Report No. 30749-AZ Azerbaijan Raising Rates: Short-Term Implications of Residential Electricity Tariff Rebalancing December 10, 2004 Europe and Central Asia Region Environmentally and Socially Sustainable Development Document of the World Bank CONTENTS EXECUTIVESUMMARY ......................................................................................... 2 INTRODUCTION ..................................................................................................... 3 THE CHALLENGEAHEAD ..................................................................................... 3 HOWPEOPLEWILL FARE ..................................................................................... 5 IDENTIFYINGTHE MOSTVULNERABLE ............................................................ 8 LINKSTOAGRICULTURE ..................................................................................... 11 LINKSTO FORESTRY ........................................................................................... 13 STAKEHOLDERANALYSIS .................................................................................. 16 Annexes Annex 1:Summary o f Combined Household Survey and UtilityDatafor Four Countries ............................................................................................................................................ 18 Annex 2: Data Reliability ..................................................................... :............................ Annex 3: ReportedAverage Hours of Electricity Supplied per Day DuringWinter 21 Annex 4: Electricity DemandModel ................................................................................. .........20 22 Annex 5: HouseholdIncome Loss under Alternative Tariff Scenarios 27 Annex 6: References ......................................................................................................... ............................ 28 Tables Table 1:Tariffs are lower and consumption is higher ........................................................ Table 2: Electricity consumption and service quality varies widely bylocation Table 3: Differences between the poor and non-poor are small (Baku only) ..................... ................45 6 Table 4: Average monthly metered consumption mayfall to 140kwhper households.....7 Table 5: Risingincomewill offset the blow o f tariff increases on household's budget shares (Baku only) .............................................................................................................. 8 Table 6: Compensation for the poor shouldbe higher (Baku only) ................................... Table 7: Households with less access to substitutes consume more electricity Table 8: Share of electricity intotal agriculture productioncost is low ............................ ................i9o 12 Maps Map 1: Risk o f increased residential wood use.................................................................. 15 Map 2: Deforestation risk is highest inthe Northeast and Southeast ............................... 16 ACKNOWLEDGEMENTS A World Bank team prepared this report under the supervision and guidance of Alexandre Marc (Sector Manager, ECSSD) and under the sponsorship of Donna M. Dowsett-Coirolo (Country Director, ECCU3). Julian Lampietti (ECSSD) managed the task. Task team members included Karin Fock, Irina Klytchnikova, Xun Wu, and Maria Shkaratan. Farid Mamedov (ECSIE) from the resident mission provided valuable information and feedback from the field. Yagut Erentlice (ECCAZ) provided support from the Azerbaijan Country Office. The team acknowledges the helpful comments of the peer reviewers Maureen Cropper (DECRG) and Masami Kojima (COCPO). Additional comments from Bjorn Hamso (ECSIE), Jane Ebinger (ECSIE), Peggy Wilson (ECSIE), Aarvo Kuddo (ECSHD) and Peter Thompson (ECSIE) were also greatly appreciated. The data collection process would not have been possible without extensive in country collaboration. This incluldes MacNeil Watkins and Vilayet Vilayev from PA Consulting and Michael Trainor from USAID. FaigJalilov and Yashar Pashafrom the State Institute of Statistics. Sabit Bagirovfrom the Entrepreneurship Development FoundationinBaku, and representatives from Barmek and Biawa. The Poverty and Environment mappingwork was financedbythe PovertyWindow of the Norwegian Trust Fund for Sustainable Development. The summary results presented here are based on a phase Ipaper written by Jane Falkingham, Craig Hutton, Ben Barton, Angela Baschieri, LouiseTricklebank, and IrinaKlytchnikova. The team is gratefulto local stakeholders including members of the Working Group who participated in two consultative workshops and various discussions to help guide the research. Particular thanks are due to representatives from the Ministry of Economic Development, Ministryof Fuel and Energy, Ministryof Labor and Social Assistance, and Parliamentary Committee on Energy. The team also thanks the Ministry of Agriculture, the State Statistical Committee, and the State Amelioration and Water Management Committee for their close collaboration and support. 1 EXECUTIVESUMMARY Tariffs are low in Azerbaijan and need to be raised to finance badly needed network maintenance and to balance supply and demand. This study presents an analysis of the short-term impacts of a 50 percent electricity tariff increase on residential consumers. Metered residential consumption in Baku is on average 200 kwh per month. Households inBaku spend about 2 percent of their income on electricity. Unfortunately, reliable data on electricity consumption outside Baku, where supply is rationed (especially duringwinter), are not available. In Baku, where service quality is already high, the income loss from a 50 percent tariff increase is about US$2 per household per month. The poor will be hit hardest by the combination of higher tariffs and collections and will require slightly more compensation (closer to US$3 per month). Recommendations are to evaluate the costs and benefits of different social transfer mechanisms such as lifeline tariffs and direct income support, paying careful attentionto targeting effectiveness. Once a preferredmechanismhas been identified, introduce it alongside a program to gradually raise higher tariffs and increase metering. Develop a parallel program to encourage households heatingwith electricity to switch to less expensive alternatives and improve demandside management. Outside Baku, there are large potential social benefits from lifting the rationing constraint. The key to minimizing the negative effects is closely linking the improvements in service quality to the timing of tariff increases. Interms of agriculture, higher electricity tariffs will not tangibly affect farmers' direct and indirect operating costs. The benefits from reliablesupply are likelyto outweighthe impact of higher tariffs on agro-processors. Irrigation costs are a concern, but there is already a plan inplace to cushion potential adverse impacts. Recommendations are to explicitly link tariff increases to improvements in service quality. This will mitigate negative social impacts and generate popular support for the reform. It also requires investments in metering, rehabilitation, and maintenance to prevent future supply shortages. Key to minimizing negative impacts i s systematically monitoring service hours of service received, number of outages, and frequency and voltage stability. Also, giving households without access to gas or wood efficiency increasing technology or less expensive fuels for heatingwill ease the transition. On environmental management, the risk of increased residential wood use is greatest in Khachmaz, the Southeast, and the Southwest. Forest areas at greatest risk are in the Northeast and Southeast. Promoting access to alternative energy sources and more efficient wood stoves inhigh-risk areas, as well as preparing and implementing spatially explicit forest management plans and encouraging participatoryforest management may helpreducethis risk. On process management, there is general consensus that the reform can be further improved by providing a transparent forum for dialogue on key actions. Particular attention should be paid to wide dissemination of information on the potential benefits of reform. Finally, providing key analysts inthe government with information on the full set of available policy choices and experiences inother countries that have gone through similar processeswill facilitate evaluation of different social transfer mechanisms. 