Incorporating Regional Rice Production Models in a ... · and relative price of rice. Because 1991...

Philippine Journal of DevelopmentNumber 66, First Semester 2009

Volume XXXVI, No. 1

Incorporating Regional Rice Production Models in a Simulation Model of Rice Importation: A Discrete Stochastic Programming Approach

CeLia reyes, CHristian mina, Jason Crean, rosaLina de guzman, and Kevin Parton*

ABSTRACTIn the Philippines, importation has remained as one of the most feasible options for the government to meet the growing demand for rice. It is thus imperative for the government to develop a strategy that would ensure adequate supply and minimum importation costs. One of the critical factors in import decisionmaking is rice production. The Inter-Agency Committee on Rice and Corn (IACRC), of which the National Food Authority (NFA) and the Bureau of Agricultural Statistics (BAS) are members, decides on importation when there is an impending production shortfall in the coming season. However, because Philippine agriculture is vulnerable to extreme climate events and climate change is expected to further intensify climate

* Celia Reyes and Christian Mina are Senior Research Fellow and Supervising Research Specialist, respectively, at the Philippine Institute for Development Studies (PIDS). Jason Crean is a Technical Specialist in the Economics Policy Research Unit at the New South Wales Department of Primary Industries (NSW DPI). Rosalina De Guzman is a Supervising Weather Specialist at the Climate Information Monitoring and Prediction Services Center at the Philippine Atmospheric, Geophysical and Astronomical Services Administration (PAGASA). Kevin Parton is the Head and Professor at Charles Sturt University (CSU, Orange Campus). The authors acknowledge the assistance provided by the following: Remedios Liwanag and Edna Juanillo, of PAGASA, for the climate-related data and other relevant information; Ma. Dolores Fernandez, Esperanza Tecson, Ma. Fioretta Estoperez, and Myrna Reburiano, of the National Food Authority (NFA), for their valuable comments on the paper as well as data on importation, distribution and storage, and basic information on NFA marketing operations; Antonette Natividad and Plenee Castillo, of the Bureau of Agricultural Statistics (BAS), for data on rice production and prices as well as background information on rice production forecasting systems.

102 PhiliPPine Journal of DeveloPment 2009

variability, rice production forecast is becoming more uncertain. Inaccurate production forecasts could lead to incorrect volume and ill-timing of rice imports, which in turn, could result in either a waste of resources for the government or a burden to consumers. Contraction of rice imports in the early 1990s and over-importation in 1998 illustrate how inaccurate forecasts of the volume and timing of rice importation, especially during El Niño and La Niña years, could result in substantial economic costs. This paper evaluates the significance of seasonal climate forecast (SCF) in rice policy decisions of the government, particularly on importation. It presents an alternative method of forecasting the level of rice production through regional rice production models. The rice production models systematically incorporate SCF and could be used in support of the current practice of forecasting rice production based on planting intentions. The paper also demonstrates how SCF, together with these production estimates, could be incorporated in the rice import decisions of the government through the Rice Importation Simulation (RIS) model, which was developed using a Discrete Stochastic Programming (DSP) modeling approach. The RIS model, which recommends a set of optimal rice import strategies, could serve as guide for the government in its rice import decisions in the face of seasonal climate variability and could be used in estimating the potential value of SCF.

INTRODUCTIONSeasonal climate variability exposes crop production to different kinds of production risk. These include lack of water supply during the critical crop growth stage due to an El Niño-induced drought, submerging of seedlings in floodwater because of typhoons associated with La Niña, and other perils attributable to extreme climate events. These are particularly true for the Philippines as the country is often severely affected by extreme climate events primarily due to its geographical location (Fraisse et al. 2007; Lansigan 2004; Dawe et al. 2006). Also, production risk in the Philippines is relatively greater compared to other countries because irrigation systems have not yet been fully developed as there are still many areas that need to be irrigated and a number of existing national irrigation facilities have to be rehabilitated (Reyes et al. 2009).

Recent climate change studies predict that there will be an overall change in global climate patterns (Lansigan 2004). This is expected to further intensify climate variability with effects not only on rice, but on agricultural production in general. There is anecdotal evidence that this might already be occurring. Rice

103reyes, mina , Crean, De guzman anD Parton

farmers in Nueva Ecija related that scheduling of farm operations was a lot easier decades ago because climate patterns were more predictable then.

Adverse effects of extreme climate events on rice production, which have significantly affected rice supply and demand in the Philippines, can be exemplified by what happened during the 1990s. The prolonged El Niño episodes from the third quarter of 1990 to the first quarter of 1995 led to consistent decline in rice production (Intal and Garcia 2005). To protect farmers’ incomes, the government tightened rice importation during this period. This resulted in what has been known as the rice crisis. During 1997 and 1998, the worst episodes of El Niño and La Niña, respectively, were experienced by the Philippines. These resulted in significant shortfalls in production, amounting to about 2 million metric tons, as estimated by the Department of Agriculture. Learning from the 1995 rice crisis, the government decided to import more rice than what was required in 1998.

The Inter-Agency Committee on Rice and Corn (IACRC), of which the National Food Authority (NFA) and the Bureau of Agricultural Statistics (BAS) are members, decides on importation when there is an impending production shortfall during the coming season. The rice production forecast is made by BAS based on the estimates of farmers, which are captured by the quarterly palay production survey (PPS). Unfortunately, this production forecasting system does not systematically incorporate seasonal climate forecasts (SCF). Moreover, IACRC does not systematically link SCF to rice importation decisions, which could have otherwise resulted in a more appropriate volume and timing of rice importation.

Contraction of rice imports during the early 1990s and over-importation during 1998 illustrate how inaccurate forecasts of the volume and timing of rice importation, especially during El Niño and La Niña years, could result in substantial economic costs. These costs arise through higher rice prices during periods of shortages and excessive storage costs (Kajisa and Akiyama 2003; Ramos 2000; Unnevehr 1985). During recent years, rice imports of the Philippines continue to exceed the required amount resulting in an increasing level of stock inventory and associated stockholding costs. While it may be safe to always import more than what the country needs, a substantial amount of money could be saved by the government if the optimal volume of imports could be determined.

The principal aim of this study was to estimate the potential economic value that could be obtained from the use of SCF in rice importation decisions of the government. An innovative approach was adopted that showed how SCF could be systematically linked to rice importation decisions. Using this method, results were presented that estimate the potential value of SCF for such decisions. An alternative method of forecasting the level of rice production using SCF was also outlined.


