€¦  · web viewpolicy makers attempt to ensure price, exchange rate and overall macroeconomic...

49
The Dynamic Effect of Monetary Policy Shock on Main Macroeconomic Aggregates in Ethiopia: Evidence from a Small Open Economy. Yohannes Ghebru Alemayehu E-mail: [email protected] Cell phone: +251920231316 Ethiopian Policy Studies Institute September, 2019

Upload: others

Post on 01-Oct-2020

1 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: €¦  · Web viewPolicy makers attempt to ensure price, exchange rate and overall macroeconomic environment stability. Therefore, the most significant question in monetary policy

The Dynamic Effect of Monetary Policy Shock on Main Macroeconomic Aggregates in Ethiopia: Evidence from a Small Open Economy.

Yohannes Ghebru Alemayehu

E-mail: [email protected]

Cell phone: +251920231316

Ethiopian Policy Studies Institute

September, 2019

Page 2: €¦  · Web viewPolicy makers attempt to ensure price, exchange rate and overall macroeconomic environment stability. Therefore, the most significant question in monetary policy

Abstract

This paper investigates how shocks to monetary policy, measured in terms of interest rate and money supply, impact key macroeconomic aggregates such as real GDP, price level, and exchange rate in Ethiopia. An effort has been made to analyse the dynamic relationship among the variables included in the model. Johansen’s (1995) co-integration techniques and error correction models are employed. To analyse the reaction of macroeconomic variables to a shock in monetary policy measures, the paper estimates impulse response graphs based on the stable vector error correction model. The study uses annual data from 1981 to 2015.The results provide evidence that there is a stable long run relationship among the variables in the model. The findings of the study also indicate that interest rate adjusts at a faster speed as compared to the exchange rate, money supply and GDP deflator on the examined period. Furthermore, there is significant evidence of exchange rate puzzle when interest rate is used as a measure of monetary policy and evidence of liquidity puzzle when money supply is used as a measure of monetary policy. The study also confirms that the interest rate and exchange rate channels are strong monetary transmission channels, both in the short run and long run. The results of the study further suggest that monetary policy measured in terms of interest rate have more impact when the central bank target is promoting output growth. However, when the central bank target is inflation handling, exchange rate or money supply seems to be more suitable and effective instruments, especially in the long run. The findings also imply that monetary policy plays an important role in stabilizing both real and nominal parts of the economy.

JEL code: Y16

Keywords: Monetary policy, Shocks, Error correction model, Impulse response function, Variance decomposition, Macroeconomic aggregates, Long-run relationship

Page 3: €¦  · Web viewPolicy makers attempt to ensure price, exchange rate and overall macroeconomic environment stability. Therefore, the most significant question in monetary policy

1. Introduction

Perspectives on monetary policy as a tool to macroeconomic management has come in to the picture of the world policy makers intensively at the end of World War II since the time of Keynes. In the 1970s, 1980s, and early 1990s, economic models for the analysis of monetary policy combined the assumption of nominal rigidity with the quantity theory of money function, which links the quantity of money to the aggregate expenditure (Carl E. Walsh, 2010). While the theoretical foundation and framework of these models is weak, they were helpful in addressing a wide range of monetary policy issues. Recently, the standard approach in monetary policy analysis integrates nominal wage and price rigidities in to standard dynamic general stochastic equilibrium model (DSGE)1 frameworks, which are based on optimization behaviour by agents.

As in many developed economies, monetary policy is expected to be a useful policy instrument in shaping the general social and economic development framework of developing countries. In developed countries, monetary policy models focus on central bank activities, whereas, those in developing countries, in contrast, do not get the same attention owing to the belief that central banks in these countries were created with the primary goal of financing the government deficit (Magda Kandai, 2014).

Policy makers attempt to ensure price, exchange rate and overall macroeconomic environment stability. Therefore, the most significant question in monetary policy is how increases or decreases in money supply and interest rates influence the economy, in terms of prices, exchange rate, real GDP and other macroeconomic aggregates. Unexpected shocks happen in an economy, and the response of macroeconomic aggregates depends on the level of financial development and the strength of monetary policy transmission mechanisms. In emerging small open economies, like Ethiopia, the impact of shocks of monetary policy on key macroeconomic aggregates and the associated monetary transmission mechanism is usually thought to be by far less straight forward.

It is evident that monetary policy plays a crucial role in the performance of the economy. However, the effectiveness of monetary policy on achieving the desired outcome largely depends on the institutional setup that facilitates the implementation process (National Bank of Ethiopia, 2009). Furthermore, the achievement of monetary policy management depends largely on the ability of policymakers to identify the changes in the strength of monetary policy transmission channels on time, and to make adjustment to the monetary frameworks accordingly (Adam Mugume, 2011). Besides the effectiveness of monetary policy, monetary policy in Ethiopia is hampered by the excess liquidity, fiscal dominance and lack of secondary markets.

The existing theory and evidence on the effects of monetary policy are not consistent, and there are different views on the effects of monetary policy on macroeconomic variables. This makes the impact of monetary policy on the economy uncertain, and the unpredictable nature

1DSGE are basically based on the optimization behaviour of agents.

Page 4: €¦  · Web viewPolicy makers attempt to ensure price, exchange rate and overall macroeconomic environment stability. Therefore, the most significant question in monetary policy

of the effectiveness of monetary policy is a huge problem in the monetary policy making process.

Monetary policy has been progressed on theoretical and empirical grounds in the last three decades with the central banks relatively independent of the government influence. Since the influential work by Friedman (1968), the role of monetary policy has become an inconclusive issue in stabilization of the macroeconomic environment. In addition to the advances in the theoretical dimension of monetary policy, a considerable body of empirical literature on the impact of monetary policy shocks on macroeconomic variables has contributed to the unending debate by providing evidence on how a monetary policy shock measured by interest rate or money supply affects GDP growth, price level and exchange rates. It is clear that the significance of monetary policy in the economy is enhanced by the implementation of a floating exchange rate system, the motto of financial reforms, trade liberalization and more independent central banks. Therefore, both academics and professionals in central banks, who are responsible for making monetary policy, are keen to understand how, when, and to what extent macroeconomic variables respond to a monetary policy shock.

Measuring the effects of monetary policy shock has an immense role to play in the policy making by the monetary authority in charge. The central bank has the responsibility of making monetary policy to stabilize the economy and foster economic growth. The monetary authority can impact the economy when monetary policy has predictable effects. In developing countries like Ethiopia where the market does not function well and the financial development is still in its infant stage predicting monetary policy impact on key economic aggregates become tricky., Many studies utilize different methodologies to address the issue of how macroeconomic aggregates respond to a monetary policy shock ranging from the recent dynamic stochastic general equilibrium models to the traditional structural models. Recent empirical studies by Bernanke et al. (2005); Bernanke et al. (1998); Eichenbaum et al. (1995) and Sims (1992) have played an immense role in contributing to the measurement of monetary policy and its innovations. These studies applied vector autoregressive models (VAR) to analyze and measure the impact of monetary policy shock on macroeconomic aggregates. The findings of these studies indicate that once a monetary policy shock occurs, economic growth, price level and exchange rate respond to the shock. However, there are many measurement problems and various anomalies. Moreover, there are some inconsistencies in price puzzle, exchange rate puzzle and liquidity puzzle found in these results.

In developing countries, monetary policy is more complicated and challenging owing to the lack of well-developed and deep financial markets, and weak channels of transmission in developing countries, the relationship between monetary measures and macroeconomic aggregates tends to be weak and inconsistent. Moreover, most of the empirical studies have focused on developed countries. Thus, we know relatively less about how macroeconomic aggregates, respond to monetary policy shocks in developing countries. Yet, understanding the effects of monetary policy on macroeconomic aggregates of developing countries is of a great significance to policymakers and academics for enhancing macroeconomic stability. Evidence on the effectiveness of monetary policy in developing countries would enhance our

Page 5: €¦  · Web viewPolicy makers attempt to ensure price, exchange rate and overall macroeconomic environment stability. Therefore, the most significant question in monetary policy

overall understanding of using monetary policy as a tool of macroeconomic stabilization in developing countries.

In an emerging and small open economy like Ethiopia, looking at the impact of monetary policy shocks to generate real economic effects is less clear. The ambiguity may come from the inherent imperfections in the goods market, money market and labour markets, and the unpredictable nature of prices, among others. As a result, monetary policy shocks may just pass quickly through to price, and have little impact or no effect at all. Most of the previous studies on the impact of monetary policy shocks on macroeconomic aggregates have focused on developed countries. In contrast, this paper focuses on a developing country and a relatively small open economy, namely, Ethiopia. Specifically, the paper empirically analyzes and investigates the impact of monetary policy shock on key macroeconomic aggregates in Ethiopia. The study uses annual data covering the time span of 1981 to 2015. First, the study analyses the short run and long run dynamics of the variables. Second, the study examines how the key macroeconomic variables respond to a one-standard deviation shock to the monetary policy.

