ESSAYS ON THE ECONOMICS OF THAI RICE POLICIES By Uchook Duangbootsee A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of Agricultural, Food and Resource Economics – Doctor of Philosophy 2014 ABSTRACT ESSAYS ON THE ECONOMICS OF THAI RICE POLICIES By Uchook Duangbootsee Recent debate among Thai policymakers has focused on trade-offs between two rice policies, the price support program (PSP) and the deficiency payment program (DPP). The PSP is currently operating but has been criticized for large budgetary costs, corruption in implementation, and unequal distribution of program benefits. To inform this debate, this dissertation provides quantitative measures of the trade-offs between these programs, and the ranking of farmers’ preferences towards them. In addition, the relationship between program participation, production technology choice, and levels of technical efficiency are also investigated. The first essay in this dissertation investigates welfare impacts of PSP and DPP measured in terms of changes in producer surplus, consumer surplus, and deadweight loss by applying a partial equilibrium model to calculate counterfactual market prices and quantities. The 2005/06 cropping season is used as a base for the calculations because of data availability. The results indicate that DPP is more efficient than PSP because the program results in a larger percentage increase in producer surplus and smaller deadweight loss generated for a given amount of Government spending. The increase in producer surplus and the deadweight loss are estimated to be93-97% and less than 1%, respectively under DPP compared to 28-51% and 11-14% under PSP. Consumers, especially domestic consumers, are much worse off under PSP as their surplus shrinks considerably, while it increases under DPP. The second essay investigates the preference ranking of a representative rice farmer in Thailand towards PSP and DPP using stochastic efficiency with respect to a function (SERF). SERF ranks the farmer preferences based on the certainty equivalent (CE) values associated with the stochastic profits under each policy scenario. Profit is stochastic because of yield variability, price volatility, and the risk of delayed payments under PSP. The preference ranking by SERF indicates that a risk-averse farmer clearly prefers DPP to PSP when support and target prices are much higher than the market price. However, the farmer is largely indifferent between the two programs when the price differential is small. In addition, it is shown that the farmer is better off under PSP compared to a no intervention scenario, even if their choice is to not participate, because of increases in the market price brought by the program. The third essay investigates production technologies and levels of technical efficiency among program participants and non-participants of the PSP, as well as key determinants of farmers’ decisions to participate in the program. The participation decision is included in the model to account for possible selection bias in estimating stochastic production frontiers and technical inefficiency levels. Results indicate that key factors in the participation decision include land size and the financial position of the farm. Results also show there is no strong evidence to support the presence of selectivity bias in the stochastic frontier estimates. In addition, a likelihood-ratio test indicates that participants and non-participants use the same frontier production technology. However, the analysis of technical efficiency reveals that participants are more technically efficient than non-participants. The findings therefore suggest that larger farmers participate more in the PSP and that these program participants tend to be more technically efficient farmers, although the analysis was not able to determine the direction of causality for this association. TABLE OF CONTENTS LIST OF TABLES…………………………………………………….……………………..… vi LIST OF FIGURES …………………………………………………………….…………...... viii KEY TO ABBREVIATIONS ………………………………………………………………..... ix CHAPTER 1: A COMPARISON OF THE WELFARE IMPACTS OF THAI RICE PRICE SUPPORT AND DEFICIENCY PAYMENT PROGRAMS……………….…..… 1 1.1 Introduction………………………………………………………….………......... 1 1.2 Background on Thai Rice Economy…………………………………………….... 4 1.2.1 Rice Production in Thailand.……………………………………….…...... 4 1.2.2 Rice marketing chain in Thailand……………………………….……....... 7 1.2.3 Government intervention in the rice market…………………….……....... 9 1.3 Model analysis…………………………………………………...……….…........ 12 1.3.1 Welfare impact under a price support program (PSP)……………..…..... 12 1.3.2 Welfare impact under a deficiency payment program (DPP)…….…..…. 16 1.4 Methods and data………………………………………………………………… 20 1.4.1 Calculating welfare impacts under the PSP using price elasticities……... 21 1.4.2 Calculating welfare impacts under the DPP using price elasticities...…... 22 1.4.3 Data…………………………………………………………………........ 24 1.5 Results……………………………………………………………………...…..... 27 1.5.1 Estimates of the impact of PSP on economic welfare……………............ 27 1.5.2 Estimates of the impact of DPP on economic welfare when the target price is equal to the support price..……………………………................. 32 1.5.3 Estimates of the impact of DPP on economic welfare when total deficiency payments are equal to total Government expenditures under the PSP....................................................................................................... 33 1.5.4 A comparison of the effects of PSP and DPP on economic welfare……….………………………………………………………...... 36 1.6 Conclusions and policy implications………………..……………………...….... 38 REFERENCES…………………………………………………...…………………....... 42 CHAPTER 2: THE RANKING OF FARMER PREFERENCE TOWARDS ALTERNATIVE FARM POLICIES IN THAILAND ………………………………………….… 45 2.1 Introduction……………………………………………………………………... 45 2.2 Literature Review…………………………………………………………....….. 48 2.2.1 Stochastic dominance criterion…………………………………….......... 48 2.2.2 Stochastic dominance with respect to a function (SDRF)………………. 49 2.2.3 Stochastic efficiency with respect to a function (SERF)…………….….. 50 2.3 Methods and data………………………………………………………….…….. 53 2.3.1 Estimating the impacts of PSP and DPP on market price…....………...... 54 iv 2.3.2 Simulating counterfactual market prices………………………………... 57 2.3.3 Simulating random yields………………………………………….……. 58 2.3.4 Generating simulated profits under the five selected scenarios……….… 58 2.3.5 Ranking a representative farmer preference by SERF…………….…….. 61 2.3.6 Data……………………………………………………………….…....... 62 2.4 Results ………………………………………………………………………........ 64 2.4.1 Simulation of market prices……………………………………………... 64 2.4.2 Simulation of rice yields……………………………………………........ 68 2.4.3 The preference ranking by SERF in the 2012/2013 season…………..…. 69 2.4.4 The preference ranking by SERF in the 2006/2007 season ……….….….72 2.4.5 Sensitivity analysis of the preference ranking………………………....…75 2.5 Conclusion and discussion ……………..……………...………………………... 79 APPENDIX…………………………………………………………………………........ 82 REFERENCES……………………………………………………………………........... 83 CHAPTER 3: TECHNICAL EFFICIENCY OF THAI JASMINE RICE FARMERS: COMPARING PRICE SUPPORT PROGRAM PARTICIPANTS AND NONPARTICIPANTS…………………………………………………………….….. 86 3.1 Introduction…………………………………………………………………........ 86 3.2 Theoretical model…………………………………………………………….…. 89 3.2.1 Stochastic frontier model (SFM model)……………………………….... 89 3.2.2 SFM estimation using Heckman’s approach……………………………. 90 3.2.3 SFM estimation using Greene’s approach………………………………. 92 3.3 Model specification and data……………………………………………………. 94 3.3.1 Model Specification……………………………………………………... 94 3.3.2 Data…………………………………………………………………….... 96 3.4 Results………………………………………………………………………........ 98 3.4.1 Differences in input allocation and farmers’ characteristics………..…… 98 3.4.2 Determinants of PSP participation……………………………………..... 99 3.4.3 Frontier production technologies…………………………………...…....101 3.4.4 Technical efficiency of PSP participants and non-participants……….... 106 3.5 Conclusion and policy implications………………………………………………... 108 REFERENCES……………………………………………………….……………....… 110 v LIST OF TABLES Table 1.1: Calculation of PSP impacts…………………………………………………………. 15 Table 1.2: Calculation of DPP impacts…………………………………………………………. 18 Table 1.3: Summary of the data observed during the implementation of PSP in 2005/06……... 28 Table 1.4: Estimates of the impacts of the price support program (PSP)………………………. 31 Table 1.5: Estimates of the impacts of DPP when the target price is equal to the support price under PSP…………………………………………………………………….... 34 Table 1.6: Estimates of the impacts of DPP when total deficiency payments are equal to total Government expenditures under PSP………..……...…….………….……... 35 Table 2.1: Regression result from the equation of Thai rice price in the VAR model…….…… 65 Table 2.2: Mean and standard deviation of counterfactual farm prices………………….……... 68 Table 2.3: OLS regression of Thai rice yield………………………………………………….... 68 Table 2.4: Summary statistics of simulated rice yields…………………………………………. 69 Table 2.5: Summary statistics of simulated profit plus initial wealth in 2012/13……………… 69 Table 2.6: The CE values at five given values of CRRA coefficient in 2012/13……………..... 72 Table 2.7: Summary statistics of simulated profit plus initial wealth in 2006/2007………..….. 73 Table 2.8: The CE values at five given values of CRRA coefficient in 2006/07……………..... 75 Table 2.9: Sensitivity analysis of the SERF ranking in 2012/13……………………………….. 77 Table 2.10: Sensitivity analysis of the SERF ranking in 2006/07…………………………….... 78 Table A2.1a: Regression result from the equation of Vietnam export price in the VAR model.. 83 Table A2.1b: Regression result from the equation of Pakistan export price in the VAR model.. 83 Table 3.1: Average input used and farmers' characteristic variables………………………….... 99 Table 3.2: The estimated parameters of the probit model for participation decision………..... 101 vi Table 3.3: Estimated parameters of the SFM model for PSP participants…………………….. 102 Table 3.4: Estimated parameters of the SFM model for PSP non-participants……………...... 103 Table 3.5: Estimated parameters of the nested-model stochastic production frontier………... 105 Table 3.6: Estimated parameters of the full-model stochastic production frontier………….... 106 Table 3.7: Distribution of technical efficiencies…………………………………………….... 107 vii LIST OF FIGURES Figure 1.1: Rice marketing system in Thailand……………………………………….…………. 7 Figure 1.2: Impacts of the price support program (PSP)……………………………….………. 13 Figure 1.3: Impacts of the deficiency payment program (DPP)……………………….……….. 16 Figure 2.1: Profit CDFs of the representative rice farmers in 2012/13………………………… 70 Figure 2.2: The SERF ranking of the five scenarios in 2012/13 based on CE values………….. 71 Figure 2.3: Profit CDFs of the representative rice farmers in 2006/07………………………… 74 Figure 2.4: The SERF ranking of the five scenarios in 2006/07 based on CE values………….. 74 viii KEY TO ABBREVIATIONS AC Additional costs of program participation under PSP AR Autoregressive model BAAC Bank of Agriculture and Agricultural Cooperatives CE Certainty equivalent CS Total consumer surplus CSD Domestic consumer surplus CSF Foreign consumer surplus DIT Department of Internal Trade DPP Deficiency payment program DWL Deadweight loss EXPS Operating expenses under PSP FAO Food and Agriculture Organization FMO Farmer Market Organization FSD First-degree stochastic dominance LOAN Total BAAC loans issued to farmers under PSP OAE Office of Agricultural Economics PSP Price support program PS Producer surplus PMT Total deficiency payments PWO Public Warehouse Organization PE Loss in production efficiency ix RD Loan repayments/redemption of pledged rice SALEP Sales of non-redeemed rice in paddy-rice-equivalent value SALEM Sales of non-redeemed rice (milled rice) SALEB Sales of rice byproduct SDRF Stochastic dominance with respect to a function SERF Stochastic efficiency with respect to a function SFM Stochastic frontier model SSD Second-degree stochastic dominance TS Loss in trade surplus TE Technical efficiency USDA United States Department of Agriculture VAR Vector autoregressive model x CHAPTER 1: A COMPARISON OF THE WELFARE IMPACTS OF THAI RICE PRICE SUPPORT AND DEFICIENCY PAYMENT PROGRAMS 1.1 Introduction Rice is the most important sector in Thai agriculture in terms of area planted and number of farm households and Government policy has played a significant role in influencing rice prices and farmer returns. During the 1970s and 1980san export tax was applied to keep domestic prices low in an environment of rising food prices as the economy was thriving. However, manufacturing and other sectors eventually surpassed rice and other agricultural products in terms of export revenue share, and rising food prices became less of an issue. So the orientation of rice policy changed towards stabilizing farm incomes through a price-support program. Government involvement in the price-support program has increased significantly since 2000 due at least in part to the intensity of political competition for farmer votes. Two main policies emerged as a result, namely the price support program (PSP) and deficiency payments program (DPP). Debates among policymakers and politicians about PSP and DPP are centered on the questions of which program is most “suitable” in the current rice market environment, and which has the most exposure in terms of Government expenditure. Under the PSP, farmers are allowed to sell their paddy rice to the Government at the support price, which is administratively determined. Then farmers are given four months to redeem the pledged paddy (pay the Government back), otherwise they have to deliver the paddy to the Government. Unlike PSP, DPP requires the Government to make deficiency payments to farmers when the market price falls below a specified target price. The Government does not buy rice under DPP. The deficiency payment amount equals the product of a provincial fixed yield and the difference between the target price and the estimated market price. Thus, the DPP 1 program payments depend on how much land farmers have in rice production, and the Government’s estimates of yield and market price. Critics argue that the Government not only has to bear the high costs running the PSP, but that it also creates market distortions throughout the rice supply chain. Furthermore, most program benefits are likely to accrue to large-sized farms and wealthy farmers due to the nature of program participation. Due to these differences in program attributes and operations, it would be valuable to compare the impacts of both programs on the Thai rice market, rice farmers, and Government expenditures. One approach to assessing the economic impacts of PSP and DPP is to use aggregatelevel data on prices and quantity, together with estimated elasticities of supply and demand, to calculate the changes in producer surplus (PS), consumer surplus (CS), Government expenditure, and deadweight loss (DWL) associated with each program. Poapongsakorn and Charupong (2010) have provided estimates of these welfare components under PSP for the 2005/06 cropping season. They noted that only 624,428 farm households participated in the program. According to their study, the estimates of the changes in CS and PS respectively range from -11,106 to -19,871 and 12,514 to 22,391 million baht. The DWL less sales of Government stocks of rice was estimated between 16,609 and 17,720 million baht. Although, the results from their study suggest that the PSP is very costly and inefficient in term of the distribution of program benefits, no comparison between the PSP and other alternative policies has been made. In addition, their results are limited in that no distinction is made between white rice, jasmine rice, and glutinous rice. Failure to disaggregate results by rice type could affect the welfare results since jasmine and glutinous rice represent a significant portion of rice production in the country. The objective of this study is to compare welfare impacts under PSP and DPP by applying a computational model to calculate counterfactual values of quantity and price that 2 would have been observed had the former been replaced by the latter. The 2005/06 cropping season is used as a base for the calculation as the information on revenue and cost of the PSP is available in this period, and it was a typical year for rice production. In addition, the weightedaverage values of price and quantity from all three types of rice are used in the calculation in order to improve accuracy. 3 1.2 Background on the Thai rice economy This section provides a brief background of Thai rice economy including its production, marketing, and policy interventions. First we discuss the production environment of major rice varieties grown in different geographical locations across the country, and the importance of the rice sector in the overall economy. Next we discuss the rice market system in Thailand. Then we conclude by giving a historical background of major Government interventions in the rice market. 1.2.1 Rice production in Thailand Rice production in Thailand used to be mainly for household consumption and a small marketable surplus but many farmers nowadays have become more commercialized. For instance, 90% of harvested crop was sold in the market in 2000, while the share was only 60% in 1980 (Shigetomi, 2009). Thailand currently produces approximately 30 million tons of paddy rice each year. Nearly half of total production is used for domestic consumption and the rest is exported or kept in stock. Although Thailand’s rice production represents only a small portion of total world production (4.45% in 2011), the country had been the world’s top rice exporter for almost three decades (30% of exports in 2011). However, Vietnam surpassed Thailand in exports in 2012 due to a large amount of unsold stock held by the Thai Government. Interestingly, the Thai ending stock in 2014 is projected to be around 14.11 million tons (milled basis) which can feed the country for at least 15 months, while the world average stock is less than 3 months (USDA and author’s calculation). Rice exports have lost their dominance in the share of total export value in Thailand; it accounted for only 2% of the Thai gross domestic product in 2011, down from 6% in 1980. The 4 share of rice was only 3% of total exports in 2011, a considerable decline from 15% in 1980 (Shigetomi, 2009). Despite the declining importance of rice in terms of the macro economy, a considerable percentage of the population still engages in rice cultivation. In 2003, agricultural labor for wet-season rice production accounted for more than 40% of the total labor force, and nearly 70% of farm households were engaged in paddy cultivation. In term of land use, as much as 60% of total cultivated area was used for rice production in 2001-2007. Rice farming in Thailand is comprised of mostly small-scale farmers (total of 3.7 million farm households) with an average landholding of 3.7 hectares per farm household and a family size of about 4 persons per household (Office of Agricultural Economics, Thailand Ministry of Agriculture and Cooperatives). Rice is grown throughout the country. The Northeastern part of the country has the largest share of area planted and production. Since the region is located in a plateau, its rice production relies on rainfall which allows only one crop per year. Household rice production in the region is mostly for food self-sufficiency with a small surplus which can be sold in the market. In contrast, rice farms in the Central Plain and Lower Northern parts tend to be large scale commercial operations. Only 23.9% of cultivated areas are irrigated, most of which are in the Central Plain and Lower Northern part of the country. In fact, only 5% of agricultural lands in the Northeastern part of the country are irrigated, yet its planted area and number of farm households are the largest in the country. The share of consumer expenditure on rice has been decreasing. According to Isvilanonda and Kongrith (2008) the average per capita rice consumption of Thai households is 101 kg which is two-third of what was consumed in the 1980s. 5 In Thailand, farmers used to produce only two crops: wet-season (June to August) and dry-season (January to May) crop. But farmers in well-irrigated areas now are able to grow a second dry-season crop. The main variety used to be those which take 120 days until harvest. Farmers now prefer the short maturity date variety of 90 days, especially following the implementation of the PSP (Poapongsakorn, 2010). The wet-season paddy is cultivated from June to August, and harvested during October to January. The dry-season paddy used to start from February and end in April, and there was no rice cultivation from September to December due to the high water level during the rainy season, an unfavorable condition for the dry-season variety. However, thanks to the introduction of pumps and drainage facilities, farmers now can lower the water level enough to grow rice during this time period which becomes the first dryseason paddy. The second dry-season paddy is cultivated from January to April. This means many farmers can now grow rice all year long (Shigetomi, 2009). Since farmers usually bring paddy to market immediately after harvest, the larger supply of paddy causes low paddy market price in December. There are two reasons that explain why farmers rush to sell their paddy. First, most farmers borrow money to purchase inputs so they need to pay off such debt as well as other debts. Second, the lack of storage facilities makes it impossible to delay sales of paddy later in the season when the market price rises. Paddy is brought to market in around December, during which the sales price of paddy is the cheapest. Then the price gradually increases until April as there is high demand from the international market, while production is less. The price falls again when the paddy from the dry season is brought to market around July. However, the price of dry-season paddy is usually lower than that of the wet season, because the production of the former is lower. The price increases sharply in August due to limited supply (Poapongsakorn, 2010). 