2 INTRODUCTION Despite being a net energy exporter, one of the most vexing problems in Azerbaijan i s domestic power supply. Outside Baku power is supplied a limited number o f hours per day. This is due to a combination of problems including badly maintained infrastructure, high commercial losses, high non-payment rates, and low tariffs. These problems are getting worse as strong economic growth increases demand for electricity. The opportunity cost o f supplying the sector with low cost domestic oil and gas is also rising, as international oil prices increase, contributing to a significant expansion instate subsidies to the sector. A reform program has been launched to improve supply and reduce subsidies to the sector. It includes unbundling generation, transmission and distribution, introducing private management contracts, and increasing investment in the network. Other key components are establishment o f an independent regulator, raising tariffs to cost recovery levels, and designing a program to mitigate any potential social and environmental impact o f these efforts. This study presents a partial equilibrium analysis of the social impacts of electricity tariff increasesinAzerbaijan. The study starts by reviewing electricity tariffs, consumption levels, and expenditure patterns compared to neighboring countries. It then considers the welfare effects of raising tariffs, with particular attention to the poor and other vulnerable groups. It concludes with potential links to agriculture, the environment, and stakeholder analysis. Electricity is an important input into the production of many goods and services. Therefore reform can profoundly influence the cost of the basic consumption basket. The direction and magnitude o f this effect is unclear. Itmaybe negative ifhigher costs are passed on to consumers or it may be positive if service quality improvements lead to increased competition between producers and efficiency gains. Ideally the reforms should be analyzed in a general equilibrium framework that takes into account these linkages. However, general equilibrium modeling is data intensive and very sensitive t o a large number of assumptions. Thus, there are strengths andweaknesses to both approaches and the analysis presented here should be thought of as one of many inputs to the decision-making process. THE CHALLENGEAHEAD Tariffs are low in Azerbaijan. Residential tariffs are US$0.0196 (manat 96) per kWh, putting them well below other countries inthe region (Table 1) as well as international norms. Electricity tariffs need to be raised to finance badly needed network maintenance and to balance supply and demand. Otherwise the condition o f the networkwill continue to decline, demand will outpace supply, and service quality will fall. While Azerbaijan may be able to afford lower tariffs than net energy importers, it must still raise prices to cover generation, transmission, and distribution costs if the network is to be financially viable. International norms suggest cost recovery will be approximately US$0.05 (manat 288) per kWh. 3 Table 1:Tariffs arelower andconsumption ishigher Mean Household Country Tariff CollectionRate (US$/kWh) (payment/billing) Consumption (kWh/month) Azerbaijan (Baku - 2002)' 0.0196 71% 198 Moldova (Chisinau- 2003) 0.0529 98% 58* Armenia (Yerevan - 1999) Georgia (Tbilisi - 2002) 0.0564 go% 158 0.0475 82% 16gt Notes: * Figures for Baku are based on records for 1,094 metered households inthe 2002 HouseholdBudget Source: SeeAnnex I. Survey. * January-November. + January-June. Lowcollectionratesfurther compoundproblems associated withlowtariffs. Collections from metered households in Baku are lower than in other countries (71 percent - Table I). Yet collections from metered households are significantly higher than general collection rates for urban (50 percent) and rural (30 percent) households in Azerbaijan.' Low collections effectively lower the tariff by one-half or more and result from a combination of poor service quality, weak enforcement, theft, lack o f metering, and non-payment. Enforcement has been improving in Baku over the last few years due to the presence o f a private operator (Barmek). Collection rates are supposed to rise to 100percentby2008. Average electricity consumption in Baku is well above basic minimum needs. Consistent with having lower prices and collections, average consumption is higher than inother countries for which similar data are available (Table I). However, it is not significantly higher (even though the price is much lower). This may be because, compared to other countries, many households in Azerbaijan (particularly Baku) have access to a reasonably reliable supply o f inexpensive natural gas. Average electricity consumption for metered households in Baku is anywhere from 2,376' t o 2,952 kWh3 per year. The lower figure is more reliable because it is based o n a larger, more representative, sample. To put these figures incontext, supplying a household with basic lighting and refrigeration requires approximately 1,500 kWhper year.4 Reliable data on household electricity consumption outside Baku are not available. This is due to a combination of factors including lack of metering, frequent service interruptions, and high non-payment rates. Table 2 shows average household consumption based o n household level data provided by Bayas ranges from a high o f 960 kWh per month in Imishly to a low of 260 kWh per month in Mingecevir. The reliability of these figures is highly questionable. For example, one expects households with more hours of supply to consume more, but the data show the reverse. One explanation is that households outside Baku are billed based o n norms, and therefore these figures represent expected rather than actual consumption. The true electricity consumption level outside Baku is unknown. Ifit is as highas the data suggest there may 1 * 2003 Household Budget Survey Data merged byhouseholdwith 2002 Barmek data (n=1,106). Energy Survey (non-random) merged byhouseholdwith 2003 Barmek and Bayva Data (n=n,ooo). 2002 3 2003 Energy Survey (non-random) merged byhouseholdwith 2003 Barmek data (households n=443). 4 A refrigerator (manual defrost, 5-15 years old) consumes about 95 kWha month and 3 incandescent light bulbs another 30 kWha month. 5 The Barmek service area is Baku and the Northeast and Bayiva covers everything else. Barmek records are for the month of November 2003 only. 4 be opportunities to substantially increase the efficiency o f electricity use outside Baku without causing major welfare losses. Electricity supply is rationed outside Baku, especially during the winter. No data is available from the utilities o n the number of hours o f electricity delivered to different locations. However, when asked directly, households in Baku and Sumgait report electricity is available 24 hours a day. Inother areas supply is worse inwinter (16 hours per day) than in summer (21 hours per day).6 This is generally attributed to difficulties in supplying higher loads associated with residential consumption of electricity for heating. The results on hours of service are internally consistent - in different locations the majority of households report similar hours of service. For example, all 150 households interviewed in Sumgait report 24 hours o f service (Annex 3). The results are also consistent with a number of other surveys undertaken in Azerbaijan showing poor service outside of the capital is a major impediment to economic development? Table 2: Electricityconsumption andservice qualityvaries widely bylocation Mean Household Winter Supply Summer CollectionRate Location Method Billing Consumption (hoursper Supply (Payments/ (kwh per month) day) (hoursper day) Billing) Alibayramly Norms 628 17 22 25% Baku Meters 265 24 24 63% Ganja Norms N a 10 22 Na Goycay Norms 503 15 18 42% Guba Norms N a 9 15 Na Imishly Norms 960 8 20 7% Ismailly Norms N a 18 21 Na Mingecev Norms 260 9 21 28% Sabirabad Norms 447 8 20 35% Sumgait Meters 374 24 24 24% Source: 2003 Energy Survey (non-random) merged byhouseholdwith 2003 Barmek and Bayva Data (n=n,ooo). HOWPEOPLEWILL FARE Improving supply and raising tariffs presents consumers with a series of tradeoffs. Consumers gain from the removal of rationing. They may gain or lose from changes inservice. For example, installation of meters may allow households to manage consumption better, but it makes non-payment more difficult. All else equal, consumers lose from tariff increases. In this section, we first examine h o w a tariff increase would affect household electricity consumption in Baku, where there is no rationing constraint. We then go on to use this information to calculate the size o f the income loss from a 50 percent tariff increase, keeping everything else constant. We don't assess the impact of relieving the rationing constraint because the data on household behavior outside of 6These are averageover the 2000 households inthe 2003 Household Energy Survey. 