2. Literature review 2.1. Theoretical literature review

Theoretical models of monetary policy have been around since the early days of the classical doctrine. Since then, the conceptual framework and shape of monetary policy has evolved overtime with different school of economics. According to Livingston (1975), monetary policy can briefly be defined as “the deliberate use of monetary system by the government and central bank to handle and impact the economy”. Moreover, McConnel (1984) described monetary policy as a tool which helps in achieving full employment of output, with a stable macroeconomic environment, and especially low level of inflation.

In the last 20th century different school of economic thoughts have been developed overtime with different views on macroeconomic problems and their solutions. In the classical quantity theory of money framework monetary policy is neutral. According to classical model, monetary policy shock is predicted as neutral (or near neutral) with respect to real variables (Jordi Gali, 2015). This is against the widely accepted idea that central bank can use monetary policy to influence output and employments at least in the short run. This leads to many economists to question the normative classical monetary framework implications as a relevant framework for monetary policy analysis. On the other hand, in the Keynesian system monetary policy plays an immense role in affecting the economic activity. Accordingly, shocks to money supply can affect permanently interest rate, aggregate demand, level of employment and output. The Keynesians assume that the economy is not at its full employment and this implies that an increase in money supply can bring about an increase in output permanently. Moreover, money supply can influence the price level depending on the influence of aggregate demand and the elasticity of aggregate output.

After the work of well-known economist Milton Friedman, monetarists’ theory enjoyed academic and government acceptance in the 1980s. In the monetarist monetary policy system money supply is what controls the economy. Monetarists believe in the management of the

Page 6: €¦  · Web viewPolicy makers attempt to ensure price, exchange rate and overall macroeconomic environment stability. Therefore, the most significant question in monetary policy

supply of money, and subsequently letting the market fix the rest itself. According to their theory, controlling the supply of money directly affects inflation and interest rate. After the seminar papers by Prescott (1986) the real business cycle (RBC) became the main theoretical framework for the analysis of economic fluctuations, and became the core of macroeconomics. Methodologically, RBC theory introduces new methods, namely the use of dynamic stochastic general equilibrium (DSGE) models as a central tool for the analysis of macroeconomic issues. The RBC framework explains economic fluctuations with no reference to monetary factors, even abstracting from the existence of a monetary sector. There were some attempts by Cooley and Hansen (1989) and others to introduce monetary sector to the RBC model while fully sticking with the assumption of perfect competition and fully flexible price and wages. However, the RBC model show that technological advancements bring’s real changes in the economy.

A recent generation of macroeconomic models such as the new Keynesian model becomes workhorse for the analysis of monetary policy fluctuations and welfare (Jordi Gali, 2015). The new Keynesian economists are different in two key assumptions from the classical economist’s monetary policy framework assumptions such as the imperfect competition in the goods market and nominal rigidities. The key points that separate the real business theory with the new Keynesian model is that there is explicit inclusion of nominal price and wage rigidities that are needed to make the framework suitable for evaluation of monetary policy (Richard Clarida et. al., 1999). Recently the new Keynesian framework become the basis for the new generation of models developed in the central banks and applied extensively for simulation and forecasting purposes (Smets, f. and R. Wouters. 2007, 2003). The new Keynesian model, based on stickiness of wage and price, argues that monetary policy is non-neutral, i.e. monetary policy has strong influence on economic activity.

2.2. Empirical literature review

The empirical literature on the impact of monetary policy shock on key macroeconomic aggregates contains many studies in developed countries, and fewer studies conducted in developing countries (see e.g. Cushman and Zha, 1997; Christiano et al., 1999; and Bernanke and Mihov 1998; Khan et al., 2002; Berument, 2007). These studies showed different outcome from their analysis, and there seems to be an agreement about the impact of monetary policy shock on non-policy variables of output and prices in developed economies as Christiano et al (2002) argued. On the other hand, it seems there is conflicting results regarding the impact of monetary policy shock in developing countries. The following section discusses some recent empirical studies regarding the impact of monetary policy shock on macroeconomic aggregates.

Forni and Gambetti (2010) analyzed the impact of a dynamic monetary policy shock, by applying the standard recursive scheme with a dynamic structural factor model for the United States of America (USA), spanning the period 1973:3 – 2007:10. They apply the same variables which are employed by Watson and Stock (1998). According to them the factor analysis model is by far superior to the factor augmented VAR (FAVAR) proposed by Bernanke et. al. (2005) since it helps in solving the problem caused by puzzles in monetary

Page 7: €¦  · Web viewPolicy makers attempt to ensure price, exchange rate and overall macroeconomic environment stability. Therefore, the most significant question in monetary policy

policy analysis. According to their findings a positive shock to the federal funds rate (FFR) appreciates the real exchange rate, which further confirms the overshooting hypothesis of Dornbusch (1976). By estimating impulse response graphs, they reveal the absence of price puzzle. On the other hand, they showed that with a positive shock to the federal funds rate (FFR), the industrial production declines temporarily to a large extent. Bjørnland (2008) examined the effect of a monetary policy shocks , including the exchange rate in the model specification, with a quarterly data covering the period 1993 – 2004.To determine the order of the variables he uses Cholesky ordering and the Kim and Roubini’s (2000) identification procedures. He showed that the interest rate temporarily increases and converges in four quarters to its normal path. However, his analysis did not show any statistically significant evidence of the price puzzle or exchange rate puzzle. Ansari et al. (2007) analyzed the association among money income and domestic price by applying the vector error correction (VEC) model with a quarterly data. They include both broad money and narrow money as measures of monetary policy in to their econometric specification. They reveal that once disequilibrium happens, output adjust 6 per cent towards its long run equilibrium. The speed of adjustment is 6 percent per quarter towards long run equilibrium. Moreover, they declare that a positive shock to monetary policy have an impact on output after 5 quarters i.e. a shock to monetary policy leads to adjustment in output after 5 quarters.

Bernanke et al. (2005) introduced methodologies which are a combination of factor model and VAR model to analyze large information sets. They think that simple VAR models are unable to incorporate large information sets in their analysis. They applied the diffusion index employed by Stock and Watson (2002) to estimate the factors by using a panel of 120 monthly macroeconomic time series data covering the period 1959:1 – 2001:8. They assumed a recursive structure regarding identifying the non-contemporary response of unobserved factors to a monetary policy shock. They compare the results of a 3-variable VAR model and two different FAVAR model specifications. The results reveal that the standard VAR model shows a significant price puzzle and reveal the inconsistent production reaction to a one-standard deviation monetary policy. On the other hand, the factor augmented VAR (FAVAR) specification improves the results revealed by the VAR model, as the price puzzle disappear after one year, real activity decreases, monetary aggregates decline and exchange rate appreciates for USA. As in the case of Forni and Gambetti (2010), to reach the identification a large number of economic restrictions are imposed. Furthermore, the restrictions are also imposed on impulse response functions (IRFs), not on the IRF of variables.

Jang and Ogaki (2004) examined the impact of monetary policy shock on exchange rates, prices and output level for the USA. The empirical analysis is carried out by applying the structural vector error correction (SVEC) model, and vector autoregressive model, following the model of Jang and Ogaki(2004). The results indicate that appreciation of exchange rate is the outcome of a tight monetary policy. Moreover, a decline in price is observed as a result of a contractionary monetary policy. They also reveal the existence of price puzzle. Wong (2000) applied a time-varying parameter model for the USA to empirically investigate the impact of monetary policy on macroeconomic variables, with a monthly data for the period covering 1959:1 – 1994:12. FFR and reserves have a contemporaneous effect while output

Page 8: €¦  · Web viewPolicy makers attempt to ensure price, exchange rate and overall macroeconomic environment stability. Therefore, the most significant question in monetary policy

and price are assumed to have a lagged effect. A VAR model with a maximum of three lags has been estimated. The results from the VAR model suggest that output increases due to a contractionary shock to monetary policy. The responsiveness of output is different at different policy targets of the central bank. Output is less responsive to monetary policy shocks when the central bank targets fostering economic growth, whereas, it is more responsive to shocks during the periods of when the central bank adopts inflation controlling policy. Above all according to the results of IRF there is evidence of the presence of price puzzle.

Bernanke and Mihov (1998) applied a VAR methodology to measure the effects of a monetary policy shock on macroeconomic variables. They applied a new measurement for monetary policy in which they measured the monetary policy from a model of Central Bank’s operating procedures and the market for commercial bank reserves which in turn make the measurement of monetary policy more consistent than previously used instruments of monetary policy. A VAR model for different time periods of post 1965 – 1996 has been estimated for USA. A generalized method of moments is applied to the standardized VAR method and the exogenous shocks are computed with a recursive ordering of variables in which the policy variables are placed last in variable ordering. According to the IRFs results output increases in response to an expansionary monetary policy. Moreover, the IRF plots show the evidence of slower but persistent increase in the price level. However, the results differ for various measures of monetary policy indicating the results are sensitive to the measurement of monetary policy. Horváth and Rusnák (2009) analyzed based on block-restriction VAR model the effects of monetary policy shock (domestic and Euro area monetary policy shock) in Slovakia. They use output gap to avoid the problem of price puzzle in their analysis. According to the results of their model, after a domestic tightening of monetary policy the monetary policy transmission mechanism functions very well, price level declines, nominal exchange rate appreciates and output level drops. Moreover, their result indicate that the effect of European central bank monetary policy shock on price level is larger than the impact of the equivalent national bank of Slovakia monetary policy shock on price level. However, the large effect of European central bank monetary policy shock does not work on the rest of variables’ response.