6 1.2.2 Rice marketing chain in Thailand The marketing chain for rice in Thailand comprises of two levels, namely, the paddy market and the milled rice market as shown in Figure 1.1 Figure 1.1: Rice marketing system in Thailand Source: Wiboonponse and Chaovanapoonphol (2001) Paddy traders collect paddy from farmers and sell to millers or the central market. They usually own barns for keeping paddy and trucks to arrange transportation. The majority of farmers do not have trucks so that they have to depend on the traders to transport the paddy to the market. This is especially true in remote areas. These traders play an important role in the Lower Northern and Central region where the highest volume of paddy is produced. They sometimes provide credit and farm inputs to farmers at the beginning of the planting season. Repayment of their loans can be made either with paddy or in cash with an interest rate which is 7 normally higher than those charged by formal lending institutions. Some farmers organize into farmer groups or cooperatives, with the purpose of cooperating and supporting members’ production and marketing needs, such as financial transactions, purchase and transportation, buying equipment, and building rice barns. The difference between these organizations is that the former sell paddy to millers while the latter often deliver paddy to bigger agricultural cooperatives with a milling house. Some cooperatives are capable of milling and marketing rice. Central markets are both established by Government agencies, such as the Bank of Agriculture and Agricultural Cooperatives (BAAC), or the business sector, and located primarily in main production areas. The BAAC, through the Farmer Market Organization (FMO) and the Public Warehouse Organization (PWO), provides facilities such as weighing machines, drying lawns, and warehouses under the PSP. Private marketplaces are used as a meeting place for farmers, paddy traders, and millers to negotiate and make transactions. Millers play a role in both production and marketing. As a production unit, millers turn paddy into milled rice. As a marketing unit, millers purchase paddy from farmers, paddy traders, and the central market before distributing milled rice to consumers, brokers, wholesalers, retailers, and Government agencies. Brokers connect millers to wholesalers or millers to exporters by searching for rice with certain quantity and quality as demanded by exporters. In fact, most millers market rice to wholesalers and exporters through brokers (Maneechansook, 2011). Export of Thai rice is conducted in two forms; Private to Private sales and Government to Government sales. 8 1.2.3 Government intervention in the rice market Prior to the 1980s, the Thai Government had targeted rice price intervention policy to assist domestic consumers through several measures that effectively lowered the domestic price. This policy was perceived as a way to support the growth of the industrial sector which was very labor intensive. Among these policies, the most effective device for pushing down price was the rice premium (1954-1986), under which exporters were required to pay a premium in order to get an export license. Until the 1970s, the rice premium accounted for 20-30% of the export price in Bangkok (Shigetomi, 2009). Other frequently applied programs include the rice reserve requirement (1960-1985) and export tax (1952-1985). The former required rice exporters to sell a certain amount of rice export volume to the Government at an officially fixed price which was usually lower than the market price (Maneechansook, 2011). The latter was an ad valorem tax levied on rice exports. However, these programs were abolished by the mid-1980s when the Government changed its rice price intervention goals from consumer assistance to producer assistance. The PSP was first introduced in 1982 but it was not until 2001 that it was implemented at the national level. The Thai Rak Thai party, founded by former Prime-Minister Thaksin Shinawattra, had won a landslide general election in 2001 and changed the face of the PSP in several important respects. New features included increasing the target quantity purchased by the Government from 2.5 million tons to 8.7 million tons; allowing farmers to receive loans worth up to 100% of the paddy value with a maximum payment of 350,000 Baht per household, and raising the support price 30 percent above the market price. The 2006 military coup, followed by the ban of the Thai Rak Thai Party by the Constitutional Court of Thailand due to violations of electoral laws during the 2006 legislative elections, put the Democrat Party back in power after 9 almost a ten year hiatus. The Democrats repealed the PSP and replaced it with the DPP, which lasted from 2009 to 2011 before the PSP was brought back after the “new” Thai Rak Thai Party had won a general election in 2011. The PSP is currently operational and has been a stalwart of Thai farm policy since 2001, except for 2009-2011 when the DPP was used. The objective has clearly been to increase farm income. As a result, the program has been criticized for its populist agenda aimed at gaining political support from the poor, particularly rice farmers, and for its large economic and budget costs. By observing current market price and the support price, farmers choose whether to sell their product in the market or to the Government. In the latter case, the Government assigns BAAC to lend farmers money equivalent to the value of the pledged paddy times the support price. The paddy serves as collateral and can be redeemed within 4-6 months at the net rate of interest of 3% per annum; the Government has to subsidize 5% interest to the BAAC to make up for the total loan rate of 8% per annum. Farmers with barns or storages also earn additional income from storage fees from the Government. For those lacking storage facilities, pledged paddy are either kept at Government-authorized mills or at public storage facilities organized by the FMO and the PWO. The Government pays storage fee as well as milling fee to these millers Poapongsakorn and Charupong (2010). The DPP was first implemented in 2009. The objective was to stabilize farm income without severely distorting the market. Unlike the PSP, farmers do not have an option to sell to the Government. The Government sets a target price calculated in a way that yields farmers a guaranteed profit equivalent to 40% of production costs after accounting for transportation cost. The Government uses the estimate of market price, which is calculated based on the net market value of milled rice and the average of seven-day historical prices, to determine whether the 10 market price exceeds the target price. Then the difference between the target price and the estimated market price establishes the per unit deficiency payment for participants. Total payment is a product of the per unit deficiency payment and the quantity produced under the program. The latter is calculated as the total planted areas multiplied by average provincial yield. This means farmers earn revenue primarily from market sales sometimes supplemented by deficiency payments. The large increases in support prices since 2001 not only resulted in a large subsidy to farmers at taxpayer expense, but also created rent-seeking activities by agents throughout the rice supply chain, including rice millers and exporters. Farmers receive an interest subsidy as well as the difference between the support price and the market price. It is also argued that wealthy farmers have captured more benefits from the program than poor farmers since their production is larger and located in irrigated areas where crops can be planted more than twice a year. Rice millers receive milling fees and do not need to borrow money to procure paddy, since the Government does it for them. Exporters receive storage fees, as well as the difference between the export price and the successful bidding price offered for Government rice. Poapongsakorn and Charupong (2010) argued that the rents captured by the exporters came from collusive bidding since the bidding prices from all bidders were unrealistically low and the winner of the big lots always comes from the same group. 11 1.3 Model Analysis In this section we develop a graphical analysis which conceptualizes the impact of both PSP and DPP on economic welfare, including producer and consumer surpluses and deadweight loss. The magnitude of the impacts is determined by the changes in price and quantity which are depicted through a shift in demand and supply curves. 1.3.1 Welfare impact under a price support program (PSP) Since Thailand is a major rice exporter, a change in the domestic price of Thai rice is assumed transmitted to the world market and vice-versa. In Figure 1.2, prior to implementation of the PSP, a market equilibrium is represented by the intersection of total demand (DT) and total supply (S) at point E, for which corresponding equilibrium quantity and price are (Q*, P*). The total demand (DT) is an aggregate sum of domestic demand (represented by DD) and foreign demand (not shown in the figure). Domestic consumption is 𝑄1𝐷 . The amount exported is the difference between total supply and the amount consumed domestically (Q*-𝑄1𝐷 ). Total consumer surplus, which combines the consumer surpluses from both domestic and foreign consumers, is represented by the area AEP*; the consumer surplus of domestic consumers is represented by the area HJP* while that of foreign consumers is represented by the area AEJH. Producer surplus is represented by the area P*EO. When the Government implements the PSP and sets a support price (PS) at a level that is higher than the equlibirum market price (P*), the support price will effectively become the market price. In other words, not only the Government but also other buyers in the market will have to buy from farmers at a price equal to the support price. As a result, the market equlibirum will shift from point E to D.The corresponding quantity and price at the new equilibrium are (QS, PS). In case of Thai rice market, however,farmers incur additional costs when participating in the 12 PSP. These costs include transportation costs for delivering rice to Government depots and transaction costs associated with delayed payment of loans. Assuming that these costs are such that farmers are indifferent between participating and not participatingin the program, this causes the observed market price to be below the support price. In Figure 1.2, the effective price for non-participants is simply the market price observed, which is equal to PM. The effective price for the program participants is the support price less the additional costs, which is also equal to the market price PM. It is important to note that this effective market price is observed after the Government has released some rice stock into the world market. This means the effective market price would have been higher than PM (but smaller than PS) had the Government stored all of the pledged rice. Thus, an impact of the PSP on economic welfare is a net impact of both Government purchase sand sales. Figure 1.2: Impacts of the price support program (PSP) 13 At the market price PM, total output increases to QM. Total consumption is QCwhile the Government has net purchases of (QM-QC). Domestic consumption is then 𝑄0𝐷 .Private exports are (QC-𝑄0𝐷 ) denoted as 𝑄0𝐸𝑥,𝑃𝑟𝑣 (not shown in the figure) while the Government exports 𝑄0𝐸𝑥,𝐺𝑜𝑣 (not shown). Thus, total exports are 𝑄0𝐸𝑥,𝑃𝑟𝑣 +𝑄0𝐸𝑥,𝐺𝑜𝑣 denoted by 𝑄0𝐸𝑥 (not shown). Total amount of program loans (LOAN) are represented by the area BCQMQC. These loans are issued to farmers by BAAC when the farmers sell rice to the Government. Producer surplus has increased from P*EO to PMFO after the implementation of the PSP. Thus, the change in total producer surplus (∆𝑃𝑆) is equal to the areaPMFEP*. Total consumer surplus has decreased from AEP*to AGPM. Hence, the change in total consumer surplus (∆𝐶𝑆𝑇 ) is equal to the areaPMGEP*, which can be further decomposed into the change in domestic consumer surplus (∆𝐶𝑆𝐷 ) represented by the area PMIJP* and the change in foreign consumer surplus represented by the area IGEJ. The additional costs of program participation (AC) are represented by the area BCFG. While some part of the additional cost can be considered a transfer to other sectors, such as transportation and banking services, a portion could also be considered deadweight loss. However, it is not possible to determine these portions so the conservative assumption is made that there is no deadweight loss (all of the costs involve transfer only). Deadweight loss (DWL) from the PSP is therefore calculated by subtracting total program loans by the sum of the additional cost of program participation, the Government revenues from the redemption of pledged paddy rice (RD), the Government revenue from sales of non-redeemed paddy rice (SALEP), and the net change in producer and consumer surplus (∆𝑃𝑆+∆𝐶𝑆𝑇 ). RD includes loan principal and interest charged. SALEP includes sales of milled rice (SALEM) and its byproducts (SALEB) less operating expenses (EXP). Formulas for calculating the economic welfare impacts under PSP are summarized in Table 1.1. 14 Table 1.1: Calculation of PSP impacts Welfare Impact Notation Formula Representation in Figure 1.3 BAAC loans issued to farmers LOAN PS (QM − QC ) BCQM QC Change in producer surplus Change in total consumer surplus ∆PS ∆CST 0.5(PM − P∗ )(QM + Q∗ ) 0.5(P∗ − PM )(QC + Q∗ ) PM FEP∗ PM GEP∗ Change in domestic consumer surplus ∆CSD D 0.5(P∗ − PM )(QD 1 + Q2 ) PM IJP∗ Change in foreign consumer surplus ∆CSF ∆CST − ∆CSD IGEJ AC (PS − PM )(QM − QC ) BCFG DWL LOAN − (AC + RD + SALEP ) Not shown Costs of program participation Deadweight loss −(∆PS + ∆CST ) 15 1.3.2 Welfare impact under a deficiency payment program (DPP) This study uses a modified version of the analytical framework proposed by Schmitz and Chambers (1986) to analyze the welfare implications resulting from a deficiency payment program (DPP). The impacts of the DPP on aggregate welfare in the domestic and export markets are illustrated in the left and right panel of Figure 1.3, respectively. Figure 1.3: Impacts of the deficiency payment program (DPP) Prior to an implementation of the DPP, an equilibrium in the world market is at point J defined by the intersection of excess demand (𝐷𝐹 ) and excess supply(𝑆𝐸 ) curves as shown on the right panel. The corresponding price is 𝑃∗ , which in turn constitutes an equilibrium in the domestic market at which quantity 𝑄 ∗ is produced, quantity 𝑄1𝐷 is consumed domestically, and quantity 𝑄1𝐸𝑋 is exported (𝑄1𝐸𝑥 = 𝑄 ∗ -𝑄1𝐷 ). Total consumer surplus, which combines the consumer surpluses of both domestic and foreign consumers, is represented by the area MEP*. The consumer surplus of domestic consumers is represented by the area ABP* while that of foreign consumers is represented by the area MEBA in the left panel or the area NJP* in the right panel. 16 Producer surplus is represented by the area P*EO. The area BET represents the trade surplus, or net gain from trade. Now consider the imposition of a DPP such that the target price, 𝑃𝑇 , is above the free market equilibrium price,𝑃∗ . For any observed market price 𝑃𝐶 below 𝑃𝑇 , producers will supply quantity 𝑄 𝑇 and receive payments from the Government equal to estimated output and the difference between the target and market prices. The domestic supply curve now becomes 𝑄 𝑇 𝐹𝑆 in the left panel. Corresponding to this new domestic supply is a new excess supply curve 𝐿𝐻𝑆𝐸 in the right panel, with the segment below H corresponding to the perfectly inelastic portion (𝑄 𝑇 𝐹) of the new supply curve in the domestic market. The introduction of the DPP leads to higher quantity produced, quantity traded, and lower world price (𝑃𝐶 ). Specifically, the quantity produced increases from 𝑄 ∗ to 𝑄 𝑇 , causing the market-clearing price to drop to 𝑃𝐶 . Domestic consumption increases from𝑄1𝐷 to 𝑄2𝐷 and the quantity exported increases from 𝑄1𝐸𝑥 to𝑄2𝐸𝑥 . Total deficiency payments (PMT) are represented by the area 𝑃𝑇 𝐹𝐺𝑃𝐶 . It is important to note that deficiency payments are paid to farmers based on estimated output calculated as a product of program yield, which is fixed and varies across provinces, and the amount of rice land registered to the program. By assuming all rice land is registered and using an average national yield as program yield, deficiency payments can be calculated based on aggregate output instead of the estimated output by multiplying total output by the difference between the target price and the market price. This amounts to assuming that farmers respond to the incentive to increase production by increasing land size while keeping yield unchanged. Both producers and consumers gain as a result of the DPP. The producer surplus increases from 𝑃 ∗ 𝐸𝑂 to 𝑃𝑇 𝐹𝑂 which means the gain of producer surplus (∆𝑃𝑆) is equal to𝑃𝑇 𝐹𝐸𝑃∗ . Domestic consumers gain as their consumer surplus increases from ABP* to ACPC, a gain (∆𝐶𝑆𝐷 ) of 𝑃∗ 𝐵𝐶𝑃𝐶 .Similarly, 17 consumer surplus of foreign consumers increases from MEBA to MGCA, a gain (∆𝐶𝑆𝐹 ) of BEGC. Hence, total change in consumer surplus (∆𝐶𝑆𝑇 ) is equal to 𝑃∗ 𝐸𝐺𝑃𝐶 . Producers sell more at a higher price while consumers buy more at a lower price. According to Coffin and Henning (1989), the increase in foreign consumers’ surplus is composed of a loss in trade surplus (∆𝑇𝑆) and a loss in production efficiency (PE), represented by the areas BEDC and EGD in the left panel, respectively. Note that these areas correspond to the areas 𝑃∗ 𝐽𝑈𝑃𝐶 and JKU in right panel, respectively. The former is related to the loss that results from the target price inducing higher output and hence lowering the world price without reducing the true cost of production. The latter is the loss that arises from producing beyond the optimal level of output at which marginal cost equals marginal revenue. Since the deficiency payments are greater than the net gain from both consumer and producer surpluses combined, there is a deadweight loss (DWL) of FEG. The formulas to calculate the welfare impacts under DPP are summarized in Table 1.2. Table 1.2: Calculation of DPP impacts Welfare Impact Notati on Formula Representation in Figure 1.3 Total deficiency payments 𝑃𝑀𝑇 (𝑃𝑇 − 𝑃𝐶 )𝑄 𝑇 𝑃𝑇 𝐹𝐺𝑃𝐶 Change in producer surplus ∆𝑃𝑆 0.5(𝑃𝑀 − 𝑃∗ )(𝑄 𝑇 + 𝑄 ∗ ) 𝑃𝑇 𝐹𝐸𝑃 ∗ Change in total consumer surplus ∆𝐶𝑆𝑇 0.5(𝑃∗ − 𝑃𝐶 )(𝑄 𝑇 + 𝑄 ∗ ) 𝑃 ∗ 𝐸𝐺𝑃𝐶 Change in domestic consumer surplus ∆𝐶𝑆𝐷 0.5(𝑃∗ − 𝑃𝐶 )(𝑄1𝐷 + 𝑄2𝐷 ) 𝑃∗ 𝐵𝐶𝑃𝐶 Change in foreign consumer surplus ∆𝐶𝑆𝐹 ∆𝐶𝑆𝑇 − ∆𝐶𝑆𝐷 Loss in production efficiency 𝑃𝐸 0.5(𝑃∗ − 𝑃𝐶 )(𝑄 𝑇 − 𝑄 𝐻 ) Loss in trade surplus ∆𝑇𝑆 ∆𝐶𝑆𝐹 − 𝑃𝐸 Deadweight loss DWL 0.5(𝑃𝑇 − 𝑃𝐶 )(𝑄 𝑇 − 𝑄 ∗ ) 18 BEGC or 𝑃∗ 𝐽𝐾𝑃𝐶 EDG or JKU BEDC or 𝑃∗ 𝐽𝑈𝑃𝐶 FEG Unlike in the case of PSP, farmers are assumed to incur no additional cost when participating in DPP. Recall that the additional cost under PSP includes transportation cost accrued when delivering harvests to the Government’s designated depots and delayed payments due to lack of Government funds. Since farmers must sell rice on the open market, there is no additional transportation cost to the Government depots. Because there is no evidence of delayed payments reported during the course of DPP implementation, the cost of delayed payment is assumed negligible. This assumption is supported by the facts that the payment made to each household is much smaller under DPP, and that the Government has knowledge regarding the amount of funds needed to be allocated to each branch of the BAAC on a daily basis, which reduces the chances of having insufficient funds for making deficiency payments to farmers. Recall that the Government only has to pay the difference between the target and market prices to compensate farmers instead of buying rice from them. The target price is known to farmers prior to a start of each cropping season but the market price is unknown. The Government announces the estimated market price on every Monday during harvesting season. When signing the program contract, farmers are required to specify the date at which they want to exercise the right to receive deficiency payments. Farmers then receive deficiency payments only when the estimated market price announced on Monday of the same week as the chosen date is below the target price. 