Money demand function for malawi- implications for monetary policy conduct

Transcription

1 Money demand function for malawi- imlications for monetary olicy conduct AUTHORS ARTICLE INFO JOURNAL Mark Lungu, Kisu Simwaka, Austin Chiumia, Arnold Palamuleni, Wytone Jombo Mark Lungu, Kisu Simwaka, Austin Chiumia, Arnold Palamuleni and Wytone Jombo (1). Money demand function for malawi- imlications for monetary olicy conduct. Banks and Bank Systems (oen-access), 7(1) "Banks and Bank Systems (oen-access)" NUMBER OF REFERENCES NUMBER OF FIGURES NUMBER OF TABLES The author(s) 17. This ublication is an oen access article. businessersectives.org

2 Banks and Bank Systems, Volume 7, Issue 1, 1 Mark Lungu (Malawi), Kisu Simwaka (Malawi), Austin Chiumia (Malawi), Arnold Palamuleni (Malawi), Wytone Jombo (Malawi) Money demand function for Malawi imlications for monetary olicy conduct Abstract This aer analyzes the money demand function for Malawi during the eriod of using monthly data. During the samle eriod, several structural changes occurred in the economy. Most of these changes were ignited by the structural adjustment rograms that started in the late 198s, then came the move to lural olitics in the early 199s. This was followed by structural reforms in the financial sector. In the very recent ast, there has been an increase in real economic activity as measured by strong growth in real GDP in the years after and the financial innovations within the banking system after the year. These factors do not seem to have affected the stability of the demand for money and hence increasing the robability of success for the conduct of monetary olicy. Cointegration test results indicate a long-run relationshi amongst real money balances, rices, income, exchange rate, Treasury bill rate and financial innovation. While all variables significantly influence money demand in the long-run, short-run olicy must be directed at increasing financial innovation, oen market activities and imroving the roductivity of the economy to rovide higher return on alternative investments. Keywords: cointegration, money demand, monetary olicy. JEL Classification: E1, E3, E. Introduction Desite the consensus that money demand function has little role under an interest rate-based (Taylor-ruletye) monetary olicy, it is still believed that money demand is imortant for both macroeconomic modelling and monetary olicy. This is esecially imortant for countries like Malawi where monetary authorities continue to emhasize the role of money demand function on their monetary olicy oerations. Duca and VanHoose (3), argue that monetary olicy does not work only through the interest rate channel, but the money demand function does also rovide useful information about ortfolio allocations. Theoretical research and emirical analysis using rimary data on develoing countries have shown that the money demand function can become unstable as a result of financial innovations and financial sector reforms. Partly, because of the instability in the money demand functions, most central banks in recent years have switched from money suly targeting which focused on monetary aggregates as the intermediate target to inflation targeting which seeks to stabilize rices by adjusting interest rates based on inflation forecasts. Malawi targets money suly by focusing on monetary aggregates as intermediate targets. With the resence of structural changes in the economy and innovations in the financial sector, it remains an emirical question whether targeting money suly remains relevant in the conduct of monetary olicy. Source: Reserve Bank of Malawi. Consequently, examining the characteristics of money demand function in Malawi should bear significant meaning for resent and future conduct of Mark Lungu, Kisu Simwaka, Austin Chiumia, Arnold Palamuleni, Wytone Jombo, 1. 5 Fig. 1. Velocity of money monetary olicy. Broad money velocity has fluctuated a lot moving from an average of 6. in 1985 to 9.8 in 1998 before declining to an average of.1 in 1. These fluctuations have often been attributed to unexlained shifts in money demand. Considering the fact that a stable money demand function is considered essential for the formulation and