There are few studies done on the effects of monetary policy shock in macroeconomic variables in Ethiopia. Temesgen (2014) analysed the effects of monetary policy shock on the economy of Ethiopia with time series data spanning form 1981/82 to 2011/12 for a closed economy by including monetary aggregate, consumer price index (CPI) as a proxy for inflation and real gross domestic product (GDP) in the model. He also analysed through which channels these shocks transmitted to the economy by applying the error correction model. His results showed the exchange rate to be an important channel through which the monetary policy transmits into the economy. Furthermore, his investigation of impulse response graphs indicate that CPI responds to shocks in real GDP and broad money, output responds to shocks in inflation and interest rate, and interest rate is unresponsive to shocks in real GDP and price level. Nuru (2013) studied the monetary transmission mechanism of monetary policy in Ethiopia. He analyses long-run equilibrium relationships, adjustment

Page 9: €¦  · Web viewPolicy makers attempt to ensure price, exchange rate and overall macroeconomic environment stability. Therefore, the most significant question in monetary policy

mechanisms and short run influences between the output, price level and transmission channels: namely, the exchange rate; the interest rate; money supply; and credit channel. The results of econometric analysis suggest that monetary policy in Ethiopia had a relatively significant influence on the real activity through the direct monetary transmission and exchange rate channel. Besides, the results of statistical tests suggest that the interest rate channel is not active.

3. Methodology 3.1. Data

Due to the nature of the study; the source of the data is secondary sources time series data. To meet the objective of the study, time series data on two policy variables and three non-policy variables is collected. In this case, a yearly time series data on money supply, real gross domestic product, GDP deflator (as a proxy to inflation), exchange rate and minimum deposit rate is collected spanning the period 1981 to 2015. The choice of the time period is entirely based on the availability of data. The relevant data for the study is collected from National Bank of Ethiopia, Ministry of Economic Development (MoFED), Central Statistical Agency (CSA), International Monetary Fund data base, World Bank, International Financial Institutions data base and other sources which are perceived to be relevant and reliable.

3.2. Econometric approach 3.2.1. Unit root test

Since most time series data, especially in economics, are non-stationary, the first step is to look at the nature of the data with appropriate statistical tests for the presence of unit roots, to avoid the problem related to spurious regression. Before we test for the presence of long run associations with the Johansson co-integration test and estimation of the vector error correction model, the well celebrated techniques in econometrics Augmented Dickey – Fuller (ADF) and Philips- Peron (PP) unit root tests are carried out to determine the order of integration for each variable. It is to make sure that the variables are integrated order zero or integrated of higher order. In other words, detect each variable included in the model weather it is stationary or non-stationary. Philips and Peron test corrects for any serial correlation and hetroscedasticity in the errors (Ut) non- parametrically by modifying the dickey fuller test statistics. The ADF test can be given as follows:

Page 10: €¦  · Web viewPolicy makers attempt to ensure price, exchange rate and overall macroeconomic environment stability. Therefore, the most significant question in monetary policy

Hypothesis is:

Ho: β= 0, implies, there exists a unit root (the series is not stationary)

Ha: β<0 ,the series is stationary (no unit root)

Where y is the vectors of variables under consideration and ε t is a white noise error term. The model in equation (2) is a random walk model without drift, equation (3) is a random walk model with drift, and equation (4) is a random walk model with drift and trend. Furthermore, the computed value will be compared with Mackinnon (1996) critical values to determine whether the series is stationary or not.

3.2.2. Lag selection

In a standard time series analysis, it is often important to include lagged values of the endogenous variables, in which it is named in typical VAR models. In this case, while including lagged values, the choice of the lag should be done carefully. Including too many lagged values of a variable in the regression model, will consume too much degree of freedom leaving the model to be estimated with fewer observations. On the other hand, including too few lagged values will also lead to specification error or decrease the estimation accuracy. Therefore, selecting the appropriate lag length using the commonly and widely applicable lag selection information criteria is important. Accordingly, in this study the lag structure of the vector error correction model specification will be determined by Akakie Information Criteria (AIC) since it controls the problems of autocorrelation and it is also advantageous for a small sample size. Based on this lag selection criterion, the lower the AIC value, the better the model fits.

3.2.3. Johansson Co-integration

Since most of the time series data has a special properties it needs special treatment. Time series variables may be non-stationary but their linear combination is stationary. In such cases, we say there is co-integration (long run relationship) between the variables. Engle and Granger (1987) were the first to formalize the idea of non-stationary variables sharing an equilibrium relation. Thus, testing for co-integration is almost mandatory.

In this study to check for the presence of long run relationship among the endogenous variables, Johansson Co-integration test is applied to test for the presence of long run relationships. The Johansen co-integration test method employs the maximum likelihood procedure to determine the presence of co-integrating vectors in the vector autoregressive system. The Johansen’s method is based on the relationship between the rank of matrix and its characteristic roots (Enders, 2010).Johansen methodology is given by the following vector autoregressive (VAR) of order p form,

Page 11: €¦  · Web viewPolicy makers attempt to ensure price, exchange rate and overall macroeconomic environment stability. Therefore, the most significant question in monetary policy

In the original work of Johansen and Juselius (1990), the model incorporates a vector non

stochastic variable ( ) orthogonal to the constant term such as seasonal dummies, ‘dummy type’ variables and/or stochastic “weakly exogenous” variable. Thus the model can be given by:

In general, most of the economic data on time series are non-stationary process and the above VAR model is expressed in its first differenced form which is given as follows:

and represent short run adjustment and long run relationship among the variables

respectively. At the same time the rank of reveals the number of linear combinations of variables which are stationary.

3.2.4. Vector error correction model

The general representation of the base line vector error correction (VEC) model with p lags, VEC (p) is specified in its reduced form as follows:

Where is a (kx1) vector of constants; is (kx1) vector of linear time trend; t= 1, ,

T; are (KxK) coefficient matrices, K being the number of endogenous variables in the

system and is a vector of endogenous

variables. The (Kx1) vector consists of reduced

form residuals ordered in their corresponding observed endogenous variables in vector , Furthermore, each residual is a mean zero white noise process that is serially uncorrelated,

i.e. . After the VEC model is estimated and passed all the critical diagnostic tests, impulse response functions and variance decomposition are estimated. Basically, impulse response functions show the effects of shocks i.e. effects of shocks of monetary policy in the case of this study on the adjustment path of the non-policy variables. On the

Page 12: €¦  · Web viewPolicy makers attempt to ensure price, exchange rate and overall macroeconomic environment stability. Therefore, the most significant question in monetary policy

other hand, to look at the contribution of each type of shock to variables in the system the variance decomposition is estimated. Both computations are useful in assessing how shocks to economic variables reverberate through a system or how the shocks to economic variables impact in the system. Impulse response functions (IRFs) and variance decompositions (VD) can be produced after a VEC model is estimated.

4. Results and discussion

Time series variables need special treatment while doing econometric analysis to bring about meaningful results. Some special properties of the time series have to be taken in to consideration before estimation of an econometric model. Preconditions for the estimation of a VEC model are: (1)that the variables under consideration must be integrated of order one, and (2) there must be a long run stable relationship among the variables. So in order to make sure the variables are integrated of order one the study carry out a unit root test.

While dealing with time series data, testing for unit root is a fundamental condition. As mentioned in the methodology part of the study, working with non-stationary variables gives rise to spurious regression (seemingly related variables) results, from which further inference is no more meaningful or leads to making the wrong inference about economic relationships. To obtain reliable and consistent results, either, non- stationary data should be transformed into stationary, or a model that treats non stationary time series data such as the VEC model should be applied. The non-stationary process has variance which varies overtime, and a mean that does not overtime return to the long run equilibrium. On the other hand, a stationary process has a mean which revert around constant long run equilibrium and constant variance which is independent of time.

Appropriate tests of stationarity should be employed on variables of interest in order to avoid problems of spurious correlation normally associated with the inclusion of non-stationary series in regression models. Two types of formal tests are conducted to examine whether the data series is stationary or not. These tests are the conventional Augmented Dickey-Fuller test (ADF) and the Phillips-Perron test (PP). These two tests allow for three options while conducting the tests; i.e., without intercept and trend, with only intercept and with both intercept and trend. The null hypothesis for the test claims that the data series under investigation has unit root. Conversely, the alternative hypothesis claims that the series is stationary. In addition, the result of the test for the variables at level and at their first difference is presented in the following table 4.1 and 4.2 respectively. We check for the stationarity condition of our variables and compare the estimated ADF and PP statistics with the Mackinnon (1991, 1996) critical values. Below, the results of ADF and PP tests are presented and both intercept and trend are included in the tests.

Page 13: €¦  · Web viewPolicy makers attempt to ensure price, exchange rate and overall macroeconomic environment stability. Therefore, the most significant question in monetary policy

4.1. Augmented Dickey Fuller unit root test

The following table summarizes the results of the augmented dickey fuller test at level and at first difference and then determines their level of integration. In the processes the test results make sure all the variables are integrated of order one which is a basic requirement for the estimation of VEC models.