19 1.4 Methods and data In order to compare the impact of the Thai rice PSP and DPP on economic welfare (i.e. producer surplus, consumer surplus, trade loss, and deadweight loss), one needs to find counterfactual values of prices and quantities that would have been observed had the policy regime implemented in any given period been different. In this study, the 2005/06 cropping season, in which the PSP was operational, is used to evaluate how alternative policies would have performed. This period was chosen mainly because it has the most detailed information on revenue and cost of the PSP which typically are not released by the Government. Fortunately, Poapongsakorn (Poapongsakorn and Charupong, 2010), a former member of Thailand National Rice Committee, was able to compile this information of the opertion of PSP in 2005/06 cropping season. Two counterfactual scenario, namely no Government intervention (NG) and the DPP, are investigated under the following circumstances:(i) when the target price is set equal to the support price observed in the 2005/06 cropping season; and (ii) when total deficiency payments are set equal to the same Government expenditures as observed under PSP during the 2005/06 cropping season. The comparison when target price and support price are set equal is designed to give insight into the relative effects when farmers face the same “minimum price” under each program, while the comparison when Government expenditures are set equal is designed to give insight into the “cost neutral” performance of the programs when Government costs are the same under each program. In each case, the changes in economic welfare can be calculated geometrically from relevant values of price and quantity observed in the selected period, assuming knowledge of key elasticities of supply and demand. 20 1.4.1 Calculating welfare impacts under the PSP using price elasticities Referring to Figure 1.2, the effective market price (PM) and observed production and consumption levels observed in the 2005/06 cropping season are substituted into equations representing demand and supply to compute for the relevant values of unknown variables including equilibrium price and output (Q*, P*) and the corresponding domestic and foreign consumptions (𝑄1𝐷 and 𝑄 ∗ - 𝑄1𝐷 )when there is no Government intervention. The unknown variables that must be calculated in order to compute the changes in economic welfare under the PSP include the equilibrium market price (P*), total output (Q*), and domestic consumption (𝑄1𝐷 ) under no Government intervention. Here, P* and Q* (located at point E in Figure 1.2) can be found by solving the following equations that represent price-elasticity of total demand (𝜀 𝑎 ) and price-elasticity of supply (𝜀 𝑠 ), respectively.1 𝜀 𝑎 = [(𝑄 𝐶 − 𝑄 ∗ )/𝑄 ∗ ]⁄[(𝑃𝑀 − 𝑃∗ )/𝑃∗ ] (1.1) 𝜀 𝑠 = [(𝑄 𝑀 − 𝑄 ∗ )/𝑄 ∗ ]⁄[(𝑃𝑀 − 𝑃∗ )/𝑃∗ ] (1.2) PM and QM, which respectively represent the equilibrium price and quantity under the PSP, are observed. So, P* and Q* are the only unknown variables and can be found directly by solving the system of two equations with two unknowns. In order to solve for 𝑄1𝐷 , which represents domestic consumption at point J on the domestic demand curve DD in Figure 1.2, the elasticity of domestic demand expressed in the following equation must be solved. 𝜀 𝑑 = [(𝑄0𝐷 − 𝑄1𝐷 )/𝑄1𝐷 ]⁄[(𝑃𝑀 − 𝑃 ∗ )/𝑃∗ ] 1 𝑎 𝜀 is a weighted average of price elasticity of domestic and export demand (𝜀 𝑑 and 𝜀 𝑥 , respectively) 21 (1.3) Again, all variables except 𝑄1𝐷 are known.Note that 𝑄0𝐷 is known and equals total production(𝑄 𝑀 ) less the sum of Government purchases (QG) and total exports (𝑄0𝐸𝑥 ); 𝑄0𝐸𝑥 is equal to a sum of private exports (𝑄0𝐸𝑥,𝑃𝑟𝑣 ) and Government exports (𝑄0𝐸𝑥,𝐺𝑜𝑣 ). Once all these unknown variables are determined, the impact of price support program on economic welfare can be computed directly. 1.4.2 Calculating welfare impacts under the DPP using price elasticities For the first DPP evaluation the target price (PT) is set equal to 8,465 baht/ton which is the support price (PS) observed in the 2005/06 season. To evaluate DPP effects three variables need to be computedtotal supply (𝑄 𝑇 ), equilibrium market price (𝑃𝐶 ), and domestic consumption (𝑄2𝐷 ). A system of three equations, which represent price-elasticity of total supply (𝜀 𝑠 ), price-elasticity of total demand (𝜀 𝑎 ), and price-elasticity of domestic demand (𝜀 𝑑 ), can be used to compute these unknowns. First, 𝑄 𝑇 can be calculated by substituting the target price (PT) and the set of information (𝑄 ∗ , 𝑃∗ , 𝑄1𝐷 , 𝑄1𝐸𝑥 ), which are known from previous calculation in the case of PSP, into the equation representing the price-elasticity of supply (𝜀 𝑠 ) evaluated at point E in Figure 1.3.The target price (PT) is set equal to the support price (PS).2 This means QT is the only unknown in the equation and hence it can be solved directly for a given value of 𝜀 𝑠 . 𝜀 𝑠 = [(𝑄 𝑇 − 𝑄 ∗ )/𝑄 ∗ ]⁄[(𝑃𝑇 − 𝑃 ∗ )/𝑃∗ ] 2 (1.4) Generally, the price that farmers receive when selling to the government is lower than the support price, because the price is discounted depending on moisture content and product-byproduct ratio. Thus, PS is set equal to the effective support price defined as total BAAC loans issued to farmers under price support program divided by total quantity of pledged rice. 22 Similarly, 𝑃𝐶 is the only unknown in the equation below that represents elasticity of total demand (𝜀 𝐴 ) at point E in Figure 1.3, so it can be solved directly for a given value of 𝜀 𝑎 . 𝜀 𝑎 = [(𝑄 𝑇 − 𝑄 ∗ )/𝑄 ∗ ]⁄[(𝑃𝐶 − 𝑃∗ )/𝑃∗ ] (1.5) The last unknown variable to solve for is the domestic consumption under deficiency payment program (𝑄2𝐷 ). It can be found by solving the equation of price-elasticity of domestic demand (𝜀 𝑑 ) at point B in Figure 1.3 in which 𝑄2𝐷 is the only unknown variable. In addition, subtracting domestic consumption from total supply yields a value of total export (𝑄2𝐸𝑥 ) under the DPP. 𝜀 𝑑 = [(𝑄2𝐷 − 𝑄1𝐷 )/𝑄1𝐷 ]⁄[(𝑃𝐶 − 𝑃∗ )/𝑃∗ ] (1.6) For the case where DPP Government expenditures are set equal to those under the PSP, total deficiency payments (PMT) are set equal to 51,758 million baht which is the sum of total loans paid to farmers and the Government’s operating expenses of the PSP in the 2005/06 season. The calculation of the unknown variables (𝑄 𝑇 , 𝑃𝐶 , 𝑄2𝐷 ) are similar to that discussed previously except that the target price (𝑃𝑇 ) is now another unknown to solve for. A system of four equations consisting of the equations from (1.4)-(1.6) and the equation representing total deficiency payments can be used to solve for these four unknowns. The latter equation expresses total deficiency payments as a product of total supply and the difference between the target price and market price. 𝑃𝑀𝑇 = 𝑄 𝑇 (𝑃𝑇 − 𝑃𝐶 ) (1.7) 23 1.4.3 Data This study uses secondary data obtained from various agencies. The quantity and price data that covers the period from November 2005 to February 2006 is used to estimate an impact of the PSP on economic welfare during the first-cropping season of 2005/06 production year. This particular period was chosen mainly because it is the only period that the data on revenue and cost of the PSP are available. The production environment in this selected period is described as a typical cropping season in which average annual temperature and rainfall were 26.64 degree Celsius and 101.17 millimeter.3 Similarly, the market condition was stable as described by a comparable ratio of total world supply and exports relative to that of the previous year. Specifically, total world supply of paddy rice was 621.40 million tons compared to 596.60 million tons in previous year while total world exports of milled rice were 29.10 million tons compared to 29.00 million tons in previous year.4 The market price (PM), support price (PS), and target price (PT) are the weighted average of prices calculated from three types of rice, including white rice, jasmine rice, and glutinous rice. Total supply (QM) and private export quantity (𝑄0𝐸𝑥,𝑃𝑟𝑣 ) are the sum of monthly quantities of all rice types. The support and target prices are obtained from Thailand Department of Internal Trade Office (DIT). Total production, total private exports, and market prices are obtained from Thailand Office Agricultural Economics (OAE). According to Poapongsakorn and Charupong (2010), the sales of Government rice stock from the 2005/06 first-cropping season did not happen until one year later and it took the Government as long as three years to sell all of the rice. Thus, exports of the Government rice stock (𝑄0𝐸𝑥,𝐺𝑜𝑣 ) in 2005/06 are calculated as an 3 Source: World Bank. Available at http://data.worldbank.org/indicator/ [accessed September 1, 2014] Source; Rice Outlook, USDA Economic Research Service. Available at http://usda.mannlib.cornell.edu/MannUsda/viewDocumentInfo.do?documentID=1285 [accessed September 1, 2014] 4 24 aggregate sum of monthly exports from November 2004 to February 2005 of non-redeemed rice from the operation of the PSP in previous cropping season (2004/05).5 No attempt has been made to estimate the relevant price elasticities since implementation of the PSP. One reason is lack of sufficient data and another is because it is difficult in practice to capture the dynamics of demand and supply through interactions of a support price, market price, quantity of pledged rice, and the Government’s rice exports. The present study does not attempt to estimate price elasticities directly either but uses estimates from existing literatures instead. Two sets of estimates of price elasticity of total supply, domestic demand, and export demand are drawn from the same sources as referenced in the study by Poapongsakorn and Charupong (2010). The combination of these estimates is used to create eight scenarios for sensitivity analysis. Estimates of price-elasticity of total supply (𝜀 𝑆 = 0.086) and domestic demand (𝜀 𝐷 = −0.392) are obtained from the study by Isvilanonda and Kongrith (2008). This study provides the most recent estimates of these elasticities in case of the Thai rice market. In their study, price-elasticity of demand was estimated using an Almost Ideal Demand System (AIDS) model while price-elasticity of supply was estimated by seemingly unrelated regression estimation (SURE) in which the dependent variables are production share of major crops. The models were estimated using price and quantity data from 1970-2000. The other set of estimates of priceelasticity of total supply and domestic demand come from the studies by Konjing (1980) (𝜀 𝑆 = 5 Unfortunately, the information on the government exports of non-redeemed rice retained from the operation of PSP in 2004/05 is not available. Instead, the government exports of non-redeemed rice retained from the operation of PSP in 2005/06, made available by Poapongsakorn and Charupong (2010), are used to derive the amount of government exports in 2005/06. Specifically, we assume that it also takes the Government three years to sell the rice stock retained from the 2004/05 PSP as did in 2005/06. Furthermore, the monthly exports by the government throughout the course of the three-year span are assumed equal. For a given rate of export, the amount of government exports in 2005/06 can be calculated accordingly. 25 0.0453) and Siamwala and Pattamasiriwat (1989) (𝜀 𝐷 = −0.12). Estimates of price-elasticity of export demand are obtained from two sources-- Siamwala and Pattamasiriwat (1989) (𝜀 𝑋 = −4) and Suntayoom (1981) (𝜀 𝑋 = −7.04). These ranges of price elasticity are quite reasonable for several reasons. The estimates of price-elasticity of supply indicate an inelastic supply of Thai rice, which is consistent with the fact that land is limited in Thailand so that an increase in production by land expansion is difficult. Because rice is the only staple food in the Thai diet, domestic demand for rice consumption is more inelastic than export demand and hence the priceelasticity of domestic demand is smaller than that of the export demand. 26 1.5 Results 1.5.1 Estimates of the impact of PSP on economic welfare Table 1.3 shows the summary of market data under the PSP including prices and quantities, and the program expenditure/revenue observed during the 2005/06 cropping season. The calculated variables include prices and quantities under no Government intervention, which are the counterfactual values of price and quantity that would have been observed in this period had there been no PSP (nor DPP). The support price was set at 8,465 baht/ton, a weighted average of the actual support price of jasmine rice, white rice, and glutinous rice during the season. Total supply (QM) was 23.34 million tons while market price (PM) was 6,614 baht/ton, a weighted average from all three types of rice. Recall that, in order to account for additional costs associated with program participation, both program participants and non-participants are assumed to receive the same realized/effective price when selling their rice. Thus, the effective price for PSP participants equals the observed market price (PM). The Government purchased a total of 5.29 million tons of paddy rice from farmers and exported a total 1.29 million tons of non-redeemed rice withheld from previous seasons.6 Total private exports were 4.54 million tons. Thus, total export (𝑄0𝐸𝑥 ) and domestic consumption (𝑄0𝐷 ) were 5.83 and 12.21 million tons, respectively.7 6 Government exports are calculated by multiplying total non-redeemed rice in the previous season by the Government rate of export. Since neither the redemption rate nor the export rate of the non-redeemed rice from the operation of PSP in 2004/05 is known, they are assumed equal to their 2005/06 counterparts found in Poapongsakorn & Charupong (2010). Specifically, the redemption rate is 21.93% of total pledged rice while the export rate is 6.50% of total nonredeemed rice. Given the total amount of pledged rice of 5.10 million tons, Government exports in 2005/06 are estimated at 1.29 million tons of paddy rice. 7 𝐷 𝑄0 is calculated from the identity equation that expresses total supply as the sum of domestic consumption, government purchase, and exports. 27 Total loans made to farmers by the Government through the BAAC (LOAN) were 44,797.02 million baht. Total operating costs (EXPS) were 6,961.24 million baht. Hence, total cost of the PSP is 51,758 million baht. The Government has two sources of revenue including the loan repayment/redemption (RD) of pledged rice worth 10,706 million baht and the sales of nonredeemed rice (𝑆𝐴𝐿𝐸𝑀 ) and its byproduct (𝑆𝐴𝐿𝐸𝐵 ) worth 24,760 million baht. Note that the Government purchases paddy rice from farmers, processes the rice, and sells it as milled rice. In order to calculate loss/gain of rice sales measured in paddy-rice-equivalent value, operating expenses (EXPS) are subtracted from the value of milled rice and its byproduct (𝑆𝐴𝐿𝐸𝑀 +𝑆𝐴𝐿𝐸𝐵 ) sold by the Government. This gives the sales value of pledged rice (SALEP) of 17,799 million baht. Table 1.3: Summary of the data observed during the implementation of PSP in 2005/06 Variables Price & Quantity under PSP (Unit: baht/ton & million tons) Support price (PS) Quantity of pledged rice (QG) Total supply (QM) Effective market price (PM) Total consumption (QC) Domestic consumption (𝑄0𝐷 ) Private exports (𝑄0𝐸𝑥,𝑃𝑟𝑣 ) Government exports (𝑄0𝐸𝑥,𝐺𝑜𝑣 ) Total export s(𝑄0𝐸𝑥 ) Program Cost & Revenue (Unit: million baht) BAAC loans (LOAN) Value of redeemed rice (RD) Sales of non-redeemed rice and its byproduct (𝑆𝐴𝐿𝐸𝑀 +𝑆𝐴𝐿𝐸𝐵 ) Operating expenses (EXPS) Value of the sales of non-redeemed paddy rice (SALEP) 28 Value 8,465 5.29 23.34 6,614 18.05 12.21 4.54 1.29 5.83 44,797 10,706 24,760 6,961 17,799 Table 1.4 reports the estimated effects of the PSP and DPP under eight combinations of supply and demand elasticities used in evaluate sensitivity of results to the elasticity assumptions. Specifically, the combinations of price-elasticity of domestic demand (-0.392 and0.120) and export demand (-4.00 and -7.08) constitute four different values of price-elasticity of total demand. Putting these cases together constitutes total of eight cases that are investigated in this study. The last two columns of the table report the upper and lower bound of computed values of welfare change. These values are the minimum and maximum corresponding to the welfare changes reported in the table. The counterfactual values of total supply and market price under no Government intervention (Q*and P*) range between 22.99-23.23 million tons and 5,485-5,991 baht/ton, respectively. These values are lower than those observed under the PSP, as expected. In other words, the PSP has raised total supply by between 0.45-1.54% and the market price by between 10.40-20.58% compared to the no intervention case. Because all outputs must be consumed domestically or exported, total exports (𝑄1𝐸𝑥 ) and domestic consumption (𝑄1𝐷 ) under no intervention are larger than those under the PSP. Specifically, exports and domestic consumption are estimated between 10.00-10.85 million tons and between 12.37-13.04 million tons, respectively. Given total exports by private exporters of 5.83 million tons in 2005/06, this means implementation of PSP caused private exports to fall by 41.70- 50.20%. An increase in market price under the PSP results in an increase in producer surplus (∆𝑃𝑆) ranging between 14,487-26,237 million baht and a decrease in total consumer surpluses (∆𝐶𝑆𝑇 ) ranging from -23,250 to -12,838 million baht. Consumer surplus of domestic consumers (∆𝐶𝑆𝐷 ) falls by -13,944 to -7,764 million baht while that of foreign consumers (∆𝐶𝑆𝐹 ) falls by 9,306 to -5,073 million baht. Despite receiving the support price on rice sales, participants of the 29 PSP incur these additional costs of participation which are then transferred to other sectors such as banking services in the form of interests paid on loans borrows for rice production and transportation services in the form of extra transportation cost for delivering rice to the designated Government depots. The additional costs of program participation (AC), calculated as a product of the difference between the support price (PS) and market price (PM) times the quantity of non-redeemed paddy rice, is estimated at 7,639 million baht. Lastly, the deadweight loss (DWL) associated with the program is estimated to range from 5,665-7,002million baht. The amount of additional costs and deadweight loss together represent the total amount from the total cost of PSP that are wasted as neither of them contributes to an increase the welfare surpluses nor Government revenue. This means as much as 10.94-13.53% of total Government spending on the operation of PSP are wasted as deadweight loss while 14.76% accounts for the additional cost of the program accrued in order to keep the program running. 