DETERMINANTS OF INFLATION IN BOTSWANA:

Transcription

1 Determinants Asian-African of Journal Inflation of Economics in Botswana: and Econometrics, Vol. 14, No. 2, 2014: DETERMINANTS OF INFLATION IN BOTSWANA: M. Raboloko *, G. R. Motlaleng ** and M. Bakwena ** ABSTRACT The paper identifies the determinants of inflation and examines the stability of the inflation function in Botswana. To achieve this, a Vector Autoregression (VAR) model is applied to quarterly data from 2003 to 2011 in order to estimate an inflation function for Botswana. The results show that the inflation rate for Botswana is determined by its lagged values, specifically the first lag of prices. The findings are consistent with those in the literature (e.g. Kuijis (1998) and Abidemi and Malik (2010) found results similar to those reported here for the case of Nigeria). Specifically, the results from these previous results showed that the first lag of inflation had a positive effect on inflation in Nigeria. The results of the current paper also suggest that the inflation function for Botswana is stable all innovations tend to converge to zero through the impulse response function analysis. I. INTRODUCTION An understanding of the nature of inflation is a precondition for effective policy action. The paper is intended to analyse the determinants of inflation in Botswana from 2003 to Despite the remarkable rates of growth of the economy over the years, inflation remains one of the major challenges for government. In fact, inflation was consistently above the mediumterm objective range of 3-6 percent in It increased from 7.4 percent in December 2010 to 9.2 percent in December 2011 (Bank of Botswana, 2012). The historic rates of inflation also suggest that inflation has always been a challenge for Botswana. According to Jefferis (2008), concerns about rising inflation in the late 1980s and early 1990s led to a gradual tightening of monetary policy. It led to the maintenance of higher real interest rates than had previously been the case. Despite success in lowering inflation during the 1990s, inflation remains a problem. It remained above international levels and in particular higher than the average inflation rate of the country s trading partners. It is claimed that these inflation concerns were part of the reason for the change in exchange rate policy around in year It is important to note that over the years, the government of Botswana has embarked on a number of strategic policies to address inflation. Jefferis argues that since 2002 much more effort has been geared towards ensuring a transparent monetary policy that is able to influence inflation expectations. Since then the Bank of Botswana has been publishing * Associate Researcher, Botswana Institute for Development Policy Analysis, Gaborone, Botswana ** Department of Economics, University of Botswana, P/Bag 0022, Gaborone, Botswana, motlaleg@mopipi.ub.bw; thokweng@mopipi.ub.bw

2 196 M. Raboloko, G. R. Motlaleng and M. Bakwena yearly monetary policy statements (MPS). The MPS indicates a moderately narrow range of inflation rates as a target to be achieved during the years ahead. Notwithstanding these initiatives, and high growth rates of the economy over the past four decades, inflation remains a concern for Botswana. Despite the fact that Botswana is among the few countries able to record a single digit inflation rate, for a long time the inflation rate has been beyond the range set by the Bank of Botswana (BoB). According to the Southern African Customs Union (SACU) Inflation Report (2012), amongst the SACU member states, Botswana recorded the highest inflation rate of 7.3 percent in July after Swaziland (8.7 percent). In fact, Botswana has recorded average annual inflation rate of between 6.9 and 12.7 percent between 2007 and 2011 rates that are relatively high compared to those prevailing in developed countries. The question is what are the underlying factors behind this rising inflation despite the concerted efforts by the monetary authorities to keep the price levels low and stable? The most prominent research on inflation in Botswana has been carried out by among others, Atta et al (1996) and Chakrabarti (2006). For exchange rate, monetary and fiscal policies to be effective in controlling inflation, adequate knowledge of the factors that determine inflation in any given economy is crucial (Atta et al, 1996). One of the major contributions of this paper is to use an up to date data set for the period 2003 and 2011 to find inflation determinants in Botswana. By using the vector autoregression (VAR) model, the study differs from previous studies on Botswana. Actually, the previous studies Atta et al (1996) and Chakrabarti (2006) used the autoregressive distributed lag (ADL) model in order to model the determinants of inflation. Through the use of a VAR model and hence a different method of analysis this paper adds to the literature on inflation in Botswana. A VAR provides a systematic way to capture rich dynamics in multiple time series such as inflation. Through the use of a VAR model, the response of inflation to various exogenous impulses is discovered. As economic conditions continue to evolve, continuous study of the nature of inflation seems imperative to adequately address it as an economic problem. This paper aims to contribute to the understanding of the nature of inflation in Botswana and to inform policy intervention with regards to keeping inflation low and stable The paper proceeds as follows: Section II gives the background of inflation in Botswana with emphasis on the trend of inflation with monetary policy variables. Section III focuses on the review of previous studies. Section IV outlines the empirical model. The findings of the study are presented in Section V. Conclusions of the paper are given in Section VI. II. INFLATION IN BOTSWANA AND MONETARY VARIABLES 2.1. Inflation in Botswana Botswana s annual inflation rate has averaged 7 9 percent since the mid 1990s (Bank of Botswana, 2012). This was despite a period of historically low inflation among its main trading partners: the Republic of South Africa (RSA), the United States (US), and the United Kingdom (UK). In 2005, year on year inflation exceeded single digits. This was triggered by the May 2005 pula devaluation and high international oil prices (IMF, 2007). Figure 2.1 below plots the

