Only 14 percent of small enterprises, and 10 percent of large enterprises suggested that the passthrough should be complete. Whereas 34 percent of small enterprises suggested that their profit margins should be squeezed in response to the depreciation, only 15 percent of large enterprises suggested this to be the correct response. At the same time, 75 percent of the large enterprises but only 52 percent of the small enterprises suggested that prices should be kept in check by seeking to reduce other costs or increase productivity.8

17. Although the CNI survey focused more on normative aspects—how should prices be formulated in light of the devaluation—it is supported by evidence in different sectors. Many industries were indeed reluctant to seek a high passthrough. Several importers, for example, suggested that they were seeking to have a reasonable split between reductions in profit margins and price increases.9

18. To some extent, at least, the reluctance of the entrepreneurs to seek a higher passthrough may have had to do with the generally sluggish domestic demand in light of a projected negative economic growth in 1999. It also may indicate that the exchange rate levels which prevailed in February and March 1999 were viewed as a temporary overshooting and would not justify dramatic price increases; if price increases were carried out anyway, they could potentially harm the company’s market share and, thus, have adverse long-run repercussions. In light of these considerations, many entrepreneurs seem to have adopted a “wait and see” attitude, at least initially. With fears of a prolonged economic downturn receding, companies that were initially reluctant to increase prices may possibly now seek to do so.

19. Overall, wholesale prices have been growing more rapidly than consumer prices since the Real was left to float in January (Figure 1.4).10 Recent estimates by IPEA11 show that profit margins have behaved differently depending on industry, but that most industries saw their profit margins shrink in the initial stages of the devaluation (Table 1.2); in some cases, however, shrinking profit margins had little to do with the depreciation of the Real per se. In 19 out of 26 industries for which calculations were made, profit margins in February 1999 were lower than immediately before the devaluation in December 1998. Industries that increased profit margins (5 out of 26) mostly have a strong export orientation (e.g., mineral extraction). However, some industries with a strong export orientation experienced a drop in profit margins that probably reflects adverse international price developments (e.g., the coffee industry).12 Also, in some industries where profit margins have fallen dramatically (e.g., oil refinery), much of the drop probably reflects the fact that energy prices were, initially, not fully adjusted to the exchange rate developments.

Figure 1.4.Brazil: Monthly Wholesale and Consumer Price Inflation

(cumulative increase in prices from January 1995, in percent) 1/

Citation: 1999, 97; 10.5089/9781451805895.002.A001

Source: Getulio Vargas Foundation.

1/ Reflects FGV-IPA (wholesale prices) and FGV-IPC (consumer prices) indices, published by the Getulio Vargas Foundation (FGV).

Table 1.2.Brazil: Estimates of Profit Margins in Different Sectors (August 1994=1)
Jan-98Feb-98Mar-98Jun-98Sep-98Dec-98Jan-99Feb-99
Agriculture and fisheries0.990.991.000.970.980.970.960.98
Mineral extraction0.810.810.810.790.800.760.810.85
Nonmetallic minerals0.890.910.920.930.940.880.880.85
Iron and steel0.960.960.960.960.950.930.910.89
Metallurgy (noniron)0.960.950.950.940.930.900.880.89
Other metallurgy0.870.860.870.860.860.810.820.83
Machines and tractors0.790.790.800.800.810.770.750.72
Electric materials0.790.780.790.790.790.750.750.75
Electronic equipment0.690.680.690.680.670.630.580.57
Cars, trucks, and busses0.830.830.830.810.800.780.780.76
Other vehicles and vehicle parts0.880.880.890.890.890.860.850.82
Wood and furniture0.850.850.850.850.850.810.820.79
Paper and graphics (printing)0.860.860.850.880.870.830.810.85
Rubber0.900.900.900.910.920.890.880.85
Oil refinery1.011.021.031.031.011.030.960.91
Pharmaceuticals and perfumes0.890.900.920.910.910.880.840.84
Plastics0.860.860.870.850.840.800.810.83
Textiles0.920.910.920.920.920.900.880.88
Clothing0.830.820.830.820.820.790.780.76
Footwear0.790.790.780.790.770.740.730.68
Coffee0.890.900.940.990.980.940.930.88
Animal slaughter0.950.950.960.960.970.950.950.95
Milk products1.021.021.011.041.020.990.980.96
Sugar0.840.850.850.840.800.750.740.70
Vegetable oils1.221.181.131.081.041.051.061.02
Food products0.970.980.971.051.061.041.010.98
Source: IPEA, Boletim de Politico Industrial, No. 7, April 1999, based on an elaboration by IPEA/DISET.

