POLICY DIALOGUE ON AGRICULTURE MODERNIZATION IN UZBEKISTAN STUDY OF WHEAT AND FLOUR MARKET INTEGRA- TION IN UZBEKISTAN1 JULY 29, 2020 1 This Report was prepared by Sergiy Zorya, Lead Agriculture Economist, the World Bank, and Stephan von Cramon-Taubadel, Yali Mu and Carlos Barrantes, Department of Agricultural Economics and Rural Development, University of Göttingen, Germany, under support of the Multi-Donor Trust Fund for the Program of Environmen- tally and Socially Sustainable Cotton Production Development in Uzbekistan funded by the European Union, Ger- many, Switzerland, and the United States of America. Executive Summary 1. The year 2021 will see abolishment of the state production targets, procurement, and price setting for wheat in Uzbekistan. This was announced on March 6, 2020 by the Presidential Decree PP-4634. These reforms will complete price liberalization of the wheat- flour-bread value chain, which started with liberalization of ‘social bread’ prices in September 20182 and continued with liberalization of flour prices in October 20193. Market-based wheat price formation will bring significant economic benefits to farmers. In most years they received much less for wheat sold to the state than they received from selling it on the market4. During 2010-2018, the total loss of wheat farmers was estimated at 15 trillion soms or 1.7 trillion soms a year (World Bank, 2019). In 2018 alone the loss was 3.3 trillion soms, an equivalent of $290 per hectare and 0.9 percent of GDP. Note that ‘wheat farmers’ every second year become ‘cot- ton farmers,’ being the part of compulsory cotton-wheat rotation under the state order. In 2018, farmers lost on cotton about 1.5 percent of GDP (World Bank, 2018). Thus, no wonder that land and labor productivity of Uzbek farmers, producing cotton and wheat, remains very low. On the positive side, in 2019 the state farm-gate prices for both cotton and wheat were in- creased, largely reaching the level of market prices, for the first time in Uzbekistan’s history. The anticipated liberalization of wheat sector in 2021 should cement this achievement. 2. Abolishment of the state order should be complemented by other measures to en- sure wheat and flour market efficiency in Uzbekistan. This is because the empirical evidence generated by this Report shows that these markets have not been functioning efficiently so far. Spatial wheat and flour price differences in the regions, arising from supply-demand mismatch, have not been corrected quickly and widely enough as would have been expected in the market economy. In other words, price transmission between market pairs in Uzbekistan has been low and slow, often twice as slow as in the peer countries. In addition, vertical relation between wheat and flour prices in Uzbekistan is different from that in other countries. And, wheat and flour prices in Uzbekistan are decoupled from prices in the neighbouring countries, including Kazakhstan, which represents the world market price for Uzbekistan. When domestic prices are decoupled from world market prices, production in Uzbekistan does not respond to global mar- ket opportunities. So far, the main reason of these market inefficiencies has been the price con- trols imposed by the state order system. But other reasons, including: (i) the lack of accurate and publicly available information about regional prices; (ii) lacking data on supply, demand, and stocks; and (iii) likely interventions of the authorities in formation of regional wholesale and retail prices, could also contribute to this inefficiency. 3. Inefficiency of markets reduces gains of the market economy. First, it takes away business opportunity for private sector to engage in arbitrage, i.e. moving commodities between markets or store commodity in anticipation of future price increases, thereby reducing private investments in transportation, logistics, and storage. Second, inefficient markets create risks for politicians. Having access to affordable bread and pastries is strongly associated with achieve- ment of food security in Uzbekistan. In February 2020, according to the Central Bank of Uz- bekistan, the households spent about 14 percent of their food expenditures for bread and pas- tries. As a result, when wheat and flour prices rise too high for too long, because of the slow movement of commodities from surplus to deficit regions, the authorities tend to ‘intervene’ by fixing prices and penalizing ‘speculators,’ which undermines market efficiency by itself and reduces the allocative efficiency of resources in the economy. And third, inefficient markets 2 Cabinet of Ministers’ Resolution No. 731 “On Measures to Secure the Supply of Wheat, Flour, and Bread for People and Economy based on Competition and Introduction of Market Instruments,” dated September 13, 2018. 3 Cabinet of Ministers’ Resolution No. 866 “On Measures for Comprehensive I mplementation of Market Tools in Wheat, Flour and Bread Supply System,” dated October 14, 2019. 4 Under the state order system, the state purchased about 3 million tons a year (i.e. roughly half of total wheat production) at the fixed farm-gate prices 2 reduce the effects of disbursement and procurement of public grain stocks, which will be used more often after the end of the state procurement system in 2020. Public stocks are often dis- bursed with an objective of smoothing prices in all regions, which would be difficult to achieve if price signals between markets are low and slow. 4. Market efficiency is best presented by the concept of price transmission/market integration. It relies on arbitrage conditions, which dictate that prices that wander too far apart trigger spatial activities that act to draw prices together, i.e. by buying where the commodity is cheap and selling where it is demanded. Thus, arbitrage ensures that price differentials (i.e., the difference between two prices at the same point in time) will be disciplined so as to not wander arbitrarily in excess of transport or processing costs. Adherence to perfect price transmission is often termed as the “Law of One Price.” 5. The Report used cointegration methods to look for evidence of integration on Uz- bekistan’s markets for wheat and flour during January 2014-February 2020. Three types of market integration were explored: (i) spatial integration between regional markets within Uzbekistan; (ii) vertical integration between wheat and flour on individual regional markets in Uzbekistan; and (iii) spatial integration between Uzbek wheat and flour markets and those in neighboring states. Main results of the empirical analyses are the following: a. Only 50 percent of the regional wheat markets display evidence of integration. This is a sign of the low market efficiency in Uzbekistan. Wheat market integration appears to take place in two separate spatial clusters - one in the Northeast and the other in the South-Central part of the country. b. For comparison, about 70 percent of the regional maize markets in Uzbekistan display evidence of integration. Maize regional clusters are also larger and include more indi- vidual markets than the wheat ones. Overall, the evidence of market integration is stronger for maize than for wheat, which was included in the Report to contrast this free-market commodity with wheat, prices of which are influenced by the state order. c. Compared with several other Central Asian countries, for which results are available in the literature – such as Kazakhstan, Afghanistan and Tajikistan, the spatial integration of wheat markets within Uzbekistan is weak. d. Spatial integration of flour markets within Uzbekistan is even weaker than that of wheat markets. e. There is evidence of vertical integration between wheat and flour prices on only 3 re- gional markets in Uzbekistan. But the observed wheat and flour price relationships do not resemble those that are reported in the literature for other countries and that would be expected to emerge under market conditions. f. Wheat markets in Uzbekistan are not integrated with wheat market in Kazakhstan. No evidence was also found that flour markets in Uzbekistan are spatially integrated with flour markets in Afghanistan, Kazakhstan, Kyrgyzstan or Tajikistan. g. The currency liberalization in September 2017 has not affected or improved spatial (wheat, maize and flour), vertical (wheat-flour), or international (wheat and flour) mar- ket integration in Uzbekistan. 6. These findings call for actions beyond the planned wheat market liberalization. While the 2021 reform by itself should help strengthen spatial and vertical market integration, monitoring will be needed to confirm that. In addition, complementary measures to improve market efficiency could include: (i) improvement in price collection and accuracy of regional production-consumption data; (ii) making these data publicly available in real time; (iii) re- frainment from imposing local price controls and regional restrictions on movement of wheat and flour; and (iv) effective management of public stocks, which will replace the state wheat procurement in 2021. Without successful implementation of these complementary measures, the 2021 wheat market reform could deliver much less than is anticipated in Uzbekistan. 3 1 Introduction 1. Uzbekistan is undergoing transition from a planned to a market economy. The gov- ernment has initiated reforms of the state order system for cotton and wheat, including the lib- eralization of wheat, flour, and bread prices, which are important for the country’s food security. Efficient markets are critical to maximize the benefits of a more market-oriented economy, as strong integration would allow markets to respond to regional shortages/surpluses, thereby helping reduce price volatility. Efficient markets are also critical for private investments, which in turn are the key to the creation of better paid jobs along the entire value chain. 