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

We investigated wheat price relationships between the import-dependent countries in Central Asia and the South Caucasus and the Black Sea wheat exporters to assess wheat market efficiency. This is crucial for ensuring availability and access to wheat and for reducing food insecurity. Results from linear and threshold error correction models suggested a strong influence of trade costs on market integration in Central Asia, while those costs were of minor importance in the South Caucasus. In particular, wheat trade in Central Asia is characterized by higher transportation costs but unofficial payments also play a large role. In addition, wheat price volatility is substantially higher in the wheat importing countries of Central Asia compared to the South Caucasus. To foster market functioning, wheat trade should be facilitated by policies that reduce the costs of trade. These include investments in grain market infrastructure, eliminating unofficial payments, and resolving geopolitical conflicts. Additionally, large distances characterize wheat trade in this region, low scope for import diversification and repeated export restrictions by Black Sea exporters. Therefore, trade-enhancing policies should be complemented with policies that increase wheat self-sufficiency in order to improve food security.

are strongly integrated, we can also infer that they are compe((ve, since price differences are quickly arbitraged in strongly integrated markets (Dillon and Dambro 2017).
As an example, in a strongly integrated market, a regional grain harvest shor]all triggers price increases, which are quickly transmiged to other markets, thereby inducing concomitant trade flows that eventually act to stem rising prices (Goodwin and Piggog 2001). By contrast, a region that is only weakly integrated in regional and world wheat markets might be restricted from accessing export markets, and then only at high costs (Jamora and von Cramon-Taubadel 2016). In this case, rising regional prices will induce only limited trade inflows, thereby nega(vely affec(ng the availability and access to a sufficient, reasonably priced grain supply.
We analyze how prices observed within the Central Asian wheat markets of Kyrgyzstan, Tajikistan, and Uzbekistan and the South Caucasian wheat markets of Azerbaijan, Armenia, and Georgia relate to prices of the Black Sea wheat export markets (Russia, Ukraine, and Kazakhstan) and the world markets (France and the USA). We complement this price transmission analysis with the analysis of historical wheat price vola(lity in these markets.
Wheat markets in Central Asia and the South Caucasus have only been studied in a rudimentary fashion in the exis(ng literature and their degree of efficiency is clearly an under-researched ques(on. This can, at least in part, be explained by the limited availability of and accessibility to suitable data (Brück et al. 2012). Unlike the case of the Black Sea wheat expor(ng countries, where strong interest in their respec(ve markets from interna(onal agricultural trading companies has spurred private data collec(on efforts, this kind of data is oien not publicly available for the wheat import-dependent countries of Central Asia and the South Caucasus. Exis(ng studies on wheat markets in the South Caucasus region have found that domes(c wheat markets are well integrated into the world market system and are characterized by a symmetrical adjustment of price devia(ons from the equilibrium (Bluashvili and Safaryan 2014;Djuric et al. 2015;Katsia and Mamardashvili 2016). In contrast, grain price rela(onships across the Central Asian countries indicate more heterogeneous pagerns of price transmission, ranging from well-integrated to completely segregated wheat markets (Bobokhonov et al. 2017;Chabot and Dorosh 2007;Ilyasov 2016;Ilyasov et al. 2016).
Differing from exis(ng studies, we follow a compara(ve approach and inves(gate market integra(on in the six selected countries of Central Asia and the South Caucasus within a unified price transmission modelling approach. A compara(ve approach may permit a more comprehensive interpreta(on of the es(mated parameters. In price transmission analysis, the es(mated parameters themselves enable judging how well a market is func(oning to a limited degree only. We tackle this issue by inves(ga(ng markets with differing characteris(cs within a similar modelling approach, allowing the es(mated model parameters to be directly compared. By using the non-linear, threshold-type price transmission model approach (Greb et al. 2013), we explicitly account for trade costs that strongly influences market integra(on (Fiamohe et al. 2013;Jamora and von Cramon-Taubadel 2016;Moser et al. 2009;Svanidze and Götz 2017;van Campenhout 2007). Poor transporta(on infrastructure and high shipping costs, as well as excessive bureaucra(c requirements, are problema(c throughout Central Asia (ADB 2006;Pomfret 2016;World Bank 2011). This is less of a concern for the South Caucasian countries, as the export markets in the Black Sea region can be accessed through Georgia's ports.
The paper is organized as follows: Sec(on 2 provides a general overview of food security and domes(c wheat markets in Central Asia and the South Caucasus, while sec(on 3 introduces the model framework and research ques(on. In sec(on 4, we discuss the data and in sec(on 5 we share our empirical results. Policy recommenda(ons and a discussion are provided in sec(on 6, followed by our conclusions in sec(on 7. 2