30 Table 1.4: Estimates of the impacts of the price support program (PSP) Cases by Elasticities d x a s Case 1 Case 2 Case 3 Case 4 Case 5 Case 6 Case 7 Case 8 -0.39 -0.39 -0.12 -0.12 -0.39 -0.39 -0.12 -0.12 -4.00 -7.08 -4.00 -7.08 -4.00 -7.08 -4.00 -7.08 -1.31 -2.09 -1.10 -1.88 -1.31 -2.09 -1.10 -1.88 0.09 0.09 0.09 0.09 0.05 0.05 0.05 0.05 Price & quantity under NG Total supply (Q*) Market price (P*) Domestic consumption (𝑄1𝐷 ) Export (𝑄1𝐸𝑋 ) (Unit: baht/ton & mil. ton) 23.02 23.14 22.99 5,629 5,991 5,510 13.02 12.70 12.49 10.00 10.44 10.50 Welfare Impacts of PSP Change in PS (PS) Change in total CS (𝐶𝑆𝑇 ) Change in domestic CS (𝐶𝑆𝐷 ) Change in foreign CS (𝐶𝑆𝐹 ) Additional cost of participation (AC) Deadweight loss(DWL) (Unit: mil. baht) 22,853 14,487 25,586 15,539 23,368 14,688 26,237 15,770 14,487 26,237 -20,245 -12,838 -22,663 -13,769 -20,709 -13,019 -23,250 -13,978 -23,250 -12,838 -12,438 -7,764 -13,642 -8,221 -12,687 -7,858 -13,944 -8,327 -13,944 -7,764 -7,807 -5,073 -9,021 -5,548 -8,022 -5,161 -9,306 -5,651 -9,306 -5,073 7,639 7,639 7,639 7,639 7,639 7,639 7,639 7,639 7,639 7,639 6,043 7,002 5,729 6,882 5,993 6,983 5,665 6,859 5,665 7,002 31 23.13 5,946 12.37 10.76 23.17 5,610 13.04 10.13 23.23 5,984 12.70 10.53 23.15 5,486 12.49 10.65 23.23 5,937 12.37 10.85 Min Max 22.99 5,486 12.37 10.00 23.23 5,991 13.04 10.85 1.5.2 Estimates of the impact of DPP on economic welfare when the target price is equal to the support price The counterfactual quantities and prices under no Government intervention are used in conjunction with the assumed values of price elasticity of demand and supply to find the counterfactual values of quantities and prices under the DPP assuming that the target price (PT) is set equal to the support price (PS). This means that total supply (Q*), market price (P*), domestic consumption (𝑄1𝐷 ), and export quantity (𝑄1𝐸𝑥 ) under no Government intervention (NG) are as previously calculated. The target price is set equal to 8,465 baht/ton. These values of price and quantity, and estimates of the DPP impact are presented in the top part of Table 1.5. The bottom part of the table contains estimates of prices and quantities under the DPP. Like the case of the PSP, a sensitivity analysis on the calculation of welfare impacts is conducted using the combination of price elasticities of total demand and supply that constitutes eight different cases. The last two columns of the table report maximum and minimum of the welfare changes drawn from all eight cases. Supply is estimated to have increased under DPP while market price has fallen compared to no intervention. The counterfactual values of total supply and market price under DPP (QT and PC) range between 23.61-23.87 million tons and 5,358-5,946baht/ton, respectively. Equivalently, the DPP is estimate to have caused total supply to increase by 1.61-3.85% and the market price to fall by 0.62-2.75%. Because all outputs must be consumed domestically or exported, total exports (𝑄2𝐸𝑥 ) and domestic consumption (𝑄2𝐷 ) increase; exports rise to between 10.52-11.46 million tons while the increase in domestic consumption ranges between 12.38-13.13 million tons. Since the target price is higher than the market price under no intervention (PT>P*), there is a gain in producer surplus; ∆𝑃𝑆 ranges between 58,121-69,673 million baht. Similarly, there is a 32 gain to consumers in term of an increase in consumer surplus as the market price under the DPP is lower than that under no intervention (PC 0for x[a,b]. Major problems of applying FSD include low discriminating power and the left-hand tail problem, resulting in a very large efficient set. The former refers to a situation in which it is unlikely to reject distributions that most risk-averse agents would eliminate from the efficient set. The later refers to a situation in which the distribution which first accumulates probability can never dominate other distributions that accumulate probability later by all decision makers who prefer more to less, no matter what happens later on. Despite higher discriminating power, SSD still faces the left-hand tail problem. As a result, the efficient set still remains very large under SSD (Robison and Myers, 2001) 2.2.2 Stochastic dominance with respect to a function (SDRF) Generalized stochastic dominance, more generally referred to as stochastic dominance with respect to a function (SDRF) (Meyer, 1977a, 1977b) is capable of eliminating the left-hand tail problem. SDRF orders risky alternatives for a class of decision makers whose utility function is defined by lower bound risk aversion coefficients denoted by 𝑟𝐿 (𝑥) and upper bound risk aversion coefficients denoted by 𝑟𝑈 (𝑥). SDRF eliminates inefficient alternatives by determining the utility function𝑢(𝑥) which satisfies (2.2) and minimizes (2.1), and checking whether the minimum is non-negative or not. 1 ∫0 [G(x) − F(x)]u′ (x)dx , ∀x ∈ [1,0] 𝑟𝐿 (𝑥) ≤ −𝑢′′ (𝑥)⁄𝑢′ (𝑥) ≤ 𝑟𝑈 (𝑥) (2.1) (2.2) Since (2.1) equals the expected utility from F(x) minus the expected utility from G(x), we know that if we minimize it over a set of agents and the minimum is greater than or equal to zero then the set of agents unanimously prefer or are indifferent between F(x) and G(x). If the 49 minimum is less than zero then the set of agents is not unanimous in choosing F(x) over G(x). Thus, the set of agents with risk aversion measures between 𝑟𝐿 (𝑥) and 𝑟𝑈 (𝑥)unanimously prefer or are indifferent between F(x) and G(x) if and only if the minimum of (2.1) subject to (2.2) is greater than or equal to zero. 2.2.3 Stochastic efficiency with respect to a function (SERF) The practical application of SDRF is limited because the method usually results in a large efficient set of risky alternatives. The reason is that SDRF requires unanimous agreement from all decision makers with all possible utility functions. The ranking by SDRF is also sensitive to marginal changes in the upper and lower bounds of risk aversion. Hardaker et al. (2004) proposed stochastic efficiency with respect to a function (SERF) as an alternative procedure to SDRF. Unlike SDRF, SERF requires that all risk aversion measures are of the same functional form as the lower and upper bound function. For instance, assuming constant bounds requires considering only decision makers with constant risk aversion coefficients (e.g. constant absolute risk aversion (CARA) or constant relative risk aversion (CRRA). Instead of eliminating an inefficient alternative by determining whether the minimum of (2.1) subject to (2.2) is greater than or equal to zero as SDRF does, SERF converts expected utility derived under each alternative into certainty equivalent values (CE) over a specified range of risk aversion as specified in (2.2). Because SERF makes a simultaneous rather than a pairwise comparison of risky alternatives, the efficient set obtained from SERF is smaller than that of SDRF. It is important to note that SDRF and SERF do not necessarily deliver the same efficient set (Meyer et al., 2009). For the set of agents defined by the upper and lower bounds of risk aversion in (2.2), and for a chosen form of the utility function, the function for utility in terms of the stochastic 50 outcome x and risk aversion r(x) is denoted as U(x, r(x)). The corresponding expected utility EU(x, r(x)) is defined as: 𝐸𝑈(𝑥, 𝑟(𝑥)) = ∫ 𝑈(𝑥, 𝑟(𝑥))𝑑𝐹(𝑥) ∑𝐼𝑖 𝑈(𝑥𝑖 , 𝑟(𝑥))𝑃(𝑥𝑖 ) (2.3) The first term on the right-hand side of (2.3) represents expected utility in a case when x is continuous while the second term represents its approximation evaluated over several discrete values of x. F(x) is the probability density of the stochastic outcome when x is continuous. In the discrete case, 𝑃(𝑥𝑖 ) is the probability for states i and there are I states for each risky alternative. According to Hardaker et al. (2004), starting with CDF data for a set of risky alternatives, equation (2.3) implies the following computational steps for evaluating SERF: - Select points on each CDF for a finite set of values of x. - Convert each of these x vales to its corresponding utility using the selected form of the utility function and the selected values of the risk version coefficient. - Multiply each finite utility by its associated probability to calculate a weighted average of the utilities of outcomes or expected utility (EU). - Converting each calculated value of expected utility into CE values which can be done by taking the inverse of the utility function evaluated at the expected utility: 𝐶𝐸(𝑥, 𝑟(𝑥)) = 𝑈 −1 [𝐸𝑈(𝑥, 𝑟(𝑥))] (2.4) The discrete function is then evaluated for a sufficient number of discrete points of r(x) to describe the relationship between CE and r(x) for each alternative. Partial ordering of alternatives by CE is the same as partially ordering them by utility values. Using this approach we end up with a vector of CE values for each of the n alternatives calculated for several values of r(x) within the bound as specified in (2.2). Only those alternatives which have the highest CE values 51 for some value in the range of r(x) are considered efficient. All other alternatives are dominated in the SERF sense. The method also provides a cardinal ranking at each level of risk aversion by interpreting differences in CE values as risk premiums. It is argued that the additional assumptions made in SERF are often reasonable because relative risk aversion is more or less constant for moderate variations in wealth (Hardaker and Lien, 2010). In other words, it is not likely that there will be large kinks in utility functions for wealth, causing sudden, large changes in relative risk aversion as wealth varies. Furthermore, when there is a change in the efficient set with a change in the form of the utility function, the cost of being wrong, measured by the difference in CEs, is almost always trivially small. For these reasons, SERF is used in this present study to rank farmer preferences. SIMETAR is a user-friendly program capable of performing the SERF ranking of risky alternatives.11 The program allows users to choose either CARA or CRRA as decision makers’ risk aversion function, which respectively entails to assuming a negative-exponential or power utility function. 11 Simulation and Econometrics to Analyze Risks; www.simetar.com 52 2.3 Methods and data In this study constant relative risk aversion (CRRA) coefficients of 0.5 to 4 are used as the lower and upper bounds of risk aversion to define a class of risk-averse farmers. A representative farmer is assumed to grow rice on a 25-rai rice land farm (the national average of land used for rice production per household) and to face stochastic profits caused by random yield and random market prices. Random yields are simulated from the estimated mean and variance obtained from the first-order autoregressive model (AR (1)) of yield with a time trend included. This choice of specification was chosen based on test statistics which reject the null hypothesis of unit root and support the null hypothesis of first-order autocorrelation of yield. The price distribution under each policy regime is simulated from the mean and variance obtained from a vector autoregressive (VAR) model which captures the relationship between the farmgate price of Thai rice and the rice prices from other exporting countries as well as the effect of policy intervention (PSP and DPP). The VAR model was chosen over other alternative models, such as a structural model of demand and supply, mainly due to data limitations. Because the preference ranking depends on the relative differences between the support/target price and market price, which varies across years, one could make a ranking of the alternative policies for all the years that these policies were implemented in order to draw an overall conclusion. However, a reasonable alternative is to investigate only the year in which the maximum and minimum of the price differential were observed. By evaluating these two extremes we can determine the sensitivity of the preference ranking to the gap between the support/target price and the market price. To achieve this goal, preference rankings in the two first-cropping seasons, 2006/07 and 2012/13, are investigated. These periods represent the period in which the price differential is smallest and largest, respectively. SERF is employed to 53 investigate the ranking of farmer preferences under the following five scenarios; (1) no government intervention denoted as NG, (2) participating in PSP denoted as PSP, (3) declining PSP when the program is available denoted as NPSP, (4) participating in DPP denoted as DPP, and (5) declining DPP when the program is available denoted as NDPP. 2.3.1 Estimating the impacts of PSP and DPP on market price Given that only small portions of total world rice supply are traded in the world market, and that Thailand is the major exporter of rice, any change in the Thai rice price is likely to have significant impacts on the world rice prices and vice-versa. A VAR model can account for these interactions. In this study, the VAR model is specified as 𝐾 𝐽 𝐽 𝑙𝑜𝑔𝑷𝑡 = ∑ 𝑨𝑖 𝑙𝑜𝑔𝑷𝑡−𝑖 + ∑ 𝑩𝑖 𝑿𝑡−𝑗 + ∑ 𝑪𝑖 𝒁𝑡−𝑗 + 𝑼𝑡 𝑖=1 where Pt 𝑗=0 (2.5) 𝑗=0 is a vector of rice prices in period t Xt is a vector of exogenous variables Zt is a vector of dummy variables representing different policy regimes Ut is a vector of error terms assumed distributed as N(0,); variance of the error term in each equation is represented by a symmetric matrix12 A, B, C denote matrices of coefficients ̂𝑡2 was regressed on the same set of To investigate potential heteroskedasticity in variance of the error terms, 𝑙𝑜𝑔𝑈 explanatory variables that appear on the right-hand side of (2.5). LR-test indicates that these variables are not jointly significant at 5% significant level. The slope coefficients of the policy dummies are also individually insignificant at 5% level of significance. Thus, the null hypothesis of homoscedastic variance cannot be rejected. 12 54 P includes the farm-gate price of Thai white rice, the export price of Vietnam 5% white rice, and the export price of Pakistan 25% white rice. Several statistical tests are employed to justify the number of lagged prices included in the model.13 The optimal number of lagged prices suggested by each test varies and ranges from two to four. Due to small sample size, two lags of each price variables are chosen for analysis here. X includes a set of dummy variables representing the seasonal difference in production environment and marketing (Season), the impacts of the food price crisis which took place in 2008 (Crisis2008and Post-Crisis), and the difference in political incentives held by the elected Government and military Government regarding the implementation of PSP (Coup2006). The variable Season is equal to 1if it is the first-cropping season and zero for the second-cropping season.14 Because production of the first crop is typically twice as large that of the second crop, the market price tends to be lower during the first season. The variable Crisis2008 is included to control for the unprecedented increase of rice prices in 2008; it is equal to 1 for both seasons in 2008 and zero elsewhere. Rice prices have not fully returned to normal levels observed prior to the crisis, so the variable Post-Crisis included to account for this recovery process; it is equal to 1 for all seasons after 2008 and zero elsewhere. Although the interim government selected by the military leaders did not discontinue PSP following the military coup that took place in September 2006, the support price was substantially cut to a level close to the market price. This probably reflects a contrasting view held by the elected Government and military government. Hence, the variable Coup2006 is included to control for this change; it is equal to 1 for both seasons in 2007 and zero elsewhere. 13 These tests include log-likelihood-ratio test, final prediction error (FPE), Akaike's information criterion (AIC), Schwarz's Bayesian information criterion (SBIC), and Hannan and Quinn Bayesian’ information criterion (HQIC). 14 First-cropping season starts in June and harvests between October and December. Although, planting and harvesting time for second crop vary by regions, most crops are delivered to the market around March and April. 55 Z contains a set of policy variables representing periods in which PSP and DPP were implemented.15 These dummy variables are denoted by PSP and DPP, respectively. Note that, by setting both PSP and DPP to zero, the resulting regime represents no government intervention (NG). Ideally, we would want to measure the marginal impact of the changes in actual values of the support/target price on market price. To do that, we need to have a sufficient number of observations from the periods during which the programs were operating as well as enough variation in the support/target prices over time. Unfortunately, PSP has been implemented only for ten years while DPP was terminated after two years of operation. Furthermore, the variation in the support/target price over those years is small. For these reasons, the dummy variables representing each policy regime are used instead of the actual values of the support/target prices. Similarly, a small number of observations limits the adoption of a more flexible model in which policy choices appear as endogenous variables. However, the issue of endogeneity may be less of a concern in this application because the policy choice is arguably less dependent on economic factors but heavily driven by the face of the political party in power. There are two major political parties in Thailand who strongly disagree on the choice of policies for the rice market. One party has always supported the implementation of PSP while the other has a strong stand against the program in favor of DPP. When elected, each of these parties have a history of repealing a program previously implemented by its opposition in order to implement one of its own. 15 PSP started in the second-cropping season of 2001/02 production year and has continued until now with two years of interruption by DPP from the first-cropping season of 2009/10 production year until 2010/11. 