2. There are only few empirical/statistical estimates of market integration for wheat and wheat flour within Uzbekistan and between Uzbekistan and other Central Asian mar- kets. Uzbekistan buys wheat and flour from Kazakhstan and sells flour to Afghanistan (World Bank, 2020). As it plans to liberalize domestic grain markets and replace the state procurement of wheat by the strategic public stocks in 2021, the government of Uzbekistan needs a stronger understanding of domestic and international market integration in order to design policies and investments that improve how markets function and foster food security. 3. This Report analyzes integration of wheat, corn, and flour markets in Uzbekistan. It assesses the short-term and long-term interactions between market pairs to better understand the state of market efficiency. The Report uses monthly retail prices in all regions of Uzbekistan from January 2014 to February 20205 from the World Bank database. Data on wheat and flour prices in Kazakhstan, Tajikistan, Kyrgyz Republic, and Afghanistan are available in the FAO GIEWS database. Using this data and with a view to filling the knowledge gaps outlined above, the objective of this study is to evaluate the extent of spatial and temporal wheat and flour market integration in Uzbekistan, both domestically and in relation to relevant regional markets. 4. The Report presents the results of eight specific tasks. As presented in Table 1, the analysis starts with an assessment of integration of wheat markets and its comparison with in- tegration of maize markets. It is followed by the assessment of integration of flour markets, and wheat and flour prices in Uzbekistan. Examples from other countries provide a reference point for wheat and flour market integration in Uzbekistan. After that, there is an assessment of rela- tionship between prices in Uzbekistan and in the neighboring countries. The final task investi- gates whether changes in Uzbekistan’s currency policy in September 2017 influenced any of the above-mentioned dimensions of market integration. 5 Note that this price timer series has a comparatively small number of observations ( = 1, 2, … 74). Thus, the cointegration methods that employed below and detailed in Annex 1 must be interpreted with some caution. Not- withstanding, even this relatively short time series generates useful results. The unit root tests were applied to each of the individual Uzbek wheat, corn, and flour prices. These tests suggest that the individual prices are integrated and based on this evidence the conclusion was made that all of the price series are candidates for cointegration. This allows proceeding to test for cointegration and estimate error correction models (ECMs) to cast light on the long-run and short-run interactions between pairs and groups of prices (see Annex 1). 4 Table 1: Objectives and study tasks Domestic market integration within Uzbekistan Integration between Uzbekistan and neighbouring markets 1. Assess the short and longer-term integration of wheat markets in Uzbekistan. 6. Analyse the relationship between 2. Assess the short and longer-term integration of wheat prices in Uzbekistan and Ka- maize markets in Uzbekistan, which, unlike zakhstan, the sole wheat exporter to wheat markets, have been free of government in- Uzbekistan. tervention. Compare with wheat market integra- tion, which has been subject to state order. 3. Assess the short and longer-term integration of flour markets in Uzbekistan. 4. Determine the short and long-term relations 7. Analyse the integration of flour and transmission between wheat and flour prices markets in Uzbekistan, Afghanistan, in Uzbekistan. Kyrgyz Republic, Tajikistan, and Ka- 5. Present examples of market integration on zakhstan. wheat and flour markets in other countries to pro- vide a reference point for Uzbekistan. 8. Analyse the impact of Uzbekistan’s currency liberalization in September 2017 on market integration. 2 Assessment of the short and longer-term integration of Uzbekistan’s wheat markets 6. The national average retail wheat prices in Uzbekistan grew from 965 Soms/ton in 2014 to 2,244 Soms/ton in 2019. In nominal terms, wheat prices grew by 133 percent. In real terms, the price growth was 55 percent. The lowest prices were recorded in Karakalpakstan and Jizzakh, while the highest prices were in Tashkent city (Figure 1). Figure 1: Retail wheat prices in Uzbekistan, ‘000 Soms/kg Source: World Bank database. 7. The level of regional prices greatly depends on production and consumption rates. Where production exceeds consumption, prices are expected to be lower than where production is less than consumption. Consumption data by region are not readily available in Uzbekistan to test this hypothesis. Thus, the estimate of regional consumption was made, based on 5 population numbers and the total consumption of food and feed wheat in the country. The US Department of Agriculture estimates Uzbekistan’s total wheat consumption at 9 million tons. Unofficial estimates of the local ministries, such as the Ministry of Agriculture and the Ministry of Economic Development and Poverty Reduction, are lower, about 7.5 million tons. Using this range of total consumption, Table 2 presents the regional consumption estimates and shows that only in three regions (i.e. Jizzakh, Kashkadarya, and Syrdarya) does production exceeds con- sumption. Bukhara is also a net producer when total wheat consumption is assumed at 7.5 mil- lion tons. But when it increased to 9 million tons, this region turns into a net wheat consumer. Table 2: Estimates of net consumption of wheat by region, Uzbekistan, 2019 Wheat Popula- Estimated wheat con- Self-sufficiency ratio produc- tion, sumption, tons (production/consump- tion, ‘000 tion) tons people At 7.5m At 9.0m At 7.5m At 9.0m tons con- tons con- tons con- tons con- sumption sumption sumption sumption Karakalpakstan 170 652 1 890 420 290 504 348 0.41 0.34 Andijan 499 501 3 110 691 648 829 977 0.72 0.60 Bukhara 475 900 1 916 426 095 511 314 1.12 0.93 Jizzakh 484 306 1 375 305 672 366 806 1.58 1.32 Kashkadarya 923 570 3 261 725 295 870 354 1.27 1.06 Navoi 210 785 993 220 720 264 864 0.95 0.80 Namangan 397 314 2 796 621 729 746 075 0.64 0.53 Samarkand 608 533 3 857 857 749 1 029 299 0.71 0.59 Surkhandarya 566 411 2 612 580 965 697 158 0.97 0.81 Syrdarya 374 006 842 187 206 224 647 2.00 1.66 Tashkent 463 861 2 930 651 573 781 888 0.71 0.59 Fergana 575 247 3 733 830 195 996 234 0.69 0.58 Khorezm 263 354 1 856 412 840 495 408 0.64 0.53 Tashkent city 0 2 554 568 022 681 627 0.00 0.00 Uzbekistan, ‘000 6 013 33 725 7 500 9 000 0.80 0.67 Source: World Bank estimates based on the data from the State Statistics Committee of Uzbekistan. 8. No strong correlation was found between regional wheat prices and regional self- sufficiency in wheat. Karakalpakstan usually has the lowest wheat prices, although it is a large net consumer of wheat. This might be partially explained by the import of cheaper wheat from Kazakhstan – Karakalpakstan shares the border with it – but this hypothesis needs to be con- firmed6. Jizzakh is a net wheat producer and its price level is among the lowest, but then in Bukhara and Tashkent regions, other two net producers, wheat prices are relatively high. Over- all, the inverse relationship between prices and net production, commonly seen around the world, does not appear to prevail in Uzbekistan. 9. Wheat prices in Uzbekistan display evidence of seasonality. The Friedman test (Friedman 1937, 1939) was applied to the national average wheat, corn, and flour prices, con- firming a stable seasonality of prices at the 1 percent level in all cases. Wheat prices generally peak in April-May and reach their lowest values in August-October, while corn prices peak in July-August and bottom out in November-February. 10. Between 2014 and 2017, the growth of retail wheat market prices exceeded that of state procurement farm-gate prices. This increased the wedge between what farmers received 6 The main border-crossing crossing for Kazakh export of wheat and flour is Saryagash Station near Tashkent City. Kazakhstan does export to Iran via Aktau port on the Caspian, so some wheat might be taken off there and enter Uzbekistan by the West. If it indeed happens, the wheat in question would need to be transported hundreds of extra kilometers and should be correspondingly more expensive than the prevailing prices in Karakalpakstan. 6 for their wheat from the state and what they received on the market (Table 3). In 2018 and 2019, however, the state procurement prices increased faster than retail market prices, and in 2019, farmers were reported to have received the same farm-gate price from the state as from the market. The recent increases of state procurement prices seem to have moderately boosted mar- ket prices, subject to the limited purchasing power of Uzbek consumers and export prices of Kazakh wheat and flour. Table 3: State farm-gate prices and retail wheat prices in Uzbekistan, ‘000 Soms/ton 2014 2015 2016 2017 2018 2019 State procurement farm-gate prices 418 460 503 550 750 1,450 (increase in % compared to previous year) (10%) (9%) (9%) (36%) (93%) Retail market prices* 965 1,224 1,335 1,501 1,929 2,244 (27%) (9%) (12%) (29%) (16%) Price ratio 2.31 2.66 2.65 2.73 2.57 1.55 Note: * Note that retail price is not the same as farm-gate price. Market farm-gate prices are not available, thus they are not reported here. Source: World Bank estimates based on the data from the Ministry of Finance and the State Statistics Committee of Uzbekistan. 