In the below table 4.1 five variables are tested for the presence of unit root at level and first difference. The variables include minimum deposit rate (MDR), log of money supply (LMS), log of GDP deflator (LGDPD), log of exchange rate (LEXR) and log of real GDP (LRGDP). At first level all the variables are non-stationary, meaning that we fail to reject the null hypothesis and the variables have a unit root. At first difference, we reject the null hypothesis and accept that all the variables are stationary at 1% level of significance except for LMS in which we reject the null hypothesis of non-stationarity at 5% level of significance. The critical values employed in this study for the test of unit root are Mackinnon critical values.

Table 4. 1: Augmented dickey fuller unit root test

Variables

Level First Difference

Intercept InterceptMDR 1.52 1.742 5.304*** 5.224*** I (1)LMS 2.474 0.443 3.268** 3.568** I (1)LGDPD 1.853 0.702 4.64*** 5.145*** I (1)LEXR 0.095 1.977 3.663*** 3.615*** I (1)LRGDP 1.899 1.673 5.578*** 5.63*** I (1)

Makinnon Critical Values Intercept Intercept and Trend Significance

1% 3.605593 4.226815 ***5% 2.936942 3.536601 **

Order of Integration

Intercept and Trend

Intercept and Trend

Makinnon Critical Values

Source: Author’s computation based on data

***, **, * show significance at 1%, 5% and 10% level

These findings of ADF unit root test are robust across different lag lengths used in the estimation of the ADF equation. Thus, the results suggest that the variables are integrated of order one.

4.2. Philippe Peron unit root test result

The second unit root test result applied in this study to cross check the results of augmented dickey – fuller test result is the philips and peron unit root test. As the philips and peron (1988) indicated, philips and peron unit root test employs non parametric statistical methods

Page 14: €¦  · Web viewPolicy makers attempt to ensure price, exchange rate and overall macroeconomic environment stability. Therefore, the most significant question in monetary policy

and further corrects for the existence of serial correlation in the residuals, while ignoring the lagged changes as in the case of augmented dickey fuller test.

Table 4.2: Philippe and Peron unit root test results

Variables

Level First Difference

Intercept Intercept Intercept and TrendMDR 1.52 1.72 5.304*** 5.224*** I (1)LMS 2.474 0.443 3.268** 3.568** I (1)LGDPD 1.853 0.702 4.64*** 5.145*** I (1)LEXR 0.095 1.977 3.663*** 3.615** I (1)LRGDP 1.899 1.673 5.578*** 5.63*** I (1)

Makinnon Critical Values

Intercept Significance 1% 3.605593 4.226815 ***5% 2.936942 3.536601 **

Order of Integration

Intercept and Trend

Intercept and Trend

Makinnon Critical Values Source: Author’s computation based on data

***, **, * show significance at 1%, 5% and 10% level

According to Gujarati (2004) for a large number of observations i.e. asymptotically the phillips-peron (PP) test has the identical distribution with the augmented dickey fuller unit root test. The augmented dickey – fuller statistics have been made robust by taking in to account the serial correlation by employing the Newey–West (1987) autocorrelation and heteroskedasticity consistent covariance estimator. The PP tests correct for any serial correlation and heteroskedasticity in the errors ut non-parametrically by modifying the dickey fuller test statistics. The table above summarizes the PP unit root test results.

The test results as indicated in the above table shows that, like the ADF test for unit root, the PP test report that all the variables included in the model are integrated of order one or I(1). The results are similar to the ADF unit root result and it can be seen from the table that the null hypothesis of non-stationarity is rejected for all at 1% level of significance except for LMS in which the null hypothesis is rejected at 5% level of significance. In other words, the PP test results do not provide any significant evidence to reject the null hypothesis of non-stationary at level for all the variables. However, the first differences of the series are stationary. Therefore, all the variables included in the study are integrated of order one or I (1).

4.3. Lag Length determination result

Before the stable long run relationship among the variables is checked, one has to select an optimal lag length. It is already known that including too much lags in the estimation consumes the degree of freedom which indicates loss of power. On the other hand, including

Page 15: €¦  · Web viewPolicy makers attempt to ensure price, exchange rate and overall macroeconomic environment stability. Therefore, the most significant question in monetary policy

too few lagged values will also lead to specification errors or decrease the estimation accuracy; therefore, selecting the appropriate lag length using the commonly and wildly applicable selection information criteria is important.

Table 4. 3: Lag Length selection result

Selection Order Criteria

Sample 1983 – 2015Number of Observation: 33

Lag LL LR df P FPE AIC HQIC SBIC0 52.77 0.00023 3.50146 3.5775 3.72821 134.3 374.2 25 0.000 1.3e-09 6.32534 5.86759* 4.96488*2 164.1 59.49 25 0.000 1.1e-09* 6.61293* 5.77372 4.11875Endogenous: LRGDP, LGDPD, LMS, LEXR, MDR

Source: Author’s computation based on data

Notes:* indicates the lag order selected by the criterion, LR: sequential modified LR test statistic (each test at 5% level), FPE: Final prediction error, AIC: Akaike information criterion, SC: Schwarz information criterion, and HQ: Hannan-Quinn information criterion

These are the Log Likelihood (LL), the Akaike information criteria (AIC), the Schwarz information criteria (SIC) and the Hannan-Quinn information criteria (HIC). The optimal lag length for this study is determined by using the Akaike Information Criteria (AIC) as this method has been proven in most empirical papers to be superior to other tests when we have small sample.

The above table 4.3 shows the results and majority of selection information criteria such as LR, FPE and AIC recommend to use two lags. On the other hand HQIC and SBIC recommend using one lag in our model. However, as it is indicated in the methodology part of the study, AIC gives optimal lags when the sample size is small, which is the case in this study. Therefore, in this study, according to AIC recommendation two lags are employed.

4.4. Co-integration test result

The best way to avoid spurious regressions is the application of a co-integration method which allows the estimation of non-stationary time series with non-spurious regressions. Any long run stable equilibrium relationship among the variables included in the model, which is non-stationary variables, shows that their stochastic trends must be linked. If the variables have a relationship in the long run, that means they cannot move independently of each other. This co-movement in the long run which shows a linkage among the stochastic trend needs the variables are co-integrated (Brooks, 2002).

The co-integration technique is based on the assumption of an equilibrium relationship among the variables, which implies that two or more variables that are individually non-stationary but are integrated of the same order possess a linear combination of a one degree lower order of integration. Therefore, if all the variables are I (1) and are co-integrated, then

Page 16: €¦  · Web viewPolicy makers attempt to ensure price, exchange rate and overall macroeconomic environment stability. Therefore, the most significant question in monetary policy

their co-integrating equation would yield a composite variable of order I (0), i.e. it would be stationary. Thus, co-integration among the variables reflects the presence of long run relationship in the system.

In section 4.1, the unit root tests of ADF and PP show that all the variables included in the model are non-stationary at levels and they become stationary at first difference. An econometric analysis of non-stationary time series data is not meaningful unless the linear combination of the variables in the model results stationary series, in which, at this time we say there is long run stable equilibrium among the variables. However, the econometric analysis of non-stationary but co-integrated variables is meaningful. The test of co-integration in this section tests for the existence of such a relationship among the non-stationary variables considered in this study. The table below presents the result of Johansen co-integration test as follows.

Table 4. 4:Co-integration test result

Johansen test for Co-integration

Trend: Constant Number of Observation: 33 Lags: 2

Sample: 1983 – 2015 Max Rank Parms LL Eigen Value Trace Statistic 5% critical

0 30 128.23346 - 71.7598 68.521 39 146.51978 0.66987 35.1872* 47.212 46 154.21364 0.37268 19.7995 29.683 51 159.97467 0.29471 8.2774 15.414 54 163.44126 0.18949 1.3442 3.765 55 164.11337 0.035992

Source: Author’s computation based on data

Note: * indicates the rejection of the null hypothesis of no co-integration equation at 0.05 significant level and maximum rank: represents the number of co-integration equations for both the trace statistics.

Once all the variables included in the econometric model are non-stationary or integrated of order one as it can be seen from the results of ADF and PP unit root test which are presented in table 4.1 and 4.2, the next thing to be done is to check for the existence of long run stable equilibrium among the variables, or to check for the presence of co-integration among the variables. To test the number of co-integrating relationships among the variables of real GDP, exchange rate, minimum deposit rate, money supply and GDP deflator, a Johansen co-integration test is considered. To determine the number of co-integrating vectors, two test statistics called the maximum eigen value (λmax) and trace statistics (λtrace) are computed.

For k-endogenous variables which are non-stationary at levels, there is a possibility to find from zero to k-1 linearly independent co-integrating relations among them. The type of test statistics which is used to determine the rank of the model in this study is called the trace test.