Food security, the wheat trade, and transporta)on costs in Central Asia and the South Caucasus
Food insecurity is chronic in most of the Central Asian countries and cri(cal in the South Caucasus region (Akramov 2012;Bobojonov et al. 2017;Chabot and Tondel 2011;Swinnen and van Herk 2011). Stun(ng in children less than five years of age averages 22% and 17% in Central Asia and the South Caucasus, respec(vely. In addi(on, underweigh(ng occurs in 7% of child popula(ons in Central Asia and 4% in South Caucasus children (see Table 1 for individual shares). The UN's World Food Programme, a humanitarian organiza(on figh(ng hunger worldwide, also operates in the Central Asian countries of Kyrgyzstan and Tajikistan, as well as in Armenia in the South Caucasus. In addi(on, households in these regions spend a large por(on of their income on food, as much as 49% on average in Armenia and 63% in Tajikistan, for example (Table 1). Among all food items, wheat, mainly in the form of bread, accounts for a large share of total daily food calories, ranging from 40% to 60% in both regions. Since household welfare largely depends on the level of food prices, increased food prices oien lead to social and poli(cal unrest. During recent food price hikes, organized public protests were observed in Uzbekistan in September 2007(Or(z et al. 2013; in Tajikistan in February 2008 (RFE/RL 2008); and in Kyrgyzstan in April 2010 (Swinnen and Van Herck 2013).
Most governments in Central Asia have emphasized wheat self-sufficiency as an important goal that they aspire to within their na(onal food security policy (FAO 2015). In Uzbekistan, Svanidze, Götz, Ðjurić, Glauben -Food security and the func=oning of wheat markets in Eurasia 5 for example, wheat produc(on is s(ll centrally planned. The government, through its land leasing contracts, sets quotas for the land area under wheat cul(va(on and defines yield and produc(on targets to be met by farmers. Although input subsidies are provided, the government also obliges farmers to sell 50% of their produce to state enterprises at the predetermined fixed price. State procurement prices are by about three to five (mes lower than counterfactual market prices (Pugach et al. 2016). For the case of Kyrgyzstan and Tajikistan, even though their Na(onal Food Security Programs aim to achieve wheat self-sufficiency, these countries apply more liberal agricultural policy measures and remain heavily depended on wheat imports.  Among the South Caucasian countries, the level of government support is the lowest (prac(cally non-existent) in Georgia and Armenia, and rela(vely high in Azerbaijan. In par(cular, farmers in Azerbaijan receive subsidies for fer(lizers, fuel, machinery, and seed produc(on, as well as monetary transfers (Robinson 2008); however, contrary to Uzbekistan, the government of Azerbaijan does not oblige farmers to sell their grain to state procurement agencies. Investments from Kazakhstan also play an important role in the development of the wheat trade and processing sector in Azerbaijan (FAO/EBRD 2009).
Nonetheless, in the countries of Central Asia and the South Caucasus, domes(c wheat produc(on falls very short of mee(ng local wheat demand. On average, imports account for 41% of wheat consump(on in Central Asia and 63% in the South Caucasus (Table 1). With the increasing impact of climate change and the growing water shortages associated with it, wheat yields are forecasted to decline over (me. Also, variability in wheat produc(on, and ul(mately wheat imports, are expected to increase in Central Asia (Sugon et al. 2013).
Central Asian countries import their wheat almost exclusively from Kazakhstan, whereas wheat to the South Caucasian countries is mainly imported from Russia, Kazakhstan, and, to a lesser extent, from Ukraine ( Fig. 1). In the recent past, the Black Sea region's wheat exporting countries experienced severe harvest shor]alls and implemented various export control systems during periods of high and vola(le prices . During wheat export restric(ons, wheat imports to countries in Central Asia and the South Caucasus from Russia, Ukraine, and Kazakhstan were subs(tuted by imports from more distant countries, such as Iran and European countries.  Asian countries, wheat is shipped mainly by train and secondarily by truck. Northern Kyrgyzstan and Uzbekistan can import wheat directly from Kazakhstan, whereas most rail shipments to southern Kyrgyzstan and Tajikistan must first pass through Uzbekistan.
In contrast, the South Caucasian country of Georgia u(lizes its Black Sea ports, through which wheat can be imported directly from Russia and Ukraine (Appendix, Fig. 2). Armenia depends on Georgia's rail network for transpor(ng imported wheat from Georgia's Black Sea ports to its border. Georgia may also import wheat from Kazakhstan by freight train, which passes through Russia and Azerbaijan. Azerbaijan relies on rail shipments of wheat directly from Russia and u(lizes the Russian railroads as well to access Ukrainian and Kazakh wheat. Due to the military conflict between Armenia and Azerbaijan, the border between the two countries is closed, forcing Armenia to import Kazakh wheat through Georgia, significantly increasing the price of Kazakh wheat and making it less compe((ve for Armenia compared to purchasing wheat from other Black Sea export markets. Countries in the South Caucasus import wheat from Russia and Ukraine nearly twice as cheaply than from Kazakhstan (Table 2). Higher freight rates for wheat imports from Kazakhstan result from large distances and inefficient and outdated logis(cs systems in Kazakhstan inherited from Soviet Union (mes. Shipping costs (official rates) of wheat from Kazakhstan to the Central Asian countries of Kyrgyzstan and Tajikistan are quite comparable to shipping costs to the South Caucasian countries of Azerbaijan and Georgia (Table 2). However, due to unofficial payments, the total cost of transporta(on could be double the official payments in Central Asia (ADB 2006;Chabot and Tondel 2011;World Bank 2005).
Unofficial payments are paid at custom checkpoints and to the traffic police. For example, a test conducted by the Asian Development Bank (ADB 2006) shows that unofficial payments paid by truck drivers on the route between Kyrgyzstan and Kazakhstan are three to four (mes higher than the official transporta(on costs. Another experiment by the World Bank (2011) demonstrates that unofficial payments for in-country transporta(on of cargos from the northern to the southern part of Kyrgyzstan may account for 9% of total transporta(on costs. Payments were extracted by transport control authori(es and traffic police. Pomfret (2016) points out that trade in Central Asia is not only characterized by high transporta(on costs, but also by inadequate regional trade infrastructure, resul(ng in slow movement of cargos and long delays at the border crossing points in this region.
In summary, while the levels of official grain transporta(on costs in Central Asia and the South Caucasus are rather similar, total transporta(on costs are substan(ally higher in Central Asia due to the high unofficial payments. 3

Methodological framework and model es)ma)on
We inves(gate the rela(onships of wheat prices observed in countries in Central Asia and the South Caucasus with Black Sea wheat export markets and world wheat markets in France and the USA within both linear and non-linear price transmission model frameworks.