3 Banks and Bank Systems, Volume 7, Issue 1, 1 conduct of efficient monetary olicy, considerable effort has been made in the emirical literature for both industrialized and develoing countries to determine factors that affect the long-run demand for money and assess the stability of the relationshi between these factors and various monetary aggregates. In Malawi, to our knowledge, no attemt has been made to study the money demand function, articularly on the stability of the estimated coefficients. There are a number of studies on money demand in other sub-saharan African countries, but most of them have used traditional estimating methods, by alying the artial adjustment model like Goldfeld (1973) when analyzing the money demand. The focus of the money demand studies has been to analyze stability of demand functions, esecially when major structural changes have taken lace. Source: Reserve Bank of Malawi. Fig.. Narrow money and financial deth Money demand is an imortant element of monetary olicy analysis. This follows the monetarists view that inflation is in the long run, everywhere and always, a monetary henomenon. If the money demand function is stable over the long run, money suly changes are closely related to rices and income, and it should be ossible therefore for olicy makers to control inflation through aroriate adjustments to the money suly. If, on the other hand, the money demand function is unstable over the long run, changes in money suly are not closely related to rices and income and it is not ossible for authorities to aroriately control inflation through adjustments to money suly. Source: Reserve Bank of Malawi. This study therefore examines the behavior of the demand for money in Malawi for the eriod of The hyothesis is that there exists a stable relationshi among narrow money, income and a vector of return rates over this samle eriod. The first objective is to estimate a demand for money function for Malawi using cointegration and error correction modelling. The second objective is to identify relevant factors in demand for money for Malawi. The third objective is to test for the stability of demand for money, in view of its imortance in Fig. 3. Narrow money and exchange rate develoments the conduct of monetary olicy. Owing to nonavailability of high frequency GDP data, annual data has been transformed into monthly data using a quadratic formula. The rest of the aer is organized as follows. Section 1 reviews literature on demand for money. Section outlines the econometric methodology used in the study. Section 3 discusses results. The final section rovides conclusion and recommendations for the conduct of monetary olicy. 51

4 Banks and Bank Systems, Volume 7, Issue 1, 1 1. Literature review 1.1. Theoretical review. There is a diverse sectrum of money demand theories. The theoretical underinnings of the demand for money have been well established in the economic literature with widesread agreement that the demand for money is rimarily determined by real cash balances. Keynes ostulated three motives for holding real money balances: transactions, recautionary and seculative motives. Following the emergence of liquidity reference theory, several authors have questioned Keynes s rationale for a seculative demand for money and have contributed to the theoretical literature by distinguishing broadly between the transactions demand (Baumol, 195; Tobin, 1956) and the asset motive (Tobin, 1958; Friedman, 1956). Emirical studies of money demand have however tended to converge to a secification where demand for real money balances is a function of scale variable e.g., income or exenditure, own rate of return on money, the oortunity cost of holding money (notably the domestic interest rate and/or exected rate of inflation). The domestic interest rate and exected inflation are roxies for the rates of return on alternative financial and hysical assets, resectively. The inclusion of exected rate of inflation has been emhasized esecially for develoing countries where, given the existence of underdeveloed monetary and financial systems and non-market determined interest rates, hysical assets reresent one of the major hedges against inflation and an alternative asset in the ortfolio of the non-bank ublic. In addition, with increasing financial globalization and emirical evidence on ortfolio balance models in oen economies, the exected rate of return on foreign securities has often been added as an exlanatory variable. Theory ortrays that money demand deends ositively on real GDP and the rice level due to the demand for transactions. Money demand deends negatively on interest rates due to seculative concerns. This relationshi can be deicted as follows: M f P Y i D ( ) ( ) ( ) (,, ), (1) where M D is the aggregate, economy-wide money demand, P is the current rice level, Y is the scale variable (real GDP, wealth, or exenditure in real terms) and i is the average interest rate. The ositive sign indicates a ositive relationshi while the negative sign indicates negative relationshi between the deendent variable and the exlanatory variable. The money demand function is often reorted in real terms as follows: M D / P L( Y, i). () Equation () can be viewed as a liquidity reference function. This secification reresents the desired or long-run real money demand function and assumes a long-run unitary elasticity of the nominal cash balances with resect to the rice level. Many commentators have argued that inflation is a monetary henomenon in the long run and the emirical relation between money and rices is usually discussed in the money demand framework. Monetary olicy is arguably effective as a means of controlling inflation. If the money demand function is stable over the long run, money suly changes are closely related to rices and income, and it is ossible for olicy authorities to control inflation through aroriate adjustments to the money suly. If, on the other hand, the money demand function is unstable over the long run, changes in money suly are not closely related to rices and income and it becomes difficult for olicy makers to aroriately control inflation through adjusting the money suly. Thus, another fundamental issue in studying the money demand function is to examine whether there exists the equilibrium relation of money demand. 1.. Emirical review. In a study covering 35 Sub- Sahara Africa countries, Hamori (8) finds a stable money demand function after finding a cointegrating relationshi of money demand in individual countries. Using Johansen s maximum likelihood and dynamic modeling rocedure, Randa, (1999) finds equilibrium in the long run and stable money demand function for Tanzania. Nell (1999) investigates money demand function for South Africa and finds a stable long run demand for money function for M3, while the demand for M1 and M dislay arameter instability following financial reforms since 198. In Ghana, results by Andoh (1999) shows no evidence of instability of both narrow and broad money demand suggesting that reforms did not affect the financial sector deely or long enough to destabilize the demand for M1 and M. Evidence gathered by Kumar, Webber & Fargher (1) through their emirical investigation into the level and stability of money demand (M1) in Nigeria between 196 and 8, suggest that money demand is stable effectively imlying that Nigeria could effectively use the suly of money as an instrument of monetary olicy. Using an error correction model to examine the stability of demand for money for the traditional monetary olicy, evidence from Uganda as reorted by Kararach (1) suggests that money demand function is unstable. From this finding, he draws recommendations that there is a need for monetary olicy to be used in conjunction with other olicies to achieve the goal of rice stabilization and adjustment. 5