11. For the study of market integration of wheat (as well as later in the Report corn and flour) within Uzbekistan, prices are expressed in Soms and selected for markets that are connected by road or rail. The analysis looks for evidence of clusters of more than two integrated markets in the Northeast, South-Central, and Western regions of Uzbekistan. The focus is given not to all possible pairs of Uzbek prices but rather on neighboring pairs as these are most likely to be integrated in the economic sense. In addition, this reduces the volume of test results that must be presented7. Results are presented schematically in maps, and detailed estimation results are contained in tables in the Appendix. 12. Figure 2 presents the results of the analysis of spatial integration on Uzbek wheat markets. More details are provided in Appendix Table 1. In Figure 2, evidence of cointegration between wheat prices is found for those pairs of markets that are connected by red arrows, with arrowheads indicating which markets adjust to correct deviations from the respective long-run price equilibrium. 13. The results suggest that there are two separate spatial clusters of integrated wheat markets in Uzbekistan, one in the Northeast of the country, and one in the South-Central region. Detailed estimation results in Appendix Table 1 indicate that in those cases in which markets are integrated, long-run elasticities of price transmission are generally close to 1, with the exception of Navoi, which is linked to the neighboring regions Samarkand and Kashkadarya with long-run elasticities of 0.69. Price transmission is sometimes unidirectional, sometimes bi- directional, and aggregate speeds of adjustment to deviations from long-run equilibrium range between 17 and 52 percent per month. In the most rapid cases (Navoi-Samarkand and Fergana- Andijan), it takes roughly one month to correct one-half of a deviation from the long-run price relationship. The median half-life of a shock that creates a deviation from long-run equilibrium over all wheat market pairs is 2.3 months. 14. For the market pairs Kashkadarya-Bukhara and Surkhandarya-Kashkadarya there is some evidence of cointegration, but only at the 13 and 12 percent levels8 of signif- icance, respectively. In other words, there is some weak statistical evidence that these pairs 7 If the analysis was made of each possible pair of the 14 wheat markets in Uzbekistan, for example, it would require presenting results for 91 unique pairs. 8 The threshold for statistically strong level of significance is 10 percent. Above this threshold, the significance is usually considered to be weak. 7 might also be integrated, which would extend the reach of the South-Central cluster of inte- grated wheat markets. Figure 2: Price transmission relationships for wheat in Uzbekistan Notes: Arrows indicate that the two market prices are cointegrated. Arrowheads point to markets that respond to disequilibrium. Source: World Bank estimates. 15. The results for Tashkent City with its neighbors Tashkent and Syrdarya regions are inconclusive. There is some evidence of cointegration, but it is contradictory and suggests that prices in Tashkent City behave fundamentally differently from the other regional wheat prices in Uzbekistan (this is confirmed in Figure 1). A possible interpretation is that the price in Tashkent City is especially influenced by the price of imported Kazakh wheat and by the much higher purchasing power of Tashkent City’s residents compared with the residents of other regions of Uzbekistan, factors which distinguish it from the other regional wheat prices in Uzbekistan. 16. Multivariate cointegration tests were also carried out to confirm the existence of integrated wheat market clusters within Uzbekistan. In the Northeast, these tests indicate that the three markets Namangan, Andijan and Fergana form an integrated cluster as their prices are linked by two cointegrating relationships.9 In the South-Central region evidence is found that the four markets Navoi, Kashkadarya, Samarkand, and Surkhandarya form an integrated cluster, as their prices are connected by three cointegrating vectors. 3 Assessment of the short and longer-term integration of Uzbekistan’s maize markets 17. The national average retail maize prices in Uzbekistan grew from 1,200 Soms/ton in 2014 to 2,704 Soms/ton in 2019. In nominal terms, wheat prices grew by 122 percent, a bit less than wheat. In real terms, the maize price grew by 43 percent. The lowest prices were recorded in Jizzakh, while the highest prices were in Navoi, Surkhandarya, and Tashkent city (Figure 3). 9 See Annex 1. A set of ‘n’ markets is considered integrated, if there are ‘n-1’ cointegrating relationships between the prices on these markets (see, for example, Gonzales-Rivera and Helfand, 2001). 8 Figure 3: Retail maize prices in Uzbekistan, ‘000 Soms/kg Source: World Bank database. 18. Regarding clusters of market integration, the results for maize are similar to those for wheat, generating evidence of two spatial clusters of integrated markets (Figure 4 and Appendix tables). However, the South-Central cluster is more extensive for maize than for wheat, as it includes links to Bukhara and to Jizzakh, and between Kashkadarya and Sur- khandarya. Figure 4: Price transmission relationships for maize in Uzbekistan Notes: Arrows indicate that the two market prices are cointegrated. Arrowheads point to markets that respond to disequilibrium. Source: World Bank estimates. 19. The multivariate cointegration tests confirm that the spatial maize market integra- tion clusters are more extensive than those for wheat. In the Northeast, these tests indicate that the four markets Tashkent, Namangan, Andijan and Fergana form an integrated cluster as their prices are linked by three cointegrating relationships (in the case of wheat, Tashkent was not included in this cluster). In the South-Central region, the four markets Navoi, Kashkadarya, Samarkand and Surkhandarya form an integrated cluster, as their prices are also connected by three cointegrating vectors. In addition, test results indicate that prices on the five markets Navoi, Kashkadarya, Samarkand, Jizzakh, and Syrdarya are linked by four cointegrating vec- tors. A plausible interpretation of the stronger market integration observed for maize compared with wheat is the fact that maize prices in Uzbekistan are determined on the market, while wheat prices are heavily influenced by the state order system that covers roughly one-half of the coun- try’s wheat output. 9 20. There is also evidence, of cointegration for the market pairs Navoi-Bukhara and Namangan-Fergana, however only at the 12 and 11 percent levels of significance. This additional evidence of market integration, although it is statistically weaker, strengthens the overall impression that maize markets are more integrated than wheat markets in Uzbekistan. As in the case of wheat, results for the market pair Tashkent City-Tashkent are contradictory, suggesting that the Tashkent City market is unique in Uzbekistan for maize as well as for wheat. 21. In the cases in which maize markets are integrated, long-run elasticities of price transmission are generally close to 1. As was the case for wheat, pairs involving Navoi are the exception, with long-run transmission elasticities below 0.8. Why price transmission elas- ticities Navoi are unusually low both for wheat and for maize is a puzzle. Theoretically one would expect price transmission elasticities to fall with increasing distance between two mar- kets, but the distance between Navoi and its neighboring markets is not unusually large. If es- pecially poor infrastructure is leading to inflated transport costs between Navoi and its neigh- bors, this could explain the low observed price transmission elasticities. Another reason could be more interference of local governments into price formation of agricultural products there. 22. Price transmission is sometimes unidirectional, sometimes bi-directional, and ag- gregate speeds of adjustment to deviations from long-run equilibrium range between 13 and 46 percent per month. In the most rapid case (Fergana-Andijan), it takes just over one month to correct one-half of a deviation from the long-run price relationship. The median half- life of a shock that creates a deviation from long-run equilibrium over all maize markets pairs is 2.2 month, which is slightly lower than for wheat (2.3 months). 4 Assessment of the short and longer-term integration of Uzbekistan’s flour markets 23. Uninterrupted flour price data for January 2014-February 2020 are only available for eight regional markets in Uzbekistan. There are no obvious signs of spatial integration apparent in Figure 5. The ranking of the prices changes frequently over time, as prices that were among the highest over some extended periods, are among the lowest over others. In a stable spatial equilibrium, one would generally expect prices in some regions (net importers) to con- sistently rank higher than prices in others (net exporters). It is striking that from January 2019 onward the regional flour prices all move in a very similar manner, first remaining constant for roughly 9 months, then increasing sharply before levelling off again. The sharp increase may reflect the liberalization of flour prices in October 2019; otherwise, such co-movement more likely reflects some form of market regulation rather than integration due to trade and arbitrage. Figure 5: Retail flour prices in Uzbekistan, first grade, ‘000 Soms/kg Source: World Bank database. 