Botswana: Selected Issues and Statistical Appendix

This Selected Issues paper and Statistical Appendix on Botswana underlies that diamond reserves are not adequate to generate enough permanent revenue to support the current level of expenditure. Despite strong overall growth, in Botswana, a pattern of dependence on diamond revenue and high unemployment persists. Botswana, as a typical small open economy, is closely linked to a large neighboring economy. This linkage means Botswana’s monetary and exchange policies must consider the external economic environment, particularly the pula’s exchange rate against the rand.

Abstract

This Selected Issues paper and Statistical Appendix on Botswana underlies that diamond reserves are not adequate to generate enough permanent revenue to support the current level of expenditure. Despite strong overall growth, in Botswana, a pattern of dependence on diamond revenue and high unemployment persists. Botswana, as a typical small open economy, is closely linked to a large neighboring economy. This linkage means Botswana’s monetary and exchange policies must consider the external economic environment, particularly the pula’s exchange rate against the rand.

III. A Note on Inflation

A. Introduction15

1. Botswana’s annual inflation rate has averaged 7–9 percent since the mid-1990s, despite a period of historically low inflation among its main trading partners, South Africa, the United States, and the United Kingdom (Table III.1). In 2005, year-on-year inflation exceeded single digits, reaching a 13-year high of 14 percent by April 2006, triggered by the May 2005 pula devaluation, high international oil prices, and one-off factors.

2. This note explores Botswana’s recent inflation developments, focusing on monetary aggregates and the exchange rate, the most powerful long-run determinants of Botswana’s inflation rate. Using a unit-root econometric technique, we estimate a simple reduced-form inflation equation with quarterly 1993–2005 data. According to the results, South African inflation has the greatest influence on price movements in Botswana. The paper also finds that, while both exchange rate depreciation against the South African rand and monetary expansion are inflationary, the impact of exchange rate depreciations is much larger.

Table III.1.

Annual Inflation, 1993–2005

(Percent; period average)

article image
Sources: Botswana authorities; and International Financial Statistics.

B. Inflation Trends in Botswana, 1993–2005

3. After the high inflation of 1992–93, Botswana experienced gradual disinflation towards 6 percent until recently, despite some inflationary pressures, including high international oil prices owing to the 1999 cuts in OPEC crude oil production and a new value-added tax (VAT) in 2002 (Figure III.1). While inflation for fuel and power items peaked at 25½ percent in mid-2000, the 10 percent VAT pushed overall inflation to 12 percent by mid-2003.

Figure III.1.
Figure III.1.

Inflation, Jan. 1993–May 2006

(12-month percentage change)

Citation: IMF Staff Country Reports 2007, 228; 10.5089/9781451806465.002.A003

Sources: Botswana authorities; and Fund staff estimates.

4. More recently, inflation has trended downward, falling to 6 percent in early 2005. However, two factors—a 12 percent devaluation of the national currency against the basket and the May 2005 introduction of a crawling peg—once again pushed inflation beyond the authorities’ targeted 4–7 percent range. On May 29, 2005, the Botswana government devalued the pula by 12 percent against the basket (comprising the South African rand and the SDR) and adopted a forward-looking crawling peg exchange rate arrangement, in which the pula’s exchange rate to the basket would adjust continuously rather than in steps.16 It also increased the margin between the buy and sell rates for currencies quoted by the Bank of Botswana (from ±0.125 percent to ±0.5 percent around the central rate). By year-end 2005, inflation had risen to about 11 percent. Additional inflationary pressures, including higher administered fuel and transport prices and the reintroduction of secondary education fees, pushed inflation to about 14 percent in the first half of 2006.17

5. The underlying relationship between inflation and growth in monetary aggregates has been weak, possibly owing to time lags in the transmission of monetary policy and noisy data. In the 1993–98 disinflation period, for instance, monetary aggregates generally increased. The M1 12-month growth rate peaked at about 50 percent in November 1998, and broad money (M3) grew more than 30 percent at the end of 1998 (Figure III.2). In contrast, when inflation picked up in 1999–2000, the growth in monetary aggregates quickly shrank. Since 2001, the monetary aggregates have been highly volatile, moving independently of the price level.