Methodological framework
We assume that prices in the spa(ally separated markets in the wheat import and export markets are linked by spa(al price equilibrium, which is represented by where " # and " ( denotes the natural logarithm of domes(c and regional/world export prices and ε * is a sta(onary disturbance term. The long-run price equilibrium is characterized by the intercept α and the long-run price transmission elas(city β. If the prices in the domes(c and regional or world markets are not in their equilibrium, then traders will make use of this price difference by trade arbitrage and sell wheat on the market with the higher price level. Through price adjustment processes, prices are brought back to their price equilibrium level.
Since Central Asian and the South Caucasian countries are net wheat importers and wheat is traded only in one direc(on from the Black Sea region to those countries, the wheat price observed in a domes(c market ( # ) is considered the dependent variable and regional and world market export prices ( ( ) are exogenous variables. Therefore, in this study we use a one-equa(on error correc(on model (linear or non-linear) rather than a vector error correc(on model, which is a system of equa(ons capable of addressing endogeneity. 3 Unlike the linear error correc(on model, the threshold error correc(on model explicitly accounts for the role of transac(on costs. According to the spa(al trade arbitrage theory (Goodwin and Piggog 2001), trade arbitrage between two spa(ally separated markets will take place only if the price difference exceeds transac(on costs. Thus, a "regime dependent" price adjustment process may be observed, which can be depicted in a threshold error correc(on model, where the threshold corresponds to the size of transac(on costs.
We use linear (Engle and Granger 1987) and threshold (Hansen and Seo 2002) cointegra(on tests to iden(fy the existence of spa(al price equilibrium and to determine whether the price adjustment mechanism is of a linear or non-linear type.
If the price series are linearly cointegrated, then a linear error correc(on model developed by Engle and Granger (1987) is es(mated to quan(fy the short-run price dynamics in the next step ∆ " # = "23 + 4 6 ∆ "26 # 7 683 + 4 6 ∆ "26 where ∆ is the first difference operator and ε "23 represents the error correc(on term (ECT) variable which is equal to the residuals from equa(on [1] lagged by one period. denotes the speed of adjustment parameter which measures the speed at which devia(ons from the longrun equilibrium are corrected by trade arbitrage. ∆ "26 # and ∆ "26 ( represent lagged values of the first difference of the domes(c and regional/world price series of lags = 1, … , , ensuring that the model residuals are serially uncorrelated. 6 and 6 contain dynamic short-run parameters; " is a conven(onal residual term with " ~ N(0, E ).
If threshold cointegra(on is iden(fied between prices, then we es(mate the threshold error correc(on model. Since wheat trade between a wheat impor(ng and a wheat expor(ng country is uni-direc(onal, we apply a model framework with one threshold and two regimes where denotes the threshold value es(mated by the model. The error correc(on term "23 serves as a threshold variable as well. The parameter is interpreted as an es(mate of transac(on costs from the world market to the domes(c markets. It includes not only observed transporta(on costs and customs clearance, but also other unobserved costs, such as physical and ins(tu(onal infrastructure, ease of accessing market informa(on, and price discounts or premiums paid due to quality differences. In a threshold error correc(on model, the threshold variable "23 and corresponding threshold parameter determine the state of the regime r, r=1, 2. If the magnitude of devia(on from the long-run equilibrium is larger than the size of threshold, then the ECT observa(ons are agributed to the "outer" regime (r =2), where strong price adjustment takes place corresponding to the profitable trade arbitrage. However, if the magnitude of disequilibrium, expressed by "23 term, does not exceed the size of threshold, then observa(ons are agributed to the "inner" regime (r =1), where the speed of adjustment is much weaker or price adjustment does not occur at all (prices may move independently of each another due to the unprofitability of trade arbitrage).
To obtain threshold parameters, we apply the regularized Bayesian es(mator recently developed by Greb et al. (2013) instead of the classic profile likelihood es(mator (Hansen and Seo 2002;Lo and Zivot 2001). 4 The former is superior due to its beger small sample proper(es and avoidance of arbitrary trimming parameter to generate a threshold es(mate. As a result, the Bayesian threshold es(mate is well-defined over the en(re domain of the threshold parameter. In contrast, a profile likelihood es(mator requires a trimming of sample observa(ons to ensure sufficient degrees of freedom for the es(ma(on of model parameters. This procedure might lead to biased model es(ma(on results if the true value of threshold parameters is excluded from the sample. The regularized Bayesian technique, on the other hand, succeeds in retaining all sample observa(ons in the es(ma(on process by penalizing differences between regimes and keeping them small when data contains ligle informa(on.
Though an error correc(on model became the benchmark in examining spa(al price linkages and market integra(on in empirical studies, this model approach yet faces several limita(ons. First, it is based on the assump(on that transac(on costs are sta(onary over (me and are equal to a constant propor(on of commodity prices. On the other hand, if this assump(on fails, implying that actual transac(on costs are indeed nonsta(onary, then the lack of linear or threshold cointegra(on can be wrongly interpreted as evidence of market inefficiencies (Fackler and Goodwin 2001). Second, the spa(al price transmission analysis does not account for the actual trade flows and transac(on costs data (Barreg 1996). The parity bounds model is an alterna(ve approach to studying market integra(on with actual transporta(on costs being accounted for; however, as con(nuous (mes series data on transporta(on costs are not available for Central Asia and the South Caucasus, we use more parsimonious price transmission models, allowing us to analyze market integra(on based on the price series data only. Third, we conduct a price transmission analysis in a bivariate setup, allowing for a pairwise price analysis only, whereas several prices at different loca(ons across space may also be simultaneously determined, which can be analyzed within a mul(variate price transmission model. Nonetheless, the mul(variate analysis of spa(al price linkages so far has only been possible for the linear modelling of price linkages. In contrast, the analysis of the spa(al integra(on of grain markets, par(cularly in Central Asia, explicitly requires accoun(ng for the influence of trade costs, which is achieved by using the threshold error correc(on model, which can only be implemented in a bivariate setup.