REVIEW OF RICE PRODUCTION MODELS IN THE PHILIPPINESClimate is one of the major determinants of rice production and yield. Many studies have found a significant relationship between climate-related factors and rice production/yield using econometric models. Some of these investigated the impact of climate-related factors on rice production, in addition to other farm inputs such as fertilizer and labor. Kaul (n.d.) found that rice production was significantly affected by fertilizer use and some climatic variables such as actual rainfall, maximum temperature, and minimum temperature. Excessive rains and very high temperature, particularly during the critical growing period of rice, were found to be detrimental to rice production. Abedullah and Pandey (2007) noted that aside from farm inputs, rainfall and its interaction with farm inputs (especially fertilizer) also had significant effects on rice production.

Other existing models analyzed only the effects of climate-related factors on rice production. Sarma et al. (2008) found that rice yield was relatively lower in El Niño than in La Niña years, and that Sea Surface Temperature (SST) was more significantly related to rice yield than Southern Oscillation Index (SOI) was. The study also noted that rice yield and rainfall were not significantly correlated because of the presence of irrigation, which weakened the relationship between the two. Tao et al. (2008), on the other hand, highlighted that rice yield was significantly influenced by the growing-season climate. In particular, growing-season temperature, which had a generally significant warming trend, translated into an increased rice yield. Furthermore, Zubair (2002) found that rice production was significantly correlated with the average SST-based El Niño Southern Oscillation (ENSO) index and aggregate rainfall, both for the current and previous seasons.

Some studies further limited their analyses to investigating the effects of ENSO indices on rice production. Falcon et al. (2004) was able to find a significant relationship between the SST index and paddy rice output. Similarly, Naylor et al. (2001) found that ENSO-related climate variability caused fluctuations both in rice plantings and production. Moreover, Delos Reyes and David (2009) discovered that the annual growth in rice production was linearly related to El Niño-induced drought. Deviations in rice production were found to be dependent on the strength and time of occurrence of the warm episode.

METHODOLOGYIn carrying out the objectives of this paper, econometric and discrete stochastic programming (DSP) models were developed. A set of econometric models of rice production was developed for each of the geographical regions in the Philippines. This set of models, where SCF was taken into account, could be used as an alternative method of forecasting the level of rice production. The DSP model was


then developed, using the forecast level of rice production and SCF, to come up with an optimal volume and timing of rice importation. The DSP model was also used to assess the potential economic value of SCF to rice importation decisions of the government.

The rice production modelProduction forecasts are an important input to the rice importation decisions of the government. However, the government currently relies on the production forecasting system of BAS, which is based on farmers’ estimates that are captured by the PPS. An alternative method of forecasting the level of rice production, which systematically incorporates SCF, is through the development of rice production models. Thus, a set of econometric models of rice production was developed for each of the geographical regions in the Philippines.

Model variablesVariables considered in developing the rice production models were as follows: rainfall, ENSO classification, area planted, irrigated area, fertilizer usage, and relative price of rice. Because 1991 is the earliest year for which data on area planted are available, data on all variables used in developing the rice production models only covered the period 1991–2008 (refer to Table 1 for variable definitions).

Agriculture-related variables (i.e., area planted, proportion of area planted that is irrigated, fertilizer use, relative price) were hypothesized to have a positive effect on production. In this study, increasing the level of inputs (e.g., area of land devoted to rice, volume of seeds planted, recommended amount of fertilizer, and water supply from irrigation facilities) was assumed to translate into a higher level of output, ceteris paribus.1 Also, an increase in the price of domestically-produced rice relative to that of rice in the world market was hypothesized to induce local farmers to increase their production, take advantage of a relatively higher price, and gain higher profit.

Climate-related variables, however, were assumed to have either a positive or negative effect on production, depending on geographical location. In the Philippines, the impact of ENSO varies substantially across region. Prolonged drought caused by El Niño, for instance, could inhibit the growth of the dry season rice crop in certain areas, resulting in a lower volume of production. In other areas, however, such conditions might cause the delay in the planting of wet season crop, particularly those in rainfed areas (David n.d.). On the other hand, La Niña could be beneficial to some rice areas where there are no irrigation systems

1 Holding other things constant.


but could also be devastating to some which are flood-prone. Meanwhile, the expected impact of ENSO on rainfall patterns for a particular region is not often realized and varies across region.

Model estimationHistorical data on the identified variables from the first cropping period of 1991 up to the first cropping period of 2008 were used. Using a multiple linear regression technique, a number of candidate rice production models were developed for each of the sixteen regions.2 Diagnostic tests were employed on these candidate models and necessary adjustments were undertaken to ensure that the basic

2 National Capital Region (NCR) is excluded because rice area is non-existent.

Variable Description

Dependent

prod Volume of total well-milled rice production (in metric tons); paddy rice converted to well-milled rice using 0.63 conversion factor (since well-milled rice is the most popular rice variety for consumers nowadays)

Independent

rain_dry Dummy for ‘below normal’ rainfall; if percentage deviation from normal rainfall (or the average amount of rainfall for 1971–2000) is below 80%

rain_avg Dummy for ‘near normal’ rainfall; if percentage deviation from normal rainfall falls within 81–120% range (base category)

rain_wet Dummy for ‘above normal’ rainfall; if percentage deviation from normal rainfall is above 120%

enso_dry Dummy for El Niño; a full-fledged El Niño is considered when Oceanic Niño Index (ONI) met or exceeded +0.50C for at least five consecutive months

enso_avg Dummy for Neutral; if ONI is less than the absolute value of 0.50C (base category)

enso_wet Dummy for La Niña; a full-fledged La Niña is considered when ONI is less than or equal to -0.50C for at least five consecutive months

irrig Proportion of area planted to paddy rice that is irrigated (in hectares); area planted to paddy rice that is irrigated divided by total area planted to paddy rice, both in hectares; proxy to irrigated area due to unavailability of data on the latter by cropping season

aplanted Area planted to paddy rice (in hectares)

fert Average fertilizer applied (in metric ton per hectare); total quantity of fertilizer applied (in 50-kilogram bags) divided by total area applied with all types of fertilizer (in hectares)

relprice Relative price of rice; domestic (retail price of well-milled rice) over world price (export price of Thai rice), both expressed in PhP per kilogram

Table 1. Variables used for the regional rice production models


assumptions in regression modeling were satisfied. All of the continuous variables were transformed logarithmically to satisfy the basic assumptions in simple regression (such as normality and constancy in variance) and time-series modeling (such as constancy in variance caused by serial correlation and nonstationarity). Also, lags of the production variable were included in the models to correct for serial correlation.

The final production models satisfied the following set of criteria: (i) residuals were normally distributed; (ii) perfect collinearity among the independent variables did not exist; (iii) serial correlation (or correlation among observations from different time periods) was not present; and (iv) signs of the estimated model coefficients were consistent with the hypothesized relationships between production and the independent variables. However, a robust regression was employed to ensure that the model estimates would not be greatly affected by the presence of outliers.3

The final set of production models had reasonable R-squares; ranging from 0.55 to 0.96, with a high mean value of 0.78 (refer to Table 2 for the regional rice production models). On average, therefore, more than 75 percent of the variation in production data could be explained by the final set of predictors in the model. This observation implies that the developed production models should be adequate in forecasting the volume of rice production in the Philippines.