Figure III.2.
Figure III.2.

Inflation and Monetary Growth, Jan. 1993–May 2006

(12-month percentage change)

Citation: IMF Staff Country Reports 2007, 228; 10.5089/9781451806465.002.A003

Sources: Botswana authorities; and International Financial Statistics.

6. Given that Botswana is a small open economy, one might expect inflation to be influenced by exchange rate movements. Botswana’s current consumption basket is 24 percent domestic tradables, 47 percent imported tradables, and 29 percent nontradables. As shown by the data, exchange rate depreciations against the South African rand (Botswana imports most of its household goods and food from South Africa) seem to have played an essential role in explaining changes in the country’s domestic prices, especially since 1995 (Figure III.3 and Table III.2).

Figure III.3.
Figure III.3.

Inflation and Exchange Rate against the Rand: Jan. 1993–May 2006

(12-month percentage change)

Citation: IMF Staff Country Reports 2007, 228; 10.5089/9781451806465.002.A003

Sources: Botswana authorities; and International Financial Statistics.1 The exchange rate is shown in pula per foreign currency terms.
Table III.2.

Correlation between Inflation and Exchange Rate Depreciation

article image
Source: Author estimations.

7. By contrast, only recently have changes in the exchange rate against major industrial country currencies, such as the U.S. dollar and the euro, affected Botswana’s inflation rate (Figure III.4). The relationship, after being fairly weak, appears to have strengthened in the past three years, according to a simple correlation of inflation and the pula’s depreciation against the U.S. dollar (about 0.7) and against the euro (0.5) (Table III.2). Similarly, after appearing to move independently of domestic inflation, the pula-SDR exchange rate, which comprises part of the currency basket of Botswana, has demonstrated a certain correlation with inflation in recent years (Figure III.5 and Table III.2).

Figure III.4.
Figure III.4.

Inflation and Exchange Rates against the US$ and Euro: Jan. 1993–May 2006

(12-month percentage change)

Citation: IMF Staff Country Reports 2007, 228; 10.5089/9781451806465.002.A003

Sources: Botswana authorities; and International Financial Statistics.1 The exchange rate is shown in pula per foreign currency terms.2 The ECM was replaced with euro at a 1:1 rate in January 1999.
Figure III.5.
Figure III.5.

Inflation and Exchange Rate against the SDR: Jan. 1993–May 2006

(12-month percentage change)

Citation: IMF Staff Country Reports 2007, 228; 10.5089/9781451806465.002.A003

Sources: Botswana authorities; and International Financial Statistics1 The exchange rate is shown in pula per foreign currency terms.

C. Empirical Model and Data

Vector error-correction (VEC) model

8. To explore the relationship between money, the exchange rate, real GDP, interest rates, and prices, we estimate a simple inflation model. The price level (CPI) is, in general, a weighted average of tradable prices (CPIT) and nontradable prices (CPIN):18

lnCPI=λ1nCPIT+(1λ)lnCPIN,(1)

where λ is the weight of tradables in the consumption basket. For Botswana, the weight amounts to 70.8 percent of total household expenditure. As specified by the law of one price of tradables, the price level of tradables is determined by the world price in foreign currency terms (CPI*) and the exchange rate (ER), defined in units of foreign currency per one unit of domestic currency:19

lnCPIT=lnCPI*lnER.(2)

9. The price level of nontradables is supposed to be determined by disequilibrium between the money supply and the demand in the domestic money market. The money supply (MS) (which is, in principle, a policy variable), and an increase in Ms would inflate domestic prices. On the other hand, the increased demand for money (Md) mitigates inflationary pressures, and is assumed to be a function of real GDP growth (RGDP) and the nominal interest rate (INTR). While higher interest rates make holding money more costly and reduce money demand, real economic expansion increases the transaction demand for money, leading to disinflation.20 Thus, inflation of nontradables can be written as:

lnCPIN=(MS,Md(RGDP,INTR)).(3)

10. We estimate the following reduced-form equation for inflation, derived from equations (1) to (3):

CPI=f(CPI*+,ER,MS+,Md(RGDP,INTR))+.(4)