Page 17: €¦  · Web viewPolicy makers attempt to ensure price, exchange rate and overall macroeconomic environment stability. Therefore, the most significant question in monetary policy

The trace test (λtrace) tests has the null hypothesis of r co-integrating vectors against the alternative hypothesis of k co-integrating vectors, where k is the number of endogenous variables, for r=0,1,2…,k-1. The trace test shows that the null hypothesis of r=0 co-integrating relation is rejected and the alternative r≥0 co-integrating equations is accepted. This means that there is one co-integrating equations because the null hypothesis of r≤ 1 could not be rejected in the next step. Therefore, the trace statistic as shown in table 4.4 confirms that there is one co-integrating equation among the variables.

4.5. Vector Error Correction model estimation results

In time series econometrics the vector error correction model has proved to be very powerful and popular organizing principle and has been applied widely. The vector error correction model has encouraged a wide range of statistical developments, especially the concept of co-integration by Engle and Granger (1987). The vector error correction model is applied on non-stationary data where there is a long run co-integration among the variables. The vector error correction model is interpreted as the method of adjusting a policy instrument to maintain a target variable close to its desired value. There are some reasons to choose the vector error correction model over the vector autoregressive model and the primary motivation is to avoid the potential misspecification bias which is natural in the VAR in first difference. The VAR which is incapable of explaining long term relations is also deficient in discovering short term relations in the presence of co-integration.

The results of the vector error correction model are sensitive to the ordering of variables. Therefore, following the empirical literature, such as Forni and Gambetti (2010); Bernanke et al. (2005) and Bjørnland (2008), this study assumes a recursive structure of ordering for the small open Ethiopian economy in which policy variables are ordered at last. The same ordering of variables is applied while estimating the impulse response graphs.

The short-run dynamic model (vector error correction mechanism) includes simultaneous current effects, short-run adjustment effects to lagged changes to the variable and previous equilibrium errors in the system. Identification of the short-run structural equations often requires that the residuals are uncorrelated, or at least not significantly correlated since residual covariance matrix plays an important role in the identification of the short-run structure. When the residuals of a short-run structural model are approximately uncorrelated, it might be possible to label them as estimated shocks. In contrast, if the residuals are correlated it will not be easy to make impulse response analysis since it is not clear which shocks leads to which effect (Reade, 2006). The following table summarizes the results of the estimated vector error correction model.

In the error correction model result for exchange rate, the ECT is negative and significant at 10% level of significance and the speed of adjustment towards equilibrium is 12.5% per annum. Once disequilibrium occurs, the adjustment process takes a lot of time with 12.5% of the deviation from the equilibrium being corrected within one year. This shows that the

Page 18: €¦  · Web viewPolicy makers attempt to ensure price, exchange rate and overall macroeconomic environment stability. Therefore, the most significant question in monetary policy

exchange rate responds to temporary disequilibrium. It takes around 11 years to correct for disequilibrium. LMDR and LEXR are significant at 1% level of significance while the LMS is significant at 10% level of significance and the constant is significant at 5% level of significance. On the other hand, Real GDP (LRGDP) is not significant at any level of significance.

Table 4. 5: Vector error correction model estimates result

Variable DLEXR DLRGDP DLGDPD DLMS DLMDR

L._Ce1 (ECT)

-0.125*

(0.0673)

0.305

(0.190)

-0.106**

(0.0532)

-0.155***

(0.0301)

-0.226*

(0.115)

LD.LEXG

0.436***

(0.159)

-0.769*

(0.449)

0.160

(0.126)

0.254***

(0.0711)

0.199

(0.273)

LD.LRGDP

-0.0130

(0.0707)

-0.0405

(0.222)

-0.109*

(0.0621)

-0.0415

(0.0352)

0.00407

(0.135)

LD.LGDPD

0.0705

(0.255)

1.653**

(0.719)

-0.138

(0.201)

-0.125

(0.114)

0.0685

(0.437)

LD.LMS

-0.802*

(0.425)

-0.893

(10199)

-0.00548

(0.335)

-0.0514

(0.190)

0.0480

(0.729)

LD.LMDR

0.357***

(0.111)

0.0365

(0.314)

-0.0125

(0.0879)

-0.00149

(0.0497)

-0.0849

(0.191)Constant 0.122**

(0.0539)

0.0774

(0.152)

0.0653

(0.0426)

0.108***

(0.0241)

-0.0673

(0.0925)Observation 33 33 33 33 33

Standard Errors in Parentheses

Significance Level: *** p<0.01, ** p<0.05, * p<0.1

Source: Author’s computation based on data

In the error correction model result for real GDP, the positive and insignificant ECT, while it is between zero and one, indicates that real GDP does not respond to any disequilibrium. That is, real GDP does not appear to respond to any deviations from the equilibrium. This again shows that the ECT is explosive and not reasonable. In the meantime, LGDPD and LEXG are significant at 5% and 1% level of significance level. LMDR, LMS, LRGDP and the constant are insignificant at any level of significance.

Page 19: €¦  · Web viewPolicy makers attempt to ensure price, exchange rate and overall macroeconomic environment stability. Therefore, the most significant question in monetary policy

In the error correction model result for GDP deflator, the negative and significant ECT coefficient for GDP deflator indicates that GDP deflator responds to temporary disequilibrium. The coefficient for ECT is significant at 5% level of significance .The speed of adjustment towards equilibrium is 10.6 % annually and this shows that every year 10.6% of the deviation from equilibrium is corrected. Once a shock happens it takes GDP deflator (proxy for inflation) to return to equilibrium around 10 years. LRGDP is significant at 1% level of significance while LMDR, LMS, LGDPD, LEXG and constant are insignificant at any level of significance level.

In the error correction model result for money supply, the ECT is negative and highly significant at 1% level of significance. This shows that money supply responds to temporary disequilibrium or any deviations from the equilibrium. The speed of adjustment is15.5% per annum, i.e. once the disequilibrium happens 15.5% of the deviation from the equilibrium is corrected in one year in the process towards equilibrium. This takes almost the process towards equilibrium 6 years. The coefficient for LEXG and constant is significant at 1% level of significance while the rest LMDR, LGDPD, LMS and LRGDP are insignificant at any level of significance.

In the error correction model result for minimum deposit rate, the ECT for minimum deposit rate is negative and significant at 1% level of significance and it indicates that minimum deposit rate responds to any temporary disequilibrium or deviations from the equilibrium. The speed of adjustment towards equilibrium is 22.6% per annum which shows 22.6% of the deviation from equilibrium is corrected within one year and this it takes to return to equilibrium almost around 4 years. The coefficients for LRGDP, LMDR, LMS, LGDPD, LEXG and the constant are all insignificant at any level of significance.

Above all the negative and significant ECT coefficients in the model above indicate that exchange rate, GDP deflator, money supply and minimum deposit rate respond to a temporary disequilibrium. The real GDP does not appear to respond to disequilibrium; the t-ratio on ECT is insignificant.

4.6. Diagnostic test

In this section testing of the robustness of the model is performed using the diagnostic test. After estimation is done, it is a must to check whether the model has achieved the desired properties. In this study, various diagnostic checks are performed. Model stability test, residual autocorrelation test and normality test are carried out.

4.6.1. Stability test result

Page 20: €¦  · Web viewPolicy makers attempt to ensure price, exchange rate and overall macroeconomic environment stability. Therefore, the most significant question in monetary policy

After the estimation of vector error correction model the structural stability test is performed to examine the stability of the model throughout the entire study period. The stability of the model and post estimation diagnostics could affect the validity of the estimated model; therefore, it should be tested before preceding it further. Appendix A3 presents the companion matrix with all the roots of the characteristics polynomial and their corresponding modulus. Furthermore, the stability test is executed based on the estimates of the vector error correction model. A stable model is a precondition for good model and to make further analysis such as the impulse response graphs. The graph in Appendix A3 summarizes the test for stability of the vector error correction model.

The graph in appendix A3 summarizes the stability test result and the result indicates that all the points are inside the circle. If the points were out of the circle we can conclude that the vector error correction model is not stable during the study period. But in our case all the points are inside the circle and the test statistics indicate that the parsimonious equations were stable in the estimation period. Thus, the parameters of the estimated vector error correction model do not suffer from any structural instability over the period of study. Therefore, the results reveal that there is no structural instability in the model during the sample period. Explained in a different way the figure indicated that all characteristic roots of the polynomial lie inside the unit circle. As a result of the report under appendix A3 suggests that the model under consideration satisfies the stability condition.

4.6.2. Residual autocorrelation test

Residual autocorrelation test is performed based on the estimated vector error correction model which is one of the diagnostic tests after estimation of a model. The table below presents the results of the lagrange multiplier (LM) test for residual serial correlation with a null hypothesis of no problem of autocorrelation in the estimated model. This decision is based on the probability values derived from the test.

Table 4.6: Residual autocorrelation test result

Serial Correlation TestLag Chi2 df Prob> Chi2

1 15.1283 25 0.938332 30.4937 25 0.206333 21.1061 25 0.686684 35.7185 25 0.075985 8.4055 25 0.99922Ho: No autocorrelation at Lag Order Source: Author’s computation based on data

As it can be seen from the above table, in the test for serial correlation or autocorrelation of the error term the Lagrange multiplier test fail to reject the null hypothesis of no autocorrelation at any lag order at 5% level of significance. This indicates that there is no serial dependency among the errors at any lag order. The vector error correction model

Page 21: €¦  · Web viewPolicy makers attempt to ensure price, exchange rate and overall macroeconomic environment stability. Therefore, the most significant question in monetary policy

estimated above did not suffer from the serial correlation among the error terms as it is shown in the above table in which the researcher accepts the null hypothesis.