7Azerbaijan Rural Infrastructure Note, World Bank, Mimeo, 2002. ForeignInvestment Advisory Service Report onAzerbaijan, 2002. Azerbaijan Rural Investment Project SocialAssessment, World Bank, Mimeo, 2003. 5 Baku are not reliable.8 We conclude by identifymg who will be the most affected by the tariff increase and the implications of potentialmitigating actions. Metered households in Baku are spending about 2 percent of their income on electricitys (Table 3). Spending at this level is similar to households in the United States (2.3 percent) and below those in the United Kingdom (4.0 percent) and most o f the transition economies (generally inthe 4-6 percent range). Not as much separation in the quintiles is observed as might be expected because income is presented o n a household, not a per capita, basis. Very low shares of income on electricity suggest there may be room to raise tariff inBakuwithout severely limitingconsumption of other goods and services. There is little difference inconsumptionpatterns betweenthe poor and the non-poor. That shares are similar across per capita income quintiles is unusual. In most countries, the bottom quintile spends a larger share of income o n electricity than the top. One explanation for this unusual pattern is that collections are lower for the poor (Table 3). This means they face a lower effective tariff and consume proportionally more than they would ifthey faced the fulltariff. Table 3: Differencesbetweenthe poor andnon-poor are small (Baku only) Quintiles Household income* consumption (kwh Shareof income on Collectionrate (per capita) (US$/month) per month) electricity (payment/bill) Bottom 20% 123 190 2.1% 65% 2 137 202 1.9% 61% 3 154 192 1.9% 74% 4 161 201 1.9% 68% Top 20% 189 200 2.2% 81% Total 158 198 2.0% 71% Source: 2002 HouseholdBudget Survey, 2002 Barmek Records. Note: * Figures for Baku are basedon records for 1,094 meteredhouseholds inthe HBS. All else equal, a 200 percenttariff increase will causemeteredconsumption of electricity to fall close to basic minimum needs. Understanding household responses to tariff increases requires knowing how much they will reduce consumption. We assume a range of elasticities (low=-o.i5, medium=-o.so, and high=-o.75) based on estimates available in the literature and simulate the impact of alternative tariff scenarios o n household consumption ( Table 4). The results show large tariff increases combined with high elasticities cause a dramatic falls in consumption. This is unrealistic, since demand is likely t o get more inelastic (less sensitive to tariff changes) as consumption approaches basic minimum needs. Also, the price elasticity of demand may change over time, with less elastic behavior over the short-run than long run. A realistic short-run scenario is that the elasticity is -0.15 and consumption falls from 200 to about 140 kwh per month with a 200 percent tariff increase. 8As noted earlier, the only data available on consumption in rationed areas is based on norms, not actual consumption. 9Throughout the report we use household expenditures from the HouseholdBudget Survey as a proxy for income. 6 Table 4: Averagemonthlymetered consumptionmayfallto 140kwhper households (Baku only) Consumption Predicted Predicted Predicted Tariff Elasticity @96manat consumption consumption consumption @i44manat 0192 manat @288manat kWh/month kWh/month kWh/month kWh/month -0.15 200 185lO 170 140 -0.50 200 150 100 Na -0.75 200 125 50 Na Source:Authors' calculationsbased on average consumption of 200 kWhper month. These results are consistent with more sophisticated modeling approaches. In order to assess reliability of the analysis, four capital city data sets (Baku, Tbilisi, Chisinau, and Yerevan) are pooled and used to estimate a household electricity demand model. Inthe model, household electricity consumption depends o n the tariff, household income, the household's access to substitute energy sources (natural gas, central heating, Liquid Petroleum Gas) and other household characteristics. .Other important factors include location, daily temperature, and cross country differences such as economic growth and inflation. The model is estimated using multivariate regression techniques and a common feature of this type o f modeling exercise is that the results are more reliable for small than for large price changes. A detailed description o f the data and modelis includedinAnnex 4. A i o percent increase in tariff will produce a 2 percent decrease in householdelectricityconsumption. The model, which applies to urban households with meters, fits the data well and produces plausible results. The price elasticity of demand (-0.20) is very close to the lower range o f the sensitivity analysis presented earlier. The model also indicates that a IO percent increase in income will produce a 1.2 percent increase in consumption o f electricity. Surprisingly, model testing revealed no plausible significant differences inthe price and income elasticity o f demand for the poor and non-poor. Consistent with expectations, the model indicates use o f central gas and LPG are negatively correlated with electricity consumption. Also, as expected, increasing the collection rate (more enforcement) is negatively correlated with consumption. Future household income growth will help offset the blow of a tariff increase. Incomes are expected to grow rapidly inthe next few years inAzerbaijan. An increase in the minimum wage is being contemplated and civil servant wages were recently increased fifty percent. Therefore, calculating the impact of tariff increases without taking into account changes in income is the worst-case scenario. Assuming current income of US$158 per household per month and a price elasticity o f demand of - 0.20, under a variety of tariff scenarios IO percent income growth will keep the share of income on electricity in the 4 percent range (Table 5). Depending o n how quickly incomes grow and, more importantly, h o w growth is distributed between the poor and non-poor, this will bringshares of income o n electricity inAzerbaijan closer to the level inother transition countries.I1It also suggests that small gradual tariff increases rather than abrupt large ones will soften the blow to household income. 10Illustrative calculation: 185kwh = 200 kwh- (0.50*o.i5*200 kwh). 11Moldova (5 percent), Georgia (5 percent), and Armenia (8 percent). 7 Table 5: Risingincome will offset the blow oftariff increases onhousehold's budget shares (Baku only) Tariff Household income Householdincome Householdincome growth @ 0% growth @ 5% growth @ io% 96 manat 2.5%" 2.4% 2.3% 144manat 3.3% 3.2% 3.0% 192manat 4.0% 3.8% 3.6% 288 manat 4.5% 4.3% 4.1% Source: Authors' calculations based on average consumption of 200 kWh per month and price elasticity of demand of -0.20. IDENTIFYINGTHEMOSTVULNERABLE Identifyingthe mostvulnerable requiresidentifying who will loosethe most as a result of tariff increases. This does not imply that everyone must be compensated for the income loss resulting from a tariff change. The loss i s measuredby linear approximation. The upper bound of this loss is the additional amount of money that the consumer would have to pay after the tariff increase, holding electricity consumption constant. The lower bound is the additional amount of money that the consumer would have to pay at the newtariff, allowing electricityconsumption to fall. The averageincomelossfrom a 50 percenttariff increase inBakuis closeto two dollars per household per month.13 Assuming pre-tariff consumption of 200 kwh,a price elasticity of demandof - 0.20, and 100 percent collections, then the upper bound on the income loss from a 50 percent tariff increase is US$i.g5 (manat 9,600) per month and the lower bound is US$1.76 (manat 8,640).'4 This is the amount of money that would have to be given to a householdto make it no worse off than it was beforethe tariff increase. The poor are morevulnerable than the non-poor when risingcollections are taken into account. Non-payments, or arrears, are one of the most vexing problems for utilities. Understanding who accumulates arrears has important implications for the welfare effect of reforms. If it is mainly the poor, affordability may be a problem and special care must be taken by the state to provide adequate assistance to the poor. If it is all households, free riding may be the problem and stricter enforcement will not disproportionately hurt the poor. As noted earlier, inAzerbaijan collections are lower for the bottom quintiles than the top. Assuming uniform enforcement of collections, this impliesthe poor will face abigger effective tariff increase than the non-poor (Table 6). It also implies that the poor will require slightly more compensation (closer to US$3 per month) than the non-poor (closer to US$2.5 per month) ifboth tariffs and collections are increasing at the same time. 12Illustrativecalculation 2.5%=(200 kWh*US$0.o2)/ US$158 per month. Calculations of income loss under alternative tariff increases are presented inAnnex 5. Upperbound = manat 9,600= (manat 144-manat 96) x 200 kWhper month. Lower bound = manat 8,640 = (manat 144- manat 96) x 180kWhper month. 