Specific inflation developments

95. The analysis above suggested that inflation in Eritrea was driven by a combination of general domestic and external factors, including, in particular, agricultural production, fiscal policies, and exchange rate movements. To shed more light on them, inflation developments are examined in further detail by disaggregating the CPI index. Figure V.1 (upper panel) shows that food and nonfood prices diverged quite strongly during the drought in 1997, which was followed by a significant increase in food prices while the rest of the economy enjoyed relative strong macroeconomic stability. Diverging trends persisted for most of the 1998-2000 war. However, the differences narrowed from 2000 and onwards, when nonfood inflation increased strongly in parallel with war financing.

Eritrea: Inflation and its Subcomponents, January 1997-December 2002

(Percentage change from previous year)

Citation: IMF Staff Country Reports 2003, 166; 10.5089/9781451811971.002.A005

Sources: Source: Bank of Eritrea; Eritrea statistical and evaluation office, and IMF staff estimates.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Eritrea: Inflation and its Subcomponents, January 1997-December 2002

(Percentage change from previous year)

Citation: IMF Staff Country Reports 2003, 166; 10.5089/9781451811971.002.A005

Sources: Source: Bank of Eritrea; Eritrea statistical and evaluation office, and IMF staff estimates.

• Download Figure
• Download figure as PowerPoint slide

Eritrea: Inflation and its Subcomponents, January 1997-December 2002

(Percentage change from previous year)

Citation: IMF Staff Country Reports 2003, 166; 10.5089/9781451811971.002.A005

Sources: Source: Bank of Eritrea; Eritrea statistical and evaluation office, and IMF staff estimates.

• Download Figure
• Download figure as PowerPoint slide

96. Prices of a number of essential goods are administered in Eritrea mainly for stabilization purposes. They include bread and flour, petroleum products, and pharmaceuticals, comprising 3 percent, 13 percent, and 1 percent, respectively, of the total CPI index. The middle panel in Figure V.1 shows developments in administered and nonadministered prices. For most of the period, price increases for administered goods and services were more limited than those for nonadministered goods and services. Where their rate of change exceeded the latter, this was mostly a result of incremental increases in the price of bread and the infrequent but rather large adjustments in petroleum prices.

97. A further breakdown of the CPI is possible by tradable and nontradable goods (bottom panel of Figure V.1). Tradables consist not only of goods whose prices are fully determined by foreign prices and the exchange rate, but also those domestic substitutes that are subject to strong competition from imports, hi the Eritrean case, these comprise mainly food, beverages and tobacco, clothing and footwear, and household goods. Using this definition in line with the system proposed by Celasun and Goswami (2002),³⁸ tradables account for about 73 percent of the Eritrean CPI based on the weights of the SEO (Table V.1). The bottom panel in Figure V.1 shows that tradable goods inflation has been higher than inflation in nontradable goods for most of the period. This suggests that imported inflation is a driving force for overall domestic inflation.

Theoretical background

98. The literature on the determinants of inflation in developing countries suggests three principal influences: cost-push pressures, demand-pull factors, and structural changes or rigidities. The main cost-push factors normally include wage growth and changes in import prices. Wage growth in excess of productivity growth—often caused by real wage rigidity—is a principal determinant of nontraded goods inflation, while, especially in small open economies like Eritrea, a significant weakening of the exchange rate predictably raises the price of tradables in the economy. Demand-pull factors are dominated by monetary and fiscal expansion, or by large, unsterilized inflows of foreign exchange. In the case of Eritrea, these are mainly related to remittances from the diaspora. Finally, structural factors include movements in administered prices, increases in the tax rates, or exogenous shocks, such as droughts and war.

