
Economy-Wide Effect of Fiscal Drag in Ethiopia
Figures

Income tax revenue, income, and effective tax rates by household types. Source: Own computation.

Bottom-up approach. Source: Adapted from Bourguignon and Bussolo (2013).
Tables
Percentage change in household consumption and income.
Household types | Household consumption | Household income |
---|---|---|
Urban poor | 1.31 | -0.06 |
Urban non-poor | -0.75 | -0.08 |
Rural poor | 0.70 | 0.16 |
Rural non-poor | 0.33 | 0.07 |
-
Source: Own computation.
Share of consumption and changes in household consumption and production.
Item | Share of consumption by urban non-poor | Household consumption expenditure | Production | |||
---|---|---|---|---|---|---|
Urban Poor | Urban non-poor | Rural poor | Rural non-poor | |||
Grain | 20.65 | 0.16 | - 0.09 | 0.38 | 0.18 | 0.14 |
Livestock | 28.92 | 1.67 | -0.95 | 0.87 | 0.40 | 0.10 |
Hotel | 41.58 | 1.86 | -1.01 | 0.63 | 0.30 | -0.13 |
Forest | 18.53 | 2.06 | -1.13 | 1.26 | 0.59 | 0.37 |
Agro-processing | 25.39 | 0.16 | -0.09 | 0.38 | 0.18 | 0.14 |
Other service | 60.88 | 1.87 | -1.00 | 0.64 | 0.31 | -0.14 |
Real estate | 63.86 | 1.88 | -0.99 | 0.65 | 0.32 | -0.34 |
Transport | 52.93 | 1.87 | -1.00 | 0.64 | 0.31 | 0.13 |
Alcohol | 53.91 | 0.17 | -0.09 | 0.39 | 0.19 | 0.06 |
Textile | 27.11 | 0.92 | -0.50 | 0.88 | 0.42 | 0.31 |
-
Source: Own computation.
Percentage change in factor demand by selected activities.
Items | Agricultural labor | Non-irrigated land | Irrigated land | Agricultural capital | Semi-skilled labor | Skilled labor | Other capital |
---|---|---|---|---|---|---|---|
Grain | 0.17 | 0.19 | 0.22 | ||||
Livestock | 0.20 | 0.04 | |||||
Hotel | -0.11 | -0.10 | -0.07 | ||||
Forest | 0.40 | 0.23 | |||||
Agro-processing | 0.16 | 0.18 | 0.21 | ||||
Other services | -0.15 | -0.13 | -0.11 | ||||
Real estate | -0.38 | -0.37 | -0.34 | ||||
Transport | 0.12 | 0.14 | 0.16 | ||||
Alcohol | 0.03 | 0.04 | 0.07 | ||||
Textile | 0.28 | 0.30 | 0.33 |
-
Source: Own computation.
Macroeconomic effects of an increase in the income tax rate.
Real GDP | 0.04 |
---|---|
Real domestic production | 0.04 |
Real household consumption | 0.03 |
Real domestic final demand | 0.02 |
Real import | -0.05 |
Real export | 0.03 |
Real intermediate input | 0.05 |
-
Source: Own computation.
Effects of an increase in the income tax rate on government income and expenditure.
Direct tax | 9.05 |
---|---|
Import tax | -0.02 |
Sales tax | -0.11 |
VAT | -0.38 |
Excise tax | -0.04 |
Government income | 2.89 |
Government expenditure | -0.08 |
Government savings | 6.91 |
-
Source: Own computation.
Balanced Macro SAM for the year 2015/16.
Activity | Commodity | Margin | Factor | Household | Tax | Govt | Kapital | World | Total | |||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Agri | Indus | Service | Agri | Indus | Service | Labor | Land | Capital | Urban | Rural | direct | Indirect | ||||||||||
Activity | Agri | 649 | 649 | |||||||||||||||||||
Indus | 1,124 | 1,124 | ||||||||||||||||||||
Service | 806 | 806 | ||||||||||||||||||||
Commodity | Agri | 48 | 320 | 65 | 89 | 199 | 10 | 48 | 779 | |||||||||||||
Indus | 46 | 365 | 114 | 201 | 342 | 0 | 531 | 13 | 1,613 | |||||||||||||
service | 4 | 77 | 125 | 172 | 169 | 138 | 166 | 0 | 67 | 917 | ||||||||||||
Margin | 98 | 70 | 4 | 172 | ||||||||||||||||||
Factor | Labor | 357 | 221 | 333 | 911 | |||||||||||||||||
Land | 20 | 20 | ||||||||||||||||||||
Capital | 174 | 142 | 168 | 483 | ||||||||||||||||||
Household | Urban | 468 | 0 | 150 | 1 | 66 | 686 | |||||||||||||||
Rural | 443 | 20 | 333 | 2 | 70 | 868 | ||||||||||||||||
Tax | Direct | 60 | 12 | 72 | ||||||||||||||||||
Indirect | 82 | 33 | 115 | |||||||||||||||||||
Govt | 45 | 72 | 115 | 30 | 262 | |||||||||||||||||
Kapital | 123 | 176 | 92 | 150 | 541 | |||||||||||||||||
World | 32 | 338 | 74 | 445 | ||||||||||||||||||
Total | 649 | 1,124 | 806 | 779 | 1,613 | 917 | 172 | 911 | 20 | 483 | 686 | 868 | 72 | 115 | 262 | 541 | 445 |
-
Source: Own computation.