8 Table 6:Compensation for the poor shouldbehigher (Baku only) Weljiare Effective Before Tariff After Minimum Maximum quintiles Tariff * (k W/ Increase" (kW/ Loss Loss (manat) month) (manat) month) (US$/month) (US$/month) Bottom 62 190 82 140 2.3 3.2 2 59 202 85 144 2.5 3.5 3 71 192 73 153 2.3 2.9 4 65 201 79 152 2.5 3.2 Top 78 200 66 166 2.2 2.7 Source:Authors' calculationsassuming elasticity of -0.20 Notes:* Effectivetariff = Table3 collectionratetimes 96 manat, "Tariffincrease=144manat-effectivetariff. Outside Baku, improving service quality while raising tariffs will help generate popular support for the reform. Raising tariffs and enforcing disconnections is very unpopular, and the public often views state actions in this sector with skepticism. This is especiallytrue when tariffs increasewithout any improvementin the quality o f service because there is a mismatch between the timing o f the gains (improved service) and costs (higher tariffs). Unpopularity can be avoided by explicitly linking tariffs and service quality. For households outside Baku, investments in metering, as well as rehabilitation and maintenance of the infrastructure, may prevent widespread supply shortages in the future. Such a strategy will also help generate popular support for the reform. Examples o f indicators that might be monitored to ensure better service quality include number o f outages and frequency and voltage stability. Householdswithout access to gas or wood inRayon centers and ruralareas maybe particularly vulnerable to tariff increases. There are no good substitutes for electricity for lighting, refrigeration, and television. However, wood, kerosene, LPG, and gas are viable substitutes for electricity in heating and cooking. Households that have few alternatives t o heating with electricity will have the greatest difficulty inshifting their energy consumption to less expensive fuels, making them more vulnerable to tariff increases. Dividing households around the country into groups based on location and access to gas and wood reveals that normative consumption is significantly higher (600- 700 kwhper month) amonghouseholds in'other urban' and 'rural' areas that don't have access to gas or wood (Table 7). In these areas, very high percentages o f households report heating onZy with electricity and they would be particularly vulnerable to tariff increases - especially if there are no improvement in service quality. Clearly better data is required electricity consumption and substitution behavior outside of Baku before-a definitive conclusion can be drawn on the magnitude o f the impact. One solution would be to pilot meteringinthese areas (where households have little access to substitutes) so as to observe actual electricity consumption. This data could then be used to determine the best mitigation strategy for households outside o f Baku. In Baku, 33 percent of households heat with electricity (12 percent heat onZy with electricity) and their average annual consumption is 3,363 kWh.15 This is about 615 kwh per year more than households that do not heat with electricity. In the short run, these households would require an additional US$5 to 6 per year incompensation for a 50 percent tariff increase unless they are able to start heatingwith gas. 15Thesefigures come from the 2003 Energy Survey. 9 Table 7:Householdswithless access to substitutes consume more electricity Location Access to Access to Hours of winter gas wood heating only with (kWh/month) electricitu (%I - - supply ~ Baku Yes No 246 12% 24 Other urban Yes No 403 19% 16 Other urban Yes Yes 361 2% 16 Other urban No No 713 76% 12 Other urban No Yes 427 33% 9 Rural Yes Yes 136 0% 22 Rural No Yes 504 2% 10 Rural No No 608 18% 12 Source: 2003 household energy survey Households with access to few alternatives should be given access to efficiency increasing technology or less expensive fuels for cooking and heating. A longer-term solution is to encourage households to undertake demand side management to reduce consumption o f electricity by using more efficient appliances and household insulation. Another option, assuming it is provided o n a full cost recovery basis, would be to encourage the use of clean and inexpensive substitutes for heatingand cooking such as natural gas. This can be done through a variety of instruments, as long as the government explicitly compensates the utility for any social transfers it provides. For example the Government could bid out competitive subsidies to encourage the extension o f natural gas networks to poor neighborhoods. While the households would still have to pay the full cost o f gas, the cost o f bringing the network to them could be partly financed bythe public sector. Raising electricity tariffs is not expected to have a significant impact on Internally Displaced Persons. There are 780,000 Internally Displaced Persons (IDPs) inAzerbaijan. While there are conflicting data on the incidence o f poverty within this group o f the population, mitigating the potential blow from tariff increases is a government priority. Currently IDPs receive an allowance, paid by the State Refugee Committee to Barmek and Bayva, o f 150 kWh per person per month. Assuming an average household size o f four, this is a public transfer o f 600 kwh per household per month. Since this is close to the upper limit o f household consumption (including for those heating with electricity), it is unlikely that IDPs consume more than this. Unless there is a change inthe size of the transfer, the only impact of raising electricity tariffs o n IDPs is on the state budget. There is no easy answer when considering the tradeoffs betweenalternative socialprotection strategies. The government can mitigatethe welfare effects oftariff increases by providing assistance to poor and vulnerable households and by stimulating income growth. There is much debate about the validity o f each assistance measure. Lovei and others (2000) found that instruments that perform well o n some criteria tend to perform poorly on others. Furthermore, not all subsidy mechanisms are applicable or perform equally well across all countries and utility services. Thus no single instrument has been identifiedthat outperforms all others. Proponents o f direct transfers argue that lifelines are not targeted and thus encourage inefficient energy use. Opponents claim that transfers through the general social assistance system, while theoretically attractive, can be very difficult to implement effectively. In evaluating the alternatives, careful consideration must be given to where the money for a compensation scheme will come from and what the welfare costs o f raising the money will be. Whether people will be 10 better off if they are compensated with an income transfer depends on how the money for the transfer (compensation) is raised. If it is through taxes, the distortionary effect scheme - and one that will beborne by the same households the scheme is designed to (welfare cost) of raising tax revenue must be considered a cost of the compensation compensate. Income transfers tend to be well targeted in countries with a small percentage of the population below the poverty line. In this case, so long as there are enough funds to finance the administration of social assistance and the informal sector is small, means testing is easy. Examples include Hungary and Poland. The transfers are lesswell targeted incountries where nearlyhalfthe populationis poor, budget