56 2.3.2 Simulating counterfactual market prices The distribution of counterfactual market prices under alternative policies can be simulated by changing the policy variables (Z) to appropriate values. The market price (pt) is assumed to be conditionally lognormally distributed: 𝑝𝑡 |(𝑥𝑡 , 𝑧𝑡 )~ 𝐿𝑜𝑔𝑛𝑜𝑟𝑚𝑎𝑙 (µ𝑡 , 2 ) (2.6) where µ𝑡 and 2 are the conditional mean and variance of log(𝑝𝑡 ). The conditional mean and variance of the market price are then: E(𝑝𝑡 ) = exp(µ𝑡 +2 /2) (2.7) Var(𝑝𝑡 ) = (exp(2 )-1)(exp(2µ𝑡 + 2 )) (2.8) Because the price variables in the VAR are in logarithmic form, simulating the VAR provides estimates of µ𝑡 and 2 . Then (2.7) is used to generate conditional means of price levels and (2.8) to generate conditional variances. The estimator of µ𝑡 conditional on exogenous variables (𝑥𝑡 ) and policy variable (𝑧𝑡 ) can be expressed as: ̂𝑡 ) |(𝑥𝑡 , 𝑧𝑡 ) µ̂𝑡 | (𝑥𝑡 , 𝑧𝑡 ) = 𝑙𝑜𝑔(𝑝 (2.9) Because the error variance, 2 , is homoscedastic, its estimator can be obtained directly from the VAR as residuals squared ̂2 | (𝑥𝑡 , 𝑧𝑡 ) =  ̂2 = 𝑢̂2  (2.10) where 𝑢̂ is residual from the equation of Thai rice in the VAR model. 57 2.3.3 Simulating random yields Rice yield (Yt) is assumed to be conditionally lognormally distributed with constant variance16: 𝑌𝑡 |(𝑡, 𝑌𝑡−1 )~ 𝐿𝑜𝑔𝑛𝑜𝑟𝑚𝑎𝑙 (𝑡 , 2𝑦 ) (2.11) A representative farmer is assumed to grow only one type of rice, white rice. White rice was chosen over other major rice types, including jasmine rice and glutinous rice, because it has the largest share in term of annual production. The simulation of rice yield is based on a series of annual average white rice yield in the Central Plain region, a major hub of white rice production, from 1981-2012. After testing for first-order autocorrelation and a unit root, one lag of yield and a time trend are included in the yield equation.17,18 𝑙𝑜𝑔𝑌𝑡 = 𝛼 + 𝛽𝑡 + 𝛿𝑙𝑜𝑔𝑌𝑡−1 + 𝑣𝑡 (2.12) where 𝑣𝑡 are assumed to be normally-distributed random errors. 2.3.4 Generating simulated profits under the five selected scenarios The simulated yields and counterfactual market prices are used to generate a stochastic profits stream under the five scenarios of interest. The profit function (NR) under each scenario can be described as follows. 16 White test statistic for heteroskedastic variance (F-test) including fitted value in (2.12) as independent variable is 0.33 and p-value is 0.57. Thus, the null hypothesis of rice yield has a homoscedastic variance cannot be rejected at 5% level 17 Augmented Dickey-Fuller test statistic for a unit root including a time trend and one yield lag is -3.768and MacKinnon approximate p-value is 0.0183. Thus, the null hypothesis of rice yield has a unit root is rejected at the 5% level 18 Durbin's alternative test statistic for first-order autocorrelation of the error vt in (2.12) is 0.155 and p-value is 0.6939. Thus, we cannot reject the null of no first-order autocorrelation 58 Scenario 1: No government intervention (NG) 𝑚 𝑁𝑅𝑃𝑆𝑃 = (𝐴𝑌)𝑃𝑁𝐺 − 𝐴𝐶 Because there is no financial support from the Government, profit only depends on 𝑚 market price (𝑃𝑁𝐺 ) and total production (AY) where A and Y denote land size and yield, respectively. Given a constant production cost per area (C), total production cost (AC) increases in proportion with lands size. Scenario 2: Participating in PSP (PSP) 𝑚 𝑁𝑅 = (𝐴𝑌)𝑃 𝑆 − 𝐴𝐶 − ((1 + 𝑟)𝑡 − 1))(𝐴𝑌)𝑃 𝑆 𝑖𝑓 𝑃 𝑆 > 𝑃𝑃𝑆𝑃 𝑚 = (𝐴𝑌)𝑃𝑃𝑆𝑃 − 𝐴𝐶 𝑚 𝑖𝑓 𝑃 𝑆 < 𝑃𝑃𝑆𝑃 The Government offers to buy unlimited amount of rice from farmer at the support price (PS). By participating in the program, the farmer’s profit is equal to a product of total production 𝑚 (AY) and either the support price (PS) or the market price (𝑃𝑃𝑆𝑃 ), whichever is higher, less the sum of production cost and participation cost (the last two terms on the right-hand side of the 𝑚 above profit equation). Note that simulated values of the counterfactual market price 𝑃𝑃𝑆𝑃 are 𝑚 obtained from the VAR model. When PS>𝑃𝑃𝑆𝑃 , the farmer will choose to sell to the Government at the support price and incur the additional cost of program participation expressed by the last term on the right-hand side of the above equation. This term represents an opportunity cost of the delayed loans assumed equal to forgone interests that could have been earned elsewhere if the loans were paid on time. Here, the interest rate (r) is equal to the average of maximum lending rates by major commercial banks. The length of time during which the interest is compounded is equal to number of days before farmers receive the sale proceeds from the government (t), which 59 is randomly drawn (without replacement) from survey data using bootstrapping techniques. The household survey data contains 187 participants of PSP, 98 of which have reported delays of loans which ranges from 7 to 95 days. Scenario 3: Declining PSP when available (NPSP) 𝑚 𝑁𝑅 = (𝐴𝑌)𝑃𝑃𝑆𝑃 − 𝐴𝐶 This scenario reflects what is commonly observed in the Thai rice market that some farmers choose not to participate in PSP and instead sell their harvest on the open market. In this 𝑚 case, the profit depends on market price (𝑃𝑃𝑆𝑃 ) and total production (AY). Scenario 4: Participating in DPP (DPP) 𝑀 𝑀 𝑁𝑅 = (𝐴𝑌)𝑃𝐷𝑃𝑃 + 𝐴𝑌 𝑇 (𝑀𝑎𝑥{𝑃𝑇 − 𝑃𝐷𝑃𝑃 , 0}) − 𝐴𝐶 Deficiency payments are calculated based on the difference between the target price (PT) 𝑚 and the estimated market prices (𝑃𝐷𝑃𝑃 ), times the estimated yields (YT). The farmer sells rice to 𝑚 buyers in the market and receives the total revenue of (AY)𝑃𝐷𝑃𝑃 . The Government only makes deficiency payments to the farmer when the market price falls below the target price. Again, the 𝑚 distribution of counter factual market prices 𝑃𝐷𝑃𝑃 is obtained from the VAR model. Scenario 5: Declining DPP when available (NDPP) 𝑚 𝑁𝑅 = (𝐴𝑌)𝑃𝐷𝑃𝑃 − 𝐴𝐶 If the farmer fails to participate in the program, he receives no deficiency payment and 𝑚 receives only the sale proceeds from selling in the open market at the market price (𝑃𝐷𝑃𝑃 ). 60 2.3.5 Ranking the representative farmer preferences using SERF The CRRA coefficients (RRAC) are assumed to range from 0.5 to 4 as suggested by Anderson and Dillon (1992); 0.5 is hardly risk averse, 1.0 is normal or somewhat risk averse, 2.0-is rather risk averse, 3.0 is very risk averse, and 4.0 is extremely risk averse. The initial wealth (W) of the representative farmer is assumed to equal twice the average of simulated total revenue obtained under no intervention. The power utility expressed as a function of the stochastic profit stream X under policy alternative j evaluated at the ith iteration can be written as 𝑈𝑗 (𝑋𝑖𝑗 ) = (𝑋𝑖𝑗 + 𝑊)(1−𝑅𝑅𝐴𝐶) Given the probability (p) of observing each stochastic profit, the expected utility is 𝐸(𝑈𝑗 ) = ∑ 𝑝𝑖𝑗 (𝑋𝑖𝑗 + 𝑊)(1−𝑅𝑅𝐴𝐶) 𝑖 The formula for calculating certainty equivalence (CE) under power utility function is 1 𝐶𝐸 = 𝐸(𝑈𝑗 )(1−𝑅𝑅𝐴𝐶) − 𝑊 SIMETAR converts the expected utility under each policy alternative into CE values for a set of selected values of RRAC which are bounded between 0.5 and 4. Using this approach we end up with a vector of CE values for each of the n alternatives calculated for several values of RRAC within the bound. Only those alternatives which have the highest CE values for some value in the range of RRAC are considered efficient. All other alternatives are dominated in the SERF sense. The method also provides a cardinal ranking at each level of risk aversion by interpreting differences in CE values as risk premiums. 61 2.3.6 Data The rice prices in the VAR model include the farm-gate price of Thai white rice, the F.O.B. price of 5% broken white rice from Vietnam, and the F.O.B. price of 25% broken white rice from Pakistan. The prices observed in November and April from 1997 to 2012 are used as representative of prices observed during the first- and second-cropping season, because most rice production in each season are sold in these respective months. PSP was first implemented at a national level in the second season of the 2001/02 production year and lasted until the second season of the 2008/09 production year. DPP was first implemented in the first season of the 2009/10 production year and lasted until the second season of the 2010/11 production year after which the program was replaced by the PSP. This gives four observations of the period in which DPP was implemented and 15 observations in the case of PSP. The rest of the observations represent market data under no intervention (NG). A series of farm-gate price of Thai rice is obtained from Thailand’s Office of Agricultural Economics (OAE).A series of the export prices (F.O.B.) are obtained from United States Department of Agriculture (USDA) and the Food and Agriculture Organization (FAO). A series of support prices under PSP from 2001 to 2012 and the target price under DPP from 2009 to 2011 are obtained from Thailand Department of Internal Trade (DIT). Production costs including both fixed and variable costs are obtained from OAE. The length of delayed payments, defined as the number of days until farmers received the loan payments after pledging their rice with the Government, is obtained from the Author’s 2012/13 rice farm survey in Buriram province. The sample contains 387 farm households of which 130 households participated in PSP during the 2012/13 first season. The cost of delayed payments is calculated by averaging the maximum 62 lending rates by major commercial banks. The rice yields obtained from OAE are the average yields observed in the Central Plain region from 1981 to 2012. 63 2.4 Results The preference rankings in 2006/07 and 2012/13 are reported separately. This cross comparison examines whether the ranking would change as the difference between the support and market prices increases from its historical minimum to maximum. This section is organized into five parts. The first part presents the estimation of conditional mean and variance of market price under three policy regimes (i.e. PSP, DPP, and NG) by the VAR model. The second part presents the estimation of conditional mean and variance of rice yield by the AR (1) model. Upon obtaining the estimates of conditional market prices and yields, SIMETAR uses these means and variances to simulate stochastic farm profits which are then converted to CE values for preference ranking by SERF. The third and fourth parts discuss the preference rankings by SERF in 2006/07 and 2012/13, respectively. The last part provides a sensitivity analysis of the preference ranking with respect to yield variability and levels of initial wealth. 2.4.1 Simulation of market prices The VAR model consists of three rice-price equations including farm-gate price of Thai rice and export prices of rice from Vietnam and Pakistan. For the purpose of this study, we are only interested in the parameter estimates from the equation of the Thai rice price in log form, (log (Thai farm-gate price)). Hence, only the regression results from this equation are discussed (see Table 2.1).19 All explanatory variables are jointly significant at the 1% significant level. All prices lagged by two seasons are statistically significant at 5%. This means the response of current price to changes of past prices is more pronounced within the same cropping season. 19 The results from the other two equations are reported in Table A2.1a-A2.1b in appendix A. 64 Despite their statistical insignificance, Season and Coup2006 are included in the model as their slope coefficients display signs that are consistent with our prior expectation. Dropping these variables causes the slope coefficients of other variables to display signs that are counterintuitive. The negative sign on the dummy variable Season indicates that the first crop receives lower market price than the second crop does, which is reasonable as we would expect lower market price during the first season due to higher production. The slope coefficient on the dummyCoup2006 is positive which indicates that only weak evidence was found to support that the continuation of PSP by the military-elected Government had boosted the market price. The dummy variable Crisis2008 has a positive slope coefficient and is statistically significant at 1%.As we would expect, the crisis drove the farm-gate price and export prices to unprecedented levels. Furthermore, the crisis also caused some spillover effect which persistently continues into subsequent years as indicated by a positive sign on the slope coefficient of the dummy variable Post-Crisis. Table 2.1: Regression result from the equation of Thai rice price in the VAR model Dep. Variable: log(Thai farm-gate price)t log(Thai farm-gate price)t-1 log(Thai farm-gate price)t-2 log(Vietnam export price)t-1 log(Vietnam export price)t-2 log(Pakistan export price)t-1 log(Pakistan export price)t-2 PSP dummy DPP dummy Season dummy Coup2006 dummy Crisis2008 dummy Post-Crisis dummy Constant Coeff. -0.116 0.378 -0.426 -0.603 0.614 0.663 0.143 -0.020 -0.024 0.090 0.534 0.301 5.069 65 Std. Err. 0.207 0.224 0.265 0.280 0.254 0.312 0.049 0.096 0.043 0.076 0.112 0.113 1.164 z -0.560 1.690 -1.610 -2.160 2.410 2.120 2.930 -0.210 0.550 1.190 4.750 2.670 4.360 P>z 0.574 0.041 0.108 0.031 0.016 0.034 0.003 0.832 0.585 0.233 0.000 0.008 0.000 Despite an increase in rice supply following the implementation of PSP and DPP, both programs have different effects on market price as evidenced by the differences in the direction of the price change and its magnitude. The dummy variable PSP is statistically significant at 1% which implies that PSP increases market price by 14.3% on average. In contrast, the dummy variable DPP has negative coefficient. This result suggests that market price decreases by 2% on average throughout the course of DPP implementation. However, the impact of DPP is not statistically significant which is probably due to insufficient number of observations as the program was implemented for only two years (four cropping seasons). Of course, these estimates may be fragile given the small number of total observations, especially those under DPP. However, it is interesting to note that these results are quite consistent with the price effects of DPP and PSP estimated by the partial equilibrium model reported in Chapter 1. For the partial equilibrium results the market price is estimated to increase by between 10.40 and 20.58% on average under PSP in the 2005/06 season while it is estimated to fall by between 0.62 and 2.75% on average under DPP. Because the estimates of policy impact on market price estimated by the VAR model fall within these ranges estimated by the partial equilibrium model, we argue that they are quite reasonable and use of these values to compute the market price distribution is appropriate. The slope coefficients of PSP and DPP have opposite sign as expected. As all rice production must be sold to private buyers in the open market under DPP, the increase in total rice supply tends to cause the market price to fall. In contrast, the competition between the Government and other buyers in the market drives up the market price under PSP.20 The 20 It is important to note that the market price can fall below the support price under PSP because some farmers decline to sell to the Government, even when the support price is higher than the market price, because the decision to participate in the program is governed not only by size of the support price relative to the market prices but also 66 magnitude of price change is much larger under PSP because the Government purchases large amounts of stock that are not released back onto the domestic market. The amount of rice production withdrawn from the market by the Government purchase under PSP offsets the increase in total rice supply induced by the program; the average quantity of pledged rice in the first-season from 2005/06 to 2008/09 is 4.24 million tons while total supply only increased by 1.14 million tons per year on average. The fitted values of the conditional log price and its standard errors are used to simulate the counterfactual market prices under PSP, DPP, and NG scenarios by assuming the log price is normally distributed. Summary statistics of the simulated prices in 2006/07 and 2012/13 are presented in Table 2.2. Recall that only the PSP was implemented in these selected years. The support prices in 2006/07 and 2012/13 were 15,000 and 6,500 baht/ton as shown in column 4.21 Mean and standard deviation of the simulated counterfactual prices are reported in column 5 and 6.Given the support price of 15,000 baht/ton, the corresponding market price is estimated at 10,112 baht/ton. Replacing PSP by DPP would cause the market price to fall to 8,598 baht/ton while removing all intervention programs would change market price to 8,775 baht/ton. When the support/target price is set at 6,500 baht/ton the estimated market price under PSP is approximately 6,281 baht/ton and the counterfactual market prices under DPP and NG are 5,445 and 5,336 baht/ton, respectively. Notice that the probability that the market price exceeds the support price under PSP is very small in 2012/13 because the price differential is as high as 4,888 while standard deviation is only 965. In contrast, the probability of the market price exceeding the support price is much by individual farm characteristics such as land size and financial position that may influence the costs of participation in the PSP. 21 The averages of actual market price corresponding to these support prices were 9,753 and 6,380 baht/ton, respectively (not shown in the table). 67 higher in 2006/07 as the price differential is only 219 while the standard deviation is 599.Thus, in 2012/13 the support/target price will almost always trigger and the representative farmer would base his or her decision on the high probability of receiving these prices when choosing between PSP and DPP. Table 2.2: Mean and standard deviation of counterfactual farm prices Type Year Regime Support/Target Price White White White White White White 2012/13 2012/13 2012/13 2006/07 2006/07 2006/07 PSP NG DPP PSP NG DPP 15,000 NA 15,000 6,500 NA 6,500 Market Price Mean Std. 10,112 965 8,775 837 8,598 820 6,281 599 5,445 519 5,336 508 2.4.2 Simulation of rice yields The result from the AR (1) of log yields is reported in Table 2.3. The adjusted R-squared is 0.75. All explanatory variables are jointly significant at 1%. Random yields are simulated from the fitted values of log yield in year 2006 by assuming that the log yield is normally distributed. Upon obtaining these fitted values rice yields are converted from log to level. Summary statistics of the simulated conditional yields in 2006/07 and 2012/13 are presented in Table 2.4. Table 2.3: OLS regression of Thai rice yield log(Yield) Constant Year L1.log(Yield) Coefficient Std. t P>t 4.849 0.019 0.183 1.1 0.005 0.186 4.41 3.79 0.99 0 0.17 0.333 68 Table 2.4: Summary statistics of simulated rice yields Mean Std. Min Max Yield 2006/07 536.50 61.12 377.61 739.75 Yield 2012/13 621.69 71.40 410.25 946.79 2.4.3 The preference ranking by SERF in the 2012/2013 season Table 2.5 shows summary statistics of the simulated profits plus initial wealth under the five scenarios of interest in the 2012/13 production season.22 The gap between the support/target price and the market price is largest in this period. The simulation of profit assumes that the support/target price is equal to 15,000 baht/ton, the actual observed value in this season. The result shows that PSP and DPP have raised mean farm profit by significant amount. The standard deviation of profit under PSP is much large than that under DPP, so PSP is found to be more risky than DPP. In fact, DPP yields the lowest standard deviation of profit in all cases. This is because the program guarantees to pay farmers a fixed payment for each unit of land registered regardless of actual outputs produced as long as the market price falls below the target price. Table2.5: Summary statistics of simulated profit plus initial wealth in 2012/13 Mean Std. Min Max NG 144,197 20,790 88,797 207,291 PSP 237,193 26,929 162,656 331,486 DPP 240,527 15,726 197,239 304,370 22 NPSP 164,108 23,886 104,432 262,690 NDPP 140,795 20,276 91,780 231,070 Because profit streams need to be greater than zero in order to calculate certainty equivalent values under power utility function, initial wealth calculated as mean of revenues under no intervention was added to each value of the simulated profit. Initial wealth in the 2012/13 season was calculated at 135,000 baht. Nevertheless, the term “profit” will be used to refer to “profit plus initial wealth” throughout the remaining sections of this study for convenience purposes. 