3 Determinants of Inflation in Botswana: annual inflation rate for the period The figure shows that average annual inflation reached the highest level of 12.7 percent in 2008, it the declined after this period, though it was still high compared to the inflation rate for RSA. Figure 2.1 Source: World Bank Various Issues Chakrabarti (2006) asserts that the level of inflation rate in Botswana is still high relative to those of her trading partners. For instance, in February 2006 the inflation rate in RSA was 3.9 percent while for Botswana it was 13.1 percent. Botswana s inflation has for a long time now, been way above the rates prevailing in her major trading partners. Jefferies (2008) argues that the openness of the economy has subjected it to a great deal of external influence, particularly from RSA. Atta et al (1996) argues that although not all inflation in Botswana is imported, in practice domestic prices tend to move fairly closely in line with import prices Inflation and Monetary Policy Variables in Botswana Inflation and Real GDP Growth Economic theory states that inflation and output growth move in opposite directions. The theory holds true for Botswana- Figure 2.2 shows that since 2003, inflation and real GDP growth have been moving in opposite direction most of the time. Inflation and Exchange Rate A high rate of inflation leads to loss of the country s competitiveness in international trade. International competitiveness of a country s products is important for successful export-led growth and development. Low competitiveness in international trade has adverse effects for Botswana because of a small domestic market. It discourages both domestic and foreign

4 198 M. Raboloko, G. R. Motlaleng and M. Bakwena Figure 2.2 Source: World Bank Various Issues investment. The exchange rate policy has been implemented such that there is a modest downward crawl of the nominal effective exchange rate. This was because the inflation rate is above the anticipated inflation of trading partner countries. The reason for this was to maintain external stability and competitiveness. Inflation and Money Supply (M2) Growth The Quantity Theory of Money states that an increase in money supply has a positive and direct impact on inflation. The trend of broad money supply (M2) growth rate and inflation (Figure 2.3) reveals that both variables were moving in the same direction most of the time. However, broad money supply (M2) showed a considerable expansion in Figure 2.3 Source: World Bank Various Issues