Following earlier studies (e.g., Johansen, 1995), the above relationship is specified by a vector error-correction (VEC) model with two lags:21

ΔlnCPIt=β0+kβ1kΔlnCPItk+kβ2kΔlnMtks+kβ3kΔlnERtk+kβ4kΔlnRGDPtk+kβ5kΔINTRtk+kβ6kΔlnCPItk*+β7ECMlnCPIt1+εt,(5)

where Δln is the first difference in logs of the variables, and ECM ln CPIt is an error correction term associated with disequilibrium from the long-term equilibrium in the money market:

ECMlnCPIt=α1lnCPItα2lnMtsα3lnERtα4lnRGDPt.α5lnINTRtα6lnCPIt*(6)

Data

11. The analysis uses quarterly data for the period 1993–2005. Before this period, Botswana’s high inflation makes it difficult to maintain the common structure assumption of price behavior. The sample period includes the latest quarter for which applicable data are available. The baseline model includes the following seven variables in the system: general consumer prices (CPI), broad money (M2), the exchange rates against the rand (Rand/Pula) and the U.S. dollar (US$/Pula), quarterly real GDP (RGDP), the 88-day notice deposit rate (Term deposit rate), and South Africa’s price level (CPI of SA).

12. These variables were selected over other specifications on the basis of statistical reliability and theoretical consistency (see ANNEX III.II for details). Notably, money supply, which is represented by M2 (currency in circulation plus current and time deposits), while partly interest bearing, does not include Bank of Botswana Certificates (BoBCs). The two foreign exchange variables—the rand per pula and the U.S. dollar per pula—were selected because these two foreign currencies play such an important role in domestic trade patterns. Although a large share of Botswana’s imports come from South Africa,22 some of them (e.g., oil) are denominated in U.S. dollars. Exports prices do not directly affect the CPI basket, but the international trade prices of diamonds denominated in U.S. dollars, of which Botswana is the world’s largest producer, may possibly influence the economy to the large extent.

13. For a proxy variable referred to as the level of world prices, we use the consumer price index of South Africa, given that half of all goods and materials traded in Botswana are imported, with three-quarters of them coming from South Africa. For GDP data, we use the three-period moving average of quarterly GDP.23 The average deposit interest rate—though viewed as problematic owing to current negative real interest rates—is the only interest rate variable that has enough time series data and variation over time.

D. Estimation Results

Unrestricted model

14. The augmented Dickey-Fuller unit root tests indicate that almost all variables are nonstationary in levels but stationary in their first differences (Table III.3).24 The data also suggest that the trace test statistic can reject the null hypothesis of no cointegration in favor of one cointegrating vector at the 5 percent significance level. (Table III.4).25

Table III.3.

Unit Root Tests

article image

10 percent significance level;

5 percent significance level;

1 percent significance level

Table III.4.

Cointegration Tests

article image

5 percent significance level.

15. Given the above, the cointegrating equation is estimated as follows:

lnCPI=0.142ln(0.063)M20.356ln(0.072)Rand/Pula+0.029ln(0.040)US$/Pula0.476(0.179)lnRGDP+0.035(0.005)TermDepositRate+1.562(0.129)lnCPIofSA+0.246(7)

The estimated dynamic error-correction inflation equation is presented in ANNEX III.I. Note that the cointegrating vector, which makes a set of variables in the system stationary by a suitable choice of its initial distribution, can be interpreted as a long-term equilibrium relationship among monetary aggregates, the exchange rate, real GDP, interest rate, and the price level. All but one sign, that for the exchange rate against the U.S. dollar, are as expected in theory and statistically significant.26 The overall fit of the equation is satisfactory.27 According to the conventional χ2 test, the hypothesis of all the coefficients being zero can easily be rejected. The test statistics is significantly large at 6129.7 (see Table III.7).

16. The null hypothesis of no autocorrelation in the residual (of Equation (5)) cannot be rejected at the conventional significance level. The Lagrange-multiplier (LM) test statistic is estimated at 50.62, for which the p-value is 0.409. In terms of stability, the largest modulus of other potential cointegrating vectors is at 0.5873, meaning no modulus is close to the unit root, thus indicating that the estimated cointegrating equation is stable. Finally, the normality test based on the skewness statistics cannot reject the null of normality at the conventional 5 percent significance level, though it can be rejected at the 10 percent significance level.