99. The inflation process in Eritrea can be described by a model including both a money market equation and an equation to capture the influence of the exchange rate and foreign prices. In this note, the domestic price level in the economy, p, is assumed to be a linear combination of the price level for tradable goods, p_T, and nontradable goods, p_NT, respectively. That is,

where 0 <0 <1. The purchasing power parity (PPP) hypothesis implies that the domestic price of tradable goods is determined in the world market and equal to³⁹

where e is the nominal effective exchange rate (defined as foreign currency per domestic currency) and q represents the weighted average of foreign prices, where the weights for both are given by Eritrea’s composition of imports. The price of nontradables. in turn, is determined by adding an assumption of equilibrium in the money market so that real money supply (m^s-p; both in logarithms) equals real money demand, where the latter is assumed to being a positive function of real income, y, and a negative function of the nominal interest rate, i:⁴⁰

Any excess supply of money is expected to increase demand for goods and services and put upward pressure on inflation.

Data issues and methodology

100. The empirical analysis was complicated by certain rigidities and a lack of reliable time series on some key variables. First and foremost, nominal interest rates, the normal measurement of the opportunity cost of holding money, are not a useful indicator in the case of Eritrea because they have remained fixed despite the scope, in principle, of banks to adjust them and their impact on money demand is likely to be limited. In this situation, the rate of inflation was used as a proxy for the opportunity costs of holding money.

101. The empirical analysis was carried out using quarterly data for the period 1992:Q2 to 2002:Q3. All series used in the analysis are plotted in Figure V.2. In order to examine the different impacts of broad money or narrow money on inflation, they were both included in the analysis: m1p is real narrow money, including notes and coins in circulation, and mSp is real broad money, consisting of m1p plus time and savings and foreign currency deposits. The nominal exchange rate, e, and the foreign price level, q, were calculated in effective terms, taking into account the foreign trade pattern of Eritrea; an increase in the effective nominal exchange rate means an appreciation of the nakfa (see Annex I for further data details).

Eritrea: Levels and First Differences of the Data Series, 1992:Q1-2002:Q4 ^1/

Citation: IMF Staff Country Reports 2003, 166; 10.5089/9781451811971.002.A005

Sources: Eritrean authorities; and staff estimates.^1/ Natural logarithms of levels are shown on left-hand sides of panels; first differences., in dotted line, on right-hand sides.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Eritrea: Levels and First Differences of the Data Series, 1992:Q1-2002:Q4 ^1/

Citation: IMF Staff Country Reports 2003, 166; 10.5089/9781451811971.002.A005

Sources: Eritrean authorities; and staff estimates.^1/ Natural logarithms of levels are shown on left-hand sides of panels; first differences., in dotted line, on right-hand sides.

• Download Figure
• Download figure as PowerPoint slide

Eritrea: Levels and First Differences of the Data Series, 1992:Q1-2002:Q4 ^1/

Citation: IMF Staff Country Reports 2003, 166; 10.5089/9781451811971.002.A005

Sources: Eritrean authorities; and staff estimates.^1/ Natural logarithms of levels are shown on left-hand sides of panels; first differences., in dotted line, on right-hand sides.

• Download Figure
• Download figure as PowerPoint slide

102. Traditional unit root tests (see the table in Annex II) indicated that all series were at least integrated of order one. However, the Dickey-Fuller tests gave inconclusive results with respect to foreign prices, the nominal effective exchange rate, and real broad money, which were found to be integrated of the second order in some specifications of the test. These findings imply that the first differences of these variables may be nonstationary and that the standard approach of determining a long-run relationship between these variables in levels may not be warranted (see further discussion below).

103. Given the problems involving the measurement of the opportunity cost of holding money and the second-order integrated variables, the standard approach to an econometric analysis of inflation had to be modified. Under the standard approach, long-run relationships are usually based on variables specified in levels. This was not possible, given that inflation (i.e. the first difference of the price level) was used as a measure of the opportunity cost of holding money. Furthermore, the second-order integration of domestic and foreign price inflation and the exchange rate depreciation called for a cointegration analysis of these variables specified in first differences.