Exceptional and favorable factors

20. The discussion above indicates that exceptional and favorable factors in certain sectors may have contributed to price inflation being below expectations since the Real was floated. In general, it is not clear how much of the stronger increase in wholesale prices will eventually find its way to consumer prices. In other words: it remains to be seen how much of the squeeze in profit margins that has occurred is temporary or permanent. This uncertainty also stems from the fact that some exceptional price developments in different sub sectors of the Brazilian economy have helped to prevent upward pressure on prices, but which may not necessarily be permanent features, particularly if domestic demand strengthens.

21. A look at the components of the consumer price index (INPC) reveals that, cumulatively from January 1995, price developments for different components of consumer prices have been significantly different. For example, while communication tariffs and housing rents and prices for housing materials experienced cumulative increases of over 100 percent, clothing prices experienced almost no inflation (Figure 1.5). Since January 1997, prices for clothing, and food and drinks—both basic consumer goods—experienced a cumulative inflation of -1 percent and 7 percent, respectively, while the rest of the index grew by over 14 percent (Figure 1.6). The virtually flat clothing prices may reflect a long-run trend toward permanently lower profit margins in that sector, although clothing prices cannot keep flat forever. At the same time, prices in sectors that have experienced above average price increases over the last few years (e.g., communication, housing) may not continue to rise as fast in the future. In contrast, the drop in food prices is probably an exceptional event related to the bumper harvest this year, and therefore of temporary nature. Also, energy price developments can be considered an exceptional factor, as consumer prices and tariffs were adjusted in several partial steps following the floating of the Real, and the passthrough in this sector was not complete until mid-year. Higher energy prices and tariffs are expected to put upward pressure on consumer prices during the next few months.

Figure 1.5.Brazil: Components of Consumer Price Inflation, 1995-99

(cumulative increase in consumer prices (INPC) from January 1995, in percent)

Source: IBGE.

Figure 1.6.Brazil: Components of Consumer Price Inflation, 1997-99

(cumulative increase from January 1997, INPC index)

Source: IBGE.

Economic policies and the transmission of inflation

22. The low passthrough has led to speculation that it may simply reflect a temporary delay in price adjustments that will sooner or later show up in consumer prices. Independently, questions have been raised how the government may gear its policies toward containing inflationary pressures in the future.

23. This section, which presents some very preliminary results from work that is still in progress, offers two main conclusions. First, while there may have been some delays in adjusting prices fully to the exchange rate developments that have taken place, in Brazil, the passthrough from devaluation to inflation can be expected to take place over a period of no more than six-nine months, which is a lot faster than in most industrialized economies. Also, given a history of swift adjustments of prices during the country’s long history of high and variable inflations that ended with the Real Plan in 1994, and notwithstanding the fact that formal indexation mechanisms no longer exist, it seems unlikely that there has been a significant delay in adjusting prices. The exception here were some administered prices, as discussed above. Second, key variables in determining inflation are broad monetary aggregates (M2)and wages. Containing the growth of these two variables can be expected to have considerable benefits for containing inflationary pressures.

24. To analyze the speed and determinants of price adjustments, an unrestricted vector autoregression (VAR) model was used, that includes consumer price inflation (as measured by the IPCA that was just selected as the main index for the government’s inflation targeting framework), money, wages, the exchange rate, the government’s primary deficit, and the unemployment rate, which is included as an indicator of slack in the economy.13 Three different monetary aggregates (base money, Ml, and M2) were examined to see which provides the best explanation for the inflation transmission mechanism.

25. VAR modeling does not require any a priori assumptions concerning the exogeneity of variables, and provides a convenient way to summarize the empirical channels in which different variables affect each other. The estimated VAR can be used to evaluate the strength of the estimated empirical relationships based on variance decomposition and impulse response functions. Variance decompositions show the portion of the forecast error variance for each variable that is attributable to its own innovations and to shocks with respect to other system variables. Impulse response functions show the estimated response of each variable to a one standard deviation impulse in one of the innovations. These dynamic multipliers show how new information in one of the variables causes changes in the forecast of another variable.

26. Before implementing the VAR, the various time series that are to be included are analized to determine their stationarity through standard augmented Dickey-Fuller (ADF) and Phillips-Perron (PP) unit root tests.14 A set of Granger causality tests is presented to explore likely causalities among the variables included in the VAR.