Table 3 presents the estimated levels of rice production for 2008 (specifically for the second/dry cropping season or during the period March–May) for different climatic states, which were generated using the regional rice production models and recent data on model variables. Since the actual rainfall state during that period was ‘wet’ (or above normal rainfall) and the actual level of rice production was around 1.3 million metric tons, the production models were able to predict about 95 percent of the national production. Except for a few regions, estimates at the regional level were generally close to actual levels.

Model validationTo evaluate the fit and predictive ability of the models, both in-sample and out-of-sample forecasting were conducted. For ‘in-sample forecasting’, estimation data, or data used in developing the production models, were used (i.e., data for 1993–2007). On the other hand, ‘out-of-sample forecasting’ made use of a new set of data that was not included in model estimation (i.e., data for 2008). Because there were three possible levels of production for different climatic

3 Robust regression addresses the problems associated with the presence of outliers and thus provides a better fit than ordinary least squares estimation (Kutner et al. 2004).

108P

hiliPPine Journal of DeveloPment 2009

Variable Name

Model 1: CAR

Model 2: Region I

Model 3: Region II

Model 4: Region III

Model 5: Region IV-A

Model 6: Region IV-B

Model 7: Region V

Model 8: Region VI

rain_dry –0.2151* 0.003 –0.0111 –0.0389 –0.0680 –0.0730 –0.0462 0.018

rain_wet –0.1050 –0.0306 0.032 –0.0892 –0.2406** –0.2057** 0.021 –0.0721

enso_dry –0.1029 0.048 –0.0624 –0.0527 –0.1383* –0.1010 –0.0630 –0.1749

enso_wet 0.027 0.1958** 0.125 0.014 0.022 0.107 –0.0656 0.097

ln_irrig 0.252 0.111 – 0.775 2.1407** – 1.4654** –

ln_aplanted 0.139 – 0.351 0.3632** 0.6503** 0.369 0.6499** 1.2033**

ln_fert 0.031 0.537 – – – 0.9540** 0.008 –

ln_relprice – – – – – – – 0.223

ln_prod[t-1] – – – – – – – –

ln_prod[t-2] 0.9029** 0.9949** 0.6880** 0.7730** 0.7640** 0.7328** 0.4801** 0.6780**

α –0.0344 0.819 –0.0264 –1.0643 –3.1178 0.840 –0.2226 –10.4950

R2 0.915 0.968 0.700 0.772 0.580 0.895 0.551 0.945

Table 2. Regional rice production models

109r

eyes, mina , C

rean, De g

uzman anD Parton

Table 2 ContinuedVariable

NameModel 9:

Region VIIModel 10: Region VIII

Model 11: Region IX

Model 12: Region X

Model 13: Region XI

Model 14: Region XII

Model 15: ARMM

Model 16: Caraga

rain_dry –0.8557** –0.0359 –0.1035 –0.0706 –0.0882 –0.1794 –0.4043* –0.0700

rain_wet –0.1072 0.1322* –0.0300 –0.0011 0.1476** –0.0414 –0.0109 0.0014

enso_dry 0.0722 0.0716 -0.0579 0.1196 –0.0116 0.0693 0.0062 –0.1258

enso_wet 0.2471 –0.0354 0.0372 0.0025 0.0817** 0.0200 –0.0179 –0.0091

ln_irrig 1.2797** – – – 1.3841** – 0.4283 –

ln_aplanted 0.7337** 0.4663 0.9349** 1.0536** 0.3471** 0.6996 0.9379** –

ln_fert 0.0774 – – 0.0426 0.1137 – – 0.2630

ln_relprice – 0.2263 – – – – – –

ln_prod[t-1] 0.3315** – – – – – – –

ln_prod[t-2] – 0.8770** 0.4589** 0.2414** 0.6227** 0.5660** 0.4684** 0.7412**

α 0.9699 –3.7935 –3.7665 –2.3673 1.1799 –2.6416 –3.6057 3.5089

R2 0.5823 0.7483 0.8637 0.7186 0.8578 0.8455 0.9263 0.6652

** Significant at 5% level* Significant at 10% level


states, the one corresponding to the observed climate state was compared with the actual.4 This process was done for all regions and for years that were considered in this study.

The mean absolute percentage error (MAPE) was computed for each regional model to determine how far the predicted levels were from the actual. Among the scale-independent statistics that are commonly used in measuring the forecasting ability of the models, MAPE is considered the most versatile, self-evident, and simplest to determine. It indicates, on the average, the percentage error that a given forecasting model produces for a specified period. Basically, it is the sum of the absolute value of the difference between the actual and predicted values divided by the actual values, expressed as a percentage, divided by the total number of periods (Chatfield 2000; ITSMF-NL 2006; Frechtling 1996). The formula used in calculating MAPE is as follows:

Region Dry Average Wet

CAR 56,360 81,743 74,740

Region I 113,674 106,982 129,684

Region II 479,814 522,582 625,376

Region III 629,318 698,056 641,122

Region IV-A 61,012 77,575 60,152

Region IV-B 170,689 207,005 185,538

Region V 150,961 171,386 162,792

Region VI 125,398 148,951 153,043

Region VII 32,099 83,138 97,252

Region VIII 252,165 242,000 270,418

Region IX 40,266 48,490 48,885

Region X 77,732 73,434 73,550

Region XI 80,320 90,167 116,881

Region XII 136,527 153,606 150,132

ARMM 40,565 64,879 62,776

Caraga 96,950 121,592 120,534

Philippines 2,543,847 2,891,586 2,972,875

* second or dry cropping season (March-May 2008)

Table 3. Expected level of rice production for 2008,* in metric tons, by state and by region

4 i.e., climate states during the planting period.


, (1)

where: = actual value at time t = predicted value at time t = forecast error = total number of periods t = time period

ITSMF-NL (2006) noted that the forecasting model with MAPE below 40 percent might be considered reasonably reliable. This is supported by the rule-of-thumb presented in Frechtling (1996), which essentially supports the interpretation of MAPE values in Lewis (1982) (Table 4).

Table 5 displays the computed MAPEs for each of the estimated regional rice production models. In-sample MAPEs show that half of the models are considered reasonable while the others are found to have good forecasting ability. Out-of-sample MAPEs, on the other hand, indicate that four of the models are inaccurate, four are reasonable, four are good, and four are highly accurate. Generally, the estimated regional rice production models were found to have reasonably good forecasting ability, as indicated by the average MAPEs less than 50 percent.