104. In the event, the equation applied consisted of five interrelated variables [mp, y, dpi, dq, de],⁴¹ where economic theory suggests that two long-run relationships would exist, first, a (revised) PPP, linking the (change in) domestic prices with that in foreign prices and the nominal exchange rate; and, second, an equation for money market equilibrium determined by real money, real income, and changes in domestic prices. Both the (change in) real exchange rate and real money demand would be expected to be fairly stable in the long run, and temporary deviations from long-run equilibria would be expected to cause an adjustment such that the long-run equilibria are restored.

105. More specifically, a vector of endogenous variables, x, that are integrated of order 1 is analyzed using the vector error-correction representation

where the parameters µ and Г₁,…,Г_k are allowed to vary without restrictions, k is the lag length of the model, and ε_t is the a vector of normally distributed shocks with a zero mean. The presence of cointegration is tested by examining the rank of π. In the event of reduced rank of π (i.e., when rank(π)=r, n, where n is the number of endogenous variables), there exist r cointegrating vectors and the matrix π can be written as π=αβ’, with β containing the r cointegrating vectors, and a describing the speed of adjustment to the long-run equilibrium (the error-correcting terms). If r>l, the issue of identification arises. In this study, the expected rank is two (in other words, there are two long-run relations relating to the money market and PPP), implying that (over)idenlifying restrictions should be placed on the parameters in

In the estimations below, β₁₁ and β₂₃ will be normalized to 1, while the simplest forms of the (revised) PPP and money demand relationships will be tested by adding exclusion restrictions on [β₁₄ β₁₅] and [β₂₁ β₂₂] to test the money market and PPP relationships.

Long-run relationship

106. The results from the first set of cointegration tests are summarized in Table V.3. The number of cointegrating vectors was estimated using the Johansen procedure.⁴² Cointegration tests in the Johansen setting are sensitive to the lag length of the vector auto regression (VAR). Although it is common to include four lags in the VAR when quarterly data are used, at this stage, the results reported are with two, three, and four lags, respectively, included in the VAR.

Table V.3.

Eritrea: Structural VAR Models, 1993:Q2-2002:Q3

^1/The VARs also include (unrestricted) seasonal dummy variables and a step-dummy for 1997:Q4

^2/The number of cointegrating vectors is based on Johansen’s trace statistics at the 5 percent significance level

^3/* and ** indicate rejection of the likelihood ratio-test at 5 percent and 1 percent levels, respectively.

^4/The estimations assume two cointegrating vectors. Bold figures are estimated coefficients.

Table V.3.

Eritrea: Structural VAR Models, 1993:Q2-2002:Q3

Lags	Number of Cointegrating Vectors 2/	Chi-Sqaure Statitics of Likelihood Ratio Test: 3/		Restricted Cointegrating Vectors 4/
VAR models, including real narrow money, m1p
				m1p	y	dp	dq	de
2	3	9.22	**	$\left[\begin{array}{ccccc}1& -0.51& -35.1& 0& 0\\ 0& 0& 1& 0.12& 0.15\end{array}\right]$
3	2	8.94	*	$\left[\begin{array}{ccccc}1& -1.19& -26.5& 0& 0\\ 0& 0& 1& 0.03& 0.37\end{array}\right]$
4	2	9.11	*	$\left[\begin{array}{ccccc}1& -1.53& -24.5& 0& 0\\ 0& 0& 1& -0.63& 1.61\end{array}\right]$
VAR-models including real broad money, m3p
				m3p	y	dp	dq	de
2	2	9.92	**	$\left[\begin{array}{ccccc}1& -1.15& -27.0& 0& 0\\ 0& 0& 1& 0.05& 0.22\end{array}\right]$
3	1	6.96	*	$\left[\begin{array}{ccccc}1& -2.63& -25.4& 0& 0\\ 0& 0& 1& -0.21& 0.58\end{array}\right]$
4	2	9.07	*	$\left[\begin{array}{ccccc}1& -2.88& -30.1& 0& 0\\ 0& 0& 1& -0.76& 1.20\end{array}\right]$

^1/The VARs also include (unrestricted) seasonal dummy variables and a step-dummy for 1997:Q4

^2/The number of cointegrating vectors is based on Johansen’s trace statistics at the 5 percent significance level

^3/* and ** indicate rejection of the likelihood ratio-test at 5 percent and 1 percent levels, respectively.