69 Figure 2.1: Profit CDFs of the representative rice farmers in 2012/13 1 Prob 0.8 0.6 0.4 0.2 0 75000 NG 125000 175000 PSP 225000 DPP 275000 NPSP 325000 NDPP Figure 2.1 shows the CDF plots of farm profits under the five scenarios. According to first-degree stochastic dominance (FSD), the profit streams under PSP and DPP dominate those of NPSP, NDPP, and NG because CDFs of the former scenarios are always located to the left of the latter scenarios. This means the farmer is always better off when joining either PSP or DPP. However, the shape of these CDFs suggests that PSP is more risky than DPP; the probability of observing profits falling in the upper- and lower-tail of distribution is higher under PSP. In addition, one may ask whether the farmer is still better off under the implementation of PSP compared to the no intervention scenario if he is unable to participate in the program. This situation is particularly important because significant portions of rice farm households in Thailand do not participate in PSP. Although those who fail to participate in PSP are not guaranteed a sale price as high as the support price, they still benefit from an increase in the market price brought about by the program. The tradeoff, of course, is the higher dispersion of the market price and profit. The CDF graphs clearly show that the NPSP scenario dominates NG and NDPP by FSD. This means the gain from an increase in the mean market price under PSP is 70 so large that it outweighs the cost associated with the higher variance of profit. Unfortunately, the preference ranking between PSP and DPP cannot be determined by FSD as their CDFs cross. Thus, the ranking must be justified based on the CE values of profit stream and levels of riskaversion which is done by SERF. Figure 2.2: The SERF ranking of the five scenarios in 2012/13 based on CE values Certainty Equivalent 260,000.00 DPP 240,000.00 PSP 220,000.00 200,000.00 180,000.00 NPSP 160,000.00 NG 140,000.00 NDPP 120,000.00 0 1 2 3 4 5 RRAC NG PSP DPP NPSP NDPP Figure 2.2 illustrates the preference ranking of the representative farmer towards the five scenarios of interest by SERF based on the CE values associated with profit streams calculated over a range of constant relative risk aversion (CRRA) from 0.5 to 4. These CE values jointly establish a CE line for each scenario. Notice that all CE lines have a downward slope which indicates that the sure sum that would make the farmer as well off as accepting a stream of stochastic profit gets smaller as the farmer becomes more risk-averse. Because the CE line under DPP always lies above that of PSP, the representative farmer would always prefer DPP to PSP. 71 The ranking by SERF of the other scenarios is consistent with that by FSD. That is, NPSP is preferred to NG and NDPP. By the same token, NG is preferred to NDPP. To supplement the ordinal ranking provided by SERF, the CE values computed at the five selected values of CRRA can be used to provide a cardinal ranking of these scenarios, as shown in Table 2.6. For a given value of CRRA, the difference in CE values between any pair of scenarios indicates the risk premium, or minimum amount of a sure sum that makes the farmer indifferent between alternatives. For example, a normal risk-averse farmer defined as having CRRA of 1 would prefer DPP to PSP and the risk premium is 4,344 baht. This means the farmer is willing to pay up to 4,344 baht in exchange for the profit stream under DPP instead of PSP. The risk premium increases as the CRRA coefficient increases from 1 to 4, in which case the risk premium is 7,272. Because the sizes of the premiums are not trivial, we conclude that the farmer clearly prefers DPP to PSP. Table 2.6: The CE values at five given values of CRRA coefficient in 2012/13 RRAC 0.5 1 2 3 4 NG 143,458 142,708 141,223 139,759 138,419 PSP 236,438 235,671 234,151 232,647 231,265 DPP 240,273 240,015 239,505 239,000 238,537 NPSP 163,254 162,390 160,682 159,003 157,470 NDPP 140,078 139,352 137,915 136,500 135,204 2.4.4 The preference ranking by SERF in the 2006/2007 season Table 2.7 shows summary statistics of simulated profits plus initial wealth under the five scenarios of interest in 2006/07.23 The support/target price is set at 6,500 baht/ton which is the actual observed value in this season. The gap between the support/target price and the market 23 The initial wealth in the 2006/07 was calculated at 75,000 baht 72 price is much smaller compared to the case of 2012/13. As a result, there is a much lower probability of realizing a market price that triggers Government purchases or deficiency payments. Because the mean and variance of the market price are higher under PSP, the farmer benefits in two ways when participating in the program. First, the farmer is guaranteed a minimum price equal to the support price. Second, his chances of receiving a market price that exceeds the support price is also higher. In other words, the program not only insures the farmer against low market prices but also increase the probability of receiving a higher market price. The latter benefit is smaller under DPP because mean and variance of the market price are lowest under this program. Table 2.7: Summary statistics of simulated profit plus initial wealth in 2006/07 Mean Std. Min Max NG 77,390 11,442 48,478 131,318 PSP 92,754 11,146 65,831 139,164 DPP 91,678 8,460 69,118 130,141 NPSP 88,402 12,402 56,195 139,164 NDPP 75,826 10,599 49,715 120,016 Figure 2.3 shows the CDF plots of profit under the five scenarios in 2006/07. Like in the case of 2012/13, the profit streams obtained under NDPP and NG are FSD dominated by the other scenarios. This means the farmer is better off when the Government intervenes in the market, either in the form of PSP or DPP. Again, the no intervention scenario is less preferred to PSP, even when the farmer does not participate in the program. The shape of these CDFs suggests that PSP and NPSP are more risky than DPP; the probability of observing profits falling in the upper- and lower-tail of distribution is higher under PSP and NPSP. Unfortunately, the preference rankings of the pairs PSP&DPP and NPSP&DPP cannot be determined by FSD as 73 their CDFs cross. Thus, the ranking must be justified based on the CE values of profit stream and levels of risk-aversion which is done by SERF. Figure 2.3: Profit CDFs of the representative rice farmers in 2006/07 1 Prob 0.8 0.6 0.4 0.2 0 40000 60000 NG 80000 PSP 100000 DPP 120000 NPSP 140000 NDPP The preference ranking by SERF in 2006/07 is shown in Figure 2.4. The result clearly indicates that DPP is preferred to PSP and NPSP as the CE lines of the former case are always above the latter cases. The fact that the CE lines under PSP and DPP are close to one another implies that the risk premium is small and hence the farmer prefers one to the other only by a small margin. Figure 2.4: The SERF ranking of the five scenarios in 2006/07 based on CE values Certainty Equivalent 95,000.00 PSP DPP 90,000.00 NPSP 85,000.00 80,000.00 NG 75,000.00 NDPP 70,000.00 0 1 2 3 4 RRAC NG PSP DPP NPSP 74 NDPP 5 The CE values computed at the five specific values of CRRA are reported in Table 2.8. Notice that the risk premium of the pair PSP&DPP for a normally risk-averse individual (CRRA coefficient of 1) is only 799 baht. This means a DPP participant requires only 799 baht of a sure sum in order to make him as well off as participating in PSP. Unlike in the case of 2012/13, the risk premium decreases as the CRRA coefficient increases from 1 to 4, in which case the risk premium is only 10 baht. This implies the gain from an increase in mean of the market price under PSP slightly outweighs its cost associated with delayed payments for a normally riskaverse individual while they equally offset each other for an extremely risk-averse individual. Because the sizes of the premiums are very small, we conclude that the farmer would be almost indifferent between the choice of DPP and PSP when the support/target price is close to the market price. Table 2.8: The CE values at five given values of CRRA coefficient in 2006/07 RRAC 0.5 1 2 3 4 NG 76,974 76,553 75,721 74,903 74,155 PSP DPP 92,423 91,484 92,087 91,288 91,421 90,896 90,763 90,508 90,160 90,150 NPSP 87,974 87,541 86,686 85,846 85,083 NDPP 75,459 75,086 74,346 73,613 72,942 2.4.5 Sensitivity analysis of the preference ranking The results from the preference ranking by SERF discussed above were obtained under some specific assumptions on initial wealth, the yield distribution, the price distribution, and rate of interest charged on delayed loan payments. The order of the ranking, however, could change if these parameters take on different values. Thus, a sensitivity analysis was conducted by varying these parameters within a specific range. Specifically, the standard deviation of rice yield, the standard deviation of market price, level of initial wealth, and level of maximum lending rate 75 were scaled up and down by a factor of two. The results from the sensitivity analysis in 20012/13 and 2006/07 are reported in Table 2.9 and Table 2.10, respectively. In 2012/13, the ranking is not sensitive to changes in these parameters. So, DPP is still always preferred by a risk-averse farmer, whose CRRA coefficients are bounded between 0.5 and 4, independent of initial wealth, degree of yield and price variability, and lending rate. In 2006/07, the ranking between PSP and DPP is sensitive to changes in all of the parameters while the rankings among NPSP, NG, and NDPP are not. Recall that PSP is preferred to DPP in the base case. However, the ranking switches and at least the extremely risk-averse farmer would switch his/her preference from PSP to DPP when initial wealth or price volatility is halved. This means the farmer is less willing to take on a more-risky asset (i.e. PSP) as wealth decreases. Similarly, the reduction in price volatility lowers the probability of receiving a high market price which decreases the benefit of PSP. In addition, the extremely risk-averse farmer would prefer DPP to PSP if yield variability or lending rate is doubled. This means PSP becomes too risky as yield variability and lending rate increase. Because the risk premiums associated with the pair DPP&PSP are quite small in all cases in 2006/07, we conclude that the farmer is largely in different between these programs. 76 Table 2.9: Sensitivity analysis of the SERF ranking in 2012/13 Parameter Ranking Initial wealth 1st 2nd 3rd 4th 5th Yield Std. 1st 2nd 3rd 4th 5th Price Std. 1st 2nd 3rd 4th 5th Lending rate 1st 2nd 3rd 4th 5th Scaled down by 1/2 LRAC URAC DPP DPP PSP PSP NPSP NPSP NG NG NDPP NDPP DPP DPP PSP PSP NPSP NPSP NG NG NDPP NDPP DPP DPP PSP PSP NPSP NPSP NG NG NDPP NDPP DPP DPP PSP PSP NPSP NPSP NG NG NDPP NDPP 77 Base LRAC URAC DPP DPP PSP PSP NPSP NPSP NG NG NDPP NDPP DPP DPP PSP PSP NPSP NPSP NG NG NDPP NDPP DPP DPP PSP PSP NPSP NPSP NG NG NDPP NDPP DPP DPP PSP PSP NPSP NPSP NG NG NDPP NDPP Scaled up by 2 LRAC URAC DPP DPP PSP PSP NPSP NPSP NG NG NDPP NDPP DPP DPP PSP PSP NPSP NPSP NG NG NDPP NDPP DPP DPP PSP PSP NPSP NPSP NG NG NDPP NDPP DPP DPP PSP PSP NPSP NPSP NG NG NDPP NDPP Table 2.10: Sensitivity analysis of the SERF ranking in 2006/07 Parameter Ranking Initial wealth 1st 2nd 3rd 4th 5th 1st 2nd 3rd 4th 5th 1st 2nd 3rd 4th 5th 1st 2nd 3rd 4th 5th Yield Std. Price Std. Lending rate Scaled down by 1/2 LRAC URAC PSP DPP DPP PSP NPSP NPSP NG NG NDPP NDPP PSP PSP DPP DPP NPSP NPSP NG NG NDPP NDPP DPP DPP PSP PSP NPSP NPSP NG NG NDPP NDPP PSP PSP DPP DPP NPSP NPSP NG NG NDPP NDPP 78 Base LRAC URAC PSP PSP DPP DPP NPSP NPSP NG NG NDPP NDPP PSP PSP DPP DPP NPSP NPSP NG NG NDPP NDPP PSP PSP DPP DPP NPSP NPSP NG NG NDPP NDPP PSP PSP DPP DPP NPSP NPSP NG NG NDPP NDPP Scaled up by 2 LRAC URAC PSP PSP DPP DPP NPSP NPSP NG NG NDPP NDPP PSP DPP DPP PSP NPSP NPSP NG NG NDPP NDPP PSP PSP DPP DPP NPSP NPSP NG NG NDPP NDPP PSP DPP DPP PSP NPSP NPSP NG NG NDPP NDPP 2.5 Conclusion and discussion The objective of this study is to compare the preference ranking of a representative rice farmer in Thailand towards two important government programs—the price support program (PSP) and the deficiency payment program (DPP). Under PSP the government buys rice from farmers at the support price by issuing loans based on actual production. In contrast, DPP requires farmers to sell rice on the open market. Only when the market price falls below the target price does the Government compensate farmers with deficiency payments, which depend on estimated production. It might initially seem that PSP and DPP should yield equal benefits to farmers when the support price and the target price are equal, which would lead to the conclusion that the farmers should be indifferent between these programs. However, the benefits realized under each program are distinct and can differ significantly for at least three reasons. First, the benefits of PSP are discounted due to the additional cost associated with the delayed payment of program loans. Second, the fact that deficiency payments are made based on a fixed yield implies that farmers are compensated more or less than they would be paid if the payments were based on actual yields. Third, distribution of profits could significantly differ if the impact of PSP and DPP on probability of the market price exceeding the support/target price differ. Stochastic efficiency with respect to a function (SERF) is employed to rank farmer preferences by comparing the certainty equivalent (CE) values associated with the stochastic profit streams under each policy scenario. The representative farmer is assumed to grow rice on a 25-rai rice land and face stochastic profit because of random yield and market price uncertainty. Random yield is simulated from the estimated mean and variance of yield obtained from a firstorder autoregression model with trend. The market prices under each policy regime are simulated from the conditional mean and variance of the farm-gate price of Thai white rice estimated from 79 the vector autoregressive (VAR) model which captures the relationship between rice prices and the effect of changes in the policy regime on the price distributions. The VAR model consists of three equations of rice prices including the farm-gate price of Thai white rice and the export prices of white rice from Vietnam and Pakistan. After obtaining the simulated values of yield and market prices the simulated profit under the following five scenarios are generated: PSP participation & non-participation, DPP participation & non-participation, and no intervention. Lastly, SERF is employed to rank the farmer preferences in two selected seasons: the firstcropping seasons of 2006/07 and 2012/13. These seasons represent a period in which the difference between the support/target and market price is considerably small and large, respectively. The results from the VAR model show that in 2012/13 PSP has increased the average farm-gate price of Thai rice by 14.3% while DPP would have depressed average price by 2%. The increase in the market price under PSP, however, comes at the cost of higher price variability. Price variance is largest under PSP and smallest under DPP. In both selected periods the profit streams under PSP and DPP dominate profits under NG by FSD. In addition, the farmer is better off under the PSP compared to the no intervention scenario, even if the choice is to not participate. This result is particularly important because more than half of rice farm households in the country do not participate in the PSP due to the additional costs associated with delayed payments. Although those who fail to participate in PSP are not guaranteed a sale price as high as the support price, they still benefit from an increase in the mean of market price brought by the program. The tradeoff, of course, is the higher variance of the market price and profit. The CDF graphs clearly indicate that the farmer is indeed better off under PSP even when failing to receive the support price as a nonparticipant of the program. This means the gain from 80 an increase in the mean of market price under PSP is large enough that it outweighs the cost associated with the higher degree of profit variation. The preference ranking between PSP and DPP cannot be determined by FSD because their CDFs cross. The preference ranking by SERF shows that all risk averse farmers would prefer DPP to PSP in 2012/13. This ranking is invariant to changes in initial wealth, yield variability, price volatility, and the lending rate viewed an opportunity cost of delayed payments under PSP. In 2006/07 the preference ranking switches and the farmer now prefers PSP to DPP by small margins. In addition, the results are sensitive to changes in all of the parameters. When the standard deviation of price or initial wealth is halved, an extremely risk-averse farmer would prefer DPP to PSP. Similarly, the farmer is willing to take less risks, and hence chooses DPP over PSP, as the lending rate or the standard deviation of yield is doubled. However, the difference in these rankings measured in term of risk premium is very small. Thus, one could argue that the farmer is largely indifferent between PSP and DPP in this case. Overall, DPP is ranked either higher than PSP by large premiums or lower by small premiums which allows us to conclude that the farmer would prefer DPP to PSP in most situations if both are simultaneously offered by the Government at identical support and target prices. At least two policy implications can be drawn from this study. First, given the criticisms that PSP generates enormous government deficits, has been marred by alleged corruption, and provides an unequal distribution of program benefits, switching from PSP to DPP could alleviate these problems and still be preferred by most farmers in most situations. Second, because sensitivity analysis indicates that the preference ranking is sensitive to changes in yield variability, crop insurance programs could have a significant influence on farmer preferences for alternative farm programs. 81 APPENDIX 82 Table A2.1a: Regression result from the equation of Vietnam export price in the VAR model Dep. Variable: log(Vietnam export price)t log(Thai farm-gate price)t-1 log(Thai farm-gate price)t-2 log(Vietnam export price)t-1 log(Vietnam export price)t-2 log(Pakistan export price)t-1 log(Pakistan export price)t-2 PSP dummy DPP dummy Season dummy Coup2006 dummy Crisis2008 dummy Post-Crisis dummy Constant Coefficient -0.506 0.420 -0.788 -0.667 1.146 0.794 0.093 0.053 0.012 0.399 0.842 0.502 3.707 Std. Err. 0.374 0.406 0.481 0.507 0.461 0.566 0.088 0.174 0.079 0.137 0.203 0.205 2.109 z -1.350 1.040 -1.640 -1.320 2.490 1.400 1.060 0.310 0.150 2.900 4.140 2.450 1.760 P>z 0.177 0.300 0.101 0.188 0.013 0.160 0.291 0.759 0.880 0.004 0.000 0.014 0.079 Table A2.1b: Regression result from the equation of Pakistan export price in the VAR model Dep. Variable: log(Pakistan export price)t log(Thai farm-gate price)t-1 log(Thai farm-gate price)t-2 log(Vietnam export price)t-1 log(Vietnam export price)t-2 log(Pakistan export price)t-1 log(Pakistan export price)t-2 PSP dummy DPP dummy Season dummy Coup2006 dummy Crisis2008 dummy Post-Crisis dummy Constant Coefficient -0.243 0.067 -1.166 -0.160 1.705 0.262 0.102 0.137 -0.148 0.399 0.542 0.365 3.579 83 Std. Err. 0.330 0.358 0.424 0.446 0.406 0.498 0.078 0.153 0.069 0.121 0.179 0.180 1.856 z -0.740 0.190 -2.750 -0.360 4.200 0.530 1.310 0.900 -2.140 3.300 3.020 2.020 1.930 P>z 0.462 0.852 0.006 0.719 0.000 0.599 0.191 0.371 0.032 0.001 0.002 0.043 0.054 REFERENCES 84 REFERENCES Anderson, J. R. and J.L. Dillon (1992). Risk analysis in dryland farming systems. Food and Agriculture Organization Management Series no. 2, FAO, Rome. Department of Internal Trade, Thailand Ministry of Commerce. Available at http://www.dit.go.th/contentmain.aspx?typeid=8&catid=155 [accessed January 28, 2014]. Economic Research Service, United States Department of Agriculture. Available at http://usda.mannlib.cornell.edu/MannUsda/viewDocumentInfo.do?documentID=1229 [acessesed December 5, 2013]. GIEWS Food Price Data and Analysis Tool, Food and Agriculture Organization. Available athttp://www.fao.org/giews/pricetool/ [acessesed December 5, 2013]. Hadar, J. and W.R. Russell. 