Model es)ma)on
Ini(ally, we es(mate the parameters of the long-run price equilibrium [1] by the ordinary least squares (OLS) method.
If the price series are found to be linearly cointegrated, we apply the linear error correc(on model framework following Engle and Granger's (1987) approach. If the price series are found to be cointegrated in a non-linear fashion, we then es(mate the threshold error correc(on model (Greb et al. 2013).
Next, the threshold parameters in equa(on [2'] are iden(fied using the regularized Bayesian technique. A func(on to choose the op(mal threshold value of ECTs is called the posterior median and is constructed as follows: where is a × matrix that compactly stacks together columns of ECTs and values of lagged terms. Z[ ( |Δ , ) denotes marginal posterior density, which is well defined across the space of all possible threshold parameter = { | min( "23 ) < < max ( "23 )}. In the previous expression, is the op(mal threshold that separates the space into two regimes and sa(sfies the requirement that > 0. Computa(on is based on a prior Z[ ( | ) ∝ ( ∈ ) for , where (•) is an indicator func(on providing for switching between regimes. Lastly, in choosing a threshold es(mate, we es(mate the addi(onal short-run price transmission parameters of the threshold error correc(on model in equa(on [2'] separately in "outer" and "inner" regimes with the restricted maximum likelihood method that is implemented through mixed-effects modelling using an "nlme" package in R (Pinheiro et al. 2017).

Data and data proper)es
This sec(on provides an overview of the sources and characteris(cs of the wheat prices, which serve as the basis for this price transmission and vola(lity analysis.

Data
We use a unique database covering wheat prices for 11 countries (  We use retail prices for the analysis of the grain markets' integra(on in Central Asia and producer and import prices for the grain markets in the South Caucasus. Using the various types of wheat prices may influence the size of parameter es(mates to some degree. In par(cular, agributable to the differences in price levels at the various stages of the supply chain, an analysis with retail prices may result in the underes(ma(on of the long-run price transmission parameter and the speed of adjustment parameter compared to the parameter es(mated with producer or import prices. Contras(ng, thresholds for prices pairs including retail prices are rather overes(mated compared to price pairs with producer and import prices. The domes(c prices in the Georgian grain market are well represented by the wheat import (CIF) prices since more than 90% of total wheat supplied on the domes(c market is imported.
We use wheat export prices observed in northern Kazakhstan to serve as the reference price for the South Caucasian impor(ng countries. In addi(on, a wheat export price observed in southern Kazakhstan is used as the reference price for exports to the neighboring Central Asian countries.
Wheat export prices for Russia and Ukraine have 15 and 16 missing observa(ons, accoun(ng for 16% and 17% of the sample, respec(vely. Export prices are not observed when the wheat trade was limited by wheat export restric(ons in both countries. The effect of export restric(ons on wheat prices in Russia and Ukraine is addressed by Götz et al. (2013Götz et al. ( , 2016. In order to create a con(nuous (me series, the missing observa(ons are filled using a linear imputa(on technique, making use of the Kazakh wheat export price, which is highly correlated with the Russian and Ukrainian prices. Since wheat trade is usually priced in US dollars, all local wheat price series are transformed to US dollars.
From this database, we built 30 bivariate price pairs, each consis(ng of a domes(c price of six impor(ng countries and an expor(ng price of five expor(ng countries (Fig. 3). We use the Kazakh wheat export price of the northern region with price pairs including the domes(c price in a South Caucasian country. Addi(onally, we built three price pairs by combining the domes(c price of a Central Asian country with a southern Kazakhstan wheat export price.

Figure 3 Analyzed price pairs
Source: See  In addi(on, the wheat quality and variability of yields from year to year might influence the distribu(onal characteris(cs of the wheat prices. For instance, the rela(vely low median domes(c wheat price in Tajikistan may correlate with the generally low quality of domes(cally grown wheat due to unfavorable clima(c condi(ons and lack of irriga(on systems, whereas wheat produced in Kyrgyzstan is of rela(vely higher quality.