Further refinements of the models, however, are deemed necessary to improve their predictive performance. Some of these may include: further classification of ENSO states to take into account the degree/intensity; exploration of nonlinear forms of the models through the use of interaction terms, e.g., rainfall and irrigation and inclusion of temperature and other relevant variables (if disaggregated data by region and by cropping season are available). The main limitation to future model refinement is the lack of longer time-series data for some of the relevant variables.

×

−

× ∑∑

n

t t

ttn

t t

t

APA

nAe

nMAPE

11

10011001

×

−

× ∑∑

n

t t

ttn

t t

t

APA

nAe

nMAPE

11

10011001

×

−

× ∑∑

n

t t

ttn

t t

t

APA

nAe

nMAPE

11

10011001

×

−

× ∑∑

n

t t

ttn

t t

t

APA

nAe

nMAPE

11

10011001

×

−

× ∑∑

n

t t

ttn

t t

t

APA

nAe

nMAPE

11

10011001

Interpretation Range of MAPE valuesHighly accurate forecasting less than 10%Good forecasting between 10 and 20%Reasonable forecasting between 20 and 50%Inaccurate forecasting greater than 50%

Source: Lewis (1982).

Table 4. Rule-of-thumb for MAPE values


Rice importation simulation (RIS) modelRice production in the Philippines is routinely affected by climatic variations leading to domestic shortfalls and surpluses. The set of econometric models outlined above captures the relationship between climate and rice production in the Philippines and is a critical input to considering the value of SCF. Using SCF to provide better information about future seasonal conditions offers economic value through the forecasts’ influence on decisions about the management of rice stocks.

To determine the potential economic value of SCF to government rice importation decisions in the Philippines, a DSP modeling approach was adopted. DSP is one of the most flexible approaches in handling risk because it incorporates all sources of uncertainties in model parameters while allowing adaptive decisions (McCarl and Spreen 2007). DSP captures all of the elements of decision-theoretic problems (often illustrated in the form of decision trees) which include an objective function to be maximized, a set of possible actions, a set of states of nature, probabilities of occurrence of those states, and consequences of actions under each state of nature.

Table 5. Computed MAPEs for each of the regional rice production modelsRegion In-sample MAPE Out-of-sample MAPE CAR 20.54 8.48Region I 20.67 33.41Region II 10.77 8.19Region III 13.27 14.84Region IV-A 16.13 21.77Region IV-B 16.44 12.35Region V 12.55 32.19Region VI 25.50 68.36Region VII 40.98 181.66Region VIII 23.41 15.93Region IX 25.13 76.26Region X 10.40 2.90Region XI 12.53 32.81Region XII 26.22 50.50ARMM 32.53 11.37Caraga 11.05 2.37Average 19.88 35.84


The general two-stage formulation of the DSP model is considered as the workhorse of stochastic programming (Rockafellar 2001). In a two-stage model, time is simply divided into the ‘present’ and the ‘future.’ Accordingly, a standard linear programming model can be developed into a DSP model by introducing a second-period decision. A brief overview of DSP modeling is provided in the Appendix.

The two-stage DSP formulation is of particular relevance to the rice importation problem of the country because there are two clearly identifiable decision stages. The first stage refers to the period prior to the main rice growing season where an importation decision must be made, while the second refers to the period in which further decisions can be taken once the true seasonal conditions are known.

The DSP model here is referred to as the rice import simulation (RIS) model and was developed to assess rice importation decisions in the face of seasonal climate variability. The ultimate objective of the model is to determine an optimal set of strategies with regard to rice importation (particularly on the volume and timing) and their associated costs under different climate forecast scenarios. This is formulated as a minimization problem that seeks to minimize the potential importation and storage costs of the government under conditions of climate uncertainty.

Objective function The objective function of the RIS model is as follows: (2)

The objective function aims to minimize the potential costs of importation and storage. The total cost of importation comprises two components, namely, the costs of preseason buying and storing rice (the stage 1 costs) and the combined costs of in-season buying, storing, and distributing rice (the stage 2 costs). The first component is the product of the volume of rice purchased preseason ( 1x )


and all costs associated with this preseason buying ( ). The latter comprises the price of rice preseason, other import-related costs and the costs of storing rice up to the in-season period.

The second component, on the other hand, is the sum of all costs associated with in-season buying, storing, and distributing rice, weighted by the probabilities of occurrence of each state. The total cost of in-season buying is the weighted sum of in-season costs of buying rice for all three states (Si), namely: S1 = ‘dry’ or rainfall is ‘below normal;’ S2 = ‘average’ or rainfall is ‘near normal;’ and S3 = ‘wet’ or rainfall is ‘above normal.’ This is a weighted sum of the volume of rice purchased in-season ( ); all costs associated with this in-season buying, including the seasonal price of rice plus other import-related costs ( ); and the probability of occurrence of a particular state ( ), which is dependent on the forecast accuracy and forecast type for each state. Similarly, the costs of in-season storage and distribution are the weighted sums of in-season costs of storing and distributing rice, respectively, for all states. The cost of in-season storage is the weighted sum of the excess volume of rice stored in-season ( ); cost of in-season storage ( ); and probability of occurrence of a particular state ( ) for each state. The cost of in-season distribution is the weighted sum of the volume of rice distributed in-season ( ); cost of in-season distribution ( ); and probability of occurrence of a particular state ( ) for each state.

Rice importation costs were minimized subject to a number of constraints. First, the volume of rice purchases in either period should always be non-negative, as they indicate net rice imports. Second, the total volume of rice purchased preseason and in-season must be greater than or equal to the identified volume of net import demand. Third, the total volume of rice purchased preseason should not exceed the specified rice storage capacity (b). Note that only preseason storage is constrained by the rice storage capacity as rice purchased in-season is assumed to be distributed immediately.

Based on the preseason and the in-season estimated costs of importation (price of rice plus other relevant costs) and probabilities of occurrence of different climatic states, together with net import demand, the model will then produce a set of optimal import strategies and their associated costs. These strategies contain decisions about the quantity of rice to buy and store preseason; the quantity to buy in-season for each state; and the quantity that needs to be distributed for consumption in-season for each state. The associated costs of all of these strategies were estimated for each state.

An important aspect of the RIS model is that it reflects the fact that some decisions must be taken prior to knowing the real state of nature. Like the NFA, in the preseason period, the RIS model only knows the probabilities of occurrence


of the climatic states for the coming season. Once the state has been realized, the RIS model takes a recourse decision. For instance, if the season turned out to be dry, the RIS model might suggest that additional rice be bought in-season to supplement the existing supplies. If the season turned out to be wet, the RIS model might suggest that some of the rice purchased preseason should be stored. Either way, there is a cost associated with not knowing the real state; either the RIS model buys too much preseason (which is associated with higher storage cost) or too little (which might be associated with higher in-season price). The model attempts to minimize the costs of climatic uncertainty by planning for a number of possible states rather than a single state.