resources are insufficient, and means (or proxy means) testing is very difficult because there is a large informal sector. The case for lifeline tariffs is stronger in countries with high poverty rates and poor targeting-so long as there is sufficient political will to keep the size of the blocks small (below 50 or 100kwh) and to reimburse the utility for its costs - than in the countries with low poverty rates and well targeted social assistance. LINKSTOAGRICULTURE The Government is concernedthat higher electricity tariffs could slow rural sector growth. Agriculture is the second largest sector of the economy, contributing 16 percent to Gross Domestic Product (GDP) in 2002 and employing 40 percent of the workforce. Agro-processing i s less important, accounting for 3.2 percent of total industrialproduction and less than 1percent of GDP in 2001and i o percent of the total industrial workforce.16 The Government considers agro-processing as a strategic industry withhighcomparative advantage and stronggrowth potential, and its failure as a result of higher input costs could compromise the Government's poverty reduction strategy in rural areas. Higher electricity tariffs affect farmers and agro-processors through two channels. First, the cost of operating agricultural equipment (e.g. water pumps, feed preparation, milking machines, lightening) or agro-processing equipment (e.g. pasteurizing, packaging) may rise. Second, the cost of farm inputs that are produced using electricity such as chemical fertilizers, pesticides, agricultural machinery, and agro-processing equipment, and irrigation (based on electrical pumps) may rise. If farmers and agro-processors cannot pass increased production costs to consumers, their profit margins will decline. This may force some of them out of business and slow down future sector growth, unless productivity gains from better electricity supply outweigh these losses. Higher electricity tariffs are not expected to tangibly affect farmers' direct operating costs. The agriculture sector consumes little electricity and therefore changes intariffs are not expected to have a significant impact on productivity. In2002, agriculture consumed 662 million kwh electricity, or 4.3 percent of total electricity consumption inAzerbaijan.l7 The share of electricity intotal production cost ranges from 0.5 to 2 percent (Table 8). 16State Statistical Committee ofAzerbaijan: Statistical Yearbook ofAzerbaijan 2002. 17State Statistical Committee of Azerbaijan: Balance of Fuel- Energy and Material Resources of Azerbaijan 2002. 11 Table 8:Share ofelectricityintotal agricultureproductioncost islow Cropproducts percent Livestockproducts percent Total of crops 0.5 Total of livestock products 1.7 Grain and legumes 0.4 Cattle 3.2 Potato 0.1 Pigs 0.4 Crude cotton 1.9 Sheep & goats 1.5 Tobacco 0.0 Poultry 1.9 Vegetables 0.2 Milk 0.7 Orchardproducts 0.1 Wool 2.9 Fruits &berries 0.1 Eggs (1,000) 0.4 Grape 0.0 Other livestock products 2.6 Other crops 0.3 Source: MinistryofAgriculture, 2002. Higher electricitytariffs are not expectedto tangibly affect farmers' indirect operating costs either. Use of agriculturalinputs and equipmentis at historicallylow levels in Azerbaijan and therefore higher electricity tariffs are not expected to have a significant on indirect operating costs. Total mineral fertilizer use in iggg dropped to only 1percent the 1990 level. In 1999, 13 percent of the agriculture area was fertilized (down from 64 percent in 1990). Virtually all fertilizer used in Azerbaijan is imported (domestic fertilizer production amounted to only 5 tons in 2002). Inthe last decade, the number of agricultural machines inuse dropped 20 percent for combines and go percent for milking equipment. The existing capital stock is outdated and completely depreciated. In 2002, only 275 pieces of agricultural machinery were produced in Azerbaijan.18 Irrigation is an important concern, but a plan is in place that will help cushion potential adverse impacts. Azerbaijan's agriculture is dependent on irrigation. In2003 about 1.3 millionhectares were under irrigation. This is 27 percent of agricultural land and 73 percent of all arable land. Energy accounts for 30 percent of total irrigation cost, so that an increase in electricity tariffs would raise the State Committee for Amelioration and Water Resources irrigation costs significantly. However, this would partly be offset by lower operations and maintenance costs from less frequent electricity service interruptions associated with investment inthe network. The magnitude of these gains is unclear. To the extent the Committee passes higher irrigation costs to farmers, those specializing inirrigation intensive crops will face rising production costs. Farmers currently pay only 8 to IO percent of the irrigation costs, and there is a plan in place to gradually increase irrigation fees to cost recovery levels. This plan will cushion irrigated farmers from the short-term adverse effect of higher electricity tariffs. Therefore, additional mitigation measures may not be necessary. Furthermore, better irrigation service combined with increased fertilizer application can lead to a significant increase in yields (50 to 200 percent depending on crop and location). Revenues from higher yields will help offset higher productioncosts. Benefits from reliable supply will outweigh the impact of higher tariffs on agro-processors.The share of electricity in overall production and marketing cost is low (1 to 4 percent). And there is almost no local manufacturing of agro-processing equipment, indicating that prices of capital goods used by agro-processors would not increase. Ina series of case studies conducted for this study, agro-processors stated that they would benefit from more reliable energy supply through additional revenues and 18State Statistical Committee of Azerbaijan, IndustryDepartment. 12 cost savings. They quantified this as IO to 15 percent of their overall production cost, leaving plenty of room for them to absorb higher electricity costs. Finally, the agro- processors interviewedalso estimatedconsumers would tolerate a tariff increase of up to 20 percent. This suggests the loss of competitiveness due to higher incremental production costs would beminimal. LINKSTO FORESTRY Raising the electricity tariffs (especially without better service) may push some people without access to gas to burn wood. In turn, increased wood consumption for heatingmight impose costs on society by leading to environmental and health damages. This section identifies where these damages are most likely occur and proposes mitigating measures. Due to data shortcomings the analysis is not explicitly relatedto electricity tariff increases. Wood is animportant source ofenergy for the ruralpopulationlocatednear forests. In rural areas with access to forests, 85 percent of households use wood for heating. While more than half of these households use it as their only source of energy for heating, 40 percent use it in combination with electricity. These households may increase consumption of wood when electricity tariffs increase. While wood use for heating is frequently cited as a cause of deforestation, it is not clear to what extent this is driven by household behavior. At present, 11percent of the country is covered with forest, much of which is extremely degraded and has turned to scrub and steppe. Forest degradation is often cited as the reason landslides have become more frequent since 1980.~9 Increased indoor air pollution may pose a health risk. Given a lack of epidemiological data, it is not clear to what extent the burningwood causes respiratory diseases in Azerbaijan. However, households that burn wood are aware and concerned about the potential associated respiratory illness. Of the 860 households in the 2003 Energy Surveythat reported having a wood stove, 60 percent said that the room where the stove was located sometimes filled with smoke and 88 percent of the latter were concerned that the smoke could cause respiratory diseases.According to estimates of the World Health Organization, indoor smoke from use of solid fuels is estimated to cause worldwide about 36 percent of lower respiratory infections and 22 percent of chronic obstructive pulmonary disease worldwide.'" Identifying deforestation risk associated with rising electricity tariffs requires strong assumptions. The analysis assumes geographic areas that currently have high deforestation risk will be at greater risk when tariffs rise. Deforestation is driven by the following factors: (1) residential wood consumption (therefore it neglects logging for commercialpurposes and that wood is cut and transported long distances to besold inother 1ocations);'l (2)poverty (poor people are more likely to burnwood); (3) access to forests (slope); (3) access to gas; (5) proximity of population density to forests. Poverty was measured by the head count (number of people below the poverty line), 19Azerbaijan Republic Ministryof Ecology and Natural Resources (2002): Naturalprogram on restoration and expansion of forests inthe Azerbaijan Republic. 