27. The results of the unit root tests are presented in Table 1.3 for both levels and first order differences of the different variables.15 The tests were carried out for the overall period for which data were collected,16 and for the period from the beginning of the Real Plan (July 1994). For the period since July 1994, the null hypothesis of a unit root cannot be rejected at the 5 percent level of significance under either the ADF or the PP only for the exchange rate and the unemployment rate. For the exchange rate, however, the null hypothesis was rejected for the overall period for which data were collected (from January 1994), whereas for the unemployment rate the unit root (i.e., nonstationarity) vanished when using first order differences. At the 1 percent level, the unit root tests for the other series seem to suggest stationarity in levels under at least one of the two tests. In general, the unit root tests on first order differences yielded better results, the exception being wages and the price index when the test was carried out for the overall period for which data were collected (January 1993 and January 1991, respectively).

Table 1.3.Brazil: Unit Root Tests 1/
LevelsFirst Order Differences
Augmented

Dickey-Fuller

(ADF)
Phillips- Perron

(PP)
Augmented

Dickey-Fuller

(ADF)
Phillips- Perron

(PP)
Price index (IPCA)
Since July 1994-3.2 **-6.0 ***-6.7 ***-7.6 ***
Total avail, series-2.2-3.9 ***-1.5-1.7
Base money
Since July 1994-2.7 *-3.5 ***-6.4 ***-8.4 ***
Total avail, series-4.2 ***-7.4 ***-2.8 *-3.9 ***
M2
Since July 1994-3.1 **-5.4 ***-18.4 ***-16.6 ***
Total avail, series-4.8 ***-10.9 ***-3.6 ***-3.0 **
Exchange rate
Since July 19940.50.4-11.0 ***-8.2 ***
Total avail, series-3.8 ***-6.3 ***-4.6 ***-3.5 **
Wages
Since July 1994-4.6 ***-4.9 ***-15.3 ***-15.4 ***
Total avail, series-3.4 **-6.7 ***-1.8-1.9
Unemployment
Since July 1994-1.1-0.7-4.5 ***-4.2 ***
Total avail, series-2.0-2.4-6.9 ***-5.5 ***
Primary deficit
Since July 1994-4.2 ***-7.1 ***-8.0 ***-19.1 ***
Total avail, series-5.3 ***-9.1 ***-10.2 ***-27.8 ***
Source: Fund staff estimates.

28. The correlation matrices in Table 1.4 show strong contemporaneous correlation between the levels of most of the variables. Surprisingly, for the period from July 1994, unemployment levels were positively correlated with the other variables. For the first-order differences, correlations are less strong, and unemployment now shows the expected negative correlation with the other variables. For the levels, wages seem to have the strongest correlation with the other variables included. For the first-order differences, the contemporaneous correlation between prices and M2 remains quite strong. In general, the first-order differences show a much stronger correlation for the period from January 1994 than for the period from July 1994. For example, for the period from January 1994, changes in wages and changes in prices have a correlation coefficient of 0.86, whereas for the period from July 1994, the correlation coefficient between these two variables is only 0.03. This may point to the significant deindexation of the economy that has taken place under the Real Plan.

Table 1.4.Brazil: Correlation Matrices 1/
Price levelBase moneyM1M2Exchange rateSalariesUnemploymentPrimary balance
From July 1994
Levels
Price index (IPCA)1.000
Base money0.8791.000
M10.9260.9791.000
M20.9630.9470.9481.000
Exchange rate0.7370.7810.7710.7871.000
Wages0.9920.9020.9430.9670.7111.000
Unemployment0.7570.7860.7990.7990.8130.7491.000
Primary deficit0.4020.3080.3400.3680.2010.3980.1471.000
First order difference
Price index (IPCA)1.000
Base money0.2761.000
M10.1680.9011.000
M20.7050.5240.4141.000
Exchange rate0.267-0.064-0.0790.1111.000
Wages0.0250.2530.2350.179-0.3231.000
Unemployment-0.234-0.422-0.360-0.3730.097-0.2611.000
Primary deficit0.034-0.0340.1030.038-0.0520.090-0.0251.000
From January 1994
Levels
Price index (IPCA)1.000
Base money0.9641.000
M10.9560.9921.000
M20.9890.9860.9801.000
Exchange rate0.9540.9380.9330.9591.000
Wages0.9890.9710.9740.9930.9551.000
Unemployment0.3190.4520.5020.4140.4460.4041.000
Primary deficit0.1130.1500.1810.1460.0790.1540.1431.000
First order difference
Price index (IPCA)1.000
Base money0.7481.000
M10.7640.9311.000
M20.9330.8310.8521.000
Exchange rate0.8480.6640.7100.8711.000
Wages0.8580.7680.8160.9350.8321.000
Unemployment0.018-0.177-0.1060.0200.1650.0861.000
Primary deficit-0.017-0.0610.027-0.028-0.056-0.015-0.0241.000
Source: Fund staff estimates.