Data proper)es
The lowest median wheat price is also observed in Uzbekistan, where domes(c wheat produc(on is highly supported by the government, but, as noted previously, farmers must also sell a por(on of their wheat to state-owned enterprises at rela(vely low prices fixed by the government.
The domes(c wheat price in Armenia, the landlocked country in the South Caucasus that cannot trade directly with Azerbaijan due to an ac(ve military conflict, represents the highest median price. This contrasts with Georgia, whose Black Sea ports provide direct access to the world market, where the wheat price is characterized by the lowest median value and least price varia(on.
The distribu(on of wheat prices in the wheat expor(ng countries is much more homogeneous. Minor differences in median values across countries might be explained by varying wheat quality grades. For example, the median export price is the highest for wheat of grade one from France, followed by wheat of grade two exported from the USA, and then exports of mostly wheat of grade three from Russia.
The interquar(le range and amplitude of wheat price varia(on is the widest and the median is the lowest for wheat export prices in southern Kazakhstan when compared with the northern region or even other wheat expor(ng countries. We suspect that the vola(le market situa(on in Central Asian impor(ng countries is influencing export price forma(on in southern Kazakhstan, as reflected in a rela(vely large interquar(le range. Also, due to the low consumer income levels in the Central Asian countries, the quality of the wheat exported to the Central Asian countries may be lower, as reflected by the lower median wheat price in southern Kazakhstan compared to that in northern Kazakhstan. A further basic characteris(c of the wheat price series is their vola(lity, indica(ng the degree of risk that prevails in the wheat markets. High price vola(lity results in subop(mal level of produc(on, increasing produc(on costs, and reducing incen(ves for investments. Historical price vola(lity of each individual price series is measured non-parametrically as the standard devia(on ( w ) of the returns of a price series given as: where w* denotes price return in (me t for country i calculated as w* = ( w* w*•⁄ ) with w* being the price of wheat expressed in USD/t and w deno(ng the mean of price returns for country : w =| ∑ w* | *}~.
We found price vola(lity in domes(c markets during the period 2006-2014 to be the highest in Central Asian countries, whereas it is significantly lower in South Caucasian countries (Fig. 5). This might be explained by the rela(vely inelas(c wheat supply, which is characteris(c for the markets in the landlocked Central Asian import-dependent countries. In those countries, grain storage facili(es are extremely limited (World Bank 2011) and access to interna-(onal grain markets incurs high transporta(on costs.

Empirical results
To specify a suitable price transmission model framework for each selected price pair, the existence of a meaningful spa(al price equilibrium needs to be confirmed. Therefore, we tested all price series for the existence of a unit root and the price pairs for the presence of linear and threshold cointegra(on. In sec(on 5.1 we present and interpret the results of the unit root and cointegra(on tests. This will be followed by sec(on 5.2 with the es(mated parameters of the price transmission models for 24 out of 30 analyzed price pairs evaluated against a background of comprehensive qualita(ve knowledge of the wheat markets in those countries.

Test on the existence of a unit root and cointegra)on
Results of the ADF test 6 (Dickey and Fuller 1981) suggest that all wheat prices contain a unit root in level and are sta(onary in first differences at the 5% level of significance. Results of a tradi(onal unit root test will be biased towards nonsta(onarity if structural breaks resul(ng from, for example, policy changes or macroeconomic shocks are ignored in the (me series. Therefore, we conducted the breakpoint ADF test (Perron and Vogelsang 1992) to account for the possible influence of export restric(ons implemented in the grain export markets of 6 Results are available from the authors upon request. . Results indicate that all price series again contain a unit root at the 10% level of significance, confirming that all price series are integrated of order one. Since the price series are iden(fied as nonsta(onary, cointegra(on of the price pairs is required to keep the es(mated spa(al price equilibrium regression from being spurious but rather meaningful (Granger and Newbold 1974).
We applied the linear cointegra(on test by Engle and Granger (1987) 7 with the null hypothesis of no cointegra(on against an alterna(ve of linear cointegra(on. We also applied the threshold cointegra(on test by Hansen and Seo (2002), which examines threshold cointegra(on within a one-threshold model corresponding to the market setup in Central Asia and the South Caucasus (compare sec(on 3.1).
The Engle and Granger's test confirms linear cointegra(on for all price pairs containing a domes(c wheat price of a South Caucasian country at the 5% level of significance (Appendix ,  Table 4). However, Engle and Granger's test suggests linear cointegra(on in just seven out of 15 cases for all price pairs that contain a domes(c price of a Central Asian country. Especially, the domes(c price series in Kyrgyzstan and Tajikistan are linearly cointegrated with the regional wheat export prices in southern Kazakhstan and the world wheat prices in France and the USA. Furthermore, linear cointegra(on is not verified for any of the price pairs that include Uzbekistan's domes(c wheat price. One excep(on is the price pair Uzbekistan-southern Kazakhstan, which we find to be linearly cointegrated.
Like the results of the linear cointegra(on test, the Hansen and Seo test on threshold cointegra(on indicates linear cointegra(on at the 5% level of significance for all price pairs containing a domes(c wheat price of a South Caucasian country (Appendix , Table 4). In contrast, for the price pairs containing prices from Kyrgyzstan and Kazakhstan, Kyrgyzstan and Russia, Tajikistan and Kazakhstan, and Tajikistan and Russia, this test suggests threshold cointegra-(on at the 5% level of significance. Threshold cointegra(on could not be confirmed for the 11 other price pairs constructed by combining a Central Asian domes(c wheat price with an export market's wheat price.
With the results of the linear and threshold cointegra(on tests, we form the cointegra(on pagerns for the 30 price pairs presented in Table 5. All price pairs involving a wheat price of a South Caucasian country are cointegrated linearly, sugges(ng that domes(c prices adjust uniformly to changes in an export price regardless of the level of trade costs.
By contrast, threshold cointegra(on is iden(fied between wheat prices of Kyrgyzstan and Kazakhstan, Kyrgyzstan and Russia, Tajikistan and Kazakhstan, and Tajikistan and Russia. This suggests that, in contrast to the South Caucasus, transac(on costs play a much larger role in the comovement of Central Asian domes(c wheat prices with export prices in regional markets.
Furthermore, the threshold cointegra(on test does not indicate the presence of threshold effects in price rela(onships between domes(c wheat prices in Central Asia and the South Caucasus and world export prices in France and the USA. Due to the vast distances involved and lack of well-established transporta(on infrastructure, transporta(on costs are prohibi-(ve, thus discouraging wheat imports to Central Asia and the South Caucasus from those interna(onally important wheat expor(ng countries 8 .
On the other hand, just as in the countries of the South Caucasus, in Central Asian Kyrgyzstan and Tajikistan, domes(c wheat prices are linearly cointegrated with the world wheat prices in France and the USA, highligh(ng the importance of informa(on flows from the interna-(onal to domes(c wheat markets.