10 24. Regional markets with flour price data are marked with stars in Figure 6. This allows looking for evidence of market integration on 6 pairs of neighboring markets. The results in Figure 6 and Appendix Table 3 suggest that there is only limited evidence of spatial flour market integration in Uzbekistan. Evidence of cointegration between flour prices is found be- tween Navoi and Samarkand, and between Fergana and Andijan. The estimated adjustment parameters suggest that prices in Samarkand respond to deviations from their long-run relation- ship with Navoi (11 percent per month), and that prices in Fergana respond to deviations from their long-run relationship with Andijan (22 percent per month). This means that it takes roughly 5.9 months to correct one-half of a deviation from the long run equilibrium wheat-flour relationship between Navoi and Samarkand, and 2.5 months in the case of Fergana and Andijan. Figure 6: Price transmission relationships for flour in Uzbekistan Notes: Arrows indicate that the two market prices are cointegrated. Arrowheads point to markets that respond to disequilibrium. Source: World Bank estimates. 5 Assessment of the short and long-term relations and transmission be- tween wheat and flour prices in Uzbekistan 25. The results of cointegration tests between wheat and flour prices in eight regions of Uzbekistan are presented in Appendix Table 4. There is evidence of a cointegrating rela- tionship between wheat and flour prices in only three of the eight regions – Bukhara, Fergana and Andijan. However, as discussed in the next chapter, the evidence for these three regions is puzzling when compared with the literature and theoretical expectations. First, the estimated elasticities of price transmission from wheat to flour are higher than would be expected under competitive conditions. Second, in two of the three regions it appears that wheat and not flour prices respond to deviations from their long-run equilibrium relationship. There is weak evi- dence of cointegration in a fourth region, Jizzakh (at the 11 percent level of significance), but here too the estimated results suggest that wheat prices adjust to flour prices, and not the other way around. In the market economy, one would expect flour prices to adjust to changes in wheat prices. Price controls, official and unofficial, in Uzbekistan could be a reason for this anomaly. 6 Presentation of examples of market integration on wheat and flour markets in other countries to provide a reference point for Uzbekistan 26. The literature on the integration of wheat and flour markets in other Central Asian countries is limited. Several studies estimate and contrast grain market integration within 11 individual Central Asian countries (comparable to Tasks 1, 2 and 3 above) and others model vertical transmission between wheat and flour prices (comparable to Task 4 above). In the fol- lowing these two sets of studies are reviewed. There is a much broader literature on grain price transmission in many settings outside Central Asia. While an exhaustive review is beyond the scope of this study, several studies from outside Central Asia that provide benchmarks for Uz- bekistan in the following review are also included. a. Spatial integration of wheat and of flour markets in other countries 27. The spatial integration of wheat and of flour markets has been analyzed for a num- ber of Uzbekistan’s neighbors. Brosig and Yorbol (2005) use weekly data from 1998 to 2004 to study wheat market integration on three markets in Kazakhstan: Petropavlovsk, Kokshetau, and Karaganda. Their results suggest strong wheat market integration between Petropavlovsk and Kokshetau, which are both important wheat producing regions in the north of Kazakhstan. The elasticity of price transmission between these two regions is close to 1, and deviations from the long-run equilibrium relationship between these prices are corrected at a comparatively rapid rate of roughly 22 percent per week. Wheat markets in Kokshetau and Karaganda are also found to be integrated, but adjustment to deviations from the long-run equilibrium relationship between these markets is slower at roughly 3.6 percent per week. Brosig and Yorbol (2005) find evidence of threshold effects in the integration of wheat markets in Kokshetau and Kara- ganda; when the price differential between these regions exceeds roughly US$43 per ton, ad- justment to deviations increases to 9.4 percent per week. Overall, these results point to stronger wheat market integration within Kazakhstan than found within Uzbekistan, especially in view of the greater distances that must be covered in Kazakhstan. Kokshetau and Karaganda, for example are separated by roughly 500 km, while the road between Tashkent City and Samar- kand, which passes through both Syrdarya and Jizzakh, is slightly longer than 300 km. 28. Halimi et al. (2015) focus on market integration for wheat and for flour in Afghan- istan based on monthly price data from 2000 to 2015. They find that wheat (and flour) prices in Afghanistan are cointegrated with wheat (and flour) prices in Kazakhstan and Pakistan. Since Pakistan imposed trade restrictions in response to the ‘food price crisis’ of 2007/08, relation- ships between Afghan and Pakistani prices have weakened, while relationships between Afghan and Kazak prices have strengthened. Within Afghanistan, Halimi et al. (2015) present evidence that wheat and flour markets in commercial centers are well integrated, but that markets in rural areas are less well integrated with commercial centers. For example, their results for flour prices in Kabul related to the commercial centers Kandahar, Jalalabad and Herat reveal price trans- mission elasticities close to 1 and monthly rates of adjustment to deviations from long-run equi- librium between 23 and 29 percent. For rural markets paired with commercial centers long-run elasticities of price transmission as low as 0.58 and 0.63 are observed. 29. Further results on spatial integration of wheat and of flour markets in Afghanistan are provided by Najibullah et al. (2017). These authors employ monthly retail prices of wheat and flour from 2004 to 2015 from seven markets including Kabul and six provincial centers Jalalabad, Kandahar, Herat, Maimana Balkh and Faizabad. Their results indicate that all pro- vincial wheat (flour) markets are integrated with the wheat (flour) market in Kabul. Their esti- mated long-run elasticities of spatial price transmission range between 0.7 and 0.9, and devia- tions from long-run equilibrium are corrected at rates of between 10 and 30 percent per month. They do not model vertical price transmission between wheat and flour prices, but they do report evidence that wheat (and flour) markets in Afghanistan are integrated with wheat (and flour) markets in Kazakhstan and in Pakistan. Both sets of results (Najibullah et al., 2017 and Halimi et al., 2015) suggest that wheat and flour markets are spatially integrated in 12 Afghanistan.10 This contrasts with our findings for Uzbekistan, which are mixed for wheat and reveal only two comparatively weak long-run relationships between flour markets. 30. Ilyasov et al. (2016) study market integration and price transmission on wheat mar- kets in Tajikistan using monthly data from 2002 to 2013. They find evidence of cointegra- tion between wheat prices for all possible pairs of the four markets that they study: Dushanbe, Gharm, Khujand, and Kurgan-Tyube. They also find that these prices are cointegrated with international prices (they consider wheat prices in Rouen in France, and Saryagash in Kazakh- stan). Speeds of adjustment to deviations from long-run equilibrium price relationships gener- ally fall between 20 and 25 percent per month. They also find that Tajik prices react more rapidly to increases in international prices than they do to increases in other domestic prices. 31. In a related study, Svanidze and Götz (2019b) compare spatial wheat market inte- gration in Russia with spatial maize market integration the United States. They use weekly prices from 2005 to 2013 and consider six regions within Russia (e.g. North Caucasus, Black Earth, Volga, and West Siberia) as well as a number of local markets within the Black Earth and West Siberia regions. For the United States they use maize price data from 16 federal states, and 28 and 15 local markets within the states of Iowa and North Carolina, respectively.11 While conditions in both Russia and the United States differ considerably from those in Uzbekistan (especially the distances between regions), Svanidze and Götz (2019b) provide some interesting benchmarks for the analysis of spatial wheat market integration in Uzbekistan. According to their results, long-run elasticities of wheat (corn) price transmission between regions in Russia (the US) average 0.43 (0.86). Within regions the average elasticities are higher: the authors report 0.94 and 0.81 for the Black Earth and West Siberia regions, respectively, and 0.97 and 0.95 for Iowa and North Carolina in the US. Deviations from long-run equilibrium wheat (corn) price relationships between regions in Russia (the US) are corrected at an average rate of 42 percent (81 percent) every two weeks. Within the Black Earth and West Siberia regions, the authors report average rates of adjustment to departures from long-run equilibrium of 41 and 52 percent every two weeks, respectively; for Iowa and North Carolina the corresponding value is 61 percent. 