1969. Rules for Ordering Uncertain Prospects. The American Economic Review 59(1): 25-34. Hanoch, G. and H. Levy. 1969. The Efficiency Analysis of Choices Involving Risk. The Review of Economic Studies 36(3): 335-346. Hardaker, J. B.; J.W. Richardson; G. Lien and K.D. Schumann. 2004. Stochastic Efficiency Analysis with Risk Aversion Bounds: a Simplified Approach. Australian Journal of Agricultural and Resource Economics 48(2): 253-270. Hardaker, J. and G. Lien. 2010. Stochastic Efficiency Analysis with Risk Aversion Bounds: A comment. Australian Journal of Agricultural and Resource Economics 54(3): 379-383. Meyer, J. 1977a. Choice Among Distributions. Journal of Economic Theory 14(2): 326-336. Meyer, J. 1977b. Second Degree Stochastic Dominance with Respect to a Function. International Economic Review 18(2): 477-487. Meyer, J.; J.W. Richardson and K.D. Schumann. 2009. Stochastic Efficiency Analysis with Risk Aversion Bounds: A Correction. Australian Journal of Agricultural and Resource Economics 53: 521-525. Office of Agricultural Economics, Thailand Ministry of Agriculture and Cooperatives. Available at from http://www.oae.go.th [accessed September 28, 2013]. Robison, L. J. and R. Myers. 2001. Ordering Risky Choices, in Just, R.E. and R.D. Pope (Eds.). A Comprehensive Assessment of the Role of Risk in US Agriculture. Boston: Kluwer 85 CHAPTER 3: TECHNICAL EFFICIENCY OF THAI JASMINE RICE FARMERS: COMPARING SUPPORT PRICE PROGRAM PARTICIPANTS AND NONPARTICIPANTS 3.1 Introduction In Thailand, the rice price support program (PSP) continues to be used to support rice prices and raise farm incomes. Under the PSP, farmers are allowed to sell their paddy rice to the Government at the support price, which is administratively determined. Then farmers are given four months to redeem the pledged paddy (reject the Government offer and sell their rice on the open market), otherwise they have to deliver the paddy to the Government and receive the support price. The primary objective of the program has turned from an initial focus on stabilizing rice prices to raising farm incomes as, over time, the support price has been raised more and more relative to the market price. As a result, Thailand has witnessed an enormous increase in rice production from 27.16 million tons in 2001 to 37.43 million tons in 2012. Some analysts have questioned the effectiveness of the PSP. In particular, it has been argued that large-scale commercialized farmers are the major recipients of the benefits while small-scale farm households tend to have been left out (Poapongsakorn and Charupong, 2010). This raises two important questions. First, what factors influence farmer decisions to participate in the PSP? Second, are there differences in the rice production technology being used and the level of technical efficiency among program participants and non-participants? The decision to participate will be governed by the size of the support price relative to the market price but other individual farm characteristics such as size and financial position may also influence the costs of participation for individual farms. For example, farmers who deliver rice to the Government 86 typically have to wait extra time to receive payment and the size, scope, and financial position of the farm may influence their ability to accept delayed payment. The PSP may also attract new farmers and marginal farmers who otherwise would not have brought land into rice production, and these new entrants may use different technologies and have different levels of technical efficiency. The determinants of the participation decision and the distribution of technical efficiency among participants and non-participants is important information for evaluating the full economic effects of the PSP. The stochastic frontier model (SFM) is a standard approach to evaluating the nature of production technologies and the distribution of technical efficiency among a sample of firms (Aigner, Lovell, and Schmidt, 1977). Several studies have applied the SFM to samples of Thai rice farmers (Chaovanapoonphol, Battese, and Chang, 2009; Rahman, Wiboonpongse, Sriboonchitta, and Chaovanapoonphol, 2009; Srisompun and Isvilanonda, 2012). Yet, the issue of whether PSP participants and nonparticipants use the same production technologies and have the same levels of technical efficiency has not been investigated to date. One approach would be to estimate different SFMs for each subsample of data (participants and nonparticipants) and compare results. However, the fact that farmers choose to participate or not in a way that is likely non-random may lead to sample selection bias in estimates from this naïve approach. Failure to account for such selectivity could bias the estimated parameters of both the stochastic frontier production technology and the distribution of technical efficiency. In response, we augment the standard stochastic frontier model with a participation equation explaining the decision to participate in the PSP, and then use Heckman’s two-step estimation and Greene’s sample selection stochastic production frontier model to explore levels of technical efficiency among participants and non-participants. The resulting model is used to 87 investigate two important issues: (a) what are the key determinants of farmers’ decision to participate in the PSP?; and (b) do program participants and non-participants use different rice production technologies and have different levels of technical efficiency? 88 3.2 Theoretical model In this section we first discuss the standard stochastic frontier model proposed by Aigner, Lovell, and Schmidt (1977) (here after the ALS model) that does not account for selection bias. Then we discuss the two approaches to accounting for selectivity bias in SFMs, namely Heckman’s SFM two-step estimation (Heckman, 1979) and Greene’s SFM with correction for sample selection bias (Greene, 2010). 3.2.1 Stochastic frontier model (SFM model) The standard stochastic frontier model or the ALS model (Aigner, Lovell, and Schmidt, 1977) is specified as: 𝑦𝑖 = 𝜷′ 𝒙𝒊 + 𝑣𝑖 − 𝑢𝑖 , (3.1) 𝑢𝑖 = 𝜎𝑢 |𝑈𝑖 |, 𝑈𝑖 ~𝑁[0,1] 𝑣𝑖 = 𝜎𝑣 𝑉𝑖 , 𝑉𝑖 ~𝑁[0,1] where 𝑦𝑖 ,𝒙𝒊 ,𝑣𝑖 , and 𝑢𝑖 represent output, input vector, idiosyncratic error in the production frontier, and technical inefficiency, respectively, for a sample of firms indexed by i. Technical inefficiency 𝑢𝑖 is assumed to be truncated normal and takes only non-negative values. The frontier is assumed linear in parameters but nonlinearity of the production frontier is allowed through transformations of the 𝑦𝑖 , and 𝒙𝒊 values (e.g. log transformations and including higher order terms in 𝒙𝒊 ). The standard model assumes that the mean level of technical inefficiency is invariant across observations. However, Kumbhakar et al. (1991) show how to relax this assumption by allowing the mean to be a function of exogenous variables (e.g. management skills). This specification allows a part of the technical inefficiency to be explained by farm- 89 specific factors. Econometric estimation provides estimates of the frontier parameters together with an auxiliary model of technical inefficiency as a function of farm-specific factors. One underlying assumption of SFMs is that all farmers in the sample have access to the same production technology. If some characteristics allow a sub-sample of farmers to have access to a different production technology, a separate estimation of the stochastic frontier production is needed. However, these subsample estimations may then provide biased estimation of population production functions if the farmers’ decision on which technology to use is governed by farm and farmer characteristics. Treating the observed data as if they are randomly sampled from the population and estimating the SFM of each subsample separately potentially biases the estimated parameters. There are two approaches to accounting for this selectivity bias in SFMs: (a) the Heckman’s two-step procedure to correct for sample selection bias by appending the inverse Mill’s ratio as a covariate in separate SFMs for each sub-sample (Heckman, 1979); (b) Greene’s SFMs with correction for sample selection bias. Green’s model jointly estimates the selection models and the SFMs allowing for correlated errors (Greene, 2010). 3.2.2 SFM estimation using Heckman’s approach Let 𝑑𝑖∗ be a latent variable representing an unobservable selection criterion variable which is postulated to be a function of some exogenous variables (𝒛𝒊 ): 𝑑𝑖∗ = 𝜶′ 𝒛𝒊 + 𝑤𝑖 (3.2) where 𝜶 is a vector of parameters and w is the error term distributed as 𝑁(0, 𝜎𝑤2 ).The selection criterion variable is unobserved. Instead, a dummy variable, 𝑑𝑖 , is observed and takes a value of 1 when 𝜶′ 𝒛𝒊 + 𝑤𝑖 > 0 and the decision is made to participate and zero otherwise: 90 𝑑𝑖 = 1[𝜶′ 𝒛𝒊 + 𝑤𝑖 > 0], 𝑤𝑖 ~𝑁[0,1] (3.3) where the variance of w has been normalized to one. SFM estimation by Heckman’s (1979) two-step procedure to correct for sample selection bias involves the following steps: (1) fit a probit model for the sample selection equation; and (2) estimate a SFM for each subsample but including the inverse Mill ratio (IMR) from the first step as a covariate to correct for selectivity bias and test its significance using a t-test. If the slope coefficient of the IMR from at least one of the sub-samples is significantly different from zero, there is evidence to reject the null hypothesis of no selection bias. On the other hand, failure to reject the null hypothesis indicates that selection bias is not present. The model can be specified as (3.3) plus: Regime 1: 𝑦𝑖1 = 𝜷𝟏 ′ 𝒙𝟏𝒊 + 1 𝐼𝑀𝑅1𝑖 + 𝑣𝑖1 − 𝑢𝑖1 𝑖𝑓 𝑑𝑖 = 0 (3.4) Regime 2: 𝑦𝑖2 = 𝜷𝟐 ′ 𝒙𝟐𝒊 + 2 𝐼𝑀𝑅2𝑖 + 𝑣𝑖2 − 𝑢𝑖2 𝑖𝑓 𝑑𝑖 = 1 (3.5) where 𝑢𝑗𝑖 = 𝜎𝑗𝑢 |𝑈𝑗𝑖 |, 𝑈𝑗𝑖 ~𝑁[0,1] ; 𝑗 = 1,2 𝑣𝑗𝑖 = 𝜎𝑗𝑣 𝑉𝑗𝑖 , 𝑉𝑗𝑖 ~𝑁[0,1] ; 𝑗 = 1,2 𝑗 is the parameter that detects the presence of selectivity bias The estimation of (3.4) and (3.5) proceeds as follows. First, (3.3) is estimated using the full sample and the IMR is obtained and substituted into (3.4) and (3.5). Second, (3.4) and (3.5) are estimated by sub-sample OLS. Finally, the t-test on IMR will determine whether we can reject the null hypothesis of no selection bias. 91 3.2.3 SFM estimation using Greene’s approach Greene (2010) argues that the Heckman’s switching regression is inappropriate in models that are nonlinear because: (1) in nonlinear models like the SFM the impact of selection on the conditional mean of the model of interest will not necessarily take the form of an inverse Mill ratio; (2) the bivariate normality assumption needed to justify the inclusion of the inverse Mills ratio in the second model does not generally appear anywhere in the SFM; and (3) the dependent variable, conditioned on the sample selection, is unlikely to have the distribution described by the model in the absence of selection. Greene proposed an internally consistent method of incorporating the sample selection problem in a SFM. In his model the error term in the selection model (𝑤𝑖 ) is assumed to be correlated with the noise in the SFM (𝑣𝑖 ). This correlation between 𝑣𝑖 and 𝑤𝑖 is denoted by 𝜌. Greene’s model is then written as: 𝑑𝑖 = 1[𝜶′ 𝒛𝒊 + 𝑤𝑖 > 0], 𝑤𝑖 ~𝑁[0,1] (3.6) 𝑦𝑖 = 𝜷′ 𝒙𝒊 + 𝜀𝑖 , 𝜀𝑖 ~𝑁[0, 𝜎𝜀2 ] where (𝑦𝑖 , 𝒙𝒊 )are observed only when 𝑑𝑖 = 1, 𝜀𝑖 = 𝑣𝑖 −𝑢𝑖 , 𝑢𝑖 = 𝜎𝑢 |𝑈𝑖 |, 𝑈𝑖 ~𝑁[0,1], 𝑣𝑖 = 𝜎𝑣 𝑉𝑖 , 𝑉𝑖 ~𝑁[0,1], (𝑤𝑖 , 𝑣𝑖 )~𝑏𝑖𝑣𝑎𝑟𝑖𝑎𝑡𝑒 𝑛𝑜𝑟𝑚𝑎𝑙 𝑤𝑖𝑡ℎ 𝑐𝑜𝑟𝑟𝑒𝑙𝑎𝑡𝑖𝑜𝑛  The conditional density for an observation in Green’s model is 92 𝑓(𝑦𝑖 |𝑥𝑖 , |𝑈𝑖 |, 𝒛𝑖 , 𝑑𝑖 ) = 2 1 (𝑦𝑖 −𝛽′𝑥𝑖 +𝜎𝑢 |𝑈𝑖 |) 2 ) 𝑒𝑥𝑝( 𝜎2 𝑣 𝜎𝑣 √2𝛱 𝑑𝑖 + (1 − 𝑑𝑖 )Φ(−𝛂′𝒛𝒊 ) ( [ × Φ( { (3.7) ) ρ(𝑦𝑖 −𝛽′𝑥𝑖 +𝜎𝑢 |𝑈𝑖 |/𝜎𝜀 )+𝛂′𝒛𝒊 √1−ρ2 ) } ] The unconditional log likelihood for the model in (3.6) is formed by integrating out the unobserved |𝑈𝑖 | then maximizing with respect to the unknown parameters. 𝑙𝑜𝑔𝐿(β, 𝜎𝑢 , 𝜎𝑣, 𝛼, 𝜌) = ∑N i=1 log ∫|𝑈 | 𝑓(𝑦𝑖 |𝑥𝑖 , |𝑈𝑖 |, 𝒛𝑖 , 𝑑𝑖 ) 𝑝(|𝑈𝑖 |)𝑑(|𝑈𝑖 |) 𝑖 where 𝑝(|𝑈𝑖 |) = 𝜙(|𝑈𝑖 |) Φ(0) 1 (3.8) 2 = 𝑒𝑥𝑝 (− 2 |𝑈𝑖 |2 ) √𝛱 , |𝑈𝑖 | ≥ 0 Since the integral of this function does not exist in a closed form, Greene (2010) proposes computation by simulation. The simulated log likelihood function is 𝑙𝑜𝑔𝐿𝑠 (β, 𝜎𝑢 , 𝜎𝑣, 𝛼, 𝜌) = 2 1 (𝑦𝑖 −𝛽′𝑥𝑖 +𝜎𝑢 |𝑈𝑖𝑟 |) 2 ) 𝑒𝑥𝑝( 𝜎2 𝑣 𝜎𝑣 √2𝛱 1 𝑅 ∑𝑁 𝑖=1 𝑙𝑜𝑔 𝑅 ∑𝑟=1 𝑑𝑖 [ + (1 − 𝑑𝑖 )Φ(−𝛂′𝒛𝒊 ) ( × Φ( { (3.9) ) ρ(𝑦𝑖 −𝛽′𝑥𝑖 +𝜎𝑢 |𝑈𝑖𝑟 |/𝜎𝜀 )+𝛂′𝒛𝒊 √1−ρ2 ) } ] The single equation MLE of α in the probit equation (3.6) is consistent, albeit inefficient. These estimates of α are then taken as given to simulate the log likelihood using (3.9). Finally, use the Murphy and Topel (2002) correction to adjust the standard errors in the same fashion as Heckman’s correction of the canonical selection model in (3.4) and (3.5). 93 3.3 Model specification and data 3.3.1 Model Specification This study uses both the Heckman and Greene methods to estimate stochastic production frontier models of PSP participants and non-participants while controlling for selectivity bias. Both methods require two sets of variables; one for the production frontier and the other for the probit model which models a farmer’s decision to participate in the PSP. The functional form used for the frontier is extended Cobb-Douglas so that for j=1, 2 sub-samples (participants and nonparticipants) the model is: 𝐿𝑁𝑃𝑅𝑂𝐷𝑗𝑖 = 𝛽𝑗0 + 𝛽𝑗1 𝐿𝑁𝐹𝐸𝑅𝑇𝑗𝑖 + 𝛽𝑗2 𝐿𝑁𝑆𝐸𝐸𝐷𝑗𝑖 + 𝛽𝑗3 𝐿𝑁𝐿𝐴𝑁𝐷𝑗𝑖 + 𝛽𝑗4 𝐿𝑁𝐿𝐴𝑁𝐷𝑆𝑄𝑗𝑖 + 𝛽𝑗5 𝐼𝑅𝑅𝑗𝑖 + 𝛽𝑗6 𝑇𝐸𝐶𝐻𝑗𝑖 + 𝑣𝑗𝑖 − 𝑢𝑗𝑖 where 𝑢𝑖 = 𝜎𝑢 |𝑈𝑖 |, 𝑈𝑖 ~𝑁[0,1] ; 𝑣𝑖 = 𝜎𝑣 𝑉𝑖 , 𝑉𝑖 ~𝑁[0,1]; (3.10) with i indexing farms. The dependent variable LNPROD is log of total production of jasmine rice per rai. The explanatory variables include a set of log inputs; land (LNLAND), land squared (LNLDSQ), total fertilizer used per rai (LNFERT), total seeds used per rai (LNSEED), a dummy variable indicating whether land is irrigated (IRR), and a dummy variable taking a value of one if a farmer uses transplanting and zero if they seed (TECH). Labor is not included because rice production is no longer labor-intensive and only a small variation in labor used per rai is typically observed. To evaluate sensitivity to exclusion of the labor variable we also estimated models with a labor variable (LNLAB), measured as a sum of family and hired laborers used in rice production per rai. The probit participation-decision equation is specified as 94 𝑃𝑆𝑃𝑖 = 1[𝛼0 + 𝛼1 𝐸𝐷𝑈𝑖 + 𝛼2 𝐸𝑋𝑃𝑖 + 𝛼3 𝐿𝐴𝑁𝐷𝑖 + 𝛼4 𝑇𝑅𝐴𝑁𝑆𝑃𝑂𝑅𝑖 + 𝛼5 𝑆𝑇𝑂𝑅𝐴𝐺𝐸𝑖 + 𝛼6 𝐵𝑂𝑅𝑅𝑂𝑊𝑖 + 𝛼7 𝐵𝐴𝐴𝐶𝑖 + 𝛼8 𝐷𝐼𝑆𝑇𝑖 + 𝛼9 𝐶𝑅𝑂𝑃2𝑖 + 𝛼𝑘 ∑5𝑘 𝑅𝐸𝐺𝐼𝑂𝑁𝑖𝑘 + 𝑤𝑖 > 0] 𝑤ℎ𝑒𝑟𝑒 𝑤𝑖 ~𝑁[0,1] (3.11) The dependent variable PSPi is observed and takes a value of 1 when the decision is made to participate and zero otherwise. As the distance from plots to the nearest depot (DIST) increases, farmers may have less incentive to sell to the government because they have to bear higher costs of transporting rice, especially if several trips are needed. By the same token, lack of transportation (TRANSPOR) may cause farmers to sell their harvest to other buyers located nearby instead. Farmers are expected to be more likely to participate in the program if they own a storage facility (STORAGE) because it gives farmers more flexibility in when to sell. Also, the government pays farmers storage fees if rice is kept with farmers after pledging. The variables EDU and EXP respectively denote head of household’s years of education and years of ricefarming experience. These variables are expected to positively affect farmers’ management skills and therefore influence participation. The notion of large farms having lower fixed costs associated with transporting rice to PSP depots implies that the probability of participation would increase as land size (LAND) increases. The variable BORROW indicates whether a farmer has borrowed money to finance his/her rice production. A high level of the support price relative to market price is likely to induce farmers to participate in the program especially those who are indebted. As the distance from home to the nearest Bank of Agriculture and Agricultural Cooperatives (BAAC) increases, the incentive to participate may decrease because information about the PSP is less frequently communicated to farmers. The variable CROP2 indicates whether a farmer produces during a 95 second-season rice. The dummy variable REGION representing six different districts in which the survey took place is included to account for other regional-specific factors that possibly influence the participation but are not observed. For instance, farmers in certain districts are discouraged from participating due to a lengthy processing-time for transferring money to farmers’ bank accounts from the BAAC, which tends to vary by branches. Sometimes, farmers in certain regions have to sell to other buyers because a depot has exceeded its daily storage capacity. Recall that Heckman’s method requires inclusion of the inverse Mill’s ratio (IMR). In the first step, the inverse Mill’s ratio is obtained from a pooled-probit estimation as shown in (3.11). In the second step, separate production frontier models are estimated by appending the inverse Mill’s ratio obtained from the first step as one of the covariates. The selectivity bias is present if the estimated coefficient of IMR () is statistically difference from zero at least in one of the subsamples under a t-test. In contrast, Greene’s method estimates (3.10) and (3.11) in one step by NLOGIT (version 4) for which the distributional assumptions of the error terms are as stated in (3.6). 3.3.2 Data The empirical analysis is based on a sample of 387 jasmine-rice farm households chosen from 21 villages located across Buriram province in Thailand. The province is one of the largest producers of jasmine rice in Thailand and represents approximately 15% of total area and production of jasmine rice in 2011.24 Six out of 23 districts located across the province were randomly selected. Then, two villages located in irrigated areas and two villages located in areas 24 Thailand Office of Agricultural Economics 96 with no irrigation system in place are randomly chosen from each selected district, constituting a total sample of 24 villages. Finally, 20 jasmine-rice farm households from each village were selected for interview.25 The data include inputs used, geographical location of plots, and socioeconomic characteristics of farm household members. The information collected covers the major (1st) rice season in 2012/13. The sample contains 130 farmers, who have participated in the PSP during the 2012/13 major cropping season, and 257 non-participants. 25 Due to some technical problems, however, the survey only took place in 21 villages from which data from 387 rice farm households were collected. 97 3.4 Results 3.4.1 Differences in input allocation and farmers’ characteristics Table 3.1 presents summary statistics for output, inputs, and characteristics of farmers classified by their PSP participation status. Note that the log of total inputs (fertilizer, seed, and labor) and total output are used in the estimation. The mean differences of total output and total inputs used significantly differ between the two sub-samples at 1% level but they are not significant at 5% level when measured on a per-rai basis. Land size and is significantly different between PSP participants and non-participants despite similar average yields. The difference in land size is quite large which indicates that the scale of production is much larger for participants. However, non-participants apply more fertilizer per rai. On average, participants have more years of education but less farming experience. The proportion of farmers lacking transportation and storage infrastructure is higher among non-participants. The proportion of participants who borrow money is higher among participants. In fact, total household debts are statistically higher for those who participate in PSP (not shown). The distances to nearest PSP depot and BAAC are higher for non-participants. Yet, only the former is statistically significant. Lastly, a higher proportion of the participants reported that they also produce rice in the second season. This perhaps indicates that the participants are more commercialized. 