Value of forecasting systemThe economic value of the forecasting system is determined by calculating the difference between rice importation costs with and without the forecast. If the optimal importation decision based on a single forecast is denoted by (the posterior optimal act) and the optimal importation decision without the forecast is denoted by (the prior optimal act), then the ex post utility of the forecast is simply defined as:

(3)

where is the posterior probability of state s given forecast . The value of a single forecast is important, but decisionmakers are generally

not able to either purchase or selectively invest their time in understanding a single forecast. The true measure of economic value is the value derived from the use of the entire forecasting system. The value of a forecasting system is obtained by weighting the value of all forecasts by the frequency with which each forecast occurs ( ), as below:

(4)

In this study, is equivalent to 1/3 or 0.33 as there are three possible forecasts, each with the same likelihood of occurrence. The value of the forecasting system can also be stated in terms of outcomes of importation decisions defined in the RIS model referred to in equation (2). If Z*

sFi denotes the cost of importation in state s resulting from implementing the posterior optimal act XFi, and Z*s0 denotes the cost of importation from implementing the prior optimal act in state s, then the value of an entire forecasting system is obtained by:


(5)

The value of a single forecast is non-negative and so the value of the entire forecasting system must also be non-negative, at least prior to considering any cost of accessing the forecasting system. Note that if = for all Fi, then the climate forecasting system F has no value because importation decisions remain the same with or without the forecast.

Model parameters and dataBased on the items discussed in the preceding section, two sets of data are needed in running the RIS model. The first set consists of forecast type and hit rates, referred to as “the climate” data. The second set comprises rice supply and demand as well as import-related costs, referred to as “the agricultural data.” To check the capability of the model, a simulation exercise was conducted. In particular, the exercise aimed to evaluate whether systematically incorporating SCF in rice importation decisions of the government could identify the optimal import policy strategies and thus, minimize the costs of importation in the future. In this exercise, data pertaining to rice importation in 2008 were used.

Climate forecast and forecast accuracyForecast types

The RIS model used four forecasting scenarios: (1) no forecast; (2) forecast—dry; (3) forecast—average; (4) forecast—wet. The three “with forecast” scenarios correspond to the tercile rainfall categories used by PAGASA, namely: below normal (or ‘dry’); near normal (or ‘average’); and above normal (or ‘wet’). PAGASA issues probabilistic forecasts about the occurrence of these rainfall terciles based on observed relationships between rainfall and larger-scale drivers of climate. In probabilistic forecasting systems, forecast probabilities indicate both the direction of climatic conditions as well as the uncertainty of the forecast (IRI 2008).5 A probabilistic forecasting system is used as a basis of the evaluation reported here.

Posterior probabilitiesPosterior probabilities were computed using the Bayes’ formula, which uses both the definition of conditional probability and the law of total probability (Hogg and Craig 1995). Below is the formula used in this study:

5 A forecast system is probabilistic if it expresses a probability distribution over an exhaustive set of states (Lawrence 1999). It differs from a deterministic forecast which would simply assert which climatic state would occur.


, (6)

where: = posterior probability = probability of observing a certain rainfall state ( ) given a certain rainfall forecast ( )

= prior probability = probability of receiving a certain rainfall state (Si) prior to new information

= conditional probability = probability that a certain rainfall forecast (Fi) will be obtained given that a certain rainfall state (Si) occurs; an indication of accuracy of forecast

i = rainfall category (1 = dry or ‘below normal’; 2 = average or ‘near normal’; 3 = wet or ‘above normal’)

As already mentioned in the preceding section, these probabilities served as weights in the computation of optimal costs of importation and as inputs in determining the optimal volume and timing of importation. The main requirement in the calculation of posterior probabilities is the forecast skill, which is computed based on hit rates.

Hit ratesFrom the viewpoint of the government, it is important to determine whether there is some degree of confidence in predicting the rainfall state for the December–February planting period so that it can make decisions accordingly. For instance, if PAGASA issued a forecast for an increased probability of dry conditions, and the reported hit rate for ‘below normal’ rainfall forecasts is high, then it might be rational for the government to consider increasing its level of importation because a production shortfall is more likely.

Hit rates are derived from the contingency tables of the observed and forecast rainfall. Table 6 shows a schematic of a 3x3 contingency table, indicating the common nomenclature of the individual cells of the table. The letters in the table represent the events from the sample that fit the indicated forecast-observed combination. Hit rates are then calculated based on the entries in the contingency table:

A/D = percentage of correct forecasts of ‘below normal’ rainfall (relative to the total number of observed ‘below normal’ rainfall) = hit rate for ‘dry’ or ‘below normal’

F/H = percentage of correct forecasts of ‘near normal’ rainfall (relative to the total number of observed ‘near normal’ rainfall) = hit rate for ‘average’ or ‘near normal’


K/L = percentage of correct forecasts of ‘above normal’ rainfall (relative to the total number of observed ‘above normal’ rainfall) = hit rate for ‘wet’ or ‘above normal’

Rice supply and demandIn this study, the lean season (which covers the third quarter or July, August, and September) is referred to as the ‘in-season.’ In the Philippines, the lean season is the period when the main crop is growing and potential shortages may be expected. This is perhaps the rationale behind the NFA mandate of positioning a 90-day level of stocks from all sectors (i.e., government, commercial, household) by the end of June, in time for the lean season. This 90-day stock inventory usually comprises any of the following: stocks carried over from the previous season; rice harvested from the second or dry cropping season; and imports. The carry-over stocks include production from the main or wet cropping season, which are harvested during the last quarter of the year. The rice planted during the second cropping season (which is usually from December to February), as noted earlier, is expected to be harvested from March to May (since most rice varieties grow for about 3 to 4 months). The imports, on the other hand, are expected to arrive before the end of June. Because of the new procurement law (known as Republic Act 9184), the importation process usually takes about three months.

Taking all these things into consideration, this study considered December as the ideal time to finalize decisions on importation so as to give enough time for bidding, shipping, internal transportation, and warehouse positioning. All the necessary information needed for importation decisions should be gathered during this period. Meanwhile, since the importation decision is made during December, the ‘preseason’ period is assumed to cover the period January–June.

Observed/forecast rainfall Forecast – below normal

(or ‘dry’)

Forecast – near normal

(or ‘average’)

Forecast – above normal

(or ‘wet’)

Total

Observed – below normal (or ‘dry’) A B C D

Observed – near normal (or ‘average’) E F G H

Observed – above normal (or ‘wet’) I J K L

Total M N O P

Table 6. 3x3 contingency table of the observed and forecast rainfall


Expected supplyExpected supply refers to the expected paddy rice production for the second cropping season (estimated using the regional rice production models) less an amount allocated for nonhuman consumption such as seeds, feeds and wastes, and for nonfood processing. The amount left was then converted to well-milled rice since, as indicated in the preceding chapter, it is the most popular rice variety among consumers nowadays.