32. The studies discussed above all analyze spatial market integration within other Central Asian countries, although some also consider the integration of markets in these countries with international markets. We next turn to studies that deal exclusively with the integration of domestic and international wheat markets. Bobokhonov et al. (2017) compare transmission between domestic and international prices for a number of agricultural products, including wheat, in Tajikistan and Uzbekistan. They analyze monthly data from 2004 to 2014 in Tajikistan, and from 2001 to 2009 in Uzbekistan. They find evidence of integration between Tajik and world markets for food crops including wheat but find no such evidence for Uzbeki- stan. 33. Svanidze et al. (2019b) investigate wheat price relationship between international markets (the Black Sea exporters Kazakhstan, Russia and Ukraine) and several Central Asian (Kyrgyzstan, Tajikistan and Uzbekistan) and South Caucasus (Armenia, Azerbai- jan and Georgia) countries. They work with monthly data from 2006 to 2014, except in the case of Uzbekistan for which they only have 39 monthly observations from 2006 to 2009. Their results indicate that wheat markets in the South Caucasus countries are more strongly integrated 10 Chabot and Dorosh (2007) also study wheat market integration in Afghanistan and find that “wheat prices in major markets in Afghanistan … tend to move together in the long run” (Chabot and Dorosh, 2007, p. 349). How- ever, they work with a comparatively short sample of only 43 observations from January 2002 to July 2005. 11 Svanidze and Götz (2019a) is a similar study based on weekly wheat prices in Russia from 2010 to 2016. In this study, the authors continue with a second stage estimation that relates the strength of integration between two regions to factors such as the distance between them. 13 with the Black Sea export markets than the wheat markets in the Central Asian countries. Av- erage estimated long-run price elasticities for the South Caucasus countries are 0.63 compared with 0.47 for the Central Asian countries. For Uzbekistan specifically, only one long-run equi- librium relationship is found, with the Kazakhstan export price, while all of the other countries considered display relationships with six or seven different international prices. 34. In summary, the literature12 provides ample evidence that the spatial integration of wheat and flour markets is stronger in Afghanistan, Kazakhstan, and Tajikistan than is the case in Uzbekistan (Task 1 above). Comparison with Russia leads to similar conclu- sions. The long-run elasticities of wheat price transmission that are estimated for Uzbekistan are similar to the ‘benchmark’ intra-regional estimates reported by Svanidze and Götz (2019b) for Russia. However, the speeds of adjustment to disequilibrium that are estimated for Uzbek- istan are considerably lower than Svanidze and Götz’s (2019b) intra-regional estimates for Rus- sia (and the United States); while our estimates for Uzbekistan generally indicate between 25 and 50 percent correction per month, Svanidze and Götz (2019b) report averages between roughly 40 and 60 percent every two weeks. 35. As regards integration with international markets, a number of the studies re- viewed above present evidence that wheat markets in Afghanistan, Kazakhstan and Ta- jikistan are integrated with international markets. This contrasts with the results for Uzbek- istan that will be reported below (see Task 6), as well as those provided by Bobokhonov et al. (2017). Indeed, only Svanidze et al. (2019) provide any evidence of integration between Uzbek and an international wheat market (Kazakhstan). However, their result is based on only 39 monthly observations from 2006 to 2009. b. Vertical integration of wheat and flour markets in other countries 36. Oskenbayev and Turabayev (2014) analyze vertical price transmission between wheat and flour prices in Kazakhstan, with a focus on how government policy responses to the 2007/08 ‘food price crisis’ (such as export restrictions and consumer subsidies) af- fected the relationship between wheat and flour prices. Using monthly data from 2000 to 2010 they find evidence of cointegration between wheat and flour prices. Their results indicate that the long-run elasticity of price transmission between wheat and flour fell from 0.69 before to 0.45 after August 2007. In addition, the speed with which flour prices respond to departures from this long-run equilibrium slowed slightly, from 17 percent monthly correction before Au- gust 2007 to 14 percent monthly correction afterwards. However, this change in the speed of adjustment is not statistically significant, and the observed changes in individual regions of Kazakhstan vary considerably, with adjustment to disequilibrium becoming faster in some re- gions, and slower in others. 37. Brümmer et al. (2009) analyze wheat to flour price transmission in Ukraine using weekly data from 2000 to 2004. They estimate a long-run elasticity of price transmission of 0.84 and find that flour prices adjust to correct deviations from their long-run equilibrium at a weekly rate of between 2.6 and 4.4 percent, which roughly corresponds to the monthly adjust- ment rates estimated for Kazakhstan by Oskenbayev and Turabayev (2014). 38. Katsia and Mamardashvili (2016) study vertical price transmission from wheat to flour in Georgia using monthly prices from 2004 to 2016. They find that wheat and flour 12 Two studies that also deserve mention are Chabot and Tondel (2011) and Fehér and Fieldsend (2019). Chabot and Tondel (2011) study price relationships between Kazakhstan and international markets (US, Canada, Argen- tina, Afghanistan and Tajikistan), and they provide a wealth of information on the grain trade, routes and transport costs in Central Asia. However, they only report correlation coefficients between price series. Féher and Fieldsend (2019) present a detailed analysis of the potential for increasing wheat production in Kazakhstan, including con- siderations of international trade, but they do not present any of their own empirical results on market integration. 14 prices are cointegrated, and that the long-run elasticity of price transmission from wheat to flour is 0.72. According to their results, flour prices respond to deviations from the long-run equilib- rium at a rate of 14.3 percent per month. 39. Varela and Taniguchi (2014) analyze monthly wheat and flour prices in Indonesia between 2000 and 2010. They estimate long-run elasticities of price transmission from wheat (which is imported in Indonesia) to flour of roughly 0.9 and find that flour prices in Indonesia respond to deviations from the long-run equilibrium with wheat prices by on average roughly 3.0 to 4.2 percent per month. However, they find that these adjustments are highly asymmetric, with Indonesian flour prices reacting more rapidly (roughly 10 percent per month) to increases in wheat prices than to decreases (at most 1.4 percent per month). 40. Alam and Jha (2016) analyze monthly data on wheat and flour prices in Bangla- desh from July 2008 to March 2016. They do not report estimates of the long-run elasticity of price transmission, but they find clear evidence of integration between wheat and flour prices and report that flour prices respond to departures from the long-run relationship at a rate of roughly 25 percent per month. 41. These studies do not all apply exactly the same methodology. Some focus on ques- tions of asymmetry, other look for evidence threshold effects or policy-induced structural breaks. Nevertheless, they typically find that wheat and flour prices are cointegrated and that long-run elasticities of transmission from wheat to flour prices range between 0.7 and 0.9. This is a plausible range that roughly corresponds to the cost share of wheat in flour production (Brümmer et al., 2009). In addition, most studies also find evidence that flour prices, but not wheat prices, respond to deviations from the long-run equilibrium. This is to be expected, as wheat is much more widely traded internationally than flour, and flour milling is only one use for wheat, albeit an important one. Hence, in the absence of major interventions or market bar- riers wheat prices will be driven by exogenous developments on international markets, and local flour prices can be expected follow. Finally, speeds of adjustment commonly estimated in the literature lie in a range between 10 and 20 percent per month. 42. These typical results contrast with our findings for Uzbekistan, which suggest only limited evidence of cointegration between wheat and flour prices (no evidence for aggre- gate Uzbek prices, and evidence of cointegration on only three of eight regional markets) . Furthermore, on the regional markets for which there is evidence of vertical price transmission in Uzbekistan, the estimated elasticities range from 1.01 to 1.19, which is higher than would be expected. Finally, on two of the three regional markets with the evidence of vertical price trans- mission (Bukhara and Andijan), it appears that wheat prices rather than flour prices respond to disequilibrium. Overall, our results suggest that wheat and flour markets in Uzbekistan are not integrated in the manner that is typically observed in other countries and that is consistent with integration due to market forces. 7 Analysis of the relationship between wheat prices in Uzbekistan and Kazakhstan, the sole wheat exporter to Uzbekistan 43. Appendix Table 5 presents the results of cointegration tests between the price of wheat in Kazakhstan and the markets of Tashkent, Tashkent City, Samarkand and Kash- kadarya in Uzbekistan. These Uzbek markets include those closest to the Saryagash Station border crossing with Kazakhstan, where wheat imports enter Uzbekistan, as well as the main transit corridor that runs through Uzbekistan to borders with Turkmenistan and Afghanistan. 44. The first two columns of results in Appendix Table 5 report the results of standard Johansen tests for linear cointegration. These tests produce no evidence of cointegration 15 whatsoever, indicating that there is no linear long-run equilibrium relationship between Kazakh and Uzbek wheat prices. 