This finding, though somewhat weak, confirms that the disturbance for the inflation equation is normally distributed.

Restricted model

17. Imposing some linear restrictions can generate a more concise result. Equations (5) and (6) have the following two restrictions:

β2k=β4kandα2=α4fork.(8)
β6k=β3kandα6=α3fork.(9)

Empirically, equation (7) may not support these restrictions, because the hypothesis for these linear restrictions can be rejected by the standard Wald tests; however, they have a theoretical basis; equation (8) holds under the quantity theory of money with stability of money velocity, and equation (9) implies that imported tradables prices are exactly calculated by the world price and the exchange rate (following equation (2)).

18. With the restrictions, the estimated cointegrating equation is:28

lnCPI=0.485(lnM2lnRGDP)(0.064)+0.067TermDepositRate(0.014)+0.582(lnCPIofSAlnRand/Pula)(0.141)+1.515.(10)

The significance of the coefficients is better than in the unrestricted model. The necessary estimation assumptions are also satisfied, and the dynamic error-correction model is reasonable, though it contains some margins of error, as shown in ANNEX III.I and III.II.

19. Equations (7) and (10) suggest Botswana’s prices over the long term behave as follows

  • Both statistically and economically, the strongest determinant of price movements in Botswana is South African inflation. The estimated elasticity of South Africa’s inflation relative to Botswana’s is 1.6, suggesting that prices between the two countries gets transmitted to a considerable degree. An elasticity well over one surely includes some secondary, indirect effects on domestic tradables and nontradables.29 Notably, this result appears consistent with the fact that Botswana’s inflation—which averages 9.1 percent over the sample period—has been one-and-a-half times as high as South Africa’s average rate—6.6 percent. By contrast, under the restricted model, the estimated elasticity of inflation imported from South Africa is 0.58, somewhat comparable to the share of imported tradables in the basket (i.e., about 50 percent).

  • A depreciation of the pula against South Africa’s rand also has a significant inflationary impact (the estimated coefficient is – 0.36). In the econometric model, the exchange rate is defined in foreign currency units per pula in logarithm. Similar to South Africa’s inflation and interest rate effects, this depreciation effect is relatively powerful in a statistical sense.

  • By contrast, the depreciation against the U.S. dollar has a statistically insignificant and economically limited impact on inflation, perhaps because only export prices are denominated in U.S. dollars, making such depreciations less relevant to Botswana’s domestic inflation.

  • Monetary expansion has a small but significant inflationary impact. At 0.14, the coefficient indicates a statistically significant but weak relationship between monetary aggregates and prices in the unrestricted equation; by contrast, the change in the ratio of money to GDP exerted a stronger effect in the restricted model, though it was still below that of the South African CPI, adjusted for changes in the rand-pula exchange rate.

  • As expected, inflation decreases with real money demand arising from economic expansion: 1 percent GDP growth would cause a half percent of disinflation in equilibrium.

  • Finally, higher price levels are associated with higher interest rates. The equation shows that a 1 percent increase in (term deposit) interest rates is accompanied by a 3.5 percent of inflation.30, 31

E. Conclusion

20. The analysis explores the long-term behavior of inflation in Botswana. Not surprisingly, changes in South Africa’s consumer prices largely determine inflation. Changes in the exchange rate against the South African rand also affect inflation. These findings support the view that Botswana, as a typical small open economy, is closely linked to a large neighboring economy. This linkage means Botswana’s monetary and exchange policies must consider the external economic environment, particularly the pula’s exchange rate against the rand. The inflation objective must also be consistent with South Africa’s monetary stance.

21. The empirical result also sheds light on the need for prudent monetary policy to keep inflation low. The estimated effect of monetary expansion on inflation looks very small and only marginally different from zero. Nonetheless, it is statistically significant, thus indicating that money growth is modestly inflationary. M2 is the only monetary aggregate variable that produced an estimation result consistent with theory; the estimation using other aggregates implied a negative association between money growth and inflation.