^4/The estimations assume two cointegrating vectors. Bold figures are estimated coefficients.

View Table

Table V.3.

Eritrea: Structural VAR Models, 1993:Q2-2002:Q3

Lags	Number of Cointegrating Vectors 2/	Chi-Sqaure Statitics of Likelihood Ratio Test: 3/		Restricted Cointegrating Vectors 4/
VAR models, including real narrow money, m1p
				m1p	y	dp	dq	de
2	3	9.22	**	$\left[\begin{array}{ccccc}1& -0.51& -35.1& 0& 0\\ 0& 0& 1& 0.12& 0.15\end{array}\right]$
3	2	8.94	*	$\left[\begin{array}{ccccc}1& -1.19& -26.5& 0& 0\\ 0& 0& 1& 0.03& 0.37\end{array}\right]$
4	2	9.11	*	$\left[\begin{array}{ccccc}1& -1.53& -24.5& 0& 0\\ 0& 0& 1& -0.63& 1.61\end{array}\right]$
VAR-models including real broad money, m3p
				m3p	y	dp	dq	de
2	2	9.92	**	$\left[\begin{array}{ccccc}1& -1.15& -27.0& 0& 0\\ 0& 0& 1& 0.05& 0.22\end{array}\right]$
3	1	6.96	*	$\left[\begin{array}{ccccc}1& -2.63& -25.4& 0& 0\\ 0& 0& 1& -0.21& 0.58\end{array}\right]$
4	2	9.07	*	$\left[\begin{array}{ccccc}1& -2.88& -30.1& 0& 0\\ 0& 0& 1& -0.76& 1.20\end{array}\right]$

^1/The VARs also include (unrestricted) seasonal dummy variables and a step-dummy for 1997:Q4

^2/The number of cointegrating vectors is based on Johansen’s trace statistics at the 5 percent significance level

^3/* and ** indicate rejection of the likelihood ratio-test at 5 percent and 1 percent levels, respectively.

^4/The estimations assume two cointegrating vectors. Bold figures are estimated coefficients.

107. The economic model suggests that two cointegrating vectors should be found and the cointegrating test typically picked up two stationary vectors (see column 2 in Table V.3). However, the results vary between one and three vectors, depending on the number of lags included in the model. The parameters in the restricted model were constrained to test whether the two stationary vectors could be represented by the two long-run relationships discussed above. The results were in part supportive of the theoretical arguments, although the significance was somewhat weak.

108. Turning to the parameters of the cointegrating vectors, it is noted that the estimated parameters for all variables had the expected sign except for inflation as a measure of the opportunity cost of holing money and, in some cases, foreign prices. The PPP relationship was found to be well established and suggests that, as expected, in the long run domestic inflation increases with foreign inflation and exchange rate depreciation.

109. By contrast, the results regarding the long-run money demand relationship were somewhat mixed. As expected, the coefficient for income indicates that money demand increases in line with real income. The income elasticity appears to be relatively high but is generally lower in regressions with narrow money than in those using broad money. This suggests that time and savings deposits are less directly influenced by developments in income than the components of Ml. In Eritrea, a considerable part of these deposits depend on transfers to households from the diaspora, a large part of which appear to reflect factors other than those determining money demand of residents.

110. Against expectations, when inflation was interpreted as the opportunity cost of holding money, its coefficient was negative in all cases, suggesting that the demand for money increases with inflation. This result could point to model misspecification. On the other hand, inflation may be a more reliable measure of the need to hold money for transactions purposes than of the opportunity cost for holding money as a savings instrument. If the transaction motive dominates, higher inflation would raise the demand for real balances. The likelihood that money demand in Eritrea is dominated by transactions needs appears to be plausible because the availability of goods for inflation protection is limited. More importantly, perhaps, a considerable part of the money balances reflects remittances of the diaspora, which are driven by considerations other than the usual domestic determinants of money demand. In fact, inflation may lead to increases in remittances by the diaspora to family members to protect their real purchasing power (and thereby increase bank deposits and money balances).