45. Figure 7 shows that the Uzbek wheat prices denominated in US$ declined sharply in September 2017 following the currency liberalization. As a result, it is not surprising that the Johansen test fails to find evidence of linear cointegration between these prices and the Kazakh price. Therefore, the Johansen et al. (2000) test was conducted that allows for a struc- tural break (shift in the constant term) in the long-run equilibrium relationship. However, as shown in the last two columns of Appendix Table 5, these test results provide no support for the conclusion that there is a long-run equilibrium relationship (with a structural break) between wheat prices in Uzbekistan and in Kazakhstan before and after September 2017. Figure 7: Wheat prices in Kazakhstan and selected markets in Uzbekistan, US$/ton Source: World Bank estimate. 46. Failure to find evidence of cointegration does not necessarily imply that two mar- kets are not integrated. It could be that the prices in question are linked by a non-linear long- run relationship that is more complex than was considered here. For example, changes in import duties applied at the border between Kazakhstan and Uzbekistan over time could lead to a long- run equilibrium relationship that contains a series of shifts, that affect not only the constant term but also the elasticity of price transmission (for example, if changes in ad valorem tariffs have taken place). However, the available evidence suggests that Kazakh and Uzbek wheat mar- kets were not integrated over the study period. While wheat markets in Kazakhstan are well integrated with world markets (FEWS NET, 2016, 2019; Halimi et al., 2015), wheat markets in Uzbekistan appear to be largely decoupled from international developments. 47. This is not a big surprise. Uzbekistan’s crop placement system annually allocates 1.4 million ha of farmland for growing wheat, and farmers must use this land only to produce wheat. They cannot adjust and switch between crops depending on their profitability outlook. As a result, wheat production is determined by yield only. Higher domestic wheat production usually leads to somewhat lower wheat prices, but they do not crash reflecting the fact that Uzbekistan is a net importer of wheat. This largely explains the lack of integration of Uzbek wheat market with Kazakh and wider world markets. 8 Analysis of integration of flour markets in Uzbekistan, Afghanistan, and Kazakhstan 48. Figure 8 presents flour prices in Kazakhstan (Almaty), Afghanistan (Kabul) and selected markets in Uzbekistan, all denominated in US$ . While the Uzbek prices all fall by 16 large amounts following the currency liberalization in September 2017, the Afghan price is considerably less volatile and follows a slight declining trend until the last roughly 18 months, when it increases slightly. The Kazakh price falls notably in the second half of 2015 and has trended gradually upward since 2016. Figure 8: Wheat prices in Kazakhstan and selected markets in Uzbekistan, US$/ton Source: World Bank estimate. 49. Since flour prices in Uzbekistan, Kazakhstan and Afghanistan appear to follow fundamentally different paths over the sample period, it is not surprising that test results in Appendix Table 6 provide no evidence of cointegration . As in the case of wheat in Task 6, the tests were made for both linear cointegration and cointegration with a structural break coinciding with currency liberalization in September 2017. Both sets of results indicate that flour prices in Uzbekistan are not cointegrated with prices in Kazakhstan and Afghanistan. Ap- pendix Table 6 also includes test results for cointegration between prices in Uzbekistan with Kyrgyzstan and Uzbekistan with Tajikistan. Again, there is no evidence of cointegration. Hence, the available evidence suggests that Uzbek flour markets were not integrated with neigh- boring flour markets over the time period that we study. 17 9 References Akaike, H. (1974): A new look at the statistical model identification. IEEE Transactions on Automatic Control, 19: 716-723. Alam, M.J. and Jha, R. (2016): Asymmetric Threshold Vertical Price Transmission in Wheat and Flour Markets in Dhaka (Bangladesh): Seemingly Unrelated Regression Analysis. ASARC Working Paper 2016/03. https://acde.crawford.anu.edu.au/sites/default/files/publication/acde_craw- ford_anu_edu_au/2016-07/wp2016-03alamjha.pdf Bobokhonov, A., Pokrivcak, J. and Rajcaniova, M. (2017): The impact of agricultural and trade policies on price transmission: The case of Tajikistan and Uzbekistan. The Journal of International Trade & Economic Development, 26: 677-692. Brosig, S. and Yorbol, Y. (2005): Interregional integration of wheat markets in Kazakhstan. IAMO Discussion Papers 88. Halle (Salle): IAMO (Leibniz Institute of Agricultural Development in Transition Economies, Halle, Germany). Brümmer, B., von Cramon-Taubadel, S. and Zorya, S. (2009): The impact of market and policy instability on price transmission between wheat and flour in Ukraine. European Review of Agricultural Economics, 36: 203-230. Chabot, P. and Dorosh, P. A. (2007): Wheat markets, food aid and food security in Afghanistan. Food Policy, 32: 334–353. Chabot, P. and Tondel, F. (2011): A Regional View of Wheat Markets and Food Security in Central Asia with a Focus on Afghanistan and Tajikistan. Study prepared for the United States Agency for International Development (USAID) Famine Early Warning Systems Network (FEWS NET). https://documents.wfp.org/stellent/groups/public/documents/ena/wfp238576.pdf Dickey, D. and Fuller, W. (1979): Distribution of the estimators for autoregressive time series with unit root. Journal of The American Statistical Association, 74: 427-431. Fehér, I. and Fieldsend, A.F. (2019): The potential for expanding wheat production in Kazakh- stan – Analysis from a food security perspective. EUR 29386 EN, Publications Office of the European Union, Luxembourg, JRC113009. FEWS NET (Famine Early Warning System Network) (2016): Regional Wheat Market Funda- mentals, Central Asia Summary. November 2016. https://fews.net/sites/default/files/documents/reports/Central%20Asia%20Re- gional%20Wheat%20Market%20Fundamentals_FINAL_0.pdf FEWS NET (Famine Early Warning System Network) (2019): Regional Supply and Market Outlook, Central Asia. October 2019. https://fews.net/central-asia/supply-and-market- outlook/october-2019 Friedman, M. (.1937): The use of ranks to avoid the assumption of normality implicit in the analysis of variance. Journal of the American Statistical Association 32: 675–701. Friedman, M. (1939): A correction: The use of ranks to avoid the assumption of normality implicit in the analysis of variance. Journal of the American Statistical Association 34: 109. González-Rivera, G. and Helfand, S. (2001): The extent, pattern, and degree of market integra- tion: a multivariate approach for the Brazilian rice market. American Journal of Agri- cultural Economics. 83: 576–592. 18 Halimi, G.H., Abbott, P.C. and McNamara, K.T. (2015): Price transmission in Afghanistan’s wheat markets. IPIA Report #1062015. https://pdfs.semanticscholar.org/ca23/991fee69ee2bffd51580205693795d9e9516.pdf Ilyasov, J., Götz, L., Akramov, K., Dorosh, P. and Glauben, T. (2016) : Market Integration and Price Transmission in Tajikistan’s Wheat Markets. IFPRI Discussion Paper 01547, Washington DC. Johansen, S. (1991): Estimation and Hypothesis Testing of Cointegration Vectors in Gaussian Vector Autoregressive Models. Econometrica. 59: 1551–1580. Johansen, S. (1995): Likelihood-based Inference in Cointegrated Vector Autoregressive Mod- els. Oxford University Press. Johansen, S., Mosconi, R. and B. Nielsen (2000): Cointegration Analysis in the Presence of Structural Breaks in the Deterministic Trend. Econometrics Journal, 3: 216- 249. Najibullah, H., Ito, S., Isoda, H. and Amekawa, Y. (2017): Cointegration and spatial price trans- mission among wheat and wheat-flour markets in Afghanistan, Applied Economics, 49: 2939-2955. Oskenbayev, Y. and Turabayev, G. (2014): Price Transmission Analysis along the Food Chain in Kazakhstan. Modern Economy, 5: 175-187. Pesaran, M.H. (2007): A simple panel unit root test in the presence of cross-section dependence. Journal of Applied Econometrics, 22: 265-312. Saikkonen,P., and Lütkepohl, H. (2002): Testing for a Unit Root in a Time Series with a Level Shift at Unknown Time, Econometric Theory 18: 313–348. Svanidze, M. and Götz, L. (2019a): Determinants of spatial market efficiency of grain markets in Russia. Food Policy, https://doi.org/10.1016/j.foodpol.2019.101769 Svanidze, M. and Götz, L. (2019b): Spatial market efficiency of grain markets in Russia: Im- plications of high trade costs for export potential. Global Food Security 21: 60-68. Svanidze, M., Götz, L., Djuric, I. and Glauben, T. (2019): Food security and the functioning of wheat markets in Eurasia: a comparative price transmission analysis for the countries of Central Asia and the South Caucasus. Food Security 11: 733-752. Varela, G.J. and Taniguchi, K. (2014): Asymmetric price transmission in Indonesia’s wheat flour market. Asian Development Bank Economic Working Paper Series No. 394, https://www.adb.org/sites/default/files/publication/31240/ewp-394.pdf World Bank (2020): Analysis of Regional Dynamics of Wheat and Flour Sectors in Central Asia and their Impact on Uzbekistan. Washington, D.C. World Bank (2019): Reforming the Wheat Sector in Uzbekistan. Technical Report, Tashkent. World Bank (2018): Cotton Taxation in Uzbekistan: Recent Developments and Reform Agenda Ahead. Technical Report, Tashkent. Zivot, E. and Andrews, D.W.K. (1992): Further evidence on the great crash, the oil-price shock, and the unit-root hypothesis. Journal of Business and Economic Statistics 10: 251-270. 