22. In addition to money supply, the interest rate adjustment seems effective in monetary policy transmission to a certain, but not large, extent. The evidence suggests that, in equilibrium, 1 percent of inflation would require 0.3 percent higher (term deposit) interest rates.

ANNEX III.I: Estimated Error-Correction Inflation Equation

23. The error-correction inflation equations associated with the cointegrating equations (7) and (10) are presented in Table III.5. The short-term movements in domestic real GDP and consumer prices in South Africa have significant coefficients in the unrestricted model. Inflation decelerates with real growth and tends to be stimulated by South African inflation. Exchange rate depreciations against the rand also appear to fuel inflation, though the effect is statistically ambiguous. However, the restricted model indicates that the import inflation through the pula-rand exchange rate has a significant effect. The impact of monetary supply and interest rates are also subject to a wide margin of error, a result that may reflect the limited number of sample observations as well as the relatively strong ability of the cointegrating vector to capture the relationship among the endogenous variables.

Table III.5.

Error-Correction Inflation Equation 1

article image

10 percent significance level;

5 percent significance level;

1 percent significance level.

24. To complement the above estimations, Granger causality tests were performed based on the system of seven error-correction equations (Table III.6). In the unrestricted model, South Africa’s inflation Granger-causes inflation in Botswana, and vice versa,32 revealing that the two countries’ price developments are interdependent. It is also finds that, while real growth Granger-causes disinflation, monetary supply does not cause inflation in the short run. The Granger-causality between inflation and exchange rate depreciation remains inconclusive, though the evidence supports the view that Botswanan inflation causes the pula to depreciate against the rand (the p-value is 0.197).

Table III.6.

Granger Causality 1

article image

10 percent significance level;

The degree of freedom is all equal to one.

5 percent significance level.

25. The impulse response function depicts the effect of one-standard error originating from a variable in the system on other endogenous variables thorough the dynamic structure (Figures III.5 and III.6). Despite the difficulty of assessing the results, owing to generally large standard errors (particularly in the unrestricted model), a positive inflation shock in South Africa would likely raise Botswana’s inflation rate for about 12 quarters (three years). A real shock would likely result in some disinflation. Although the impact of monetary and exchange rate shocks are difficult to assess because of their unstable transition paths, the projected response indicates that monetary shocks have a limited short-term impact on prices.

Figure III.6.
Figure III.6.
Figure III.6.

Impulse Response Function, Unrestricted Model

Citation: IMF Staff Country Reports 2007, 228; 10.5089/9781451806465.002.A003

Figure III.7.
Figure III.7.

Impulse Response Function, Restricted Model

Citation: IMF Staff Country Reports 2007, 228; 10.5089/9781451806465.002.A003

ANNEX III.II: Alternative Specifications for Cointegrating Equations

26. This Annex examines the selection of the variables used for the baseline estimation. Compared with other estimation results using different variables, the baseline model is the most significant in a statistical sense and the most consistent with economic theory. The analytical framework is the same as in the main text; the choice of the number of cointegrating equations follows the Johansen’s trace test technique, and the stationarity of almost all variables is confirmed in Table III.3.

27. First, in the baseline model, the price level is measured by a general consumer price index (CPI). An alternative measurement may be the price level, excluding exogenous factors, such as food and energy prices. The second column in Table III.7 shows the estimated cointegrating equation for CPI, excluding fuel and food items. 33 The result looks similar to the baseline model’s, but the overall fit is less favorable, though the disturbance meets the normality assumption more favorably. Meanwhile, when taking CPI for only tradables or nontradables, the estimated equations hardly satisfy the normality assumption. Notably, however, the coefficients between these two models are close, suggesting that tradable prices affect nontradables prices, and vice versa.

Table III.7.

Alternative Cointegrating Equations for Different CPIs 1

article image

10 percent significance level;

5 percent significance level;

1 percent significance level.