5 Determinants of Inflation in Botswana: III. LITERATURE REVIEW 3.2. Review of Previous Studies There is considerable amount of empirical studies done to identify the possible determinants of inflation in developing countries. For instance, Abidemi and Malik (2010), Khan and Schimmelpfenning (2005), Mwase (2006), Dlamini et al. (2001), Kuijs (1998), Lim and Papi (1997), and Chhibber et al. (1990) concurs that money supply is the main cause of inflation in most developing countries. Most of these studies also found that foreign prices, exchange rate and interest rates greatly influence domestic prices. In a nutshell it can be argued that monetary factors such as money supply, interest rates, foreign prices and exchange rates are the prime determinants of inflation. However, a budget surplus/deficit is the only fiscal factor found to have a significant impact on inflation (Abidemi and Malik (2010)). A budget deficit increases inflation while a surplus is believed to be anti-inflationary. In Botswana, earlier studies like that of Huda (1987), Leith (1991) and Ncube (1992) concentrated on price and nominal exchange rate and foreign price relationships. They established a high explanatory power for the SA price index (used as a proxy for foreign prices) suggesting strongly the hypothesis that inflation in Botswana is essentially imported. These studies also found a small impact of nominal exchange rate changes on prices in Botswana, except in the case of Ncube (1992) where it was found to be insignificant. The estimation methodology used in both studies appears to have overestimated the speed at which domestic prices adjust to exchange rate changes. The authors failed to recognize that there is a possibility that disequilibrium might persist for a considerable period during the adjustment process. In this case, the price equation is specified with a lagged adjustment mechanism whereby the change in the Botswana price index is a function of the difference between the equilibrium relationship and the actual level in the previous period. The major limitation with the above studies is that they used more traditional econometric approaches. i.e. the specific to general approach. The regressions were undertaken without testing for stationarity of the variables. Testing for stationarity is imperative because of the fact that the use of non-stationary variables in a regression brings about the issue spurious regression results.masale (1993) made an attempt to solve this problem by testing for stationarity and employing cointegration analysis in order to identify both short and long run relationships in the equation. The findings were that Botswana and South African CPIs, as well as the Rand\Pula exchange rate, were indeed found to be nonstationary. However, the cointegration tests between the relative prices and the exchange rate did not confirm the existence of a longrun relationship. Estimation of a short-term relationship between the variables using first differences and error correction factor also led to the rejection of the PPP model. Atta et al. (1996), building upon the model by Masale (1993), used monthly data from 1975 to 1993, thereby extending the sample size since the latter had used quarterly observations. The approach employed a general to specific method using an unrestricted autoregressive distributed lag model. Stationarity testing, cointegration analysis and error correction modeling were employed to distinguish clearly between short- run and long-run price relationships. The model employed by Atta et al. (1996) incorporated a wider range of variables applicable to the

6 200 M. Raboloko, G. R. Motlaleng and M. Bakwena Botswana situation. These are real income, nominal narrow money supply, nominal interest rate (proxied by the savings deposit rate), the nominal exchange rates of the Pula against the Rand, Zimbabwe dollar, USA dollar and foreign prices as represented by the South African, Zimbabwean and American CPIs. The results showed a strong pass-through cointegrated relationship of 92% between SA prices and Botswana prices. The changes in the Zimbabwean prices were also significant, with 14% after a one-year lag, but the influence of USA price changes, at 72%, was surprising. This was explained in terms of the larger proportion feeding through the RSA price changes. Other variables also proved to be highly significant and had the expected signs. Changes in real income, money supply, interest rate and Rand/Pula nominal exchange rate explained -28%, 2%, 1% and 3%, respectively, of changes in Botswana prices. The error correction term was negative 2%, and highly significant. It indicated that 2% of the past months disequilibria feeds back into the current Botswana price changes. Chakrabarti (2006) developed and estimated an econometric model using annual data over the period 1980/81 to 2003/04 for Botswana. The author used stationarity testing, cointegration analysis and error correction modeling. In the study, the CPI was treated as dependent on, RSA prices, US prices, the bilateral exchange rate, government revenue and expenditure, credit to the private sector and the private sector real GDP. The results suggest that growth of credit to the private sector, government expenditure and revenue play a significant role in determining inflation. However, the statistical performance of the fiscal variable was not uniformly good. It is evident from the foregoing that a number of studies have been done on inflation in Botswana. It is revealed that Botswana s inflation is affected by both domestic and external factors. South African inflation, exchange rates and balance of payments have been associated with the imported inflation. Money supply, interest rates and budget surplus/deficit are considered as domestic factors that determine inflation in Botswana. There have been a lot of policy changes and economic conditions have evolved since The period of the current paper (2003 to 2011) is expected to offer new insights into understanding the determinants of inflation in Botswana. Additionally, this paper intends to draw comparative lessons from empirical studies done on inflation in Botswana and other developing countries. VI. METHODOLOGY In order to model the real causes of inflation in Botswana, the paper follows the approach by Atta et al. (1996). They specified their model as shown in equation (4.1) below: P = f(p f, E, M, R, y) (4.1) Where P is Botswana CPI, y is real output, M is money supply, R is the nominal rate of interest. E is the exchange rate and P f is the foreign price. All the coefficients on the variables in the model are expected to have positive signs except for real income. The foreign price and exchange rate terms in equation (4.1) above were related to specific countries in the empirical model. The authors were interested in three countries: RSA, Zimbabwe and the US. South Africa and Zimbabwe were the two most important sources of Botswana s imports and markets for non-traditional exports. The US dollar trade was used as a proxy for