5 Although various methods have been used in modeling demand for money, an error correction model becomes articularly aealing because of its ability to uncover short- and long-run dynamics. Suose variables in this study include real money balances, real income, interest rate, inflation have a unit root and a linear combination of these non-stationary variables is stationary, then any deviation from the relation is temorary and the relation holds in the long-run. If such a linear combination exists, the variables are said to be cointegrated. If the money demand is well identified and the money suly is an integrated rocess, the money demand function can be estimated by the cointegration method. This study therefore alies the multivariate error correction model.. Data descrition and model secification.1. Data descrition. This study alies cointegration analysis and derives an error correction model to examine the behavior of the demand for money in Malawi. The variables in the model are real money demand, real gross domestic roduct, inflation, Treasury bill rate, exchange rate and financial deth (money suly/gdp). The data source for rices and gross domestic roduct (GDP) are the National Statistics Office and the International Monetary t t t t 3 t t 5 t t Banks and Bank Systems, Volume 7, Issue 1, 1 Fund, resectively. All other variables have been sourced from the Reserve Bank of Malawi Research and Statistics Database. Excet for interest rates, inflation and financial deth, other variables have been exressed in logarithm form. We use both inflation and interest rate as measures of oortunity cost... Model secification. The choice of money demand definition is emirically determined by the roblem to be modelled. Particularly, olicy makers are increasingly interested in understanding not only what haens to a articular variable when its determinants change but also whether the relationshi amongst the variables is stable and whether it means reverting and how soon the stability will be achieved. Most studies on demand for money have used an error correction model. The basis of this choice is the fact that this technique is caable of revealing more information on the long- and short-run behavior of economic variables. This study therefore emloys the Johansen Cointegration technique to uncover the long-run and short-run behavior of M1. To analyze the relationshi between money demand and its causative factors various models have been tested and the most aroriate in the context of this study is resented below: ln M b b lny b INF b ln E b FINDP btb, (3) where t reresents white noise error rocess M t is the real money demand (defined as currency outside banks lus demand deosits), Y t is the real GDP, INF t is the annualized inflation rate, Et is the US dollar Nominal Exchange Rate, FINDP t is the financial deth, TB t is the average treasury bill rate..3. Exected results. Based on the conventional economic theory, the income elasticity coefficient, b 1 is exected to be ositive imlying that higher income leads to an increase in money demand; the coefficient of inflation b is exected to be ositive, higher inflation may lead to high demand for money because of the time value of money. Arango and Nadiri (1981) argue that for the elasticity coefficient on the exchange rate variable, b 3, it can either be ositive or negative. If the increase in exchange rate (dereciation) is erceived as the increase in wealth and leads to a rise in domestic money, the coefficient on exchange rate is ositive. But if the increase in exchange rate leads to a decrease in domestic money demand (currency substitution), then the coefficient of exchange rate is negative. Imroved technology/financial roducts entails reduced demand for cash balances. Demand for money by economic agents decreases as they shift from consumtion to investment in which case the elasticity of money demand with resect to return on invested funds is exected to be negative. Hence, b and b 5 are exected to be negative... Econometric methodology...1. Unit root tests. Stationarity tests are a rerequisite before conducting most econometric works. The Augmented Dickey Fuller (ADF) test is used to determine the order of integration of the data. It is however well established in literature that the ADF test has very low ower in the resence of structural breaks as under such circumstances, it is biased towards nonrejection of a unit root. This test is, therefore, augmented by the Philli Perron (PP) unit root test. While the former uses augmentation to whiten residuals, the latter uses non-arametric correction. Besides, the samle in this study is large enough to warrant the use of the two and they remain the most widely used in literature. They both use the null of stationarity with the following test equations and null hyothesis: Y Y 1 Y 1, t 1 t t i t t t 1 H :. m ()... Co-integration tests. The concet of cointegration has been widely used to test long-run relationshis. Suosing real money balances, real income and interest rate are non-stationary variables with a unit root and a linear combination of these non-stationary variables is stationary, then any deviation from the relation is temorary and the relation holds in the long run. If such a linear combination 53

