Demand for money and shortages in Ethiopia

Abstract

The paper discusses the long-run monetary conditions in Ethiopia in the last three decades. These decades can be characterized by large political changes, leading to shocks on income and population growth, and two serious periods of drought. Both affected inflation and real demand for M ₁ through shortages of food. Shortage due to drops in rainfall might have long-run monetary consequences. Despite regime shifts we find support for stability of Ethiopian narrow money demand.

Acknowledgements

The author thanks the Economic Research and Planning Department of the National Bank of Ethiopia for its hospitality. Stefan Dercon kindly provided the coffee price index.

Notes

Moser (1994) is of special interest for this study since Moser discusses the consequences of severe agroclimatic circumstances for inflation.

M ₁ is included in the analysis and the demand for M ₂ is not studied for two reasons. First, a theoretical model that is based on the shopping time model is developed. This model, which will be discussed hereafter, is a basic model to describe the transactions demand for money. It does not directly focus on portfolio motives that might be relevant to the demand for money. The second reason for focusing on M ₁ rather than M ₂ is the relative quantity of secondary liquidity in the Ethiopian economy. Quasi money is about half of the primary liquidity, as can be seen from Table 1. M ₁ is not disaggregated further into demand for currency and demand deposits separately, because both currency and demand deposits are in active use in transactions and one does not want to complicate the model any further. Moreover, such a disaggregation would lower the applicability of the results in monetary policy. Per capita definitions are used for two reasons. First, a microeconomic framework is used to explain money demand per capita in the next section and one wants to be as close as possible to this theoretical model. Second, in Ethiopia population growth certainly varies substantially, which might trouble econometric models in neglecting it. In western economies population growth is not so relevant to include in a money demand model, since the rate of population growth is rather constant.

Quarterly data were constructed using annual observations from the World Bank Indicator-source on nominal GNP for those quarters where no quarterly data were available. The algorithm used is the Boot-Feibes-Lisman routine (see Boot et al., 1967). GNP and GDP of Ethiopia in the World Bank Indicator data set show a correlation of 0.999 for the sample 1966–1994, so the study refrains from using GDP.

In Nigeria the general price index is for 69% dependent on the food price index.

Coffee prices are converted from annual figures into quarterly figures using the algorithm developed by Boot et al. (1967). It is remarkable to notice a downward trend after 1979 until 1992.

The World Currency Yearbook (until 1983 Pick's Currency Yearbook) gives the black market rates of US Dollars.

The study refrains from corporate demand for money. Secondary liquidity is relatively small in a large part of the sample. If there is forced holdings of domestic financial assets, it is likely that firms would hold time and savings deposits.

In case of one regime shift Hubrich (1999) shows that the Phillips-Perron test with a step-dummy variable is a good alternative of the (A)DF-test.

The software package CATS was strictly followed (see Hansen and Juselius, 1995).

Being a statistical exercise the Johansen procedure picks out those combinations with the most stationary residuals, but these need not correspond to meaningful economic relationships. The cointegration vectors will however, usually be linear combinations of the underlying economic relationships and these need to be recovered using identification conditions. If only one cointegration vector is found this vector is identified.

The Trace test is not discussed.

There are basically three alternatives for the appropriate choice of the deterministic components:

1.	Model A: including a constant in the cointegration space;
2.	Model B: including a constant in the short-run model (allowing for the presence of linear trends in the levels of the data);
3.	Model C: including a trend in the cointegration space.

Model A is clearly the most restricted, followed by model B. The Pantula principle suggests to test for the rank of the long-run model by estimating the three model versions sequentially from the most restricted to the least restricted one. The decision rule is then to pick the model the first time the null hypothesis is not rejected using the trace test.

Information criteria are used to warn for potentially overparameterizing the model. The Schwarz criterion is defined by

for K regressors and a sample length T.

is the unadjusted residual variance.

A final point one can make to address the stability of the short-run coefficients is an investigation into the stability of the coefficients over regime shifts. To that purpose the assumption is made that one can focus on the conditional density of m _t given the other variables and test for a single equation model. Given the endogeneity of y and p _co one loses efficiency of the coefficient estimates. In order to stay as close as possible to the multivariate model the long-run cointegration vector is included in this experimental model. So a model was estimated in first difference with real per capita money demand as the endogenous variable including up to three lags for all determinants as shown in the VECM. The model was reduced to significant parameter estimates only. The final model specification is estimated recursively. From the recursive parameter plots it can be seen that the specification of the model shows some instability during the first break in 1975, but stability thereafter.

It is sometimes suggested to invert the time range and estimate the model from the future to the past.