Table 5 Cointegra)on paSerns and selec)on of error correc)on models
Price series Cointegra(on pagern Es(mated error correc(on model (ECM) Linear Threshold •Ž… x x None Note: "None" indicates that cointegra(on tests do not suggest linear or threshold cointegra(on; hence, es(ma(ons are not conducted for the respec(ve price pairs. Neither linear nor threshold cointegra(on is established between domes(c wheat prices in Central Asia and the wheat export prices in Ukraine. In contrast to prices in the South Caucasus, which we find to be linearly cointegrated with the Ukrainian wheat export prices, Central Asian countries do not import wheat from Ukraine because of the rela(vely long distance between the countries (compare Fig. 1).
Compared with the other Central Asian countries, empirical evidence on wheat market integra(on is the weakest for Uzbekistan. We find the Uzbek wheat price to be linearly cointegrated solely with the wheat export price in southern Kazakhstan. Long-run price equilibrium is not established between Uzbekistan and any other export market in the Black Sea region or interna(onal markets. This may be explained by the fact that the Uzbek wheat market is one of the most comprehensively regulated markets in Central Asia, with governmental input cost subsidies, wheat price controls, and state grain buying programs, among others.

Price transmission model es)ma)on results
We analyze the price rela(onships between selected domes(c wheat prices in Central Asia and the South Caucasus and export prices in the Black Sea and interna(onal markets within linear and threshold error correc(on model frameworks. The price transmission model es(mates are evaluated for characteris(cs of spa(al price equilibrium and error correc(on behavior, with the role of trade costs explicitly accounted for.