Expected demandExpected demand is equivalent to the estimated human consumption from the time the importation decisions are made up to the end of the lean season. The human consumption requirements were derived by multiplying the estimated per capita consumption of rice by the population estimates for the period.

Current stocks heldCurrents stocks held refers to the total quantity of rice stored by all sectors (i.e., government, household, and commercial) when importation decisions are made. Since December is considered as the ideal time for making importation decisions, the available data on stock inventory is the beginning stock inventory for December. It is assumed that the remainder of the harvest from the main or wet cropping season (which is either procured by the NFA or private traders, or kept by farming households for their own consumption) is already reflected in this stock inventory.

Expected net import demandThe expected net import demand reflects the expected rice supply and demand situation in the country. It is positive if the expected level of domestic production plus stocks handled by the three different sectors (i.e., NFA, household, commercial) is not enough to meet the expected domestic demand for rice. It is zero, however, if expected domestic production and stocks exceed expected domestic demand.

Rice import prices and other import-related costsRice import prices

The export price of Thai rice (5% broken) was used as the benchmark for the world price of rice because: (i) Thailand is now the leading rice exporter in the world; and (ii) data on the price of Thai rice (5% broken) are the most readily available among the rice varieties.


The preseason price used was based on the export price of Thai rice (5% broken) for December. The in-season price, on the other hand, was the estimated July price and was dependent on the rainfall state. The in-season rice prices were estimated for three different states. This was done by first grouping the years in which historical data on December and July prices are available based on ENSO state during the December–February planting period. Climate condition during this period greatly affects the supply, and thus the prices, during the lean season. Justifications for the use of ENSO state during the December–February planting period are as follows:(i) The Philippines and Thailand have relatively similar rice cropping

calendars. The planting period for the second or dry cropping season in the Philippines is from December to February, and the harvest period is usually from March to May. In Thailand, the second crop is planted from January to early March (the bulk are from January to February) and harvested from late April up to end of June. Thus, it is assumed in this study that most (if not all) of the rice imported by the Philippines from Thailand during the lean season came from the latter’s produce during the second crop.

(ii) Rainfall state during the planting period in Thailand should have been the basis for the classification of years, but data are not readily available. A good proxy is information on ENSO state during the planting period for the second crop since this is readily available at the National Oceanic and Atmospheric Administration (NOAA) website.

After grouping the years, the percentage differences between the July and December prices were calculated and then averaged per group. The averages for the three groups were then incorporated to the preseason price to calculate the estimates of the in-season prices for three different states.

The preseason and the estimated in-season prices were free-on-board (FOB) prices. Freight and marine insurance were then added to these prices to determine cost, insurance, and freight (CIF) prices. These CIF prices were increased by 50 percent to incorporate tariffs on imports.

Other import-related costsOn top of the rice import prices referred to above, other import-related costs were also included in the model. These include: an interest cost set at a rate of 10 percent per annum; unloading expenses such as trucking and handling from disport up to


the first warehouse (weighted by disport); storage costs; and distribution costs (from first warehouse up to the next warehouse only).6

Storage capacityThe storage capacity was set at a level greater than the net import demand even under the lowest rainfall state (dry) and hence, is nonbinding. The basis of this assumption is that the NFA can always lease additional warehouses when its own warehouses cannot accommodate additional stocks (especially when the total volume of imports is larger than usual).

RESULTS

The economic value of forecasting systemThis section reports on the accuracy of a climate forecasting system and whether it has the potential to influence the import strategies recommended by the RIS model. The simulation results presented here assume the following hit rates: 0.39 for a dry state; 0.38 for an average state; and 0.42 for a wet state. Using these hit rates, the set of optimal import policy strategies under the ‘no forecast’ scenario was found to be the same as those under the ‘with forecast’ scenarios. Under all of these scenarios, the model recommended a level of preseason buying that was equivalent to the net import demand for the wet state. This result may be affected by two things: (1) hit rates under ‘with forecast’ scenarios are not significantly higher than 0.33 and thus, each scenario yields the same set of optimal strategies; and (2) the preseason price is lower than the estimated in-season prices for the two states (i.e., dry and average) and thus, on average, there is an incentive to purchase higher volumes during the preseason than during the in-season.

Consequently, the RIS model showed that the potential importation cost is the same for ‘with’ and ‘without’ forecasts. This means that SCF was found to have no economic value in the context of rice importation. The strategy of only buying sufficient rice preseason to satisfy the minimum level of possible net import demand dominates all other strategies. In other words, the change in the odds of experiencing different states, as reflected in the climate forecast probabilities, is not sufficient to change the rice importation strategy.

6 Major disports in the country are located in the following areas: La Union (for Region I and Benguet); Subic (for Regions II and III, plus the rest of provinces in CAR); Metro Manila (for NCR); Batangas (for the whole Region IV); Tabaco/Legaspi (for Region V); Negros Occidental/Iloilo (for Region VI); Cebu (for Regions VII and VIII); Zamboanga City (for Region IX); Cagayan de Oro (for Region X); Surigao City (for Caraga); and Davao City/General Santos City (for Regions XI, XII, and ARMM).


Sensitivity testingThe most critical parameters in the RIS model that significantly affect the set of optimal import policy strategies and costs are the hit rates, import rice prices (both preseason and in-season), and the net import demand. A series of sensitivity analyses was carried out to further evaluate the effects of these input parameters on the outputs of the model, which include the set of optimal import policy strategies and costs as well as the potential economic value of SCF.

Sensitivity of results to changes in hit ratesThe sensitivity to changing the value of hit rates was assessed first. This provides an assessment of the potential economic value derived from improvements in the accuracy of the forecast. Under the ‘forecast-dry’ scenario at low hit rate ranging from 0.33 to 0.75, the model recommended preseason buying that was equivalent to the net import demand for a wet state. When the hit rate was increased to between 0.80 and 0.85, the recommended volume increased moderately to the net import demand for an average state (which was slightly higher than that for the wet). When the hit rate became closer to 100 percent (i.e., 90–100%), the model recommended preseason buying that was equivalent to the net import demand for dry (which was the highest among all import requirements).

A similar set of observations was found under the ‘forecast-average’ scenario. If the forecast is an average state, the model only bought the net import demand for that state during the preseason if the hit rate was 0.85 or higher. These observations indicate that the government would only buy a higher volume during the preseason (i.e., at least a volume equivalent to the total import requirements for a particular state) and take advantage of a lower preseason price, if there is a high probability of occurrence of a particular state. If the forecast is wet, on the other hand, the model discouraged preseason buying when the hit rate reached 0.80. Recall that the estimated in-season price for a wet state is slightly lower than the preseason price. Once there is a high degree of confidence in the occurrence of a wet state, it would be advisable for the government to buy the volume equivalent to the total import requirements during the in-season, and take advantage of the small difference between the preseason and in-season price.