7 Determinants of Inflation in Botswana: trade with the rest of the world. Nominal exchange rates are redefined as foreign currency per pula. Consumer price indeces were used for both Botswana and foreign prices. Money supply (M) was proxied by the narrow definition of money (M1) while the savings deposit rate was used as a proxy for R and lastly non-mining gross domestic product was used as a proxy for y. They used monthly data collected on all of the variables except for non-mining GDP which was interpolated from annual data Model Specification In view of the above model by Atta et al (1996), the inflation function is specified in logs as follows with minor modification on the explanatory variables: ln P ln ln Y ln M ln R ln SAE ln SAP µ (4.2) t 0 1 t 2 t 3 t 4 t 5 t t The model includes the CPI (SAP t ) for RSA and Rand-pula nominal exchange rate (SAE t ). The reason for this is that RSA is the major trading partner of Botswana in terms of imports. Money supply (M) is proxied by Botswana s broad definition of money (M2). All the coefficients on the variables in the model are expected to have a positive sign, except for real income, which is expected to be negative. In a monetarist model real income would be expected a priori to have a negative sign Vector Autoregression (VAR) Estimation The VAR model is used for analysis of the interrelation of time series and the dynamic impacts of random disturbances on the system of variables. A VAR is a k equation, k variable linear model in which each variable is in turn explained by its own lagged values, plus current and past values of the remaining k-1 variables. VAR models have had much success in forecasting the economy. In many cases, VARs perform better in forecasting than more sophisticated macroeconomic models. Sims (1980) argued that VARs provide a coherent and credible approach to data description, forecasting, structural inference and policy analysis. These reasons provide the justification for using VAR technique for analysis in this paper. Following Enders (1995), first order VAR that is VAR (1) model is presented as follows: (4.3)

8 202 M. Raboloko, G. R. Motlaleng and M. Bakwena The system can be represented in matrix format as follows: A i - a (k * k) matrix of parameters k - the lag length of VAR µ t - a (k * 1) the column vector of random stochastic values. An unrestricted VAR (UVAR) model involving up to t - k th lag of X t can be written as: (4.4) k t i t 1 t i 1 X A X µ where µ t ~ NID (0, 2 ) (4.5) Where X t is (k * 1) and each of A i is a (k * k) matrix of parameters. Sims (1980) used this type of VAR to estimate dynamic relationships among jointly endogenous variables without imposing strong priori restrictions. A critical element in the specification of VAR model is the determination of the lag length of the VAR. The importance of lag length determination is demonstrated by Braun and Mittnik (1993) who show that estimates of a VAR whose lag length differs from the true lag length are inconsistent. The impulse response functions and variance decompositions derived from the estimated VAR are also incosistent. Lütkepohl (1993) indicates that overfitting (selecting a higher order lag length than the true lag length) causes an increase in the mean-square forecast errors of the VAR. Underfitting the lag length often generates autocorrelated errors. Increasing number of parameters leads to loss of degrees of freedom which can result in significance of inefficient estimators. The optimal lag length is determined by allowing a different lag length for each equation at each time. The Akaike Information Criteria (AIC) and Schwarz - Bayesian Information Criteria (SBC) are the two most common methods for estimating the optimal lag length for a VAR. The model with the lowest value between the AIC and SBC values is chosen. 5. EMPIRICAL RESULTS 5.1. Unit Root Test for Stationarity The stationarity test is based on the Augmented Dickey Fuller test. The ADF test is the commonly used unit root test due to its simplicity. In our paper, the ADF test is used with the Phillips- Perron (PP) test to verify the ADF test results. The ADF and PP tests estimate the same equation, they only differ in terms of the test statistic used therefore the PP and ADF results are expected to be consistent. The ADF approach used here tests the null hypothesis that a series has a unit