4.6.3. Normality test

One of the diagnostic tests after estimation of an econometric model is the test for the normality of residuals after estimation. As in most cases normality test is performed after an estimated model. Based on the above estimated vector error correction model, normality test for residuals is executed. In the table below, the results of the Jarque – Bera test for normality of residuals are presented. The null hypothesis for the Jarque – Bera test is that the residuals are normally distributed. The decision is based on the probability value obtained from the Jarque – Bera test result.

Table 4. 7: Jarque Bera normality test result

Jarque – Bera TestEquation Chi2 df Prob>Chi2

D_LEXR 18.023 2 0.00012D_LRGDP 220.017 2 0.00000D_LGDPD 0.697 2 0.70583D_LMS 0.300 2 0.86079D_LMR 24.506 2 0.00000All 263.543 2 0.00000Source: Author’s computation based on data

The results indicate that we can strongly reject the null hypothesis of normally distributed errors at 1% level of significance. Most of the errors are both skewed and kurtotic. However, non-normality of the data is an endemic problem in a small sample given that the asymptotic property of any sample data is normally distributed. The estimated vector error correction model suffers from the non-normality of residual.

4.7. Impulse response and variance decomposition

To investigate the impulse response and variance decomposition, the first step is to check the stability of the vector error correction model. The stability test (presented in appendix A3) shows that the vector error correction model is stable. Hence it is possible to undertake impulse response and variance decomposition analysis. In this section the impulse response and variance decomposition based on the vector error correction model is analysed. First the impulse response graphs are discussed and later the variance decomposition for LGDP and LGDPD are computed and discussed.

4.7.1. Impulse response analysis

Page 22: €¦  · Web viewPolicy makers attempt to ensure price, exchange rate and overall macroeconomic environment stability. Therefore, the most significant question in monetary policy

Impulse response analysis is a method used to analyze how a one unit standard shocks to a policy variable, affect the key non policy variables in the study. Impulse response variables are estimated after vector error correction model estimation and they show changes in each variable due to the shock in other variables in the study by taking in to account all the interactions between the variables. The results of impulse responses function for stationary and non-stationary data are quite different.

The impulse response functions for stationary data estimated by vector autoregressive model die out overtime and are expected to converge at a certain point. Impulse response functions for co-integrated non-stationary data estimated with vector error correction models, however, do not always die out overtime. Each variable which is stationary has a constant mean and variance and the effect of a shock to any one of the variables must die out overtime so that the variables can revert to its mean. In contrast, the integrated order one variables modeled by vector error correction models don’t have constant means and variances, and some shocks will not die out overtime. These two situations give rise to two terms with regard to shocks. When the effect of a shock dies out overtime, the shock is called transitory. On the other hand, when the shock does not die out overtime, the shock is called permanent.

In the following two sections, impulse response functions are reported for a horizon of ten years (Fig: 4.1 and Fig: 4.2), which enables us to trace out the response of output, exchange rate, interest rate and inflation to a shock in policy variables. The shock is represented by one standard deviation of the error term in the underlying structural model for the variable. One main problem of the impulse response functions is the sensitivity with respect to the variables ordering in the system. As mentioned while estimating the vector error correction model, in these sections a recursive ordering of variables is employed in which policy variables are ordered at last. This implies that macroeconomic aggregates do not respond contemporaneously to monetary policy innovations but monetary policy might react toward any news from macro aggregates within the period. The recursive structure assumes that variables appearing first contemporaneously influence the latter variables but not vice versa.

4.7.1.1. Response of macroeconomic aggregates to a shock in money supply

The graph below shows the computed impulse response graphs to analyze the response of exchange rate, GDP deflator, minimum deposit rate and real GDP to a one-standard deviation shock to money supply.

Figure 4. 1: Impulse response graph for unit standard shock in money supply

Page 23: €¦  · Web viewPolicy makers attempt to ensure price, exchange rate and overall macroeconomic environment stability. Therefore, the most significant question in monetary policy

-.04

-.02

0

.02

-.04

-.02

0

.02

0 5 10 0 5 10

vec1, LMS, LEXR vec1, LMS, LGDPD

vec1, LMS, LMDR vec1, LMS, LRGDP

stepGraphs by irfname, impulse variable, and response variable

Source: Author’s computation based on data

The early response of exchange rate to a one-standard unit shock in money supply is negative and later after three years it turns out to be positive and dies out. The graph above indicates that an orthogonalized shock to money supply has a permanent effect on exchange rate. The response of inflation to a one-standard unit shock on money supply has mixed results. Initially a shock to money supply does not have too much impact on inflation; however, after four years the impact of orthogonalized shock to money supply produced positive effect. Moreover, a shock to monetary policy has a permanent effect on inflation.

The initial effect of a shock of monetary policy on interest rate is small but fluctuates around zero. After three years an orthogonalized shock to money supply is accompanied by an increase in interest rate which shows the existence of liquidity puzzle. Furthermore, a shock to monetary policy has a permanent effect on interest rate. The effect of a one-standard unit shock of money supply on real GDP is negative and permanent throughout the time horizon. The response of real GDP to a shock on monetary policy is negative and fluctuates below zero in the time horizon.

Page 24: €¦  · Web viewPolicy makers attempt to ensure price, exchange rate and overall macroeconomic environment stability. Therefore, the most significant question in monetary policy

4.7.1.2. Response of macroeconomic aggregates to a shock in interest rate

The graphs below show impulse response graphs to analyse the response of exchange rate, GDP deflator and real GDP to one-standard unit shock in interest rate. Looking at the response of exchange rate to a one-standard unit shock in interest rate, it can be seen that the early effect of shock to interest rate on exchange rate is positive but after five years the effect becomes negative. The effect of interest rate shocks on exchange rate is permanent. Moreover, in the short run there is evidence of the exchange rate puzzle as an increase to interest rate is coupled with depreciation rather than appreciation of the local currency (i.e. Birr). The response of inflation to a one-standard unit shock to interest rate is negative throughout the time horizone.The impact of orthogonalized shock to interest rate is permanent. Moreover, the analysis confirms there is no price puzzle as the contractionary monetary policy through positive innovations in interest rate leads to a decrease in price level.

Figure 4. 2: Impulse response graph for unit standard shock in minimum deposit rate

-.1

0

.1

.2

-.1

0

.1

.2

0 5 10

0 5 10

vec1, LMDR, LEXR vec1, LMDR, LGDPD

vec1, LMDR, LRGDP

stepGraphs by irfname, impulse variable, and response variable

Source: Author’s computation based on data

The response of real GDP to a one-standrad unit shock in interest rate is postive and increasing throughout the time horizone. The impact of orthogonalized shock to interest rate is permanent throughout the time horizon.

Page 25: €¦  · Web viewPolicy makers attempt to ensure price, exchange rate and overall macroeconomic environment stability. Therefore, the most significant question in monetary policy

4.7.2. Variance decomposition

To look at the strength of each monetary policy transmission channel, variance decomposition based on the estimated vector error correction model is estimated for output and inflation. Variance decomposition represents the proportion of movements in one variable that are due to errors in own shocks and to each other variables in the system. Fundamentally it provides information on how important is each variable in explaining variations in the variable in question in the system. For undertaking variance decomposition analysis, we enter the variables with target variables first followed by policy variables: Exchange Rate (LEXR), Real GDP (LRGDP), GDP deflator (LGDPD), Monetary Aggregate (LMS) and Minimum Deposit Rate (MDR).

4.7.2.1. Variance decomposition for output (LRGDP)

Variance decomposition for output is estimated for the time horizon of 10 years in the following table. The fluctuation in output is explained from different sources in the vector error correction system such as shocks in exchange rate, real GDP, GDP deflator, money supply and minimum deposit rate. The following table 4.8 summarizes the results of the variance decomposition for output. The results of variance decomposition for output are reported in the below table 4.8 which indicates that the interest rate channel is the most important amongst the three channels both in the short run and long run. The next important channel is exchange rate, while money supply has little impact in explaining the variation in output throughout the time horizon.

Table 4. 8: Variance decomposition for output

Variance decomposition for output (LRGDP)Perio S.E LEXR LRGDP LGDPD LMS MDR1  0.3330  3.7233  96.276  0.0000  0.0000  0.00002  0.4653  5.9430  84.106  6.8443  0.4679  2.63763  0.5653  7.2655  76.098  7.7813  0.4528  8.40144  0.6525  6.7558  75.120  6.0172  0.3509  11.7555  0.7288  5.4985  73.872  5.0990  0.7536  14.7766  0.8083  5.5598  70.531  4.5917  1.6533  17.6637  0.8918  5.5495  68.520  4.0172  2.0699  19.8428  0.9759  5.7469  66.80  3.6170  2.3640  21.4689  1.0578  6.4004  64.546  3.4572  2.7770  22.81810  1.1361  6.9682  62.774  3.3317  3.0775  23.847

Source: Author’s computation based on data

According to the variance decomposition table, after one year exchange rate shocks account for around 4 % of the fluctuations in output while others such as money supply and interest rate contributing 0% in fluctuation of output. After two years exchange rate account 6% of fluctuation in output followed by the shock in interest rate which account 3% of the variation. Impact of money supply shocks is dampened and they account for 3% fluctuation in output

Page 26: €¦  · Web viewPolicy makers attempt to ensure price, exchange rate and overall macroeconomic environment stability. Therefore, the most significant question in monetary policy

after 10 years, while the shock in interest rate steadily rose and account for 24% fluctuations in output after 10 years, with own shocks of output accounting for only 63% after 10 years which shows slight decline from the first year impact which accounted around 96%.