The results of the Granger causality tests on the levels of the variables in question are reported in Table 1.5.17 The results are somewhat inconclusive as they suggest strong Granger causalities between M2, wages, and prices running in both directions.18 Still, some interesting tentative conclusions for the relationship between prices and the other variables can be drawn from these results. First, the F-statistics for Granger causality running from M2 to prices are significantly higher than for the reverse hypothesis, for all lags, although the test results do not allow to reject the possibility that price levels may Granger-cause M2. Second, the hypothesis that wages do not Granger-cause prices can be rejected at the 1 percent level of significance at all lags, whereas the hypothesis that prices do not Granger cause wages can only be rejected at the 1 percent level of significance for lags 1, 5, and 6.19 Third, Granger causality clearly seems to run from the exchange rate to prices, and not in the other direction. Fourth, there is little evidence of any Granger causality between prices (or any of the other variables) and unemployment or the primary deficit in any direction.20 The Granger causality tests also yield some other interesting results. For example, there seems to be strong evidence that Granger causality runs from wages to M2, but not the other way around.

Table 1.5.Brazil: Granger Causality Tests on Levels, 1994-99 1/
Lag length in number of months
123456
Impact of money
Base money ==> price level (IPCA)0.36.6 ***12.3 ***9.8 ***9.0 ***7.2 ***
Base money ==> wage levels3.1 *8.5 ***4.9 ***2.8 **1.91.1
Base money ==> exchange rate2.63.1 *0.32.9 **2.00.9
Base money ==> primary deficit1.81.01.42.6 **3.7 ***3.5 ***
M2==> price level (IPCA)4.4 **63.5 ***64.3 ***52.3 ***38.6 ***27.1 ***
M2 ==> wage levels6.6 **1.71.50.70.41.8
M2 ==> exchange rate4.8 **2.4 *0.11.81.61.8
M2 ==> primary deficit1.71.03.0 **3.1 **3.1 **2.3 *
Impact of price levels
Price level (IPCA) ==> base money0.54.1 **1.91.21.13.6 ***
Price level (IPCA) ==> M28.8 ***3.4 **5.0 ***7.3 ***3.9 ***1.3
Price level (IPCA) ==> wage levels12.7 ***0.12.3 *1.93.8 ***3.4 ***
Price level (IPCA) ==> exchange rate20.5 ***1.81.72.1 *1.22.5 **
Price level (IPCA) ==> primary deficit0.71.01.12.5 *2.1 *1.9
Impact of exchange rates
Exchange rate ==> base money8.1 ***2.30.90.91.60.7
Exchange rate ==> M24.7 **2.6 *1.83.8 ***2.2 *1.3
Exchange rate ==> price level (IPCA)18.0 ***14.4 ***10.3 ***16.0 ***17.3 ***16.1 ***
Exchange rate ==> wage levels0.52.43.7 **3.8 ***7.3 ***2.6 **
Impact of wages
Wage levels ==> base money5.7 **3.6 **4.0 **4.5 ***3.5 ***2.0 *
Wage levels ==> M213.7 ***5.2 ***4.4 ***4.1 ***1.51.1
Wage levels ==> price level (IPCA)40.8 ***86.9 ***57.8 ***71.5 ***56.1 ***58.8 ***
Wage levels ==> exchange rate1.55.7 ***0.80.91.71.4
Wage levels ==> unemployment0.72.8 *1.71.31.24.1 ***
Wage levels ==> primary deficit1.20.50.80.72.8 **2.6 **
Impact of primary deficit
Primary deficit ==> base money1.01.61.21.71.51.6
Primary deficit ==> M21.61.20.81.80.60.7
Primary deficit ==> price level (IPCA)0.00.40.30.30.30.2
Primary deficit ==> wage levels0.60.30.30.60.60.7
Primary deficit ==> exchange rate1.31.20.71.00.60.4
Primary deficit ==> unemployment0.03.9 **4.2 ***3.4 **2.6 **2.2 **
Impact of unemployment
Unemployment ==> price level (IPCA)5.3 ***2.7 *1.91.61.31.2
Unemployment ==> wage levels8.6 ***5.7 ***3.3 **2.3 *1.91.7
Unemployment ==> exchange rate14.9 ***3.3 **4.6 ***2.01.70.6
Unemployment ==> primary deficit1.50.70.70.50.30.2
Source: Fund staff estimates.