6 Banks and Bank Systems, Volume 7, Issue 1, 1 exists, the variables are said to be cointegrated. The Johansen and Juselius aroach of cointegration is based on a VAR model which can be written in different form as follows: 9 t k tk t1 i1 W W W vt, (5) where W t reresents a vector of the six variables and vt is a multivariate Gaussian error term i.e. (free from autocorrelation, heteroskedasticity and multivariate normal). k is the lag length. is the 6x6 imact matrix which rovides information relating to long-run relationshi. The rank r of the imact matrix is the number of co-integrating vectors. Johansen (1995) introduced two likelihood ratio methods in order to investigate co-integrating vectors as illustrated below. 1. Trace statistic. trace m1 () r N ln(1 l ), (6) k 1 where N is the number of observations, r is the number of cointegrating vectors, j is the number of variables and the lambdas are the Eigen values. The null hyothesis of the trace statistics is as follows: H : number of C.I r against H 1 : of C.I > r where C.I imlies co-integrating equations. Maximal eigenvalue test statistics. This statistic is given as follows with the variables defined as in the trace statistics r, r 1) Nln(1 ). (7) m( r1 The null and alternative hyotheses of maximal Eigen values are as follows: H : number of C.I = r against H 1 : no of C.I = r + 1, where C.I imlies co-integrating equations...3. Vector error correction model (VECM). According to Engle and Granger (1987), if two series are co-integrated of order one i.e. I(1), then there must exist a VECM reresentation in order to govern joint behavior of the series of the dynamic system. In VECM secification, short-run as well as long-run adjustments are made. VECM also rovide information about instantaneous adjustment of the actual stock of real money balances to its desired level. The equilibrium state between real money suly and the real money demand is unlikely to be achieved given the existence of transaction costs, uncertainty and other factors. In addition, the desired level of real money balances is unobservable. A distinction has therefore to be made between long and short-run behavior in the money market by secifying an errorcorrection mechanism (ECM) of the actual real cash balances towards the desired (long-run) level. The VECM secification related to the determinants of money demand is as follows: ln M1 lngdp ln INF ln EXP lntbr ln FININ 5 1t t 3t t 5t 6t lngdp 11, i i1 i1 i1 i1 i1 ln M1 31, i 51, i i1 1, i 1, i 61, i 1, t 1 1, i i1 1, t 1 lngpd 1, t 1 lngpd lngpd 1, t 1 1, t 1 lngpd i1 i1 1, t 1 ln INF ln INF i1 i1, i ln M1 3, i 5, i i1, i 6, i, t 1 13, i i1, t 1, t 1 ln INF ln INF, t 1, t 1 ln INF i1 i1, t 1 ln EXP ln EXP i1 i1 3, i 33, i 53, i i1 3, i 63, i 3, t 1 1, i i1 3, t 1 ln EXP 3, t 1 ln M1 3, t 1 ln EXP This model can be exressed in comact form as defined in equation (3) above. The starting oint is to model changes in real money balances as a resonse to the deartures from one or two or all of the stationary linear combinations of the I(1) variables, augmented by short-term dynamics from the current and lagged first differences of the variables included in the cointegrating vector.... VECM based causality. Granger (1998), states that in Granger reresentation theorem, if two variables are stationary of order (1) and cointegrated, then either the first variable causes the second or vice-versa. In this study, multivariate granger causality test based on VECM is utilized. It rovides an 3, t 1 ln EXP i1 i1 i1 3, t 1 lntrb lntrb i1, i lntrb 3, i, i ln M1 5, i i1 6, i, t 1 15, i i1, t 1, t 1 lntrb, t 1, t 1 ln M1, t 1 ln FININ i1 i1 i1 i1 5, i 35, i 55, i i1 5, i 65, i 5, t 1 ln FININ ln FININ ln FININ lntbr 5, t 1 5, t 1 ln FININ 5, t 1 5, t 1 5, t 1 ECT t 1 ECT ECT ECT, t 1 t 1 ECT t 1 t 1 ECT t 1 1t t, 3t, t, 5t. 6t (8) additional channel for long-run causality which is ignored by Sims and Granger causality tests. Longrun causality is confirmed using the joint significance of the coefficients of lagged variables. Chi-squared test is emloyed to check the joint significance of the coefficients of the lagged variables and t-tests is used to check significance of the error term...5. Diagnostic tests. Diagonostic tests are utilized to check the validity of the fitted model. In this study, VECM based diagnostic tests are reorted. VEC residual serial correlation Langragian multilier test is used for investigating ossible serial correlation in the error term. The null hyothesis for this test is reorted below:

7 H : E (µ, µt-q) = for t q, q = 1,,..., (9) The residual normality test is carried out in order to investigate whether residuals are normally distributed or not. We use the Jarque-Bera normality test with the following null hyothesis. H : Residuals are multivariate normal. Finally, the VEC residual heteroskedasticity test is alied to check whether there is heteroskedasticity or not. The Chi-squared test will use the following null hyothesis. Table 1. ADF test results (1985-1) Banks and Bank Systems, Volume 7, Issue 1, 1 H : Variance of residuals is homoscedastic...6. Stability tests. We use the inverse characteristic roots to determine the stability of the VEC. If the characteristic roots of the variables lie within the circle, the arameters estimated are deemed to be stable. 3. Emirical results and discussion 3.1. Unit root tests. Table 1 resents ADF test results for the eriod of Variable Without constant With trend & constant In levels LnM LGDP INF LEXR LROC LFININ TBR First differences LnM ** -.137** LGDP -.111** -.17* INF * -6.99* LEXR * -11.6* LFININ * -6.15* TBR * * MacKinnon critical values 1% % Notes: Null hyothesis: series have unit roots. *Significant at 1% level. **Significant at 5% level. Table. PP test results (1985-1) Variable Without constant With trend & constant In levels LnM LGDP INF LEXR LFININ TBR First differences LnM LGDP INF LEXR LFININ TBR MacKinnon critical values 1% % Notes: Null: series have unit roots. *Significant at 1% level. **Significant at 5% level. The ADF results resented in Table 1 revealed that all the variables were nonstationary in levels. However stationarity was achieved after first differencing imlying that all variables were I(1). In light of 55

8 Banks and Bank Systems, Volume 7, Issue 1, 1 the fact the ADF is weak in times of structural breaks, for further confirmation, results from the PP unit root test are reorted alongside. Both tests indicate that all series are I(1) rocesses. This finding triggers the search for a long-run relationshi amongst the series which is done using the cointegration technique. Table 3. Johansen and Juselius test (1985-1) Trace statistic 3.. Cointegration results. Since stationarity results confirmed that all variables were integrated of order 1, before identifying the number of co-integrating vectors, we first alied VAR test in order to determine otimal lag length. The Schwartz Bayesian Criterion statistic indicated the otimal lag length of for the Johansen Co-integration Test. Null hyothesis Alternative hyothesis Test statistics Critical value P-value** H: r Ha: r > * H: r 1 Ha: r > * H: r Ha: r > H: r 3 Ha: r > Maximum eigenvalue statistics H: r = Ha: r = * H: r = 1 Ha: r = * H: r = Ha: r = H: r = 3 Ha: r = Notes: The tests indicate at least cointegrating eqn(s) at the.1 level. * Denotes rejection of the hyothesis at the.1 level of significance. **MacKinnon-Haug-Michelis (1999) -values. The maximal and trace eigenvalue statistics strongly reject the null hyothesis of both the none and at most one cointegrating vectors in favor of at least two cointegrating vectors at the 1. ercent significance level. Although we find at least two cointegrating relationshis, the interest in this study is to examine the resonse of money demand to changes in real income, inflation, financial deth and Treasury bill rate. Therefore the co-integrating vector identified is given as follows. Table. Normalized cointegrating vector LM1 C LGDP INFL LEXR LTBR LFININ [-1.8]* [-.1]* [.1]* [.5] [.58]* Note: [] denote t-statistics. *Significant at 1%. The P-values are taken from MacKinnon-Haug-Michelis (1999). With the evidence from the cointegration test, it can be interreted that Malawi s demand for money, gross domestic roduct, inflation, exchange rate, interest rates and financial innovation move together. The cointegrating vectors were, therefore, normalized by the deendent variable. From the longrun equation it can be concluded that money demand largely deends on all the variables in the model. The demand for money with resect to income changes is highly elastic with 1. ercentage change in GDP leading to a more than roortionate increase in demand for money. The ositive association suorts theory and other findings from similar constant elasticity models. Individuals are inclined to increase their holdings of money balances as their incomes rise. With regard to inflation, it is found that the semi-elasticity of money suly with resect to inflation is.7 ercent, i.e. a 1. ercent increase in inflation leads to an increase of.7 ercent in demand for money. This again conforms to economic theory; according to the time value of money theory, under inflationary ressures the real urchasing ower of the currency is eroded and a enny today is worth less than a enny tomorrow. A 1 ercent rise in Treasury bill yield will only reduce demand for money by 1. ercent where as the elasticity of money demand with resect to financial deth is.7. These findings conform to economic theory. Technological advances in the financial market will reduce demand for money. Similarly, if the financial market rovides better return of cash balances, the demand for cash balances will decrease. A 1. ercent rise in financial deth leads to a.7 ercent decline in demand for money balances. An areciation of the currency is found to have a negative relationshi with money demand Vector error correction model. The model estimates that the short run dynamics are mainly driven by lagged money balances, rices, and financial innovation. The exchange rate, Treasury bill rate and income are not significant in correcting the disequilibrium. However, the concern is on the resonse of money demand to changes in GDP, INFL, EXR, TBR and FINDP. We therefore begin by analyzing short-run dynamics contained in equation (1) in Table 5. In the short run, demand for money in the current eriod is quite sensitive to what it was in the revious eriod. A 1 ercent increase in demand for money in the current eriod leads to a further increase of.8 ercent in demand in the next eriod. Financial deth is the most imortant factor 56