Moreover, the variance decomposition result indicates that price has insignificant explanatory power of predicting the movement in real output, it accounts almost 4% of the variation after 10 years. The interest rate channel‘s is amongst all the channels is the strongest and important, as it accounted for large part of output variability throughout the time horizon.

4.7.2.2. Variance decomposition for inflation (LGDPD)

Variance decomposition for inflation is estimated for the time horizon of 10 years in the following table. The fluctuation in inflation is explained from different sources in the vector error correction system such as shocks in exchange rate, real GDP, GDP deflator, money supply and minimum deposit rate. The following table 4.9 summarizes the results of the variance decomposition for inflation.

Table 4. 9: Variance decomposition for inflation

Variance decomposition for inflation (LGDPD)Perio S.E LEXR LRGDP LGDPD LMS MDR1  0.0919  0.4057  3.7370  95.857  0.0000  0.00002  0.1427  0.6959  9.9866  87.778  0.5800  0.95913  0.1734  0.8963  7.9766  87.762  2.1363  1.22794  0.2103  1.7127  6.7573  86.050  2.4352  3.04375  0.2530  3.8477  6.6178  82.003  2.8116  4.71956  0.2939  8.3875  6.0923  75.458  3.9944  6.06737  0.3341  12.565  5.7499  69.388  4.8855  7.41018  0.3734  15.468  5.8878  64.768  5.3034  8.57119  0.4110  18.240  5.9617  60.550  5.7513  9.495510  0.4466  20.755  5.9251  56.870  6.1904  10.257

Source: Author’s computation based on data

Regarding the variance decomposition of log GDP deflator, the result reveals that the exchange rate channel is the dominant one amongst all the channels throughout the five year period. The effect of exchange rate on price is relatively small within first five years-ahead of the forecast horizon. For example, exchange rate explains only less than 2 percent of variability in inflation for the first four year-ahead of the forecast period. However, the effect accumulates over time. By 10 years ahead, exchange rate accounts for about 21 percent of the variability in inflation.

Similarly, shocks in monetary aggregate after two year account only for 2.8 percent of variability in prices. However, its importance grows over the medium term to 4 percent within six years period and increased again after six years to 6%.

Innovations in interest rate are equally important over the time horizon. Initially, shocks in interest rate account for 0.00 percent of variability in prices. However, over the medium term,

Page 27: €¦  · Web viewPolicy makers attempt to ensure price, exchange rate and overall macroeconomic environment stability. Therefore, the most significant question in monetary policy

the influence of interest rate increases to reach about 5 percent and thereafter reaches 10% in 10 years’ time. Moreover, the variance decomposition for GDP deflator (a proxy for inflation) indicates that interest rate shocks have little significant effect on inflation. Looking at the exchange rate channel, it shows strong impact over the time horizon which shows that exchange rate is the main monetary policy transmission mechanism channel. To look at all the variance decomposing of the variables included in the model see Appendix A2.

5. Conclusion

The study empirically analyzes the reaction of macroeconomic aggregates, such as real GDP, GDP deflator (proxy for inflation) and exchange rate to shocks to monetary policy measure by money supply or interest rate in Ethiopia for the duration of the 1981 to 2015. The findings of the stationary test results in the study indicate that all the variables included in the study are non-stationary at level and stationary at first difference i.e. they are integrated of order one or I (1). The lag length was determined based on the AIC and the optimal lag length selected by the AIC is lag two. To test the number of co-integrating relationships among the variables of real GDP, exchange rate, minimum deposit rate, money supply and GDP deflator, a Johansen co-integration is applied and the results of the Johansen co-integration show that there is one co-integrating equation among the variables. Furthermore, the study applied recursive ordering of variables in estimating the vector error correction model and in estimating the impulse response graphs.

The results of the vector error correction model show negative and significant ECT coefficients for the models of LEXR, LGDPD, LMS and LMDR and they indicate that exchange rate, GDP deflator, money supply and minimum deposit rate respond to a temporary disequilibrium. On the other hand, the positive and insignificant ECT for real GDP indicate that real GDP does not appear to respond to disequilibrium. Moreover, the estimated vector error correction model is stable and there is no residual autocorrelation in the estimated model. However, the model suffers from the non-normality of the data.

The graphs of the IRF provide the evidence of exchange rate puzzle when interest rate is used as a monetary policy. Furthermore, there is no evidence of price puzzle when interest rate is used as a measure of monetary policy. In contrast, when money supply is used as a measure of monetary policy, no evidence of exchange rate puzzle is witnessed. This finding is in line with the Dornbusch (1976) overshooting hypothesis. However, there is evidence of liquidity puzzle when money supply is used as a monetary policy measure.

Although the long run and short run analysis explain the relationship between monetary policy variables and goal variables, the relative strength of the monetary transmission mechanism is assessed by the variance decomposition. The results of variance decomposition indicate that interest rate and exchange rate channels are strong monetary policy transmission channels both in the short run and long run in transmitting monetary policy effects to the real sector.

Page 28: €¦  · Web viewPolicy makers attempt to ensure price, exchange rate and overall macroeconomic environment stability. Therefore, the most significant question in monetary policy

ReferenceAhmed, S. and Ansari, M. (2008), “Does money matter? Evidence from vector error-correction for Mexico”, The Journal of Developing Areas, Vol. 41 No. 1, pp. 185-202.

Benti, T., 2014. Impacts of monetary policy shocks on the Ethiopian economy (Doctoral dissertation, Jimma University).

Berument, Hakan (2007), Measuring monetary policy for a small open economy: Turkey, Journal of Macroeconomics 29, 411-430.

Bernanke, Ben, and IlianMihov (1998), Measuring monetary policy, The Quarterly Journal of Economics (3) 870-902.

Bernanke, B., Boivin, J. and Eliasz, P. (2005), “Measuring the effects of monetary policy: a factor-augmented vector autoregressive (FAVAR) approach”, Quarterly Journal ofEconomics, Vol. 120 No. 1, pp. 387-422

Bjørnland, H. (2008), “Monetary policy and exchange rate interactions in a small open economy”,Scandinavian Journal of Economics, Vol. 110 No. 1, pp. 197-221.

Brooks, C. (2002). Introductory econometrics for finance. United Kingdom, Cambridge University Press.

Christiano, Lawrence, Martin Eichenbaum and Charles Evans (2002), Nominal rigidities and the dynamic effects of a shock to monetary policy, Federal Reserve Bank of Chicago.

Christiano, Lawrence, Martin Eichenbaum, and Charles Evans (1999), Monetary policy shocks: what have we learned and to what end? In Woodford, Michael and John 128 Taylor (Eds), The Handbook of Macroeconomics Vol. 1 North-Holland, Amsterdam, pp. 65-148.

Clarida, R., Gali, J. and Gertler, M., 1999. The science of monetary policy: a new Keynesian perspective (No. w7147). National bureau of economic research.

Cooley, T.F. and Hansen, G.D., 1989. The inflation tax in a real business cycle model. The American Economic Review, pp.733-748.

Cushman, David and Tao Zha (1997), Identifying monetary policy in a small open economy under flexible exchange rates, Journal of Monetary Economics 39, 433- 448.

Dickey, D. and W. Fuller (1981), "The likelihood Ratio Statistics for Autoregressive Time Series with a Unit Root.", Econometrica, 49: 1052-72.

Dornbusch, R. (1976), “Expectations and exchange rate dynamics”, The Journal of PoliticalEconomy, Vol. 84 No. 6, pp. 1161-1176.

Page 29: €¦  · Web viewPolicy makers attempt to ensure price, exchange rate and overall macroeconomic environment stability. Therefore, the most significant question in monetary policy

Eichenbaum, M. and Evans, C. (1995), “Some empirical evidence on the effects of shocks to monetary policy on exchange rates”, Quarterly Journal of Economics, Vol. 110 No. 4, pp. 975-1009.

Engle, R.F. and Granger, Clive W. J. (1987a), “Co-integration and Error Correction: Representation, Estimation, and Testing.” Economterica, Vol. 55, pp. 251-276.

Enders, W. (2010), Applied Econometric Time Series, 3rd ed.

Forni, M. and Gambetti, L. (2010), “The dynamic effects of monetary policy: a structural factor model approach”, Journal of Monetary Economics, Vol. 57 No. 2, pp. 203-216.

Friedman, M. (1968), “Money and business cycle”, American Economic Review, Vol. 58 No. 1, pp. 1-17.

Friedman, M. (1994). Money Mischief. New York: Harcourt Brace & Company.