29. Given the strong contemporaneous correlation between the levels of different variables, we also carried out Granger causality tests on first-order differences of the variables in question, which are reported in Table 1.6. The results are slightly more conclusive, although there continues to be strong bi-directional causality between some of the variables. The main results can be summarized as follows. First, there is clear evidence of Granger causality running from changes inM2 to price changes, except for the two- and three-month lags, where causality seems to run in both directions. Similarly to the previous exercise on the levels of these variables, the F-statistics for causality running from M2 changes to price changes is much higher than for the reverse causality.21 Second, there is strong evidence that wage changes Granger-cause price changes, except for the four-month lag, where causality seems to run in both directions. Third, Granger causality clearly runs from exchange rate changes to price changes, but not in the other direction. Fourth, there is little evidence of any Granger causality between changes in unemployment or changes in the primary deficit and the other variables included, with the exception of some Granger causality running from changes in unemployment to changes in wages for lags 1–4 and from changes in wages to changes in unemployment for lags 5–6. The results also provide strong evidence of Granger causality running from changes in wages to changes in M2.

Table 1.6.Brazil: Granger Causality Tests on First-Order Differences, 1994-99 1/
Lag length in number of months
123456
Impact of changes in base money
D base money ==> D price level (IPCA)9.1 ***14.2 ***9.3 ***6.7 ***7.2 ***1.5
D base money ==> D wage levels8.4 ***7.4 ***5.1 ***3.6 **1.61.2
D base money ==> D exchange rate0.00.50.30.30.40.5
D base money ==> D primary deficit0.20.71.51.61.12.0 *
D M2==> D price level (IPCA)120.8 ***107.7 ***67.5 ***62.4 ***35.2 ***2.5 **
D M2 ==> D wage levels0.11.01.61.32.4 *2.0 *
D M2 ==> D exchange rate0.93.3 **0.51.71.91.7
D M2 ==> D primary deficit0.31.51.71.61.61.8
Impact of changes in price levels
D price level (IPCA) ==> D base money35.9 ***10.3 ***5.1 ***3.5 **3.5 ***3.2 ***
D price level (IPCA) ==> D M20.86.0 ***4.9 ***3.4 **1.81.3
D price level (IPCA) ==> D wage levels2.73.9 **2.5 *4.4 ***2.9 **2.9 **
D price level (IPCA) ==> D exchange rate0.05.0 **2.21.31.01.6
D price level (IPCA) ==> D primary deficit0.70.41.11.41.21.3
Impact of changes in exchange rates
D exchange rate ==> D base money11.8 ***3.9 **2.5 *2.1 *1.12.3 *
D exchange rate ==> D M23.1 *1.11.11.30.91.0
D exchange rate ==> D price level (IPCA)30.7 ***18.7 ***17.3 ***14.2 ***11.1 ***3.0 **
D exchange rate ==> D wage levels0.90.20.14.2 ***1.61.8
Impact of changes in wages
D wage levels ==> D base money28.4 ***13.6 ***8.6 ***6.7 ***1.84.3 ***
D wage levels ==> D M219.2 ***13.9 ***11.2 ***3.3 **1.30.9
D wage levels ==> D price level (IPCA)179.0 ***87.8 ***92.6 ***68.4 ***74.2 ***59.7 ***
D wage levels ==> D exchange rate7,9 ***8.3 ***2.4 *2.9 **1.41.6
D wage levels ==> D unemployment0.81.01.71.24.3 ***2.7 **
D wage levels ==> D primary deficit0.10.60.52.1 *1.81.8
Impact of changes in primary deficit
D primary deficit ==> D base money2.71.71.71.31.81.7
D primary deficit ==> D M20.10.10.30.40.80.7
D primary deficit ==> D price level (IPCA)0.10.10.00.00.10.1
D primary deficit ==> D wage levels0.00.00.10.10.20.4
D primary deficit ==> D exchange rate0.40.20.20.60.50.5
D primary deficit ==> D unemployment2.8 *1.61.20.80.71.0
Impact of changes in unemployment
D unemployment ==> D price level (IPCA)4.1 **1.61.41.01.81.5
D unemployment ==> D wage levels5.8 **4.0 **3.3 **2.7 **1.81.9
D unemployment ==> D exchange rate0.10.00.20.30.50.5
D unemployment ==> D primary deficit0.01.21.10.70.50.6
Source: Fund staff estimates.

30. Next, the VAR was estimated using the causal ordering money, prices, wages, and the exchange rate.22 The rationale for this particular ordering is that this seems consistent both with intuition and the general results from the Granger causality tests, where monetary aggregate affect price levels, which in turn feed through to wages and the exchange rate; the same holds for changes in these variables.23

31. The results from the variance decomposition exercises of the VAR estimates are presented in Tables 1.7 (levels) and 1.8 (first order differences). The variance decomposition exercise decomposes variation in an endogenous variable into the component shocks to the other endogenous variables in the VAR. The variance decomposition gives information about the relative importance of each random innovation to the variables in the VAR.