Percentage change in Factor prices.
Non-irrigated land | Irrigated land | Agricultural capital | Semi-skilled labor | Skilled labor | Other capital |
---|---|---|---|---|---|
0.27 | 1.83 | 0.20 | -0.07 | -0.07 | -0.12 |
-
Source: Model simulation results.
Percentage distribution of household income by factor types.
Factor types | Urban poor | Urban nonpoor | Rural poor | Rural nonpoor |
---|---|---|---|---|
Agricultural labor | 2.37 | 0.15 | 94.54 | 28.12 |
Non-irrigated land | 0.16 | 0.04 | 5.23 | 1.08 |
Irrigated land | 0.23 | 0.45 | ||
Agricultural capital | 1.07 | 0.06 | 28.84 | |
Semi-skilled labor | 41.39 | 47.22 | ||
Skilled labor | 55.01 | 25.43 | 14.83 | |
Other capital | 27.10 | 26.68 |
-
Source: Model simulation results.
Percentage change in unemployment.
Unemployment in agricultural labor |
---|
-8.24 |
-
Source: Model simulation results.
List of main Account of the 2015/16 SAM for Ethiopia.
Activities | Commodities | Factors | Households | Taxes | ||||
---|---|---|---|---|---|---|---|---|
|
|
|
|
|
|
|
|
|
-
Source: Own compilation.
Income elasticities for Ethiopia.
Commodities | Urban poor household | Urban non-poor household | Rural poor household | Rural non-poor household |
---|---|---|---|---|
Teff | 1.14 | 1.14 | 1.08 | 1.08 |
Barley | 0.33 | 0.33 | 0.06 | 0.06 |
Wheat | 0.41 | 0.41 | 0.42 | 0.42 |
Maize | 0.58 | 0.58 | 0.62 | 0.62 |
Sorghum | - 0.81 | - 0.81 | 1.00 | 1.00 |
Finger millet | - 0.81 | -0.81 | 1.00 | 1.00 |
Oats | 0.33 | 0.33 | 0.64 | 0.64 |
Rice | 0.13 | 0.13 | 0.64 | 0.64 |
Horse beans | 0.87 | 0.87 | 1.13 | 1.13 |
Peas | 0.87 | 0.87 | 1.13 | 1.13 |
Haricot beans | 0.87 | 0.87 | 1.13 | 1.13 |
Lentils | 0.87 | 0.87 | 1.13 | 1.13 |
Vetch | 0.87 | 0.87 | 1.13 | 1.13 |
Other pulse | 0.87 | 0.87 | 1.13 | 1.13 |
Neug | 2.10 | 2.10 | 0.96 | 0.96 |
Linseed | 2.10 | 2.10 | 0.96 | 0.96 |
Groundnuts | 2.10 | 2.10 | 0.96 | 0.96 |
Sesame | 2.10 | 2.10 | 0.96 | 0.96 |
Other cereals | -6.70 | -6.70 | 2.30 | 2.30 |
Other spices | 0.67 | 0.67 | 0.30 | 0.30 |
Other oilseeds | 2.10 | 2.10 | 0.96 | 0.96 |
Peppers | 0.67 | 0.67 | 0.30 | 0.30 |
Other vegetable | 0.87 | 0.87 | 0.95 | 0.95 |
Onion | 0.87 | 0.87 | 0.95 | 0.95 |
Potato and sweat potato | 0.59 | 0.59 | 0.18 | 0.18 |
Garlic | 0.67 | 0.67 | 0.30 | 0.30 |
Taro godere | 0.59 | 0.59 | 0.18 | 0.18 |
Other root crop | 0.59 | 0.59 | 0.18 | 0.18 |
Bananas | 0.87 | 0.87 | 0.95 | 0.95 |
Other fruits | 0.87 | 0.87 | 0.95 | 0.95 |
Chat | 0.85 | 0.85 | 1.39 | 1.39 |
Coffee | 0.85 | 0.85 | 1.39 | 1.39 |