9 Determinants of Inflation in Botswana: root (non-stationary) against the alternative of stationarity. A summary of the stationarity results is presented in Table 5.1 and they show that in the case of variables with intercept only, the null hypothesis of non-stationarity cannot be rejected for Botswana CPI (P t ), GDP (Y t ), money supply (M t ), nominal interest rate (R t ), Rand/Pula nominal exchange rate (SAE t ) and South African CPI (SAP t ). That is, these variables have a unit root. However, these variables become stationary after first differencing. i.e. they are integrated of order one I (1). The conclusion is that Botswana CPI, GDP, money supply (M2), nominal interest rate, Rand/Pula nominal exchange rate and South African CPI are non-stationary at levels. Table 5.1 Results of ADF Unit Root Test for Stationary Variable Unit Unit Unit Roots Roots Roots ADF with ADF with ADF without Intercept Intercept and Intercept and Only Trend Trend Levels First difference Levels First difference Levels First difference lnp *** *** *** lny *** *** *** lnm *** *** *** lnr *** ** *** lnsae *** *** *** lnsap *** *** *** Critical ADF with ADF with ADF without Values Intercept Intercept and Intercept and only Trend Trend 1% % % ***indicates stationarity at 1%, **indicates stationarity at 5% and *indicates stationarity at 10% The ADF test with intercept and trend shows that money supply (M2), nominal interest rate, Rand/Pula nominal exchange rate and South African CPI are non-stationary at levels. There is a presence of a unit root for these variables. However, Botswana CPI and GDP become stationary when a trend is included. The results of the PP test are consistent with the ADF test results Cointegration Tests Since the variables are integrated of order one, i.e. I(1), there is a need to determine if they are cointegrated. i.e., if there exists a long run relationship between the dependent and independent variables. If there is no cointegrating relationship between the dependent variable and any of the independent variables, then the next approach would be to run regression on the levels at which the variables are stationary. The Johansen and Juselius (1990) trace and maximum Eigen value test were conducted in order to test for cointegration. The Johansen Maximum Likelihood

10 204 M. Raboloko, G. R. Motlaleng and M. Bakwena procedure is an ideal approach to estimate when there are more than one cointegrating vectors or variables. To determine the number of cointegrating equations, the trace test is used. The variables tested for cointegration are Botswana CPI (P t ), GDP (Y t ), money supply (M t ), nominal rate of interest (R t ), Rand/Pula nominal exchange rate (SAE t ) and South Africa CPI (SAP t ). Table 5.2 Unrestricted Cointegration Rank Test (Trace) Hypothesised Eigen Value Trace Statistic 0.05 Critical Value Prob. Number of Cointegrating Equation (s) r = r r r r r E Trace test indicates no cointegration at the 0.05 level. The trace test results in Table 5.2 above fails to reject the null hypothesis of zero cointegrating equations at the 95% critical value. The trace test indicates zero cointegrating equations at the 0.05 level. This implies that there is no long run relationship between Botswana CPI, GDP, money supply, nominal rate of interest, Rand/Pula nominal exchange rate and South Africa CPI. The variables do not eventually move together in the long run. This means that we cannot estimate a vector error correction (VEC) model as the variables are not cointegrated. In this case, we can only estimate an unrestricted VAR model. The test was repeated using the Maximum Eigen Value test and the results from the exercise are similar to the ones presented here-i.e., the conclusion that the variables are not cointegrated remains the same Unrestricted Vector Autoregression (UVAR) Model Since none of the variables are cointegerated, the next step is to estimate a VAR model. The VAR model allows us to analyse the interrelation of time series and the dynamic impacts of random disturbances on the system of variables. In a VAR model each variable has an equation explaining its evolution based on its own lags and the lags of the other model variables. Parameter estimates and summary statistics for the VAR model are given in Table 5.3. Equation 1 shows that Botswana inflation depends only on its past values. The results show that the rate of change of Botswana inflation (P t ) is explained by only its past values (P t ( 1)). The parameter estimate for Botswana inflation is statistically significant at 5 per cent level and assumes the expected positive sign. The implication is that inflation in the current year is influenced by inflation in the past year. The results are in contrast with Atta et al (1996) who found Botswana inflation to be influenced by real income, money supply, Rand/ Pula nominal exchange rate, interest rate and largely RSA inflation. The method of analysis used by Atta et al. differs from the method used in the current analysis. They used an unrestricted