20World Health Report 2002, Reducing Risks, Promoting Healthy Life, WHO, 2002. 21According to information from officials from the Ministry of Ecology and NaturalResources, wood exports are not substantial. 13 calculated for each district applying the Elbers-Lanjouw technique to the 1999 Census and 2002 household budget survey data. Access to forests was measured by road density (withini okmof forests) and by slope.22 The proportion of ruralhouseholds in a given district that have gas connection calculated from the Census data measured access to substitutes. The population density is the rural population within a i o km buffer o f the forest, calculated using Census data. Only rural households are considered because no more than 8 percent of urban households report usingwood. The risk of increased residential wood use is greatest in Khachmaz, Southeast, Southwest, andNakhichivan.23 The results of the analysis are shown in Map i(wood use risk is highest indark red areas). To generate the map, each district was ranked according to each risk factor above, assigning the lowest rank to the lowest risk level in each case. Then the ranks for each risk factor were summed for each district to form a risk index for increased wood use. The districts were then mapped according to the risk index. The map indicates the potential for increased residential wood use in all districts. Under the assumptions detailed above, this can be interpreted as deforestation risk for districts with forest coverage. Where there is no forest cover (such as in Southwest), the map demonstrates where the total vegetation cover is threatened by increased wood use, including peri-urban trees, tree lined alleys, tree-lined rivers and hedgerows. Since wood burning is associated with health risks from indoor air pollution, the dark (red) areas in the map also show where the potential health risks from indoor air pollution resulting from wood burning are highest. Given financial constraints, improvements in service quality will be gradual. Therefore, it may make sense to direct early infrastructure investments in service quality towards the high-risk areas to reduce the threat of increased wood use and associated negative health impacts. 22The roads data was digitized from Soviet 1:ioo ooo scale maps, 1975 - 1981supplied from RISKgroup. The slope was calculated applyingthe Hornmethodto the DigitalTerrain Model data suppliedbythe Shuttle Radar Topography Mission. 23Areas at lower but still elevated risk are Khachmaz inthe Northeast; Sharur inNakchivanAR; Agdam and Barda inthe center; and Jalilabad andYardymly inthe South. 14 Map 1:Riskofincreasedresidentialwood use Source: GeoData, 2004 Forests areas at greatest risk are in Khachmaz in the North, and in the Southeast. Overlaying Map 1with the forest resources inAzerbaijan, shows the areas where deforestation will most likely occur (see M a p 2). The most efficient use of limited resources to protect the forest should be targeted to Khachmaz in the North, and the Southeast. The environmental project currently being prepared by the Government and the Word Bank will help address these resource management issues inthe North. 15 I Map 2: Deforestationrisk is highest inthe Northeast and Southeast I ISource: GeoData, 2004 I Promoting access to alternative energy sources and more efficient wood stoves in high risk areas may help reduce deforestation. Promoting access to alternatives will increase people's choice of energy sources and is likely t o increase the incentive to substitute wood for other energy sources, thereby reducing pressure o n forest resources and illnesses related to indoor air pollution. Encouraging the use of more energy efficient stoves in high risk areas would reduce the amount of wood consumed andhave positive environmental as well as health effects. Preparing and implementing spatially explicit forest management plans below the district level may improve forest management. Developing Geographic Information Systems that take into account the specific characteristics of and threats to forests indifferent locations will help improve policy targeting and impact monitoring. Promote participatory forest management that meet the short-term and long-termneeds of the people cutting wood inhighdeforestation risk. Well- designed and implemented forest management helps to maintain biodiversity and conserve the natural functions o f a forest, while also generating income for rural communities. STAKEHOLDERANALYSIS This stakeholder analysis is based on a set of in-depth interviews with key informants. The objective of the stakeholder analysis was to identify key elements of the reform package that are not supported by the stakeholders and why. A series o f in- depth interviews was conducted with representatives from the Presidential 16 Administration, Cabinet of Ministers, Ministry of Economic Development, Ministry of Fuel and Energy, Ministry of Labor, Ministry of Environment and Natural Resources, Parliament, EnergySector enterprises, the media, andNGOs. There is general consensus on the key reform measures. Stakeholders from various backgrounds generally agree that the key elements of the reform are the same. These include the need for tariff reform; that mitigation policies must be in place to cushion adverse social effects of reforms; that metering and collection rates must be improved first; that higher tariffs and collection rates should be accompanied by improved service; and that private sector participation is necessary. Consensus on the broad reform measures suggests that the risks of controversy when engaging in a more transparent dialogue are low. Provide a transparent forum for dialogue on key reform measures. Only a limited number of key players in the Ministry of Economic Development and the Ministryof Financeare fully aware of the actions beingconsideredas part ofthe reform program. Other stakeholders are less well informed and often feel excluded from the process. For example, some of the stakeholders do not think the Tariff Council, with its limitedmandateoftariff setting, isthe appropriate institutionto facilitatethe dialogue. Target non-energy enterprises and general population with information about the potential benefits of reform. While they have little influence on the decision-making, the groups have raised concerns about the reform. Non-energy enterprises are concerned about losing competitiveness due to higher production cost from higher electricitytariffs. Households are afraid that they would end up paying more and still not receive sufficient electricity supply. These concerns can be addressed through a targeted public information campaign. If properly executed, such an information campaignwill helpbuildbroadpublic support for the program. Provide information on the full set of policy choices to key analysts in Government. There is anunmet demandfrom senior government staffworking onthe reform program to learn from the experience of other countries that have undergone similar reform programs. Some of the most pressing questions relate to mitigating social impacts, tariff design, and attracting private investors. While a fair amount of technical assistance is already being provided in this area, it may be possible to enhance this assistance through a series of study tours to countries such as Hungarywhere the reform program has been very innovative. Another idea would be to provide counterparts with translations of key sector documents that provide an overview of the challenges encountered duringimplementation of different reformprograms. 