19 Annex 1: Cointegration methods 50. The evidence of integration between two markets is looked for by studying whether prices on these markets are cointegrated. ‘Cointegration’ is a term that time series econometri- cians use to describe two or more variables (such as prices) that are individually non-stationary but tend to move together over time. Since non-stationary variables are sometimes referred to as being ‘integrated’, two non-stationary/integrated variables that tend to move together are referred to as being cointegrated (i.e. ‘co-integrated’). 51. Testing for cointegration involves several steps. First, so-called unit-root tests are used to determine whether the individual time series (in our case, individual prices) are integrated. If a stationary series is shocked upwards or downwards, it will tend to revert towards its original mean in the ensuing periods; it can be said that a stationary series ‘forgets’ the shock over time. An integrated series by contrast has ‘infinite memory’; each shock results in a permanent shift in the long-run expected value of the series. As a result, integrated series do not revert to a constant mean but rather drift randomly. So-called unit root tests are designed to identify this behaviour. We employ the standard ADF test (Dickey and Fuller, 1979), as well as a panel unit root test proposed by Pesaran (2007) that can be applied to a set of prices and tests the null hypothesis that all of them are integrated/non-stationary. 52. Second, if two series are individually integrated, a cointegration test determines whether or not they share the same fundamental pattern of non-stationarity/integration. Cointegration tests essentially test whether there is a linear combination of two integrated series that is itself stationary. If such a stationary linear combination exists, then both series are driven by the same underlying set of shocks; they are individually non-stationary, but they share a common form of non-stationarity. The stationary linear combination represents a long-run equilibrium rela- tionship between the two series. To test for cointegration we employ a so-called ‘trace test’ procedure proposed by Johansen (1991, 1995). 53. If two series are cointegrated, then the third step is to estimate a so-called error-correc- tion model (ECM). The ECM describes the mechanism that ensures that the long-run equilib- rium relationship between the two series is maintained in the long run. Cointegrated variables can move apart (deviate from their long-run equilibrium relationship) temporarily, but the ECM ensures that such deviations are ‘corrected’ and thus transitory. It is as if cointegrated variables were connected by a rubber band or spring that allows them to move closer or father away from one another temporarily, but that always pushes or pulls them back towards an equilibrium distance. A key parameter in an ECM is the so-called ‘adjustment parameter’ (often represented in the literature by the Greek symbol alpha – ) that measures how quickly a series adjusts to correct any deviation that has developed between it and its cointegrated partner. 54. A simple ECM for two price series and takes the following form: ∆ = (−1 ) − 0 − 1 −1 + 1 ∆−1 + 1 ∆−1 (1) 55. The dependent variable on the left-hand-side of equation (1), ∆ , measures the change in between the previous and the current period (∆ = − −1 , the subscript counts time). The expression in brackets on the right-hand-side is the long-run equilibrium relationship between and . If these two series are cointegrated, then this expression is stationary, and = 0 + 1 in the long run. Hence, 1 = ⁄ , referred to as the long-run coefficient of price transmission, measures how will respond to a change in in the long run. If the long-run equilibrium has been temporarily disturbed, then ≠ 0 + 1 , and the adjustment parameter measures how responds in the following period to correct a proportion of this disturbance or disequilibrium. is expected to be negative, and if it for example equals -0.5, 20 then responds to correct 50 percent of any disequilibrium that existed at the end of the pre- vious period. 56. The two lagged price change terms on the right-hand-side of equation (1) capture short run dynamics. Equation (1) only includes one lag of each price change, but in applied work information criteria such as the Akaike Information Criterion (AIC) (Akaike, 1974) are used to determine the most appropriate number of lagged terms to include. 57. A similar ECM can be specified and estimated for changes in : ∆ = (−1 ) − 0 − 1 −1 + 1 ∆−1 + 1 ∆−1 (2) 58. In equation (2), is expected to be positive, and measures how contributes to the restoration of the long-run equilibrium relationship. Together, equations (1) and (2) make up a so-called ‘vector error correction model’ (VECM). The equations in a VECM are typically es- timated simultaneously using a maximum likelihood method first proposed by Johansen (1991, 1995). 59. Equations (1) and (2) make up a so-called ‘bivariate’ VECM, because they describe the cointegration of two variables. The multivariate VECM is a generalization that describes coin- tegration between more than two variables. In the market integration literature, tests for multi- variate cointegration are used to look for evidence of clusters of integrated markets. If n markets are integrated with one another, then cointegration tests should indicate that there are in total n- 1 cointegrating relationships between them (Gonzales-Rivera and Helfand, 2001). 60. In our application to market integration for wheat, maize and flour in Uzbekistan we convert all prices into logarithms. This enables us to interpret the coefficient 1 as the long-run elasticity of price transmission from one market to another. In addition, we uniformly employ a specification with two lags as suggested by the AIC. 61. The use of cointegration analysis to study market integration is justified with reference to trade and information flows and arbitrage between markets.13 If two markets are linked by trade, then in the long run the price in the importing market will equal the price in the exporting market plus the costs of trade. This defines the long-run equilibrium relationship between the two prices. In the short run the prices can deviate from this equilibrium relationship, for exam- ple if one of them is shocked by unexpected news (e.g. bad weather that reduces the expected harvest). In this case traders respond to the disequilibrium by increasing or decreasing trade flows until the long-run equilibrium price relationship is restored. The ECM describes the path that the prices take on the way to the restoration of equilibrium. In practice, of course, markets are constantly subject to smaller and larger shocks and will therefore be in a constant state of adjustment, with the long-run equilibrium relationship holding on average over time but rarely, if ever, at any given point in time. 62. Like all empirical methods, cointegration analysis has its shortcomings. Unit root tests are often unable to reliably distinguish between stationary and non-stationary series, especially over short periods of time. A series that appears to be drifting and therefore non-stationary may actually be mean reverting if observed over a sufficiently lengthy period of time. Market inte- gration analyses typically work with 100 and (sometimes many) more observations, but we only have 74 observations of Uzbek prices to work with. 13 Note that in the market integration literature, the word ‘integrated’ is used in two senses: first to describe ‘inte- grated’ time series (i.e. those that are not mean-reverting but rather drift randomly); and second to describe whether or not two markets are economically ‘integrated’ as a result of trade and information flows and arbitrage. Two price series that are individually ‘integrated’ (in the statistical sense, based on unit root test results) are candidates for cointegration, and if they are found to be cointegrated (based on the results of Johansen tests), this can be interpreted as evidence that these prices were collected on markets that are ‘integrated’ in the economic sense. 21 63. In addition, structural breaks in a time series can create the impression that it is non- stationary. In other words, an otherwise stationary series that includes a structural break can ‘fool’ a unit root test into concluding that the series is integrated. To check for this, we used the Saikkonen and Lütkepohl (2002) and the Zivot and Andrews (1992) tests. Both test the null hypothesis of a unit root against alternatives with level or trend stationarity and various forms of structural break. Similarly, a structural break in the long-run equilibrium relationship be- tween two prices could lead tests to falsely conclude that they are not cointegrated, and the markets therefore not integrated. Such a structural break could occur, for example, if trade costs between two markets fall following the completion of a major infrastructure project. In this case the two prices will move in parallel separated by a one distance prior to the break, and by a shorter distance after the break (the reduction in trade costs will lead to a reduction in the value of 0 in equations 1 and 2). The markets in question are actually integrated throughout, but a standard cointegration test may indicate that their prices are not cointegrated. To control for this possibility, we employ a test for cointegration with a structural break (shift in the constant term) proposed by Johansen et al. (2000). 