11 Determinants of Inflation in Botswana: Table 5.3 VAR Model Estimation Results Equation 1 Equation 2 Equation 3 Equation 4 Equation 5 Equation 6 (ln P t ) (ln Y t ) (ln M t ) (ln R t ) (ln SAE t ) (ln SAP t ) (lnp t ( 1)) [ ] [ ] [ ] [ ] [ ] [ ] (lny t ( 1)) [ ] [ ] [ ] [ ] [ ] [ ] (lnm t ( 1)) [ ] [ ] [ ] [ ] [ ] [ ] (lnr t ( 1)) [ ] [ ] [ ] [ ] [ ] [ ] (lnsae t ( 1)) [ ] [ ] [ ] [ ] [ ] [ ] (lnsap t ( 1)) [ ] [ ] [ ] [ ] [ ] [ ] C [ ] [ ] [ ] [ ] [ ] [ ] R-squared Adj. R-squared Sum sq residuals S.E. equation F-statistic Log likelihood Akaike AIC Schwarz SC Mean dependent S.D.dependent autoregressive distributed lag model while the current analysis used an unrestricted vector autoregression model. The difference in the method of analysis and the time period might explain the differing results. The results are however, consistent with Kuijis (1998) who applied a VAR model to analyse the determinants of the price level, exchange rate and output in Nigeria. The author found a direct relationship between the first and third lag of prices on the price level. Abidemi and Malik (2010) also found similar results on their study on inflation and its determinants in Nigeria. They used a VAR model for their analysis and found the first lag of inflation to have a positive effect on inflation. These two studies applied a similar method of analysis (VAR) to the one adopted in the current analysis and the results are consistent. The coefficients for GDP, nominal interest rate, and RSA CPI have expected signs but are statistically insignificant at 5 percent level. The positive, but insignificant relationship between GDP and inflation suggest that economic growth does not necessarily lead to reduced inflation. It can lead to increased inflation due to the existence of monopolistic or oligolopolistic elements in the economy. Nominal interest rate appears to have no significant role in the inflation function for Botswana. This is in line with the theory that in most developing countries, interest rates