Table 1.7.Brazil: Variance Decomposition on Levels of Variables, 1994-99 1/(in percent of total variance)
Variance

of

Variable
Variance

period

(months)
Explained byExplained by
M2PricesWagesFX RateBase MoneyPricesWagesFX Rate
M2/1100000100000
Base29514095032
Money384214188192
65583716812173
945173716219144
Prices11990099100
255301041239418
35317246619669
6308584315766
92476183176812
Wages1481510180820
235165082900
326174056890
6163782316783
9158733519705
FX rate1160083110385
2181180152677
3173278164971
6149116618121456
91314126118151453
Source: Fund staff estimates.
Table 1.8.Brazil: Variance Decomposition on First-Order Differences, 1994-99 1/(in percent of total variance)
Variance

of

Variable
Variance

period

(months)
Explained byExplained by
M2PricesWagesFX

Rate
Base

Money
PricesWagesFX

Rate
M2/1100000100000
Base289191880111
Money3668241794152
65220261634313
95121272604343
Prices101000039700
275121031321606
36313195812746
64525273612784
94425273611784
Wages166232093880
2551133083880
3481734075870
6442233165881
9432233165881
FX rate1211375401977
22221066203464
32161953314551
619142047334845
919142048334845
Source: Fund staff estimates.

32. There are four results from the variance decomposition exercises that are of particular interest for this study. First, money and wages explain much of the innovation in prices, except for the shortest (one month) horizon. Second, M2 seems to be a far better variable to explain variations in prices and wages than base money. Third, the contribution of the exchange rate in explaining the variations in any of the other variables is fairly small at any of the forecasting horizon that were explored (one-nine months). While, to some extent, this may reflect the fact that most of the data in the sample come from the period when the Real was closely managed under the pegged exchange rate regime that prevailed until January this year, the results did not change qualitatively when the variance decomposition exercise was carried out for different periods (e.g., restricting the data only to the period of the crawling peg exchange rate regime or using the entire period that also includes data from before and after the crawling peg).24 Fourth, the results are fairly robust across specifications (levels versus first order differences); for the first order differences, money (M2) is the dominant variable in explaining innovations in prices over any forecasting horizon (except for the one-month horizon).

33. The results from the impulse response functions of the VAR estimates are presented in Figures 1.71.10. The impulse response functions employ the same causal ordering as the variance decomposition exercise. An impulse response function traces the effect of a one standard deviation shock to one of the innovations on current and future values of the endogenous variables. A shock to the i-th variable directly affects the i-th variable, and is also transmitted to all of the endogenous variables through the dynamic structure of the VAR. While variance decomposition decomposes variation in an endogenous variable into the component shocks to the endogenous variables in the VAR, impulse response functions trace the effects of a shock to an endogenous variable on the variables in the VAR.

Figure 1.7.Brazil: Response to One Standard Deviation innovations (± 2 S.E.): Variable Levels (Using M2), 1994-99 1/

1/ All variables are log-transformed from the original data.

Figure 1.8.Brazil: Response to One Standard Deviation Innovations (± 2 S.E.): Variable Levels (Using Base Money). 1994-99 1/

1/ All variables are log-transformed from the original data.

Figure 1.9.Brazil: Response to One Standard Deviation Innovations (± 2 S.E.): First-Order Differences (Using M2), 1994-99 1/

1/ All variables are first-order differences of the log-transformed original data.

Figure 1.10.Brazil: Response to One Standard Deviation Innovations (± 2 S.E.): First-Order Differences (Using Base Money), 1994-99 1/

1/ All variables are first-order differences of the log-transformed original data.

34. There are five main results from the impulse response functions that are of particular interest for this study. First, most of the innovations in different variables work themselves through the system fairly rapidly: after nine months, the effects on other variables of innovations in any given variable are insignificantly different from zero for most pairs of variables considered. Second, the results are qualitatively fairly similar, regardless of whether the VAR was speficied in levels or first-order differences. Third, innovations to M2 have a much stronger impact on prices than innovations in base money. Prices respond the innovations in M2 with a two-three-month lag; for innovations to changes in M2, the impact on changes in the price level is strongest after two months, and the effect drops off thereafter over a nine-month time horizon. Fourth, innovations to wages have a significant impact on prices, although with a slightly longer lag than money. Finally, the effects on prices of innovations to the exchange rate are fairly small. While, to some extent, this can be attributed to the fact that much of the data comes from a period when the Real was closely managed under the pegged exchange rate regime, the impulse responses for the exchange rate did not change qualitatively when estimating the model for different time periods.