Galí, J., 2015. Monetary policy, inflation, and the business cycle: an introduction to the new Keynesian framework and its applications. Princeton University Press.

Horváth, R. and Rusnák, M. (2009): “How Important Are Foreign Shocks in a Small Open Economy? The Case of Slovakia”, Global Economy Journal, Vol. 9, No. 1.

Johansen, S. and Juselius, K., 1990. Maximum likelihood estimation and inference on co-integration—with applications to the demand for money. Oxford Bulletin of Economics and statistics, 52(2), pp.169-210.

Jang, K. and Ogaki, M. (2004), “The effects of monetary policy shocks on exchange rates: a structural vector error correction model approach”, Journal of the Japanese andInternational Economies, Vol. 18 No. 1, pp. 99-114.

Kandil, M., 2014. Borsa _ Istanbul Review On the effects of monetary policy shocks in developing countries *. Borsa istanbul Review, 14(2), pp.104–118.

Kahn, Michael, Shmuel Kandel and OdedSarig (2002), Real and nominal effects of central bank monetary policy, Journal of Monetary Economics. (49) 1493-1519.

Kim, S. and Roubini, N. (2000), “Exchange rate anomalies in the industrial countries: a solution with a structural VAR approach”, Journal of Monetary Economics, Vol. 45 No. 3, pp. 561-586.

Livingston, I. and Ord, H.W. (1975). Economics for Eastern Africa. Landon: Heinemann Educational Books Ltd.

McCornel, Campbell R. (1984). Economics. New York: McGraw-Hill Book Company.

MacKinnon, J.G., 1996. Numerical distribution functions for unit root and co-integration tests. Journal of applied econometrics, pp.601-618.

Page 30: €¦  · Web viewPolicy makers attempt to ensure price, exchange rate and overall macroeconomic environment stability. Therefore, the most significant question in monetary policy

MacKinnon, J.G., 1991. Critical Values for Co-integration Tests, Chapter 13 in Long-run Economic Relationships: Readings in Co-integration, EdRF Engle and CWJ Granger.

Mohammed, N.H., 2013. An Empirical Investigation on Monetary Policy Transmission Mechanism in Ethiopia.

Mugume, A., 2011. Bank of Uganda’s Monetary Policy: What it can do and what it cannot. Mineo, Bank of Uganda.

National Bank of Ethiopia Annual and Quarterly reports various issues (1990-2011), Ethiopia.

N Gujarati, D., 2004. Basic econometrics. The McGraw − Hill.

Phillips, P.C. and Perron, P., 1988. Testing for a unit root in time series regression. Biometrika, 75(2), pp.335-346.

Prescott, E.C., 1986, September. Theory ahead of business-cycle measurement. In Carnegie-Rochester conference series on public policy (Vol. 25, pp. 11-44). North-Holland.

Reade J. (2006). The Co integrated VAR methodology .Summer school, University of Copenhagen

Smets, f., and R. Wouters.2003. “An Estimated Dynamic Stochastic General Equilibrium Model of the Euro Area.” Journal of the European Economic Association, 1(5): 1123-1175.

Smets, f., and R. Wouters. 2007. “Shocks and Frictions in US Business Cycles: a Bayesian DSGE Approach.” American Economic Review, 97(3): 586-606.

Sims, C. (1992), “Interpreting the macroeconomic time series facts: the effects of monetary policy”, European Economic Review, Vol. 36 No. 5, pp. 975-1000.

Stock, J.H. and Watson, M.W., 2002. Macroeconomic forecasting using diffusion indexes. Journal of Business & Economic Statistics, 20(2), pp.147-162.

Wong, K. (2000), “Variability in the effects of monetary policy on economic activity”, Journal ofMoney, Credit and Banking, Vol. 32 No. 2, pp. 179-198.

Walsh, C.E., 2010. Monetary Theory and Policy, Volume 1 of MIT Press Books.

AppendixAppendix A1: The Long Run Vector Autoregressive Model Estimation

Page 31: €¦  · Web viewPolicy makers attempt to ensure price, exchange rate and overall macroeconomic environment stability. Therefore, the most significant question in monetary policy

Variable LEXR LRGDP LGDPD LMS LMDR

L.EXR 0.984***(0.144)

0.161(0.434)

0.0138(0.125)

0.161**(0.0774)

-0.0111(0.0179)

L2.LEXR -0.232*(0.132)

0.341(0.396)

-0.115(0.114)

-0.266***(0.0707)

-0.0162(0.0164)

L.LRGDP 0.0411(0.0553)

0.748***(0.166)

-0.0653(0.0477)

0.00648(0.0296)

0.000195(0.00686)

L2.LRGDP 0.0700(0.0611)

-0.178(0.184)

0.115**(0.0528)

0.0364(0.0327)

0.00168(0.00758)

L.LGDPD 0.211(0.195)

0.967*(0.585)

0.779***(0.168)

0.136(0.104)

0.0158(0.0242)

L2.LGDPD 0.0431(0.199)

-2.238***(0.598)

0.000662(0.172)

0.107(0.107)

-0.0131(0.0247)

L.LMS -0.692**(0.323)

-0.687(0.971)

0.195(0.279)

0.895***(0.173)

-0.00180(0.0401)

L2.LMS 0.713**(0.315)

1.142(0.947)

0.0234(0.272)

0.0184(0.169)

0.0157(0.0391)

L.MDR 5.859***(1.438)

1.096(4.322)

-0.981(1.242)

-1.573**(0.771)

0.676***(0.179)

L2.MDR -4.724***(1.423)

-2.860(4.278)

0.492(1.229)

-0.452(0.763)

0.191(0.177)

Constant -1.383**(0.557)

2.521(1.675)

-1.391***(0.481)

0.0860(0.299)

-0.100(0.0692)

Observation 33 33 33 33 33

Standard Errors in Parentheses

Significance Level: *** p<0.01, ** p<0.05, * p<0.1

Source: Author’s computation based on data

Appendix A2: Variance Decomposition

Page 32: €¦  · Web viewPolicy makers attempt to ensure price, exchange rate and overall macroeconomic environment stability. Therefore, the most significant question in monetary policy

Variance Decomposition for Exchange Rate (LEXR)

Period S.E LEXR LRGDP LGDPD LMS MDR1  0.1055  100.00  0.0000  0.0000  0.0000  0.00002  0.2181  88.325  1.3848  0.4552  3.5334  6.30073  0.2985  84.676  2.2035  3.1018  5.3510  4.66724  0.3437  81.233  3.0779  7.0552  4.7772  3.85635  0.3738  77.926  4.0611  10.182  4.1926  3.63736  0.3993  75.206  4.3350  13.324  3.8595  3.27497  0.4195  72.681  4.3529  16.421  3.5739  2.96968  0.4338  70.378  4.5380  18.932  3.3417  2.80969  0.4461  68.234  4.6711  21.192  3.1624  2.739010  0.4574  66.233  4.6618  23.350  3.0080  2.7462Source: Author’s computation based on data

Variance Decomposition for Minimum Deposit Rate (MDR)

Period S.E LEXR LRGDP LGDPD LMS MDR1  0.0128  7.4996  2.5446  3.1653  5.1760  81.6142  0.0165  8.3739  4.4923  3.4228  8.2619  75.4483  0.0183  6.8753  5.8604  10.256  7.4873  69.5204  0.0203  6.8523  6.1774  16.755  6.1669  64.0475  0.0221  9.7620  6.1084  21.406  5.1709  57.5516  0.0242  15.907  5.4566  24.953  4.4299  49.2527  0.0267  24.225  4.6704  25.965  4.0811  41.0578  0.0292  32.275  3.9696  25.522  3.9220  34.3109  0.0318  39.009  3.3701  24.777  3.8181  29.02410  0.0344  44.726  2.8887  23.721  3.8110  24.851Source: Author’s computation based on data

Variance Decomposition for Money Supply (LMS)

Period S.E LEXR LRGDP LGDPD LMS MDR1  0.0527  22.048  39.266  12.047  26.636  0.00002  0.0802  13.106  41.842  9.4831  25.856  9.71143  0.1117  10.276  37.172  13.234  19.330  19.9854  0.1550  17.978  27.194  14.256  16.262  24.3085  0.2104  29.241  18.980  11.875  15.612  24.2906  0.2679  36.084  15.244  9.9200  14.912  23.8397  0.3239  40.008  13.630  8.4326  14.214  23.7138  0.3785  42.920  12.484  7.1239  13.835  23.6359  0.4301  45.038  11.680  6.1283  13.592  23.56010  0.4778  46.453  11.240  5.423788  13.359  23.522Source: Author’s computation based on data

Appendix A3: Stability Test Result

Page 33: €¦  · Web viewPolicy makers attempt to ensure price, exchange rate and overall macroeconomic environment stability. Therefore, the most significant question in monetary policy

0.0000.0000.0000.0000.253

0.599

0.599

0.728

0.849

0.849-1

-.50

.51

Imag

inar

y

-1 -.5 0 .5 1Real

The VECM specification imposes 4 unit moduliPoints labeled with their distances from the unit circle

Roots of the companion matrix

Source: Author’s computation based on data