9 in determining the short-run behavior of demand for money. The error correction term (ECT) is significant and negative indicating the existence of long-run relationshi amongst the variables. The coefficient of the error term is -.8 which showed C D(LM1t-1) D(LGDP t-1) D(INFL t-1) D(LEXR t-1) D(LTB t-1) D(LFINDP t-1) (ECTt-1) Table 5. VECM results Banks and Bank Systems, Volume 7, Issue 1, 1 low seed of adjustment towards long-run equilibrium. This indicates that whenever there was a disturbance in the system, in every short eriod, only 8% correction to the disequilibrium would take lace. Eq 1. D(LM1) Eq. D(LGDP) Eq 3. D(INFL) Eq. D(LEXR) Eq 5. D(LTBR) Eq 6. D(LFININ) [ 3.878] [ 7.513] [-.656] [.383] [-.9767] [ ] [ ] [.8573] [.111] [.7] [ ] [ 6.987] [ ] [ ] [-.5168] [-.576] [.367] [ ] [.7375] [ 1.68] [ ] [ 1.361] [ 1.951] [.97195] [ ] [.883] [ ] [ ] [ ] [.38951] [.8] [.881] [.67] [.581] [-1.91] [.711] [ ] [-.55] [ ] [ ] [.3171] [.956] [-.7993*] [ 1.96] [ ]* [-.67] [-.737] [-7.5]* Summary statistics for the VECM R-squared Adj. R-squared Sum sq. resids S.E. equation F-statistic Mean de. var E-5 S.D de. var De. var Table 6. VECM based block exogeneity tests Indeendent variable DLnM1 DLnGDP DINF DLnEXR DTBR DLnFININ (1df) Joint (5df) P-values 1 (ECTt-1)1t DLnM ** * *.* -.795* DLnGDP DINF * * 3.58* DLnEXR ** DTBR ** ** ** -.75 DLnFININ.131* ** * -7.59* Source: Authors calculation using E-views. Notes: *Significant at 1% level. **Significant at 5% level. The results show that in case of the first deendent variable, the DLnM1 t lagged residual was statistically significant indicating the resence of long-run relationshi amongst deendent and indeendent variables which was confirmed by the significance of (ECT t-1 ) 1t. The significance of the is an indication of the resence of short-run causality. 3.. The air-wise Granger causality test. Aendix C shows the air-wise Granger causality test results. The air-wise Granger causality test confirms the cointegration findings. Results show that money suly does not Granger cause inflation. We also find that income Granger cause demand for money. The other channel suorted here is that the rise in demand for money has an imact on the exchange rate. Furthermore, we fail to accet the null that exchange rate does not Granger cause inflation. One of the channels suorted by this study therefore is that a rise in the level of income has had an imact on money demand which has affected the exchange rate which eventually has imacted inflation Imulse resonse functions. Aendix B resents imulse resonse functions. The study uses imulse resonse function as an additional check of 57