111. On the basis of this reasoning, all long-run coefficients of the money demand equation would display the correct sign, and the cointegrating vectors from the specification with four lags were used in the short-run analysis.⁴³ A visual inspection of these vectors suggested that they were relative stable in the equations with both narrow and broad money.

112. A test of weak exogeneity indicates that m1p, m3p, y, dq, and de are weakly exogenous in the model for both narrow and broad money, implying that the inflation rate, dcpi, is the only variable in the system that adjusts to the disequilibrium in the money market. In other words, a change in inflation increases the demand for money while reducing the real money supply, thereby putting downward pressure on prices and restoring equilibrium.

Short-run inflation dynamics

113. Given that inflation is the only variable in the money market relationship that is not weakly exogenous to the cointegrating vector, the dynamics can be consistently analyzed by estimating a single error-correction equation for the first difference of inflation⁴⁴ The model was estimated using ordinary least squares for both narrow and broad money. As explanatory variables, it included their current values and the four lags of the first difference of real money demand, output, the change in exchange rate depreciation, and the change in foreign price inflation (and including seasonal dummies), as well as the first lag of the two error-correction terms, ECM(money) and ECM(PPP), which represent excess supply of money and real depreciation, respectively.

114. By removing the variables with statistically insignificant coefficients from the general error-correction equation, the following inflation equation is obtained for narrow money for the period 1994:Q1-2002:Q2:⁴⁵

115. Except for narrow real money balances, the coefficients in the equation have the expected sign. The first three lags of inflation enter the regression with positive signs, suggesting that there is a significant amount of inertia in the formation of inflation expectations and their influence on actual inflation. An acceleration in the rate of depreciation of the effective exchange rate significantly raises the rate of inflation with a lag of one quarter. As expected, the coefficient with respect to the error-correction term, ECM(money), was positive, which suggests that an excess supply of money will cause the rate of inflation to increase in the subsequent quarter and thereby partly offsets the initial excess money supply in real terms. The coefficient for ECM(PPP) was not significant, suggesting that the restoration of PPP is not achieved over short-term periods like the ones examined.

116. For the model using broad money, the following inflation equation was obtained:

117. The coefficients in the equation have the expected sign, except, again, for real broad money balances (the first two lags). The first and third lags of inflation enter the regression with positive signs. Hence, similar to the model for narrow money, there appears to be significant amount of inertia in the formation of inflation expectations. Furthermore, foreign inflation was found to predictably increase domestic inflation with a lag of two quarters. A temporary deviation from the money market equilibrium ECM(money) was again found to have a predictable influence on the rate of inflation, interestingly, the pace of adjustment in the model with broad money was found to be only half as fast as that in the narrow money model. This may be explained by the fact that the support by the diaspora may increase in the wake of macroeconomic volatility, thereby prompting a rise in deposits, most of which are initially kept as savings. As in the model for narrow money, the coefficient with respect to ECM(PPP) was not found to be significant.

118. The most difficult result of the empirical analysis concerns the unexpected sign for real money balances in the short-term inflation equations. Normally, any increase in these balances would be expected to raise inflation, but that does not appear to be the case in the equations estimated. The reason may lie in the extremely large inflows of private transfers from the diaspora, which not only increase private sector deposits but for much of the period under review, also generated exceptionally high levels of excess reserves in commercial banks. As the latter constitute idle resources as far as the inflationary process is concerned, the relationship between their size (and change) and inflation must have been tenuous at best for much of the period under review. Since mid-2001, these excess reserves have disappeared as a result of the heavy borrowing of government from the banks. There are indications that this large deficit financing of government has contributed to the recent acceleration in inflation. However, this effect may not lasted long enough to influence the parameter values for real money balances.