12 206 M. Raboloko, G. R. Motlaleng and M. Bakwena tend to be inoperative. According to Dlamini et al (2001), it is difficult to ascertain whether interest rates affect prices or not in developing countries. This is because of less developed money market and the fact that interest rates do not necessarily reflect money market conditions but are institutionally pegged. The insignificance of the aforementioned variables implies that Botswana inflation is insensitive to first lag of these variables. The parameter coefficients for money supply and nominal exchange rate have unexpected signs and statistically insignificant. The regression results show that the goodness of fit is satisfactory. The of 0.55 implies that 55 percent of the variation in Botswana inflation is explained by the regression during the period 2003 to The overall model performs well as explained by the F statistic of Impulse Response Analysis Impulse response function (IRF) analysis is used to uncover the dynamic relationship between macroeconomic variables within the VAR model (Mitchell, 2000). Impulse responses measure the time profile of the effect of a shock or impulse on the (expected) future values of a variable. The impulse response analysis is used to answer the question of how the modeled variables respond to any innovations or shocks in the economy. The impulse responses generally tend to converge to zero, implying that the model depicts stability over time. The IRF helps to discover the effects of unanticipated shocks on the stability of Botswana inflation. Figure 5.1 presents the impulse responses of Botswana inflation to one standard deviation shock in one of the innovations of GDP, money supply, interest rate, exchange rate and RSA CPI. The results of the IRF show that past value of Botswana inflation is significant in explaining the current value of Botswana inflation. Thus, Botswana inflation is affected contemporaneously by the shocks to its past. The result shows that a shock to Botswana inflation has a positive impact on itself from the first to the sixth quarter. Own shock to Botswana inflation has a negative impact on itself from the seventh quarter and tend to converge to zero, implying that the model depicted stability over time. The response of Botswana inflation to one innovation appears to be greater in GDP than other exogenous variables. A shock to GDP appears to have a positive effect on Botswana inflation on the first and second quarters. After the second quarter, a shock to GDP has a negative effect on Botswana inflation and tends to converge to zero. Thus, GDP shock has an immediate impact on Botswana inflation. The implication is that GDP innovations play a crucial role in variation of inflation in the short run than they do in the long run. Innovations in exchange rate have a negative impact on Botswana inflation. This implies that appreciation of the exchange rate might lead to a fall in Botswana inflation. A shock to RSA CPI appears to have a minor positive effect on Botswana inflation in the first and second quarters and becomes negative after the second quarter. This result suggests that in the short run, Botswana inflation appear to be influenced by RSA CPI. Innovations in money supply and the interest rate explain relatively small proportions of variation in Botswana inflation. All the innovations tend to converge to zero which suggest that the inflation function for Botswana is stable and therefore can reliably be used for forecasting purposes.

13 Determinants of Inflation in Botswana: Figure 5.1: Impulse Responses to Botswana CPI

14 208 M. Raboloko, G. R. Motlaleng and M. Bakwena 5.5. Variance Decomposition The forecast error variance decomposition looks at what proportion of variance in one series is due to its own shock, the policy shock and other identified shocks. It shows how much of the forecast error variance of each of the variables can be explained by exogenous shocks to the other variables. The forecast error variance decomposition is performed to assess the extent to which gross domestic product, money supply, nominal interest rate, rand pula nominal exchange rate and RSA CPI affect the inflation rate for Botswana overtime. Table 5.4 presents the results of variance decomposition of Botswana CPI. The predominant source of variation in Botswana inflation s forecast errors is own shocks. Own shocks account for between 100 and 78 percent of the forecast errors in Botswana inflation. In the short term, own shocks account for 77 to 100 percent of the forecast error variance. The innovations from the nominal exchange rate and RSA CPI are also important sources of the forecast error variance in Botswana inflation. The least significant source of forecast error variance in Botswana inflation is the nominal rate of interest, followed money supply. Nominal GDP has minimal impact on Botswana inflation. The RSA CPI only starts to affect Botswana inflation marginally after five lags. This implies that changes to RSA CPI did not have an immediate impact on Botswana inflation from 2003 to Table 5.4 Variance Decomposition of Botswana CPI Period S.E. Botscpi GDP Money Supply Interest Rate Exchangerate Rsacpi CONCLUSIONS The paper established the determinants of inflation and the stability of the inflation function in Botswana through the vector autoregression (VAR) model. The lagged value of the inflation rate was found to be the only variable significantly different from zero. The implication is that inflation in the current year is influenced by inflation in the past year. The results of this paper are in contrast with those of Atta et al (1996) who found Botswana inflation to be influenced by real income, money supply, Rand/Pula nominal exchange rate, interest rate and largely RSA CPI. The method of analysis used by Atta et al differs from the method adopted in this study. They used an unrestricted autoregressive distributed lag model. Worth noting here is that through the use of a different method of analysis, the study came out with different and interesting