17 Annex I: Summary of Combined Household Survey andUtilityDatafor Four Countries Electricity Average Price of Monthly kWh electricity Monthly expenditures aggregate electricity per (utilities income aspercent of collection rate kWh records) (stated in HBS) '/ income stated 2/ paid/billed UScents/kWh kWh US$/month US$/month % Georgia (Tbilisi)3/ 2000, q1 22% 4.55 205 2.7 168 2.0% 92 24% 4.55 207 2.9 138 2.5% 93 31% 4.68 179 2.7 171 2.3% 94 35% 4.95 146 3.5 171 2.9% 2001, 91 62% 4.73 146 4.2 182 2.7% 92 56% 4.73 156 4.9 169 3.5% 93 64% 4.73 128 5.2 164 4.1% 94 73% 5.15 143 5.6 169 4.0% 2002, q1 133% 5.64 173 5.8 165 4.2% q2 77% 5.64 170 6.0 164 4.4% 43 73% 5.64 139 4.0 172 5.9% Moldova (Union Fenosa service area) 5.64 94 75% 151 5.9 189 4.4% 2001,l 100% 5.05 43 3.0 97 3.4% 2 100% 5.05 37 2.1 47 4.1% 3 99% 5.05 41 2.9 53 5.7% 4 100% 5.05 41 2.5 84 4.7% 5 100% 5.05 39 2.2 73 2.5% 6 100% 5.05 29 2.0 48 4.5% 7 100% 5.05 29 2.2 50 5.5% 8 99% 5.04 47 3.2 73 6.0% 9 100% 5.01 51 3.0 72 5.4% 10 100% 5.03 56 3.0 90 4.5% 11 100% 5.25 60 3.2 72 5.1% 12 100% 5.26 57 3.0 72 4.9% 2002,l 100% 4.99 62 3.2 76 4.3% 2 100% 4.99 53 2.8 73 4.9% 3 99% 5.00 50 2.6 70 4.7% 4 100% 5.00 49 2.7 69 5.0% 5 100% 4.98 53 2.6 74 4.0% 6 98% 4.99 45 2.5 73 4.4% 7 99% 4.98 50 2.6 77 4.3% 8 100% 4.99 50 2.8 87 4.1% 9 99% 5.26 52 3.0 109 3.6% 10 100% 5.25 52 2.6 99 3.4% 11 100% 5.26 60 3.1 94 3.8% 12 100% 5.27 64 3.5 102 3.9% 2003,i 98% 5.10 69 3.0 98 3.6% 2 100% 5.13 58 4.0 94 4.5% 3 99% 5.13 65 3.3 90 3.9% 4 99% 5.15 56 3.0 100 3.5% 5 100% 5.13 56 3.3 95 3.8% 6 99% 5.13 55 3.1 92 4.1% 7 96% 5.17 52 3.0 104 3.7% 8 97% 5.61 51 2.7 102 3.4% 9 96% 5.55 51 3.4 127 3.9% 10 95% 5.54 60 2.9 105 3.4% 11 85% 5.51 61 3.3 91 4.5% Armenia (Yerevan) 3/ Jun-Dec '98 89% 3.80 173 5.9 100 9% Azerbaijan 2002 (all months) Baku, only meteredhouseholds Poorest 20% 65% 1.96 190 2.2 123 2.1% 2 61% 1.96 202 2.1 137 1.9% 18 3 74% 1.96 192 2.3 154 1.9% 4 68% 1.96 201 2.4 161 1.9% Richest 20% 81% 1.96 200 2.6 189 2.2% Total 71% 1.96 198 2.3 158 2.0% Source: Calculated from householdsurvey data and utilitycompany billingrecords. Notes: i/Incomeproxiedbytotal monthly household expenditures. 2/ inArmenia and inAzerbaijan, electricity expenditures shown here are not stated inthe survey, butcalculated as anaverage monthly electricity payment from theutilitycompany records. 3/ Decreasingelectricity consumption despite increasing income maybe due to rationing. 19 Annex 2:DataReliability The 2002 Household Budget Survey data in Baku were merged (on a household by household basis) with 2002 and 2003 (all months) records on billing and payment from Barmek. These are records for metered households only, as records for non-metered households are not considered reliable. The table below shows the distribution o f total per capita expenditures by welfare quintile in the original HBS sample and in the sub- sample used inthe analysis. The retained sub-sample does include a higher percentage o f households inthe top quintile. 2002 HBS Samplefor Baku Subsamplefor Baku used in the analysis Quintile Income, manat/month N Income, manat/month N Bottom 20% 627,915 439 603,939 2 7271123 438 673,301 3 762,290 439 739,544 4 812,958 438 770,663 Top 20% 941,158 438 918,995 Note:Total 774,217 2,192 775,405 1,094 income proxiedbytotal expenditures. Quintiles generatedbased on per capita total expenditures. Quintiles were defined only once basedonthe entire HBS Bakusubsample. The table below shows the distribution of households by Baku rayon (district) in the original HBS sample and in the sub-sample used in the analysis. The retained sub- sample does include households from all rayons inBaku.0 Rayon 2002 HBS Sample Subsamplefor Baku for Baku used in the analysis 1 188 105 2 101 23 3 188 50 4 94 63 5 308 108 6 174 36 7 191 153 8 261 134 9 204 139 10 282 183 11 178 100 Total 2,169 LO94 Note: Bakuhas a total of 11rayons. 20 Annex 3: ReportedAverage HoursofElectricitySuppliedper Day During Winter ; * Baku, n=497 Symgayit, n=150 120 - 100 - g 8 0 - 0 v) 60 - r 40 - c 40 - CI 0 0 C CI Q 20 - 0 - 18 19 24 Q 24 average hoursof service per day average hoursof service per day Rayon Centers, n=390 Rural, n=721 f 25 yQ2 0 - 0 15;1 2 10 1 2 3 4 5 6 7 8 91011121314151617181924 1 3 5 7 9 11 13 15 17 19 average hours of service per day average hoursof service per day Source: reported in the 2003 Azerbaijan Household Energy Survey. This is response to the question: On average how many hours per day is electricity available for your household during the winter? Households that reported having more than 20 hours of supply per day ina preceding question are treated here as having 24 hours of supply on average during the winter. The summer hours of supply are higher than winter. 21 Annex 4: ElectricityDemandModel Welfare effects o f a tariff increase can be evaluated using results of an electricity demand model. Our empirical strategy is to estimate the pooled model of electricity demand using household survey data for four countries: Azerbaijan, Armenia, Georgia and Moldova. These household survey data were merged household by household with the payment and billing records provided by the electric utilities for limited samples o f households inthese in the capital cities of each country. Pooling these data sets creates enables us to estimate the price elasticity of demand. Datasources (metered householdsonly) Observations in Percent of the merged data Utility data Source of household households set data interviewed once, but not more Armenia (Yerevan) 544 6 monthsbefore and Dec. 98- Jan. 99 Energy after survey Survey 100% Azerbaijan (Baku) 1,094 2002 (all months) 2002 Household Budget Survey 100% 2000,2001,2002 Georgia (Tbilisi) 2ooo, 2o01, 2o02 Household Budget 11'985 months) 18% Survey Moldova (Union 2o01, 2o02>2003 2000,2001,2002 HouseholdBudget FenosaServiceArea) 6'553 months) 59% Survev The pooled data set is dominatedbythe data from Georgia. Estimation o f a single model o n the pooled data set combining these four countries imposes the assumption that these countries have similar conditions, particularly in the household energy sector. This is reasonable given all countries share many common characteristics and are at approximately the same stage o f the transition process. The biggest difference is in the per capita income and access to substitutes. These factors are accounted for in the model. Electricity DemandEstimation Procedure andResults We estimate the following model of electricity demand: kWhijt=f(priceijt, incomeijt, use of substitutesijt, enforcementjt, demographic characteristicsijt, seasonjt, locationjt), where idenotes an individual i,t - period t, and j - country j. In order to control for correlation in the error terms by household, we use a random effects model. The unbalancedpanel nature o f the data is automatically taken into account inthe estimation of one-way random effects using STATA. N o t surprisingly, the model is much better at explaining the differences between households than it is explaining the differences between observations in different time periods for the same household. This is evident from the relatively high R-square that explains the variation between households. The signs of the coefficients o n most variables are as expected for all variables except for use of central heating. We would expect household with central heating to have a lower demand for electricity, since part of electricity consumption is often for heating. However, the central heating and central gas variables do not reflect the quality o f these 22 services. Central heatingequipment maybe installed, but much o f the time it may not be working. With central gas, the story is similar even if the coefficient is negative as expected. Gas pressure is frequently so low even in the areas where gas is supplied, that it cannot be usedfor heating, and even cooking. Itis important to notethat the impact of LPG relies o n data from other countries, since in Azerbaijan only 1 percent of households were using LPG. Similarly the impact o f being connected to central gas relies on data from other countries since loo percent of Azeri households in the data set are connected to gas. The most policy relevant result of the model is the price elasticity o f electricity demand. According to the model, a IO percent increase in the price o f electricity results in a 2.1 percent decrease in electricity consumption. The elasticity estimate falls in the 95 percent confidence interval of -0.14 to -0.28. This estimate is reasonably consistent with the studies that have estimated residential electricity demand inother parts o f the world. ~~ ~ Countru PriceElasticity IncomeElasticity Source Ethiopia -0.74 1.005 Kebede, Bereket, Almaz Bekele and Elias Kedir. Can the UrbanPoor M o r t Modern Energy?The Case of Ethiopia. EnergyPolicy, 30,2002 Greece -0.41 1.56 Hondroyiannis, George. Estimating Residential Demandfor ElectricityinGreece, Energy Econommics, 2004. India -0.42 (winter) 0.6-0.64 Filippini,Massimo and Shonali Pachauria. Elasticities -0.51 (monsson) of electricity demand inurban Indianhouseholds. 