Spa:al price equilibrium
In general, our results suggest that the comovement of domes(c prices with export prices in the Black Sea region and world markets is stronger in the South Caucasus than in Central Asia (Table 6). Price changes in the regional Black Sea wheat export markets are on average by 16% more completely transmiged to domes(c wheat prices in the South Caucasus (0.63 on average) as compared to Central Asia (0.47 on average). By way of example, if the wheat export price in southern Kazakhstan increases by 10%, then the wheat price in Kyrgyzstan increases by 4.8%.
Comparing domes(c markets across regions, price changes are again less strongly transmiged from the world to domes(c markets in Central Asia, with the long-run price transmission elas-(city ranging between 0.40 and 0.61, compared to the South Caucasus, for which price transmission elas(city varies between 0.49 and 0.79. With respect to Central Asian countries, our results suggest that the wheat market in Uzbekistan is solely integrated with the wheat market in Kazakhstan, but, on the other hand, segregated from the wheat export markets in Russia, Ukraine, France, and the USA. Specifically, wheat prices in Uzbekistan almost perfectly comove with wheat prices in southern Kazakhstan. This might be explained by the dominance of the Uzbek state-run enterprise that centralizes the trade of wheat (Bobojonov et al. 2017).
Among other Central Asian countries, the Kyrgyz wheat market (0.48) is the most strongly integrated with the wheat market in Kazakhstan, followed by Tajikistan (0.40). Kazakh wheat is exported to Kyrgyzstan by a direct railway line through a common border, whereas Kazakh wheat is mainly exported to Tajikistan through Uzbekistan.
The Kyrgyz and Tajik markets are more strongly integrated with export markets in Russia than in Kazakhstan, although the amount of wheat imported by Kyrgyzstan and Tajikistan originating in Russia is negligibly small. Moreover, if Kyrgyzstan and Tajikistan import Russian wheat, then the railway passes through Kazakhstan, sugges(ng that the transporta(on costs of wheat from Russia are higher. Obviously, the domes(c wheat price observed in Kyrgyzstan and Tajikistan is more strongly influenced by the Russian wheat export price than by the wheat export price observed at the southern border of Kazakhstan. Within the South Caucasus region, prices in Georgia's wheat market exhibit the strongest comovement with the export prices in the Black Sea grain expor(ng countries (0.71 on average), followed by prices in Armenia (0.63 on average). On the other end of the spectrum is Azerbaijan, with the weakest price comovement on average at 0.55.
Specifically, price changes in the Russian export market are transmiged to the domes(c wheat market in Georgia by 74%, Armenia by 63%, and Azerbaijan by 49%. Wheat price changes in Kazakhstan, compared with Russia's, are transmiged to a lesser degree to the wheat prices in Armenia and Georgia, which is in line with the observed wheat transporta(on costs (compare Table 2).
Although transporta(on costs of wheat imports to Azerbaijan are higher from Kazakhstan compared to Russia, wheat prices in Azerbaijan comove more strongly with prices in Kazakhstan. This could be explained by the strong business (es between Kazakhstan and Azerbaijan, which indicates that bargaining, search and informa(on costs, as well as other parts of transac(on costs usually not subject to empirical inves(ga(on, are lower from Kazakhstan to Azerbaijan than from the other Black Sea export markets. Moreover, Azerbaijani importers prefer Kazakh wheat with its high protein content over Russian wheat, resul(ng in a higher share of wheat imports from Kazakhstan compared to Russia among total Azerbaijani wheat imports.
Es(ma(on results also indicate that long-run price transmission from wheat markets in France and the USA to the South Caucasian and Central Asian wheat markets is as high from markets in the Black Sea region, or in some cases even higher. This result is striking since neither the South Caucasian nor Central Asian countries import wheat from France or the USA. The strong comovement with wheat prices in the USA can be explained by the domina(ng role of the CBOT wheat price for price forma(on in those markets. According to informa(on provided by traders, the USA CBOT price data can usually be monitored by all market par(cipants, and it serves as a benchmark against which prices generally are nego(ated in the wheat trade.
Finally, the long-run price equilibrium is further characterized by the intercept parameter, which corresponds to the transac(on costs of the wheat trade. Our results suggest larger intercept values for the price rela(onships involving Central Asian countries and the Black Sea regional exporters (3.63 on average) compared to those involving the South Caucasian countries (2.29 on average). This supports our previous findings (see sec(on 2), indica(ng that total transporta(on costs are significantly higher in the landlocked countries of Central Asia than in the South Caucasus. Similarly, results of the threshold and linear cointegra(on tests suggest that trade costs play a large role in the wheat trade of the Central Asian countries.
The par(cularly low value of the intercept (1.68 on average) for the price pairs involving the wheat price in Georgia can be explained by Georgia's direct access to the Black Sea market via its own ports, and thus its generally lower transporta(on costs.

Correc:on of the temporary disequilibrium
Well-func(oning markets are characterized by rapid correc(on of short-run devia(ons from the long-run spa(al price equilibrium, which is reflected by the large value of the speed of adjustment parameter. Our results suggest that the speed of adjustment of prices in the South Caucasian countries is generally higher than in the Central Asian countries (Table 7). Concerning Central Asian markets, the highest speed of price adjustment is iden(fied for the wheat price in Uzbekistan, which corrects devia(ons from the long-run equilibrium with the export price in southern Kazakhstan at a speed of adjustment equal to 0.65. We explain the very quick elimina(on of price disequilibrium in Uzbekistan by the country's centralized state trading system. Wheat prices in Kyrgyzstan and Tajikistan both adjust price devia(ons from the price equilibrium with the Kazakh export prices more quickly (0.35 and 0.32 in the "outer regime") than with the Russian export prices (0.18 and 0.13 in the "outer" regime). We trace this pagern of short-run price dynamics back to the wheat transporta(on costs.
In the South Caucasian countries, we find that the speed of adjustment of wheat prices with the export prices of the Black Sea wheat expor(ng countries is the highest in Georgia (0.34), followed by Armenia (0.32), and Azerbaijan (0.18), reflec(ng respec(ve transporta(on cost levels.
The size of the thresholds iden(fied in the threshold error correc(on model for price pairs containing Tajik wheat prices (0.225, on average) are 0.05 higher than the thresholds es(mated for the price pairs containing domes(c wheat prices in Kyrgyzstan (0.175, on average). These es(mates of transac(on costs for Tajikistan and Kyrgyzstan clearly correspond with the respec(ve distance to the export markets in Kazakhstan and Russia.
The degree of market integra(on may also be characterized by the percentage distribu(on of observa(ons in the "inner" and "outer" regimes. A higher share of observa(ons in the "inner" regime indicates that fewer instances of market disequilibrium are observed and thus evidences stronger market integra(on.
The distribu(on of the price disequilibrium term in different regimes indicates that domes(c wheat prices in Central Asia are more oien in an equilibrium rela(onship with the export price in Kazakhstan (92%) than in Russia (89%). This proves that domes(c wheat markets in Central Asia are more strongly integrated with the export market in Kazakhstan than in Russia.