D. Conclusions

35. This chapter has argued that four main elements have contributed to the positive inflation outcome since the Real was left to float in mid-January this year: (i) an economy that does not rely on imported inputs to the extent of other economies; (ii) tight monetary and fiscal policies; (iii) sluggish consumption that has resulted in profit margin reductions at different stages of the production process; and (iv) a number of exceptional factors, like a good harvest and the slow adjustment in some administered prices (such as electricity tariffs), that have dampened upward pressure on prices in different subsectors of the economy (e.g., food and energy). To the extent that the latter two elements may be expected to be temporary, underlying inflation may be expected to be higher in the future.

36. With the relatively high real interest rates that have been maintained since the floating of the Real and the moderate growth of monetary aggregates, monetary policy has remained fairly tight. Similarly, notwithstanding a number of setbacks, the government has continued to deliver on the fiscal performance it promised in its adjustment program from late 1998. There is ample evidence that profit margins have indeed been squeezed in various industries, with wholesale price increases outrunning consumer price increases by a significant margin. This may be attributed, at least initially, to producers adopting a “wait-and-see” attitude in light of what was perceived as an overshooting of the exchange rate. The main goal seems to have been to preserve market share, in light of an already fairly depressed domestic demand. This was helped by the fact that only few industries have an extensive reliance on imported inputs; many industries experience the effect of exchange rate changes mainly indirectly, by buying inputs from other producers who rely on imported inputs. This, to some extent, may have slowed down the passthrough from the depreciation of the Real, even though, in general, a shock to any of the variables that were analyzed (i.e., money, wages, prices, the exchange rate, (primary) fiscal balances, and economic slack indicators) can be expected to work itself through the system during six-nine months, much faster than in most industrialized economies.

37. The VAR exercise has suggested that the impact of the exchange rate on prices may be significant, but that its magnitude is relatively small compared to other variables, particularly broad money (M2) and wages. In fact, M2 and wages help to explain much of the innovations in prices for all time periods that were tried. Changes in M2 have their strongest influence on price changes with a two-four-month lag; changes in wages have their strongest impact on price changes with a lag of about three-five months. In addition, wages seem to have a strong effect on M2. While the small effect of exchange rates on prices may, to some extent, reflect the fact that exchange rates were closely managed under the pegged exchange rate regime that prevailed until January 1999, this result was surprisingly robust across different time periods (e.g., when restricting the data only to the period of the pegged exchange rate regime or when including data from before and after as well).

38. What policy recommendations would come out of this analysis? Given the preliminary nature of this study, a very tentative conclusion would be that containing wage pressures and maintaining a tight fiscal and monetary policy stance that would help to contain the growth in broad monetary aggregates can be expected to have a strong impact on mitigating inflationary pressures, and this already over a fairly short time horizon given the lags involved. The study also suggests that, given the relatively small effect of exchange rates on other variables (money, wages, and prices), a further weakening of the exchange rate per se should not necessarily generate concerns about renewed inflation. Given, however, the large impact of monetary aggregates and wages on the exchange rate, this conclusion would clearly hinge on the reasons behind a weakening of the exchange rate.

APPENDIX Data and Data Sources

1. This appendix presents an overview of the variables used in estimating the VAR. The VAR exercise uses log-transformations of all variables, except for the primary deficit of the federal government, which was used untransformed. The following variables were used.

2. Base money, M1, and M2, where base money and Ml are measured as the daily average of a given month, and M2 reflects the end-of-month data. Data are available from the central bank; data for all variables were collected from January 1994 onward.

3. Prices, as measured by the consumer price index (IPCA) published by the Brazilian Statistical Institute (IBGE). The IPCA provides information on prices in 11 metropolitan regions for families earning between 1 and 40 minimum wages. The IPCA is the main index for the Brazilian Central Bank’s inflation targeting framework; data were collected from January 1991.

4. Wages, as measured by the nominal salary index for São Paulo, published by FIESP. Data were collected from January 1993.

5. The Real/U.S. dollar exchange rate, specified as the daily average exchange rates prevailing over a given month, as reported by the Brazilian central bank. Using end-period exchange rates yielded similar results. Data for both variables were collected from January 1994.

6. The unemployment rate, as measured by DIEESE for total (open and hidden) unemployment in São Paulo; data were collected from January 1992. Using the official unemployment rate reported by IBGE yielded fairly similar results; data for IBGE’s unemployment rate were collected also from January 1992.

7. The primary deficit of the central government, as reported by the federal treasury. Data were available from January 1991, although there were some structural breaks concerning coverage of the data. While in most months Brazil generated primary surpluses, it also experienced primary deficits; given that this variable may be either positive (for a deficit) or negative (for a surplus) it was not possible to log-transform this variable.