0.29 (summer) EnergyPolicy 32 (2004) 429-436 Norway -0.5 (short-run) 0.2 Nesbakken, Runa. Price sensitivity of residential energy consumption inNorway. Energy Economics 21 1999. Taiwan -0.15 1.04 Holtedahl, Pernille, and Frederick L.Joutz, Residential Electricity DemandinTaiwan. Energy Economics, 26, 2004. UK -0.5 0.5 Manning, D.N.Householddemand for energy inthe UK,EnergyEconomics, January 1988. USA -0.5 0.62 Silk, Julian I.Frederick L.Joutz. Short and long-run elasticities inUSresidentialelectricity demand: a co- integration approach. EnergyEconomics 19 (1997) USA -0.27 Wills, John. Residential demand for electricity. Energy Economics, October, 1981 It is important, however, to differentiate between the short term and long term price elasticities. In the short run, the elasticity is likely to be lower than in the long run, because over a longer time period a household is able to better adjust to the new relative prices of fuels and switch to substitutes. Fuel switching is only likely to occur for the more energy intensive but less essential uses o f electricity. For example, a household may be able to find such substitutes as fuel wood or even LPG for heating, but it is difficult to substitute the use of electricity for lighting. Kerosene is not used for heating and is a much inferior alternative for lighting. Therefore, we are likely to observe non- constant price elasticity of demand. At a very l o w level of consumption, the elasticity is likely to become very low, and welfare losses associatedwith further price increase very 23 high. We tested for the non-constant elasticity of demand in this model by testing flexible functional forms and using spline functions, but due to the nature of the data the model didnot produceplausibleresults.24 We found the income elasticity of demand of only 0.12, which means that a io percent increase in the income results in a 1.2 percent increase in consumption. The income elasticity is very stable in a variety of specifications, and was never higher than 0.20 in any of the models that were tested. Controlling for GDP by country may capture some of the effect of income and omitting this variable might lead to a slight increase the magnitude of this coefficient. Payratio variable is an aggregate collection rate, defined by quarter, month, or welfare quintile rather than at the individual level, depending on a country. Itis a proxy for the strength of enforcement. Since it is defined at an aggregate level, there is no problem of endogeneity. This is an important variable to control for the varying level of arrears by country. As expected, stricter enforcement of payment discipline results in a lower level of electricity consumption. 24The four country dataset contains observations at low prices only for Azerbaijan and the pre-price increase period for Armenia. Creating cross-terms or square terms usingthe price variable makes these variables collinear withthe other country-level variables. Usingthe spline function results inthe same problem: as we are separatingthe observations inone or two entire countries as the "low price" observations, and the rest as the "high price" observations. 24 DescriptiveStatistics for Variables Usedinthe Model Variabledescription Variablename Armenia Georgia Moldova Azerbaijan Variables innaturallogs: Monthly electricity consumption lnkwhm 5.21 4.79 3.56 4.90 Electricityprice lnpELE (3.27) (2.98) (2.97) (3.93) Monthlytotal expenditures lntotexp 4.6 5.1 4.5 5.1 (proxy for income) Household size lnhhsize 1.35 1.19 0.82 1.12 Same variables as above, expressedinmeasurement units: Monthly electricity consumption, kwh kwh-month 183 120 35 134 Electricityprice, US cents/kwh pELE-kwh 0.038 0.051 0.051 0.020 Monthlytotal expenditures totEXP 99 169 91 158 (proxy for income), US$ Household size hhsize 3.9 3.3 2.3 3.1 Dummvvariables: Use liquidpropane gas (LPG) useLPG 0.37 0.36 0.45 0.01 Connectedto central heating(CH) (may useCH 0.22 0.38 0.61 not necessarily be a working connection) Connectedto central gas (CG) useCG 0.08 0.32 0.44 1.00 (may not necessarily be suppliedat high enough pressure) Other variables controlling for auarterlv and cross-country differences: Aggregate collections ratio, by quarter or payratio 0.89 0.61 0.99 0.71 month and country Quarterly inflation inflation 2.00 2.70 2.17 2.00 Quarterly GDP per capita dPPC 5.05 5.03 4.60 5.22 Quarterly average temp 10.00 13.01 10.78 15.58 30year temperature Number of observationsused in the estimation 539 11,740 6,295 1,094 Note: a few observations were dropped if a household did was not billed for, or did not pay for electricity in any of the months inthe periods included inthe estimation. Zero billing and payment records indicate that a household has either moved, did not merge correctly with the household budget survey data or isn't tracked by the utilitycompany. This explains the discrepancy withthe figures reportedinTable 1intext. 25 RandomEffectsGLS RegressionResults(Log form) Dependentvariable: natural log of kWhper month i i s (hhid) x t r e g lnkwhm 1npELE 1ntotexp lnhhsize useLPG useCG useCH p a y r a t i o i n f l a t i o n lngdppc temp Random-effects GLS regression Number o f obs - - 19665 Group v a r i a b l e (1): hhid Number o f groups = 6556 R-sq: w i t h i n = 0.0000 Obs p e r group: m i n = 1 between = 0.3537 avg = 3 . 0 o v e r a l l = 0 . 2 7 4 1 max = 1 2 Random e f f e c t s u-i - Gaussian Wald chi2 (10) = 2396.77 corr(u-i, X) = 0 (assumed) Prob > chi2 = 0.0000 lnkwhm I Coef. S t d . E r r . z P > l Z / [95% Conf. I n t e r v a l ] lnpELE - .2094197 .0371998 - 5 . 6 3 0 . 0 0 0 - .28233 - .1365093 I n t otexp .1177077 .0107378 1 0 . 9 6 0 . 0 0 0 .096662 .1387534 lnhhsize .3001163 ,0162547 1 8 . 4 6 0.000 .3319749 useLPG - .150281 .015325 - 9 . 8 1 0 . 0 0 0 -..26a2577 1803173 - .1202446 useCG - .0786205 . 0 1 8 2 1 1 1 - 4 . 3 2 0 . 0 0 0 -.1143137 - .0429273 useCH .1755975 .0265015 6 . 6 3 0 . 0 0 0 .1236556 .2275395 p a y r a t i o - .2891546 .0336239 - 8 . 6 0 0.000 - .3550563 - .2232529 i n f l a t i o n - .0052019 .0025918 - 2 . 0 1 0 . 0 4 5 - .0102817 - .0001222 1ngdPPc .929897 .0445859 2 0 . 8 6 0 . 0 0 0 .8425104 temp -.0116777 .0010328 . 1 1 . 3 1 0 . 0 0 0 - .0137018 -.1.017284 0096535 -cons - 1 . 3 6 8 2 3 2 .208619 - 6 . 5 6 0 . 0 0 0 - 1 . 7 7 7 1 1 7 - .9593458 sigma-u ,79978476 sigma-e .55688425 rho .67348075 ( f r a c t i o n o f variance due t o u-i) 26 Annex 5: HouseholdIncome Lossunder Alternative Tariff Scenarios % increase Tariff Tariff Consumption Max income loss Min income loss (manat) (US$) (kwh) (US$/month) (US$/month) 0% 96 0.02 200 0 0 50% 144 0.03 180 2.0 1.8 100% 192 0.04 160 3.9 3.1 150% 240 0.05 140 5.9 4.1 200% 288 0.06 120 7.8 4.7 27 a Annex 6:References Abdel-Khalek, Gouda. Income and price elasticities of energy consumption in Egypt A time-series analysis. Energy Economics.January 1988. Bernard, Jean-Thomas, Michel Lemieux and Simon Thivierge, Residential energy demandAn integratedtwo-levels approach. Energy Economics,July 1988. Donatos, George S. and George J. Mergos. Residentialdemand for electricity: the case of Greece. Energy Economics.January 1991. Fiebig, Denzil G., James Seale and HenriTheil. The demandfor energy: Evidencefrom a cross-country demand system. Energy Economics.July 1987 Filippini, Massimo and Shonali Pachauria. Elasticities of electricity demand in urban Indianhouseholds. Energy PoZicy 32 (2004) 429-436. Flaig, Gebhard. Household production and the short-and long-run demand for electricity. Energy Economics.April 1990. Garbacz, Christopher. A model of residential demand for electricity using a national household sample. Energy Economics.April 1983. Huisman, Ronald, Ronald Mahieu. Regime jumps in electricity prices. Energy Economics25 (2003) 425-434. Pachauri, Shonali. An analysis of cross-sectional variations in total household energy requirements inIndiausingmicro survey data. Energy Policy, 2004. Manning, D. N. Household demand for energy in the UK, Energy Economics, January 1988. Nesbakken, Runa. Pricesensitivity of residential energy consumption inNorway. Energy Economics 211999. Silk, Julian I.Frederick L. Joutz. Short and long-run elasticities in US residential electricity demand: a co-integration approach.,Energy Economics 19 (1997) 493-513. Wills, John. Residentialdemandfor electricity. Energy Economics,October, 1981. 28