28. Second, the baseline model defines money supply as a broader measurement, M2, which is defined as currency in circulation plus current and time deposits. This measure covers interest-bearing accounts but does not include the Bank of Botswana Certificates (BoBCs), which are one of Botswana’s major saving instruments. 34 With narrower monetary aggregates (M0 or M1) included in the model, the impact of monetary expansion turned negative (Table IV.8), a result that defies theory. On the other hand, with broader monetary aggregates (M3 or M4), the coefficients of money supply are also negative and insignificant; however, this finding may be reasonable because the BoBCs, which are part of M3 and M4, are used to mop up excess liquidity and contain the inflationary impact of monetary growth. Nonetheless, all these alternative models violate the normality assumption.

29. For the exchange rate, the baseline adopted two bilateral exchange rates against the South Africa’s rand (Rand/Pula) and the U.S. dollar (US$/Pula). While Botswana imports most goods from a neighboring country, South Africa, the U.S. dollar figures prominently in the international trade markets, including diamonds. An obvious alternative is the euro (Euro/Pula) or the SDR (SDR/Pula), but the real growth and monetary growth results were theoretically inconsistent, and a stability concern remained. The estimation using the rand and SDR, both of which make up Botswana’s currency basket, is close to the baseline model’s, though money does not have a significant coefficient. The nominal effective exchange rate (NEER) is another way to incorporate all relevant foreign exchange rates in the model. The VEC estimation with the NEER index, which is employed from the standard IMF Effective Exchange Rate Facility database, generates a cointegrating vector that differs dramatically from that of the other specifications and is inconsistent with theory (the 11th column).

30. Finally, there are two interest rate data other than the 88-day notice deposit rate: the Bank Rate and primary lending rate, both of which are closely related to each other. The estimated cointegrating equation with the Bank Rate is more or less similar to the baseline, though the coefficients of real GDP and money supply are insignificant (though they have correct sign).35 Notably, the model indicates that 1 percent of inflation would be associated with higher interest rate increases than in the baseline estimation. This finding makes sense because the Bank Rate has varied little over the past decade, and thus tends to be used to respond relatively actively to any given price movement.36 However, this inelasticity calls into question the empirical validity of the unit-root technique used in the analysis.

Table III.8.

Alternative Cointegrating Equations with Different Independent Variables 12

article image

The dependent variable is the logarithsm of CPI.

* 10 percent significance level; ** 5 percent significance level; *** 1 percent significance level.

One of the estimated cointegrating equations is shown.

Table 1.

Botswana: GDP by Type of Expenditure at Current Prices, 2000/01–2004/05 1/

article image
Source: Central Statistics Office.

National accounts year beginning July 1.

GDP minus consumption

Table 2.

Botswana: GDP by Type of Expenditure at Constant 1993/94 Prices, 2000/01–2004/05 1/

article image
Source: Central Statistics Office.

National accounts year beginning July 1.

Table 3.

Botswana: GDP by Type of Economic Activity at Current Prices, 2000/01–2004/05 1/

article image
Source: Central Statistics Office.

National accounts year beginning July 1.

Table 4.

Botswana: GDP by Type of Economic Activity at Constant 1993/94 Prices, 2000/01–2004/05 1/

article image
Source: Central Statistics Office.

National accounts year beginning July 1.

Table 5.

Botswana: Beef Sales, 2000–2005 1/

article image
Source: Ministry of Agriculture.

Calendar year.

Table 6.

Botswana: Mineral Production and Value, 2000–2005

article image
Source: Central Statistics Office and Department of Mines.

Estimated value of production.

Table 7.

Botswana: Agricultural Producer Prices, 1998/99–2004/05 1/

(Pula per ton)

article image
Source: Botswana Agricultural Marketing Board.

Crop year beginning April 1.

Table 8.

Botswana: Formal Sector Employment, 2000/01–2004/05 1/

(Number of employees, unless otherwise indicated)

article image
Sources: Central Statistics Office.

Data for September of first year listed.

The increase in central government employment reflects the absorption of community, junior, and secondary school staff, the salaries of which were already being paid by the central government.

Data for March 2005.

Table 9.

Botswana: Statutory Minimum Hourly Wage Rates, 2000–2006

article image
Source: Central Statistics Office.

100 thebe = 1 pula.

Table 10.

Botswana: Average Monthly Cash Earnings by Sector, 1999–2003 1/

article image
Source: Central Statistics Office.

Based on the formal sector employment survey in March each year except for 1999, in which the survey was conducted in September.