10 Banks and Bank Systems, Volume 7, Issue 1, 1 the cointegration test s findings. The Cholesk-tye of contemoraneous identifying restrictions are emloyed to draw a meaningful interretation. The recursive structure assumes that variables aearing first contemoraneously influence the latter variables but not vice versa. In Aendix B, there are two salient outcomes ertinent to monetary olicy oerations. First that an increase in income raises demand for money from an initial eriod of the shock u to 9 months after which demand for money stabilizes at a new higher level. Secondly, a ositive shock to exchange rate Table 7. Model fitness temorarily raises de-mand for money until the 5th month beyond which demand stabilizes towards reshock levels. Two ways of ermanently reducing the demand for money are to increase the level of financial innovation and increasing Treasury bill rate which is a roxy to investment returns. A one standard deviation shock to inflation however is found to raise demand for transaction money balances until a 5th month beyond which demand stabilizes at a higher level Diagnostic checks. Table 7 resents model fitness. Test Test statistics P-value Conclusion LM serial correlation statistics No serial correlation Jarque-Bera statistic.9.31 Residuals are multivariate normal Chi-squared Residuals are homoscadastic Fig.. VECM characteristic roots Conclusion and recommendation This aer investigates the real demand for money in Malawi using a cointegration analysis. Tests show that the model is stable and adequate. The stability of coefficients imlies that the robability of the effectiveness of monetary olicy is high in the long run. While in the long run, real GDP, inflation, exchange rate, Treasury bill rate and financial deth all have significant imact on the demand for money, in the short run, it is financial innovation, exchange rate movements and lagged money suly that dislay causality in money demand. While in the long run demand for money is resonsive to changes in interest rate, ursuits of olicies in the short run aimed at altering level of interest rate to contain money demand are unlikely to bear fruit. The results obtained oint to roblems in imlementing monetary olicy using the bank rate in the short run. The insensitivity of demand for money to changes in interest rate in the short run is manifested in a sticky and wide sreads between the lending and savings rates. This develoment oints to the underdeveloed nature of the money market and lack of financial deeening. The long-run significant and negative relationshi between real demand for money and exchange rate suggests that dereciation leads to increased demand for money balances for transaction uroses. In the short run however, an areciation (overvalued currency) of the currency contributes to an increase in real money balances which may raise domestic absortion through imorts and leading to ersistent balance of ayment roblems. Another salient outcome of the model is that a dereciation of the Malawi kwacha leads to inflation. When the currency dereciates by 1 ercent, rices in Malawi overreact, rising by around 1. ercent immediately. 58

11 References Banks and Bank Systems, Volume 7, Issue 1, 1 From the foregoing discussion, it can be concluded that the conduct of monetary olicy must clearly distinguish between short-run and long-run objectives. The recent develoments where money suly has been increasing at a faster rate while inflation has been somewhat stable or indeed trending downwards suort the finding that targeting money suly could be ineffective in the short run. Although monetarists argue that inflation is always a monetary henomenon, in Malawi inflation has most often been observed to go down when money suly is exanding. This finding oints to the need for monetarist to rethink the monetary olicy otions in countries where inflation basket is largely skewed towards suly side like Malawi. Acknowledgements Secial thanks should go to Dr. Grant P. Kabango, Acting General Manager, Mr. Efford Goneka, and all members of the Monetary Policy Imlementation Committee of the Reserve Bank of Malawi for their comments on the aer. The views in this aer are those of the authors and do not necessarily reresent the views of the Reserve Bank of Malawi. Any comments on this working aer are welcome. 1. Andoh, S.K. and Chaell, D. (). Stability of the money demand function: evidence from Ghana, Alied Economics Letters, Vol. 9, No. 13, Arango, S. and Nadiri, M. (1981). Demand for money in oen economies, Journal of Monetary Economics, 7, Dickey, D. and Fuller, W. (1981). Likelihood ratio tests for autoregressive time series with a unit root, Econometrica, 9, Engle, R.F. and Granger, C.W. (1987). Cointegration and error correction: reresentation, estimation and testing, Econometrica, 55, Fuller, W. (1976). Introduction to Statistical Time Series, New York: John Wiley and Sons. 6. Goldfeld, S.M. (1973). The demand for money: revisited, Brookings Paers on Economic Activity, 3, Duca, John V. (). Financial Technology Shocks and the Case of the Missing M, Journal of Money, Credit, and Banking, 3,.. 8. Friedman, Benjamin M. (1988). Lessons on Monetary Policy from the 198s, Journal of Economic Persectives,, Goldfeld, Stehen M. (1973). The Demand for Money Revisited, Brookings Paers on Economic Activity, 3, Goldfeld, Stehen M. and Daniel E. Sichel (199). The Demand for Money, in B. Friedman and F. Hahn (eds.), Handbook of Monetary Economics, Elsevier, Kararach, George (1). Evidence on the Demand for Money function in Uganda, UNICEF, WP No Kevin S. Nell (1999). The Stability of Money Demand in South Africa, , Studies in Economics 995, Deartment of Economics, University of Kent. 13. Kumar, Saten & Webber, Don J. & Fargher, Scott (1). Money demand stability: A case study of Nigeria, MPRA Paer 67, University Library of Munich, Germany. 1. Randa, John (1999). Economic Reform and the Stability of the Demand for Money in Tanzania, Journal of African Economies, 8 (3),