22 Appendix Table 1: Results of cointegration tests and VECM estimation for pairs of wheat prices in neighbouring regions of Uzbekistan Trace test Cointegration relationship Speed of adjustment Market pair H0: r=0 H0: r=1 | | + Half-life Khorezm + Karakalpakstan 23.87 (0.01) 7.54 (0.10) 0.82 (0.05) 0.89 (0.00) -0.24 (0.01) 0.04 (0.44) 0.24 2.5 Khorezm + Bukhara 16.28 (0.16) 3.84 (0.45) 1.31 (0.00) 0.82 (0.00) -0.23 (0.01) 0.09 (0.13) 0.23 2.7 Kashkadarya + Bukhara 16.98 (0.13) 3.19 (0.56) 0.57 (0.29) 0.93 (0.00) - 0.20 (0.00) 0.09 (0.07) 0.29 2.0 Surkhandarya + Kashkadarya 17.47 (0.12) 3.50 (0.50) 0.20 (0.67) 0.97 (0.00) - 0.12 (0.13) 0.18 (0.03) 0.18 3.5 Navoi + Bukhara 15.75 (0.19) 4.31 (0.38) 2.64 (0.00) 0.65 (0.00) - 0.10 (0.20) 0.26 (0.01) 0.26 2.3 Navoi + Samarkand 22.04 (0.03) 4.45 (0.36) 2.32 (0.00) 0.69 (0.00) -0.01 (0.93) 0.49 (0.00) 0.49 1.0 Samarkand + Jizzakh 15.30 (0.21) 2.78 (0.63) 0.46 (0.33) 0.95 (0.00) -0.11 (0.09) 0.17 (0.05) 0.28 2.1 Syrdarya + Jizzakh 18.32 (0.09) 3.13 (0.57) 0.15 (0.74) 0.98 (0.00) -0.13 (0.11) 0.19 (0.02) 0.19 3.3 Syrdarya + Tashkent City 25.56 (0.01) 8.79 (0.06) 0.56 (0.37) 0.89 (0.00) -0.12 (0.04) 0.14 (0.00) 0.26 2.3 Tashkent + Tashkent City 20.50 (0.04) 8.05 (0.08) 0.75 (0.91) 0.72 (0.44) 0.01 (0.08) 0.01 (0.00) 0.02 34.3 Fergana + Tashkent 25.35 (0.01) 4.10 (0.41) 0.53 (0.16) 0.93 (0.00) -0.27 (0.00) 0.01 (0.91) 0.27 2.2 Fergana + Andijan 32.87 (0.00) 3.27 (0.54) 0.34 (0.34) 0.96 (0.00) -0.21 (0.00) 0.31 (0.00) 0.52 0.9 Namangan + Tashkent 14.63 (0.25) 3.78 (0.46) 0.32 (0.66) 0.96 (0.00) -0.17 (0.00) - 0.01 (0.89) 0.17 3.7 Kashkadarya + Samarkand 18.50 (0.09) 4.11 (0.41) 0.08 (0.89) 0.99 (0.00) -0.14 (0.02) 0.11 (0.03) 0.25 2.4 Andijan + Namangan 24.74 (0.01) 4.40 (0.37) 0.30 (0.50) 0.96 (0.00) -0.31 (0.00) 0.14 (0.02) 0.45 1.2 Fergana + Namangan 24.67 (0.01) 4.37 (0.37) 0.59 (0.07) 0.92 (0.00) -0.22 (0.00) 0.10 (0.23) 0.22 2.8 Surkhandarya + Samarkand 23.34 (0.02) 3.08 (0.57) 0.33 (0.27) 0.96 (0.00) -0.20 (0.01) 0.23 (0.07) 0.42 1.3 Navoi + Kashkadarya 21.96 (0.03) 5.01 (0.29) 2.28 (0.00) 0.69 (0.00) -0.11 (0.08) 0.34 (0.01) 0.45 1.2 Note: p-values in brackets. Shading highlights those market pairs for which the prices are cointegrated (p-value <10%). For those cointegrated pairs, red highlights significant adjustment parameters (p-value <10%) that have the expected sign. The summed speed of adjustment (|1 | + 2) only includes statistically significant estimates (p-value <10%). Half-life is the number of months required to correct one half of a deviation from the long-run equilibrium. See Annex 1 for a description of the methods used. 23 Appendix Table 2: Results of cointegration tests and VECM estimation for pairs of maize prices in neighbouring regions of Uzbekistan Trace test Cointegration relationship Speed of adjustment Market pair H0: r=0 H0: r=1 | | + Half-life Khorezm + Karakalpakstan 11.19 (0.53) 4.79 (0.32) 2.11 (0.13) 0.75 (0.00) -0.08 (0.01) -0.01 (0.57) 0.08 8.3 Khorezm + Bukhara 19.09 (0.07) 8.04 (0.08) 0.06 (0.91) 1.00 (0.00) -0.11 (0.11) 0.13 (0.01) 0.13 5.0 Kashkadarya + Bukhara 25.68 (0.01) 6.95 (0.13) 0.31 (0.53) 0.96 (0.00) -0.08 (0.13) 0.18 (0.00) 0.18 3.5 Kashkadarya + Surkhandarya 23.55 (0.02) 4.91 (0.30) 0.27 (0.57) 0.98 (0.00) -0.14 (0.01) 0.23 (0.00) 0.37 1.5 Navoi + Bukhara 17.45 (0.12) 5.90 (0.21) 2.69 (0.00) 0.66 (0.00) - 0.19 (0.00) 0.01 (0.89) 0.19 3.3 Navoi + Samarkand 20.35 (0.05) 4.18 (0.40) 1.68 (0.00) 0.79 (0.00) -0.29 (0.00) 0.04 (0.73) 0.29 2.0 Samarkand + Jizzakh 20.63 (0.04) 5.45 (0.25) 0.80 (0.04) 0.90 (0.00) -0.31 (0.00) 0.03 (0.75) 0.31 1.9 Syrdarya + Jizzakh 24.50 (0.01) 5.48 (0.24) 0.66 (0.12) 0.92 (0.00) -0.37 (0.00) 0.04 (0.53) 0.37 1.5 Syrdarya + Tashkent City 28.63 (0.00) 12.83 (0.01) 0.72 (0.34) 0.89 (0.00) -0.21 (0.00) 0.06 (0.08) 0.27 2.2 Tashkent + Tashkent City 27.94 (0.00) 9.64 (0.04) -1.14 (0.89) 0.94 (0.37) 0.01 (0.01) 0.01 (0.00) 0.02 34.3 Tashkent + Fergana 19.24 (0.07) 6.82 (0.14) 0.95 (0.14) 0.87 (0.00) -0.12 (0.02) 0.11 (0.03) 0.23 2.7 Andijan + Fergana 20.34 (0.05) 3.74 (0.46) 1.12 (0.09) 0.85 (0.00) - 0.28 (0.00) 0.07 (0.12) 0.28 2.1 Namangan + Tashkent 20.72 (0.04) 6.07 (0.19) 0.24 (0.69) 0.97 (0.00) - 0.30 (0.00) 0.04 (0.50) 0.30 1.9 Samarkand + Kashkadarya 26.58 (0.00) 5.40 (0.25) 0.48 (0.21) 0.92 (0.00) - 0.30 (0.00) 0.13 (0.06) 0.43 1.2 Andijan + Namangan 28.93 (0.00) 5.05 (0.29) 0.21 (0.58) 0.97 (0.00) -0.46 (0.00) 0.13 (0.18) 0.46 1.1 Namangan + Fergana 17.56 (0.11) 3.28 (0.54) 1.13 (0.06) 0.85 (0.00) - 0.23 (0.00) 0.06 (0.27) 0.23 2.7 Samarkand + Surkhandarya 19.24 (0.07) 3.60 (0.49) 0.65 (0.07) 0.91 (0.00) -0.26 (0.00) 0.11 (0.27) 0.26 2.3 Navoi + Kashkadarya 27.32 (0.00) 5.07 (0.29) 2.09 (0.00) 0.73 (0.00) -0.28 (0.00) -0.02 (0.82) 0.28 2.1 Note: p-values in brackets. Shading highlights those market pairs for which the prices are cointegrated (p-value <10%). For those cointegrated pairs, red highlights significant adjustment parameters (p-value <10%) that have the expected sign. The summed speed of adjustment (|1 | + 2) only includes statistically significant estimates (p-value <10%). Half-life is the number of months required to correct one half of a deviation from the long-run equilibrium. See Annex 1 for a description of the methods used. 24 Appendix Table 3: Results of cointegration tests and VECM estimation for pairs of flour prices in neighbouring regions of Uzbekistan Trace test Cointegration relationship Speed of adjustment Market pair H0: r=0 H0: r=1 | | + Half-life Karakalpakstan - Khorezm 11.43 (0.51) 4.66 (0.33) 2.77 (0.01) 0.65 (0.00) -0.10 (0.03) 0.04 (0.40) 0.10 6.6 Khorezm + Bukhara 13.58 (0.33) 4.09 (0.41) 1.82 (0.01) 0.77 (0.00) -0.04 (0.34) 0.13 (0.02) 0.13 5.0 Navoi + Bukhara 13.78 (0.31) 5.62 (0.23) 3.45 (0.00) 0.58 (0.00) -0.08 (0.01) 0.02 (0.59) 0.08 8.3 Navoi + Samarkand 18.70 (0.08) 6.48 (0.16) 0.27 (0.82) 0.98 (0.00) -0.05 (0.20) 0.11 (0.00) 0.11 5.9 Samarkand + Jizzakh 15.64 (0.20) 4.88 (0.31) 0.06 (0.93) 0.99 (0.00) -0.21 (0.00) -0.02 (0.71) 0.21 2.9 Andijan + Fergana 18.79 (0.08) 5.84 (0.21) 0.62 (0.00) 0.92 (0.00) - 0.13 (0.31) 0.24 (0.02) 0.24 2.5 Note: p-values in brackets. Shading highlights those market pairs for which the prices are cointegrated (p-value <10%). For those cointegrated pairs, red highlights significant adjustment parameters (p-value <10%) that have the expected sign. The summed speed of adjustment (|1 | + 2) only includes statistically significant estimates (p-value <10%). Half-life is the number of months required to correct one half of a deviation from the long-run equilibrium. See Annex 1 for a description of the methods used. 25 Appendix Table 4: Results of cointegration tests and VECM estimation for vertical wheat-flour price transmission in eight regions of Uzbekistan Trace test Cointegration relationship Speed of adjustment Region H0: r=0 H0: r=1 | | + Half-life Karakalpakstan 8.5 (0.78) 3.3 (0.53) 7.15 (0.18) 0.15 (0.84) -0.02 (0.08) -0.01 (0.08) 0.03 22.8 Khorezm 10.2 (0.62) 3.3 (0.53) -0.70 (0.65) 1.17 (0.00) -0.03 (0.29) 0.09 (0.04) 0.09 7.3 Bukhara 20.7 (0.04) 4.0 (0.43) -0.95 (0.02) 1.19 (0.00) -0.11 (0.22) 0.19 (0.00) 0.19 3.3 Navoi 13.0 (0.37) 5.4 (0.25) -228.6 (0.23) 30.38 (0.24) -0.00 (0.01) -0.00 (0.12) 0.00 >70 Samarkand 13.6 (0.32) 3.3 (0.53) 1.01 (0.23) 0.91 (0.00) -0.08 (0.13) 0.10 (0.04) 0.10 6.6 Jizzakh 17.6 (0.11) 5.0 (0.29) 1.44 (0.01) 0.86 (0.00) 0.03 (0.61) 0.26 (0.00) 0.26 2.3 Fergana 21.5 (0.03) 6.9 (0.14) -0.40 (0.57) 1.11 (0.00) -0.10 (0.00) 0.04 (0.34) 0.10 6.6 Andijan 21.3 (0.03) 7.3 (0.11) 0.34 (0.67) 1.01 (0.00) 0.04 (0.22) 0.28 (0.00) 0.28 2.1 Uzbekistan 10.5 (0.59) 5.0 (0.29) 1.89 (0.67) 0.86 (0.16) -0.02 (0.02) -0.00 (0.58) 0.02 34.3 Note: p-values in brackets. Shading highlights those market pairs for which the prices are cointegrated (p-value <10%). For those cointegrated pairs, red highlights significant adjustment parameters (p-value <10%) that have the expected sign. The summed speed of adjustment (|1 | + 2) only includes statistically significant estimates (p-value <10%). Half-life is the number of months required to correct one half of a deviation from the long-run equilibrium. See Annex 1 for a description of the methods used. 26 Appendix Table 5: Results of cointegration tests and VECM estimation for wheat prices in selected regions in Uzbekistan paired with Kazakhstan Cointegration with a struc- Linear cointegration tural break in September 2017 Market pair H0: r=0 H0: r=1 H0: r=0 H0: r=1 Uzbekistan+ Kazakhstan 7.37 (0.88) 3.02 (0.58) 18.03 (0.35) 3.36 (0.84) Tashkent + Kazakhstan 7.39 (0.87) 3.08 (0.57) 15.18 (0.61) 3.51 (0.82) Tashkent City + Kazakhstan 6.26 (0.93) 2.63 (0.66) 16.30 (0.51) 4.10 (0.75) Samarkand + Kazakhstan 7.50 (0.86) 2.89 (0.61) 18.74 (0.30) 3.06 (0.87) Kashkadarya + Kazakhstan 6.99 (0.89) 3.04 (0.58) 20.39 (0.19) 3.88 (0.78) Note: p-values in brackets. See Annex 1 for a description of the methods used. 27 Appendix Table 6: Results of cointegration tests for flour prices in selected regions of Uzbekistan paired with Afghanistan (Kabul) and other Central Asian countries Cointegration with a struc- Linear cointegration tural break in September 2017 Market pair H0: r=0 H0: r=1 H0: r=0 H0: r=1 Andijan + Afghanistan 6.41 (0.92) 2.55 (0.67) 9.86 (0.96) 3.48 (0.83) Bukhara + Afghanistan 8.23 (0.80) 2.74 (0.64) 14.41 (0.69) 3.54 (0.82) Fergana + Afghanistan 6.63 (0.91) 2.63 (0.66) 11.58 (0.90) 3.46 (0.83) Jizzakh + Afghanistan 7.16 (0.88) 2.67 (0.65) 13.74 (0.74) 4.17 (0.74) Karakalpakstan + Afghanistan 6.94 (0.89) 2.66 (0.65) 12.26 (0.86) 3.87 (0.78) Khorezm + Afghanistan 7.18 (0.88) 3.02 (0.59) 13.49 (0.77) 3.43 (0.83) Navoi + Afghanistan 7.95 (0.83) 2.89 (0.61) 16.31 (0.51) 4.83 (0.65) Samarkand + Afghanistan 6.63 (0.91) 2.89 (0.61) 19.42 (0.25) 4.14 (0.74) Uzbekistan + Afghanistan 6.28 (0.93) 2.79 (0.63) 10.95 (0.93) 3.29 (0.85) Uzbekistan + Kazakhstan 6.00 (0.94) 2.45 (0.69) 15.85 (0.55) 3.23 (0.85) Uzbekistan + Kyrgyzstan 9.88 (0.66) 3.14 (0.56) 16.80 (0.46) 2.38 (0.93) Uzbekistan + Tajikistan 9.57 (0.69) 2.45 (0.69) 14.81 (0.65) 3.13 (0.86) Note: p-values in brackets. See Annex 1 for a description of the methods used. 28