Cotton | 0.85 | 0.85 | 1.39 | 1.39 |
Hops | 0.85 | 0.85 | 1.39 | 1.39 |
Sugarcane | 0.85 | 0.85 | 1.39 | 1.39 |
Enset | - 0.39 | -0.39 | 2.12 | 2.12 |
Flower | 1.50 | 1.50 | 1.72 | 1.72 |
Livestock | 1.23 | 1.23 | 1.22 | 1.22 |
Forest | 1.50 | 1.50 | 1.72 | 1.72 |
Fishing | 0.12 | 0.12 | 0.52 | 0.52 |
Preserve | 0.12 | 0.12 | 0.52 | 0.52 |
Vegetable oil | 0.90 | 0.90 | 0.83 | 0.83 |
Dairy | 0.12 | 0.12 | 0.52 | 0.52 |
Grain | 0.12 | 0.12 | 0.52 | 0.52 |
Bakery | 0.12 | 0.12 | 0.52 | 0.52 |
Sugar | 0.96 | 0.96 | 0.16 | 0.16 |
Argo-processing | 0.12 | 0.12 | 0.52 | 0.52 |
Animal feed | 0.12 | 0.12 | 0.52 | 0.52 |
Alcohol | 0.12 | 0.12 | 0.52 | 0.52 |
Soft drinks | 0.12 | 0.12 | 0.52 | 0.52 |
Tobacco | 1.50 | 1.50 | 1.72 | 1.72 |
Textile | 0.67 | 0.67 | 1.19 | 1.19 |
Apparel | 0.67 | 0.67 | 1.19 | 1.19 |
Leather | 0.67 | 0.67 | 1.19 | 1.19 |
Wood | 1.50 | 1.50 | 1.72 | 1.72 |
Papier | 1.50 | 1.50 | 1.72 | 1.72 |
1.50 | 1.50 | 1.72 | 1.72 | |
Chemicals | 1.50 | 1.50 | 1.72 | 1.72 |
Fertilizer | 1.50 | 1.50 | 1.72 | 1.72 |
Ink | 1.50 | 1.50 | 1.72 | 1.72 |
Soap detergent | 1.50 | 1.50 | 1.72 | 1.72 |
Pharmaceutical | 1.50 | 1.50 | 1.72 | 1.72 |
Rubber | 1.50 | 1.50 | 1.72 | 1.72 |
Non-metal | 1.50 | 1.50 | 1.72 | 1.72 |
Metal | 1.50 | 1.50 | 1.72 | 1.72 |
Electrical | 1.50 | 1.50 | 1.72 | 1.72 |
Machinery | 1.50 | 1.50 | 1.72 | 1.72 |
Furniture | 1.50 | 1.50 | 1.72 | 1.72 |
Gas | 1.50 | 1.50 | 1.72 | 1.72 |
Construction | 1.50 | 1.50 | 1.72 | 1.72 |
Water | 1.35 | 1.35 | 0.86 | 0.86 |
Transport | 1.35 | 1.35 | 0.86 | 0.86 |
Trade | 1.35 | 1.35 | 0.86 | 0.86 |
Real | 1.35 | 1.35 | 0.86 | 0.86 |
Public | 1.35 | 1.35 | 0.86 | 0.86 |
Private | 1.35 | 1.35 | 0.86 | 0.86 |
Other service | 1.35 | 1.35 | 0.86 | 0.86 |
Hotel | 1.35 | 1.35 | 0.86 | 0.86 |
Health | 1.35 | 1.35 | 0.86 | 0.86 |
Entertainment | 1.35 | 1.35 | 0.86 | 0.86 |
Electricity | 1.35 | 1.35 | 0.86 | 0.86 |
Education | 1.35 | 1.35 | 0.86 | 0.86 |
Banking | 1.35 | 1.35 | 0.86 | 0.86 |
-
Source: Own compilation based on Hertel (1997) and Tafere et al. (2010).
Frisch elasticity
Urban poor household | Urban non-poor Household | Rural poor household | Rural non-poor household |
---|---|---|---|
-5.85 | -5.85 | -5.85 | -5.85 |
-
Source: Own compilation based on Hertel (1997).
Data and code availability
We used the STAGE 1 variant of the CGE model, which is open source at http://cgemod.org.uk/stage_1.html, and the Social Accounting Matrix can be accessed from the Ministry of Planning and Development of Ethiopia.