12 Banks and Bank Systems, Volume 7, Issue 1, 1 Aendix A. Residuals LM1 residuals. LGDP Residuals LGDP residuals Residuals INFL residuals. Residuals LEXR residuals TBR Residuals residuals. LFININ residuals Residuals Fig. 1. Distribution of residuals 6

13 Aendix B Resonse Resonseto tononfactorized Nonfactorizedone One S.D. S.D. innovations Innovations Resonse of LM1 to LM1 Resonse of LM1 to LGDP Resonse of LM1 to INFL Resonse of LM1 to LEXR Resonse of LM1 to TBR Resonse of LM1 to LFININ Resonse of LGDP to LM1 Resonse of LGDP to LGDP Resonse of LGDP to INFL Resonse of LGDP to LEXR Resonse of LGDP to TBR Resonse of LGDP to LFININ Resonse of INFL to LM1 Resonse of INFL to LGDP Resonse of INFL to INFL Resonse of INFL to LEXR Resonse of INFL to TBR Resonse of INFL to LFININ Resonse of LEXR to LM1 Resonse of LEXR to LGDP Resonse of LEXR to INFL Resonse of LEXR to LEXR Resonse of LEXR to TBR Resonse of LEXR to LFININ Resonse of TBR to LM1 Resonse of TBR to LGDP Resonse of TBR to INFL Resonse of TBR to LEXR Resonse of TBR to TBR Resonse of TBR to LFININ Resonse of LFININ to LM Resonse of LFININ to LGDP Resonse of LFININ to INFL Resonse of LFININ to LEXR Fig.. Imulse resonse functions Resonse of LFININ to TBR Resonse of LFININ to LFININ Banks and Bank Systems, Volume 7, Issue 1, 1

14 6 Aendix C Percent LM1 variance due to LM1 Percent LGDP variance due to LM Percent LM1 variance due to LGDP Percent LGDP variance due to LGDP Percent LM1 variance due to INFL Percent LGDP variance due to INFL Variance Decomosition Variance decomosition Percent LM1 variance due to LEXR Percent LGDP variance due to LEXR Percent LM1 variance due to TBR Percent LGDP variance due to TBR Percent LM1 variance due to LFININ Percent LGDP variance due to LFININ Banks and Bank Systems, Volume 7, Issue 1, 1 Percent INFL variance due to LM1 Percent INFL variance due to LGDP Perc ent IN FL variance due to INFL Percent INFL variance due to LEXR Percent INFL variance due to TBR Perc ent IN FL variance due to LFININ Percent LEXR variance due to LM1 Percent LEXR variance due to LGDP Percent LEXR variance due to INFL Percent LEXR variance due to LEXR Percent LEXR variance due to TBR Percent LEXR variance due to LFININ Percent TBR variance due to LM1 Percent TBR variance due to LGDP Percent TBR variance due to INFL Percent TBR variance due to LEXR Percent TBR variance due to TBR Percent TBR variance due to LFININ Percent LFININ variance due to LM1 Percent LFININ variance due to LGDP Percent LFININ variance due to INFL Percent LFININ variance due to LEXR Percent LFININ variance due to TBR Percent LFININ variance due to LFININ Fig. 3. Variance decomosition

15 Banks and Bank Systems, Volume 7, Issue 1, 1 Aendix D Table 1A. Pairwise Granger causality test Pairwise Granger Causality Tests Samle: 198M1 M1 Lags: 1 Null hyothesis Obs F-statistic Prob. INFL does not Granger cause LM LM1 does not Granger cause INFL LGDP does not Granger cause LM LM1 does not Granger cause LGDP LEXR does not Granger cause LM LM1 does not Granger cause LEXR TB does not Granger cause LM LM1 does not Granger cause TB LFININ does not Granger cause LM LM1 does not Granger cause LFININ LGDP does not Granger cause INFL INFL does not Granger cause LGDP E-16 LEXR does not Granger cause INFL INFL does not Granger cause LEXR TB does not Granger cause INFL INFL does not Granger cause TB LFININ does not Granger cause INFL INFL does not Granger cause LFININ LEXR does not Granger cause LGDP E-13 LGDP does not Granger cause LEXR TB does not Granger cause LGDP E-1 LGDP does not Granger cause TB LFININ does not Granger cause LGDP LGDP does not Granger cause LFININ TB does not Granger cause LEXR LEXR does not Granger cause TB LFININ does not Granger cause LEXR LEXR does not Granger cause LFININ LFININ does not Granger cause TB TB does not Granger cause LFININ Cointegrating relation 1 Fig.. Cointegrating grah for the demand for money 63