Policy recommenda)ons and discussion
Based on the results of our analysis, we iden(fy five points of departure for policies to improve the func(oning of wheat markets and to raise food security in Central Asia and the South Caucasus (Table 8). As our results indicate, trade costs are high in Central Asia, hindering the efficient func(oning of grain markets within the region. By reducing trade costs, the wheat trade between the wheat expor(ng and wheat impor(ng countries is spurred, which contributes to stabilizing prices and strengthening market integra(on. Investments in transport infrastructure, public or private, are fundamental for reducing transporta(on costs in Central Asia. In this context, the Belt and Road Ini(a(ve project (HKTDC 2017), which aims to facilitate intra-regional trade in Central Asia, may provide a suitable pla]orm for improving the region's transporta(on system.
In addi(on, the governments of the Central Asian countries should give priority to designing and implemen(ng effec(ve policies for elimina(ng unofficial payments, which are another significant factor impac(ng high transporta(on costs in the region. We also find that wheat price vola(lity is significantly higher in Central Asia than in the South Caucasus or the Black Sea region. An increase in domes(c wheat storage facili(es in the Central Asian countries, where the wheat storage capacity is less than a week (FEWS NET 2016), would facilitate managing the wheat price risk and contribute to stabilizing wheat prices and reducing price vola(lity. Grain stocks could also serve as a crisis measure. For example, strongly increasing wheat prices could be counteracted by releasing grain stocks (Schmitz and Kennedy 2016).
Our analysis has iden(fied that in the South Caucasus, Armenia has the least diversified grain imports and the highest trade costs compared to other neighboring countries in the region.
In Armenia wheat trade costs could moreover be reduced by resolving geopoli(cal conflict with Azerbaijan. If Armenia and Azerbaijan would open their closed border for cargo transi(ng at least, then Armenia could directly import wheat from Kazakhstan through Azerbaijan, substan(ally reducing wheat transporta(on costs. However, due to large distances to grain producing regions, the grain trade could remain challenged by rela(vely high trade costs even in more efficient markets with modern transport infrastructure. In addi(on, the landlocked posi(on of the impor(ng countries leaves ligle scope for diversifica(on of wheat imports. Also, the Black Sea wheat exporters have a history of restric(ng wheat exports in (mes of crisis and the frequency of harvest shor]alls are expected to increase with climate change. Therefore, the countries in Central Asia, but also Armenia and Azerbaijan in the South Caucasus, should complement their trade enhancing policies with agricultural policies aiming to boost domes(c wheat produc(on and to increase wheat self-sufficiency. Clapp (2017) discusses the instances when increases in domes(c food produc(on makes sense economically and poli(cally to increase food security more broadly, while Watson (2017) provides the contextual analysis of food price policies chosen by the governments in developing countries from the poli(cal economy perspec(ve. In the context of Central Asian and South Caucasian wheat markets, we advocate for increased wheat selfsufficiency because of their high trade costs, landlocked geographical loca(on, lack of diversifica(on possibili(es of grain imports (especially for Central Asia), and the high importance of food prices for the stability of poli(cal systems during the periods of rising bread prices (compare sec(on 2).
Finally, Georgia is the country with the best performing wheat market in these regions by far, resul(ng from its market-oriented policies and favorable geographic loca(on reflected in lower transporta(on costs and easy access to the grain export markets in the Black Sea region. Therefore, we see that the food insecurity prevalent in Georgia is not related to a func(oning of the wheat markets. Thus, to improve food security in Georgia, more consumer-oriented measures might play an important role.

Conclusions
In this paper we inves(gated wheat price rela(onships between the six wheat import-dependent countries in Central Asia and the South Caucasus and the three Black Sea wheat exporters to assess how well these markets are func(oning. Well-func(oning wheat markets ensure availability and access to wheat and are crucial for reducing food insecurity, which is prevalent in countries of Central Asia and the South Caucasus. Our results summarized in Table 9 suggest that Georgia is the South Caucasian country with the strongest integrated wheat market, while Uzbekistan is the Central Asian country with the weakest, confirming the findings of Bluashvili and Safaryan (2014), Djuric et al. (2015), and Katsia and Mamardashvili (2016) that grain markets in South Caucasus are well integrated. These results also confirm the findings of Bobokhonov et al. (2017), Ilyasov (2016,  that grain markets in Central Asia are either segregated (Uzbekistan) or characterized by a lower degree of market integra(on with the asymmetric structure of price adjustment (Kyrgyzstan and Tajikistan).
In addi(on, our analysis evaluates the func(oning of grain markets in a compara(ve context, providing novel insights into the func(oning of grain markets in Central Asia and the South Caucasus. From the compara(ve analysis it becomes evident that grain markets in the South Caucasus are more strongly integrated with the world wheat market compared to Central Asia. In addi(on, wheat price vola(lity is substan(ally higher in the wheat impor(ng countries of Central Asia compared to the South Caucasus.
Furthermore, our modelling approach has been made evident that trade costs significantly influence grain market integra(on in Central Asia, while those costs seem to not play a significant role in the integra(on of wheat markets in the South Caucasus. In par(cular, wheat trade in Central Asia is characterized not only by higher transporta(on costs, but unofficial payments also play a large role.
Weak integra(on of Central Asia's wheat markets into the world trade system, accompanied by high transporta(on costs and vola(le wheat prices, indicates low resilience of the food system and rather high vulnerability to food insecurity.
Based on those results, we have iden(fied five policy measures for improving the func(oning of wheat markets and food security in Central Asia and the South Caucasus.  Note: a H0: no cointegra(on | H1: linear cointegra(on. Test is applied to the regression residuals from cointegra(on equa(ons. One-sided p-values are from MacKinnon (1996). Lag length selec(on is based on Schwarz Informa(on Criterion. b H0: linear cointegra(on | H1: threshold cointegra(on. Trimming parameter is equal to 0.05; number of bootstrap replica(ons is set to 1000; fixed regressor bootstrap method. * p<0.10, ** p<0.05, *** p<0.01.