98 Table 3.1: Average input used and farmers' characteristic variables Variable Non PSP PSP Mean Difference (N=257) (N=130) (Non PSP-PSP) Production Total production (kg) 4,327 10,117 Total fertilizer (kg) 461 1,053 Total seed (kg) 349 799 Total labor (man-day) 93 231 Land (rai) 13.42 30.29 Irrigation (irrigated=1, zero otherwise) 0.33 0.39 Technique (transplanting=1,seeding=0) 0.13 0.15 Characteristics Education (years) 5.11 5.52 Farming experience (years) 35.64 33.39 Land (rai) 13.42 30.29 Transportation (own=1,none=0) 0.38 0.50 Storage (own=1,none=0) 0.80 0.92 Borrow (yes=1,no=0) 0.37 0.57 Distance to nearest BAAC (km.) 8.31 7.75 Distance to nearest PSP depot (km.) 17.06 14.64 2nd-season crop grower (yes=1,no=0) 0.17 0.23 Production & input used per rai (supplemental information/not used in SFM) Yield (kg/rai) 348.80 345.15 Fertilizer (kg/rai) 35.69 33.85 Seed (kg/rai) 26.42 25.93 Labor (man-day/rai) 6.72 7.11 ***, **, * denote 1%, 5%, and 10* significant levels, respectively Source: author's survey -5,790 -591 -450 -138 -16.87 -0.06 -0.03 *** *** *** *** *** -0.41 2.25 -16.87 -0.12 -0.12 -0.20 0.56 2.85 -0.06 * * *** *** *** *** *** * 3.65 1.84 * 0.49 -0.39 3.4.2 Determinants of PSP participation Results from the probit model of PSP participation are shown in Table 3.2. Neither education nor farming experience has a statistically significant impact on the likelihood of program participation. Similarly, owning a vehicle that can be used for transporting rice increases the probability of program participation but its effect is statistically insignificant. 99 Owning a storage facility increases the probability of participation and this effect is statistically significant at the 10% level. The ability to store may facilitate program participation because the Government’s PSP depots are often overwhelmed at the beginning of harvest season and participants have to delay delivery. Without storage farmers would have to sell immediately on the market. However, the distance to nearest PSP depot does not have a statistically significant impact on the likelihood of program participation, once regional differences are accounted for. Distance to nearest BAAC has a statistically significant (10% level) negative effect on the probability of program participation. The BAAC is a source of PSP program information and close proximity may also increase the ability of farmers to borrow money from the bank to finance the delayed payment that usually accompanies program participation. Similarly, farm borrowing has a positive relationship with program participation. Finally, land area has a positive relationship with the probability of program participation, as does the farmer’s cultivation of rice during the second growing season. These factors indicate that there is a positive relationship between the degree of commercialization of the farm and the likelihood of participating in the PSP. However, a direction of causality cannot be determined. On the one hand, large farms may tend to participate in the program. On the other hand, it is also possible that participating farms get larger overtime. 100 Table 3.2: The estimated parameters of the probit model for participation decision Variable Constant Education Farming experience Land Transportation Storage Borrow Distance to nearest BAAC Distance to nearest PSP depot 2nd-season crop grower Region 1 Region 2 Region 3 Region 4 Region 6 Model diagnostics Log likelihood Chi squared P-value McFadden pseudo R-squared ***, **, * denote 1%, 5%, and 10% significant levels, respectively. Coefficient -1.66 -0.02 -0.01 0.05 0.16 0.34 0.41 -0.03 0.01 0.41 0.34 -0.53 0.85 0.18 -0.15 *** *** * *** * * * *** -180.62 130.61 0.00 0.27 3.4.3 Frontier production technologies Table 3.3 and Table 3.4 report the parameter estimates of stochastic production frontiers of participants and non-participants, respectively. The results in column 2 are from Greene’s method while those in column 3 and 4 are from Heckman’s method. In the table GRN and HECK-N denote results from the functional form for production technology described in (3.10) and (3.11) with inclusion of the inverse Mill’s ratio in the latter. The HECK-F results use alternative specification in which labor is added. We evaluate this alternative specification because labor is considered a typical input used in rice production despite its declining 101 intensity.26 Note that HECK-N is nested in HECK-F. Results show that estimates from the two Heckman specifications are very similar but results from Green’s model are quite different. This divergence in parameter estimates using Greene’s and Heckman’s method has been noted in other studies as well (e.g., Wiboonpongse et al., 2012). Because the participation decision equation and the SFM are simultaneously estimated under Greene’s model, overparameterization could be the cause this divergence in parameter estimates. Table 3.3: Estimated parameters of the SFM model for PSP participants Variable GRN coefficient Production function Constant Fertilizer Seed Labor Land Land squared Irrigation Technique Variance parameters Log likelihood 𝜎𝑣 𝜎𝑢 (𝑣,𝑤),Heckman (𝑣,𝑤),Greene 22.272 0.1036 0.0722 1.4271 * 0.3576 -0.0002 0.1873 -1920.66 1.02 23.81 HECK-N coefficient HECK-F coefficient 6.745 *** 0.072 0.124 ** 6.539 *** 0.069 0.111 ** 0.068 ** 0.282 0.050 0.100 0.096 0.246 0.062 * 0.116 * 0.135 * -42.47 0.18 0.49 -0.20 * -40.67 0.06 0.08 -0.17 0.33 ***, **, * denote 1%, 5%, and 10% significant levels, respectively. The alternative specification did not converge using Green’s model and so results for that case are not shown. So, Greene’s model can only be compared to Heckman’s model using the nested specification (i.e. a specification in which labor is excluded) 26 102 Table 3.4: Estimated parameters of the SFM model for PSP non-participants Variable GRN coefficient Production function Constant Fertilizer Seed Labor Land Land squared Irrigation Technique Variance parameters Log likelihood 𝜎𝑣 𝜎𝑢 (𝑣,𝑤),Heckman (𝑣,𝑤),Greene 12.374 0.005 0.191 *** 0.012 ** 1.256 0.426 *** 0.264 -2015.97 0.98 5.97 HECK-N coefficient 5.617 *** 0.185 *** 0.159 *** 0.569 -0.008 0.115 ** 0.036 -196.87 0.16 0.91 -0.11 HECK-F coefficient 5.395 0.175 0.158 0.115 0.512 -0.017 0.181 0.006 *** ** *** *** *** *** -189.63 0.15 0.90 -0.11 -0.0018 ***, **, * denote 1%, 5%, and 10% significant levels, respectively. For the participants (Table 3.3), only land size is statistically significant using Greene’s method. Under the HECK-N specification, all inputs except fertilizer are statistically significant. The HECK-F specification indicates that only seed and labor are statistically significant. However, log-likelihood-ratio test (LR-test) strongly rejects joint exclusion restrictions for land and land squared (not shown). Hence, land is still a key factor of production. In case of the nonparticipants (Table 3.4), more parameters are statistically significance under Greene’s model, seed, land, and irrigation. Yet, their estimates are very different from those estimated by the Heckman’s method. Under the HECK-F specification, all variables except planting technique are statistically different from zero. Like the participants, the estimates of land and land squared are not individually significant but are jointly significant under HECK-N specification. 103 The estimates for the selectivity bias parameter () are reported at the bottom of Table 3 and 4. For the participants, we cannot reject the null hypothesis of no selectivity bias using Green’s model of HECK-F. Selectivity bias is somewhat significant under the HECK-N specification as the null hypothesis is rejected but only at the 10% level (p-value = 0.095). For the non-participants, all three models reject the existence of selectivity bias. Therefore, the conclusion is that there is no strong evidence suggesting the presence of selectivity bias. This means the stochastic production frontier for the participants and non-participants can be estimated separately using the standard SFM if their production technologies indeed differ, or by pooling the data if their production frontiers are the same. Table 3.5 and Table 3.6 report the parameter estimates of stochastic production frontier estimated by the standard SFM (or ALS) model under several alternative specifications. ALS-N and ALS-F are the standard SFM specified in (3.1) and are respectively similar to that of HECKN and HECK-F except that now the inverse Mill’s ratio is excluded. The models specified under ALS-N and ALS-F are estimated using the pooled/full sample. The other results are from sample-separated models. ALS-N1&2 and ALS-F1&2 are constrained to have the same frontier coefficients but allow variance parameters to differ. That is, technical efficiency for each subsample is estimated separately by constraining frontier parameters to be the same for participants and non-participants but allowing standard deviations of the errors (𝑢 and 𝑣 ) to differ across the sub-samples. For ALS-N3&4 and ALS-F3&4, no constraint is imposed on production technology and variance parameters. A LR-test for different production frontiers in these two subsamples rejects the null hypothesis of homogenous production frontier; i.e. testing the fullsample model against (unconstrained) sample-separated models.27 A LR-test for different 27 A likelihood-ratio Chow test gives a test statistic (a Chi-squared statistic) of 33.94 with 9 degrees of freedom (seven coefficient parameters of production technology and two variance parameters) which is significant at 1% 104 production technology while allowing the variance parameters to differ strongly supports the null hypothesis of homogeneous production technology; testing the full-sample model which allows only differences in an intercept between participants and non-participants against the full model which allows differences in both an intercept and slope coefficients.28 An LR-test for different variance parameters strongly reject the null hypothesis that technical inefficiency of these subsamples are drawn from the same distribution.29 These test results imply that participants and non-participants share the same production technology but there technical efficiencies do differ. Table 3.5: Estimated parameters of the nested-model stochastic production frontier Variable Production function Constant Fertilizer Seed Land Land squared Irrigation Technique ALS-N (Pooled) coefficient 5.645 0.145 0.157 0.475 0.027 0.150 0.103 ALS-N1 (Non PSP) coefficient ALS-N2 (PSP) coefficient constrained constrained constrained constrained constrained *** constrained * constrained constrained constrained constrained constrained constrained constrained constrained *** *** *** *** ALS-N3 (Non PSP) coefficient 5.409 0.190 0.150 0.562 0.005 0.135 0.052 *** *** *** *** ** ALS-N4 (PSP) coefficient 6.189 0.103 0.117 0.267 0.069 0.141 0.122 *** * * * ** * Variance parameters Log likelihood -258.23 -223.02 -48.02 -197.55 0.18 0.20 0.14 0.17 𝑣 0.78 0.86 0.60 0.90 𝑢 2 0.64 0.78 0.39 0.84  ***, **, * denote 1%, 5%, and 10% significant levels, respectively. -43.70 0.25 0.38 0.21 level. Thus, we reject the null hypothesis of homogenous production frontiers between participants and nonparticipants. 28 A likelihood-ratio test gives a test statistic (a Chi-squared statistic) of 3.11 with 6 degrees of freedom (six interaction-coefficient parameters of production technology) which is not significant at 5% level. Thus, we cannot reject the null hypothesis of homogenous production technology between participants and non-participants. 29 A likelihood-ratio test gives a test statistic (a Chi-squared statistic) of 29.85 with 2 degrees of freedom (two variance parameters) which is significant at 1% level. Thus, we reject the null hypothesis technical inefficiency of these subsamples are drawn from the same distribution 105 Table 3.6: Estimated parameters of the full-model stochastic production frontier Variable Production function Constant Fertilizer Seed Labor Land Land squared Irrigation Technique ALS-F (Pooled) coefficient 5.414 0.146 0.146 0.108 0.443 0.015 0.180 0.044 ALS-F1 (Non PSP) coefficient ALS-F2 (PSP) coefficient ALS-F3 (Non PSP) coefficient constrained constrained constrained constrained constrained constrained *** constrained constrained constrained constrained constrained constrained constrained constrained constrained constrained 5.178 0.179 0.146 0.120 0.504 -0.005 0.203 0.022 *** *** *** *** *** ALS-F4 (PSP) coefficient *** *** *** *** *** 6.064 0.097 0.105 0.078 0.299 0.054 0.113 0.092 *** *** * * ** * Variance parameters Log likelihood -251.05 -191.83 -47.16 -193.50 0.17 0.19 0.12 0.18 𝑣 0.78 0.85 0.61 0.88 𝑢 2 0.64 0.75 0.38 0.80  ***, **, * denote 1%, 5%, and 10% significant levels, respectively. -41.79 0.20 0.46 0.25 3.4.4 Technical efficiency of PSP participants and non-participants Summary statistics of the technical efficiency scores for PSP participants and nonparticipants under ALS-N and ALS-F are presented in Table 3.7. On average, the participants are more technically efficient than the non-participants as indicated by the fact that mean technical efficiency is higher while having lower standard deviation. The distribution of technical efficiency scores for the participants displays leftward skew while that of the nonparticipants displays rightward skew. Under the ALS-N specification 23.08% of participants have technical efficiency scores above 0.8 compared to 16.74% for non-participants. This means participants group has a higher proportion of farmers with high technical efficiency scores compared to that of non-participants. In contrast, a higher proportion of non-participants technical efficiency scores are located in the lower tail of the distribution; 35.41% of the sample 106 are located below 0.5 compared to 26.15% of participants. On the one hand, one may argue that the program tends to attract efficient farmers. On the other hand, it is also possible that the program participants have become more efficient. Unfortunately, the direction of the effect cannot be determined. The ALS-F specification also produces similar conclusions. The distribution of technical efficiency scores are not much different from those reported in Table 3.6 when estimated from the unconstrained model using the pooled sample (not shown). Table 3.7: Distribution of technical efficiencies Interval 0.91-1.00 0.81-0.90 0.71-0.80 0.61-0.70 0.51-0.60 under 0.51 Mean TE Standard Deviation Minimum Maximum Nested Model NON-PSP PSP (ALS-N1) (ALS-N2) 1.95% 3.85% 14.79% 19.23% 19.84% 13.85% 13.23% 18.46% 14.79% 18.46% 35.41% 26.15% 0.58 0.63 0.22 0.17 0.02 0.20 0.93 0.94 107 Full Model NON-PSP PSP (ALS-F1) (ALS-F2) 2.33% 6.15% 15.56% 18.46% 16.73% 13.85% 16.34% 16.15% 12.84% 16.15% 36.19% 29.23% 0.58 0.63 0.22 0.18 0.02 0.18 0.93 0.95 3.5 Conclusion and policy implications The objective of this study was to identify the factors that determine Thai jasmine-rice farmers’ decision to participate in the PSP and estimate the frontier production technology and technical efficiency of participants and non-participants. Two approaches to dealing with the selection bias problem were applied—Greene’s model and Heckman’s two-step adjustment approach. The result indicates that land size has a positive relationship with the likelihood to participate in the program. Households using loans are also more likely to participate in the program than those who do not. Barriers to program participation include distance to the nearest BAAC branch, which is a government-affiliated agency responsible for issuing loans to farmers. The difference between parameter estimates obtained from the Heckman’s and Greene’s methods are large. Estimates from Greene’s method indicate there is no statistical evidence of selection bias while only weak evidence was found under Heckman’s method. Therefore, the conclusion is that there is no selectivity bias and the production model can be estimated using the standard frontier approach without accounting for selection bias. However, the results from a likelihood-ratio test indicate that both participants and non-participants share the same frontier production technology but have a different distribution of inefficiency. So, technical efficiency scores for each group are computed separately assuming a homogeneous frontier production function. The analysis of technical efficiency reveals that PSP participants are more efficient because the mean of technical efficiency scores are relatively higher while their variance is smaller. The distribution of technical efficiency scores for non-participants also displays leftward skew compared to the rightward skew for participants. In other words, a higher proportion the participants are located in the high-efficiency range and less in the low-efficiency range. 108 The findings from this study have some important policy implications. First, there is a strong relationship between land size and the probability of participating in the program. This is consistent with the observation that most PSP participants produce rice on a large scale. Therefore, a significant portion of program benefits are captured by large farms. However, the causality between farm size and participation can also go in the opposite direction in which case it implies that participating farms get larger. Since the participants are more technically efficient in production, one can also argue that the program tends to attract efficient farmers. On the other hand, it is also possible that participating farms become more efficient overtime. Lastly, policymakers may need to investigate factors that significantly deter the farmers’ participation decision if they want to distribute program benefits more evenly to all farmers. 109 REFERENCES 110 REFERENCES Aigner, D.; C. A.Lovell and P. Schmidt. 1977. Formulation and Estimation of Stochastic Frontier Production Function Models.Journal of econometrics 6(1): 21-37. Chaovanapoonphol, Y.;G. E. Battese and H. S. Chang. 2009. The Impact of Agricultural Loans on the Technical Efficiency of Rice Farmers in the Upper North of Thailand, in Lee J-D. and A. Hesmati (Eds.). Productivity, Efficiency, and Economic Growth in the AsiaPacific Region. Berlin: Springer-Verlag. Greene, W. 2010.A SFM with Correction for Sample Selection.Journal of Productivity Analysis 34(1): 15-24. Heckman, J. J. 1979. Sample Selection Bias as a Specification Error.Econometrica: Journal of the Econometric Society 47(1): 153-161. Kumbhakar, S. C.; S. Ghosh and J. T. McGuckin. 1991. A Generalized Production Frontier Approach for Estimating Determinants of Inefficiency in US Dairy Farms. Journal of Business & Economic Statistics 9(3): 279-286. Murphy K. and R. Topel. 2010. Estimation and Inference in Two Stem Econometrics. Cambridge: University Press. Poapongsakorn, A. and J. Charupong. 2010. Rent Seeking Activities and the Political Economy of the Paddy Pledging Market Intervention Measures. A Research Study Proposed to Office of the National Anti-Corruption Commission. Bangkok: Thailand Development Research Institute. Rahman, S.; A.Wiboonpongse; S. Sriboonchitta and Y. Chaovanapoonphol. 2009. Production Efficiency of Jasmine Rice Producers in Northern and North Eastern Thailand.Journal of Agricultural Economics 60(2): 419-435. Srisompun, O., and S. Isvilanonda. 2012. Efficiency Change in Thailand Rice Production: Evidence from Panel Data Analysis. Journal of Development and Agricultural Economics. 4(4): 101-108. Wiboonpongse, A.; S. Sriboonchitta; S. Rahman;P.Calkins and T. Sriwichailumphun. 2012. Joint Determination of the Choice of Planting Season and Technical Efficiency of Potato in Northern Thailand: A Comparison of Greene’s versus Heckman’s Sample Selection Approach.African Journal of Business Management 6(12): 4504-4513. 111