Prepared by Gerd Schwartz.

Passthrough is defined here as the accumulated consumer price inflation relative to the cumulative depreciation of the Real vis-à-vis the U.S. dollar. If not noted otherwise, the IPCA of Brazil’s Statistical Institute (IBGE) is used to measure consumer prices.

Eduardo Borensztein and José de Gregorio (1999) “Devaluation and Inflation after Currency Crisis,” (draft), February.

Mark Stone (1999), “The Low Rates of Inflation in Post-Crisis East Asia” (draft).

See Thiago Rabelo Pereira and Alexandre Carvalho (1999), “O Impacto da Desvalorização Cambial Sobre os Custos Industriais: Um Estimativa dos Efeitos Cumulativos dentro da Cadeia Industrial,” (The Impact of Exchange Rate Devaluation on Industrial Costs: Estimates of the Cumulative Effects Across Industries). Boletim de Politica Industrial, No.7, April.

To the extent that the Brazilian economy may be more open now than it was in 1995, the price impact of the devaluation would be accordingly higher.

Andrew Haldane (1999), Presentation at the BCB/MAE seminar on inflation targeting, Rio de Janeiro, May 3–5.

Confederação Nacional da Indústria (1999), Sondagem Industrial Suplemento Especial (Industry Survey—Special Supplement); January/March.

See, for example, “Importados enfrentam ‘filtro’ da desvalorização” (Imports encounter “filter” from devaluation) in Gazeta Mercantil of May 21, 1999.

A more detailed comparison could be carried out by isolating the price changes in tradable goods in both indices. This was not done here.

See Boletim de Politica Industrial, No.7, April 1999. The estimates of changes in profit margins are derived by comparing actual price increases in different sectors with the “full passthrough” estimates obtained on the basis of the 1995 I-O matrix.

International prices for coffee per ton have continued to drop from around US$120 in December 1998 to about US$94 in July 1999.

A detailed specification of these variables is provided in the appendix. The analytical approach presented here has benefited (and borrowed) significantly from other research carried out in the Fund, including, in particular, Kevin Ross (1998), “Post Stabilization Inflation Dynamics in Slovenia,” IMF Working Paper WP/98/27 (March); Benedict Clements (1998), “Fiscal Policy and Other Determinants of Inflation in Bolivia” (draft); and Ramana Ramaswamy and Torsten Sløk (1998), “The Real Effects of Monetary Policy in the European Union: What are the Differences?” IMF Staff Papers, Vol. 45, No.2 (June), pp. 374–396. However, errors in this chapter should not be attributed to these sources.

The ADF test augments the Dickey-Fuller (DF) test by including higher order lag terms (in addition to AR(1) processes) so as to capture autocorrelation in the error terms. The PP test applies a nonparametric correction in estimating the variance of the error term.

The various tests and results for Ml are excluded from the presentation here as they yielded results similar to those for base money.

The availability of data differs depending on the variable in question; see the appendix of this chapter for an overview of the data. For all variables, data were available from at least January 1994; for some variables, data were available from 1991 onward.

Causalities for which no strong economic rationale could be established are excluded from the presentation; overwhelmingly these were also not significant, and none was significant at the 1 percent level.

For base money, there appears to be strong evidence of Granger causality running from money to prices, but not the other way around.

Note, also, the much higher values for the F-statistics for Granger causalities running from wages to prices.

The positive contemporaneous relation between unemployment rates and prices that was shown in the correlation matrices seems counter-intuitive and is probably spurious. In general, these results should not be taken to suggest that fiscal or economic slack variables are unimportant for the inflationary process. They are certainly important, and the inclusion of different variables that were not explored in this study may have yielded different results.

In contrast, for base money, Granger causality between money and prices are now less clear than under the tests on levels.

Given the lack of causality between unemployment and the primary deficit and all other variables, these two variables were excluded from the different VARs. Including these two variables did not change the qualitative results.

Various other orderings were tried out for comparison. In general, separating prices and monetary aggregates in the variable ordering tended to weaken somewhat the statistical relationship.

Initially, on July 1, 1994, the Real was introduced with a floating exchange rate with respect to the U.S. dollar. The float of the Real was continued for the first three months after its introduction. Between October 1994 and February 1995, the Real remained in a narrow band around R$0,85 per U.S. dollar. In March, 1995, the authorities announced that the exchange rate would be left to fluctuate within a band of R$0.88 to R$0.93 to the U.S. dollar for an unspecified period. Periodic adjustments to the band (and the bandwidth) in which the Real was managed were carried out thereafter, until the Real was left to float on January 15, 1999. Also see SM/95/299 for some background on the Real plan.

Other Resources Citing This Publication