The sensitivity analysis using hit rates as the input variable revealed that there is a threshold level of forecast accuracy that needs to be obtained for forecasts to be influential in rice importation policy. Specifically, a threshold accuracy level was needed before the forecast started to have a value if hit rates for dry, average, and wet were set to 0.80, 0.85, and 0.80, respectively. Figure 1 summarizes the results. The slopes of the curves in the graph indicate that a marginal improvement in hit rates for dry and wet states gives a higher forecast value compared to that for an average state. This implies that correctly forecasting


Assumptions:Preseason price: PhP24,857.27/MTEstimated in-season prices: PhP28,445.7/MT (dry); PhP27,320.60/MT (average); PhP24,441.11/MT (wet)Estimated net import demand: 2,100,518 MT (dry); 1,752,779 MT (average); 1,671,490 MT (wet)Assumed hit rates: 0.3868 (dry); 0.3823 (average); 0.4231 (wet)

Figure 1. Value of a single forecast for a particular state, at varying hit rates (holding two others constant at assumed hit rates)

the more extreme climate conditions (either a dry or wet state) is more valuable to NFA, which has responsibility for rice importation. Assuming that all of the parameters in the model are reasonable, the above results taken together imply that an increase of about 50 percent in the assumed hit rates is required for the forecast to become useful for the government in terms of achieving efficiency in rice importation.

Sensitivity of results to changes in in-season prices and net import demandIn doing sensitivity analyses using in-season prices and net import demand as input variables, hit rates were fixed at 0.80 because the forecast started to have a value only at this level. Also, a dry forecast was evaluated in price sensitivity analysis while a wet forecast was assessed in sensitivity analysis on net import demand.

To determine the effect of in-season prices on the value of the forecast, the in-season price was varied from PhP28,000 to PhP35,000 while holding prices for other two states constant at their estimated prices (which are both lower than that for dry) and net import demand at their estimated value. Figure 2 summarizes the results of price sensitivity analysis. Generally, increasing the in-season price (away from the preseason price) increases the forecast value. The forecast started to have a value when the in-season price was increased to PhP29,000, which was about PhP4,000 higher than the preseason price. Because the expected level of rice production varied minimally across states, such a gap between the preseason


and in-season price marked the minimum price difference that could induce decisionmakers (who are forecast users) to opportunistically vary the distribution between the preseason and in-season volume of importation, which in turn could result in a positive forecast value. Clearly, as the price difference became larger, the forecast value increased further.7

To determine the effect of net import demand on the forecast value, initial stockholding was varied from 1.0 million to 3.0 million metric tons while holding in-season prices constant at their estimated values. Figure 3 presented the results of sensitivity analysis on net import demand. Based on the results, the forecast value decreased as initial stockholding increased (giving lower net import demand). Because there is a high probability that a wet state will be realized and in-season price for a wet state is relatively lower than the preseason price, in-season buying was recommended. The amount of rice bought in-season falls in line with higher levels of initial stocks. With a constant level of in-season purchases without the forecast, the absolute differences between in-season rice bought with and without the forecast falls. Thus, as initial stockholding increased, the estimated gains from using a forecast decline as there is less and less difference between optimal import levels with and without the forecast.

Figure 2. Value of a dry forecast, at varying in-season prices (using a hit rate of 0.80)

Assumptions:Preseason price: PhP24,857.27/MTEstimated in-season prices: PhP28,445.7/MT (dry); PhP27,320.60/MT (average); PhP24,441.11/MT (wet)Estimated net import demand: 2,100,518 MT (dry); 1,752,779 MT (average); 1,671,490 MT (wet)

PhP0.00

PhP200.00

PhP400.00

PhP600.00

PhP800.00

PhP1,000.00

PhP1,200.00

PhP1,400.00

PhP1,600.00

PhP28,0

00.00

PhP29,0

00.00

PhP30,0

00.00

PhP31,0

00.00

PhP32,0

00.00

PhP33,0

00.00

PhP34,0

00.00

PhP35,0

00.00

In-season price, per MT

SCF

valu

e (m

illio

n)

7 Note that although higher rice prices were associated with higher forecast values in this instance, this may not be the case in general. The reason is that factors, other than the attributes of the forecast, can influence both with and without forecast strategies and hence, lead to variable estimates of the economic value.


PhP0.00

PhP50.00

PhP100.00

PhP150.00

PhP200.00

PhP250.00

PhP300.00

1,000,00

01,20

0,0001,40

0,0001,60

0,0001,80

0,0002,00

0,0002,20

0,0002,40

0,0002,60

0,0002,80

0,0003,00

0,000

Initial stockholding, MT

SCF

valu

e (m

illio

n)

Assumptions:Preseason price: PhP24,857.278/MTEstimated in-season prices: PhP28,445.78/MT (dry); PhP27,320.60/MT (average); PhP24,441.11/MT (wet)

Figure 3. Value of a wet forecast, at varying initial stockholdings (using a hit rate of 0.80)

SUMMARY, CONCLUSIONS, AND RECOMMENDATIONS This study demonstrated an approach that systematically linked SCF to rice importation decisions. In particular, it showed that SCF can be incorporated in rice production models to predict the total supply of rice. Together with the estimated total demand for rice, the rice production models can be used to forecast the required level of importation. Using the net import demand estimates, estimated prices, and SCF, the RIS model was able to determine the optimal level of rice to be imported preseason and in-season. The model was also used to estimate the potential economic value of SCF based on the optimal volume and timing of rice importation under different forecast scenarios.

Initial simulations suggested that the value of SCF is sensitive to both the accuracy of the forecast and attributes of the decisionmaking environment. In this context, assumptions about rice production and prices have key and complex influences on the value of the forecast. The simulation results indicated that there is a threshold for forecast accuracy to have a positive value of SCF. Specifically, forecasts for dry and wet states require hit rates of around 80 percent, while forecasts for an average state require a hit rate of around 85 percent. The simulations also found that skillful forecasts for dry and wet states could translate into relatively higher SCF value than for an average state. Intuitively, this implies that the government would benefit more from correctly forecasting the occurrence of the more extreme climate events.

Aside from hit rates, in-season prices and net import demand were also found to have significant influence on the set of import policy strategies and SCF values.


Price sensitivity analysis indicated that SCF value increased as the difference between the in-season and preseason price of rice became larger. The results also revealed that there is a minimum price difference required to opportunistically vary the distribution between the preseason and in-season volume of importation. On the other hand, sensitivity analysis on net import demand noted that as initial stockholding increased (giving lower net import demand), the estimated gains from using SCF decreased as the difference between optimal import levels, with and without a forecast also decreased.