15 Determinants of Inflation in Botswana: results. Our, findings provide a different policy advice. The results are consistent with those of Kuijis (1998) and Abidemi and Malik (2010) who applied a VAR model to analyse the determinants of the price level, exchange rate and output in Nigeria. The authors found a direct relationship between the first and third lag of prices on inflation. The model used is stable in the sense that the Impulse Response Function shows that a shock to Botswana inflation has a positive impact on itself from the first to the sixth quarter. Furthermore, own shock to Botswana inflation has a negative impact on itself from the seventh quarter and tend to converge to zero. The forecast error variance decomposition shows that inflation in the RSA (RSACPI) only starts to affect Botswana inflation marginally after five lags. This implies that changes in RSA inflation do not have an immediate impact on Botswana inflation. References Abidemi, O. I. and Malik, S. A. A. (2010), Analysis of Inflation and its determinant in Nigeria, Pakistan Journal of Social Sciences, 7(2), African Economic Outlook, (2012), Botswana. Atta, J. K., Jefferis, K. R. and Mannathoko, I. (1996), Small Country Experiences with Exchange Rates and Inflation: The Case of Botswana, Journal of African Economies, Vol. 5, No. 2, pp Bank of Botswana, (2012), Monetary Policy statement, Bank of Botswana, Gaborone. Braun, P. A. and Mittnik, S. (1993), Misspecifications in Vector Autoregressions and Their Effects on Impulse Responses and Variance Decompositions, Journal of Econometrics Vol.59 pp Chakrabarti, S. (2006), Control of Inflation in Botswana, Botswana Journal of Economics Vol.3, Issue No.6 Chhibber, A. and Shafik, N. (1990), Exchange Reform, Parallel Markets and Inflation in Africa: The Case of Ghana, World Bank Working Paper, WPS 427, World Bank, Washington DC. Dlamini, A., Dlamini, A. and Tsidi Nxumalo (2001), A Cointegration Analysis of inflation in Swaziland Central Bank of Swaziland Enders, W. (1995), Applied econometric Time Series, Wiley & Sons Inc., New York. Huda, S. (1987), A Note on Inflation in Botswana, in Bank of Botswana, Selected Papers on the Botswana Economy, Gaborone: Bank of Botswana. International Monetary Fund, (2007), Botswana Country Report, N0. 07/228. Jackman, R., Mulvey, C. and Trevithick, J. (1981), The Economics of Inflation, Oxford. Jefferies, K. R (2008), Botswana s Experience with Monetary and Exchange Rate Policy Lessons for Angola USAID/Southern Africa, Gaborone, Botswana. Johansen, S. and Juselius, K. (1990) Maximum Likelihood Estimation and Inference on Cointegration with Applications to the Demand for Money, Oxford Bulletin of Economics and Statistics, Vol. 52, No.2, pp Khan, M. S and Schimmelpfenning, A. (2005), Inflation in Pakistan: Money or Wheat? IMFWP/06/160 Kuijs, L. (1998) Determinants of Inflation, Exchange Rate and Output in Nigeria, IMF working paper no. 160, Leith, J.C. (1991), The Exchange Rate and the Price Level in a Small Open Economy: Botswana, Journal of Policy Modelling, 13 (2):

16 210 M. Raboloko, G. R. Motlaleng and M. Bakwena Lim, C. H. and Papi, L. (1997), An Econometric Analysis of the Determinants of Inflation in Turkey, IMF Working paper no. 170, Lütkepohl, H. (1993), Introduction to Multiple Time Series Analysis, Second Edition. Berlin: Springer-Verlag. Masale, S. M. (1993), Inflation and Exchange Rate Policy in Botswana ( ): an Exploratory Study, MA dissertation (unpublished), University of Sussex. Mitchell, J. (2000), The Importance of Long Run Structure for Impulse Response Analysis in VAR Models, NIESR Discussion papers No.72. National Institute of Economic and Social Research. Mwase, N. (2006), An empirical Investigation of Inflation in Tanzania, IMFWP/06/150 Ncube, A. (1992), A Study of Inflation in Botswana, Bank of Botswana Bulletin, 10 (1): Sims, C. A (1980), Macroeconomics and Reality, Econometrica vol. 48, pp Southern African Customs Union, (2012), Inflation Report.

17