However, the current versions of the RIS and rice production models have some limitations. The potential value of SCF was estimated at the national level using the RIS model. Because the modeling approach does not reflect how the NFA coordinates regional rice stocks, it might have underestimated the true value of SCF. The NFA has the opportunity to redistribute rice between regions, but this had not been modeled yet. This opportunity adds value because redistribution decisions could be made in response to forecasts. The optimal decision is likely to be climate-sensitive. Meanwhile, further refinement of the rice production models might provide better estimates of supply under different climatic states, and might be influential in refining optimal import strategies both with and without a forecast.

Notwithstanding the aforementioned limitations, these models could already be considered by the government in its import decisionmaking. The rice production estimates of the model could already be used in support of the current practice of forecasting rice production based on planting intentions. Together with these production estimates, SCF could also be incorporated in the RIS model to be able to come up with a set of optimal rice import strategies, which would serve as guide for the government in its rice importation decisions.


* This background information on DSP modeling was drawn from Crean (2009).

APPENDIX

Overview of DSP modeling*The DSP modeling approach is an extension of the standard linear programming (LP) approach with a second-period decision. The general feature of the two approaches is the two-stage model formulation where time is simply divided into ‘present’ and ‘future.’ While the order of events in the static LP model is in x → s format, the format in the DSP model is x → s→ z(s, x), where: x is a vector of stage 1 decisions; s is the state of nature; and z(s, x) is a set of stage 2 decisions, contingent upon earlier stage 1 decisions and the state of nature.

The objective function of the DSP model is composed of returns in each of these stages and can be written as:

In the first equation, cTx is the stage 1 return, which is a product of a net

revenue vector (cT) and the level of stage 1 activities (x). The decisions are made prior to determination of the state of nature. The constraints relating to stage 1 are the same as those for the general LP model and concern a matrix of resource requirements (A) and a vector of resources (b). The stage 2 return function (also known as ‘recourse function’), EsQ(x,s), on the other hand, is an expected return which is summed over all states. The return in each state is a product of a stage 2 net return vector (eT) and the level of stage 2 activities (zs). The probability of each state (πs) gives different weights to the outcomes in each state. The constraints of stage 2 consist of a technology matrix (Tsx), a recourse matrix (Ws), and a vector of state-contingent resource supplies (bs).


REFERENCESAbedullah and S. Pandey. 2007. The value of rainfall forecasts in the rainfed rice

areas of the Philippines. The Lahore Journal of Economics 12(2):69–81.Amien, I., P. Redjekiningrum, B. Kartiwa, and W. Estiningtyas. 1999. Simulated

rice yields as affected by interannual climate variability and possible climate change in Java. Climate Research 12:145–152.

Bureau of Agricultural Statistics (BAS). 2008. Metadata of BAS statistics [online]. http://countrystat.bas.gov.ph. [Accessed 10 July 2008].

Chatfield, C. 2000. Time-series forecasting. Boca Raton, FL: Chapman and Hall/CRC.

Clewett, J.F., O.S. Hwang, D.T. Owens, D.A. George, and I.J. Partridge. 2002. Rainman International Version 4: an integrated software package of rainfall information for better management. Queensland: QI02023, Department of Primary Industries.

Climate Prediction Center, National Centers for Environmental Prediction (NCEP), National Oceanic and Atmospheric Administration (NOAA), United of States Department of Commerce (USDA). 2009. ENSO cycle: recent evolution, current status, and predictions. 15 June.

Crean, J. 2009. The economic value of seasonal climate forecasts in Central West New South Wales (NSW). Doctoral thesis (unpublished). New South Wales, Australia: Charles Sturt University, Orange.

Das, H.P., D.M. Rase, and M.P. Shewale. 2006. Dependence of rice yield in Orissa on rainfall and sea surface temperature anomalies over Equatorial Pacific SST. Vayumandal 32 (1–2):28–32.

David, C.C. 2006. The Philippine hybrid rice program: a case for redesign and scaling down. Research Paper Series No. 2006-03. Makati City, Philippines: Philippine Institute for Development Studies.

David, C.C. and A.B. Inocencio. 1995. Comparative and competitive advantage of Philippine rice production: 1966–1990. PIDS Discussion Paper Series No. 95-03 (October). Makati City, Philippines: Philippine Institute for Development Studies.

David, W.P. (n.d.). El Niño Southern Oscillation (ENSO): its impact on Philippine agriculture.

Dawe, D., P. Moya, and C. Casiwan (Editors). 2006. Why does the Philippines import rice?: Meeting the challenge of trade liberalization. Los Baños, Laguna: International Rice Research Institute (IRRI) and Science City of Muñoz, Nueva Ecija, Philippines: Philippine Rice Research Institute (PhilRice).

Delos Reyes, M.L.F. and W.P. David. 2009. The effect of El Niño on rice


production in the Philippines. The Philippine Agricultural Scientist 92(2):170–185. (June).

Environmental Systems Research Institute, Inc. (ESRI). 2001. Using ArcGIS™ Spatial Analyst [online]. http://www.gis.unbc.ca/help/software/esri/Tutorials/Using_ArcGIS_Spatial_Analyst_Tutorial.pdf. [Accessed August 2009].

Falcon, W. P., R. L. Naylor, W. L. Smith, M. B. Burke, and E. B. McCullough. 2004. Using climate models to improve Indonesian food security. Bulletin of Indonesian Economic Studies 40(3):355–377.

Foreign Agricultural Service (FAS), United States Department of Agriculture (USDA). 2009. Thailand: drought affects 2004/05 rice crop. [online]. http://www.fas.usda.gov/pecad/highlights/2004/12/Thai04/index.htm. [Accessed 5 May 2009].

Frechtling, D.C. 1996. Practical Tourism Forecasting. Linacre House, Jordan Hill, Oxford: Butterworth-Heinemann.

Fraisse, C.W., J.O. Paz, and C. M. Brown. 2007. Using seasonal climate variability forecasts: crop yield risk. CIR1498 (January). Florida: Institute of Food and Agricultural Sciences (IFAS), University of Florida.

Hogg, R. V. and A.T. Craig. 1995. Introduction to mathematical statistics (5th ed.). Upper Saddle River, New Jersey: Prentice Hall, Inc.

Ignacio, J. 2005. Trading-off Philippine rice: implications of the proposed removal of quantitative restriction (QR). Quezon City: Integrated Rural Development Foundation of the Philippines.

Intal, P. and M. Garcia. 2005. Rice and Philippine politics. PIDS Discussion Paper Series no. 2005-13. Makati City, Philippines: Philippine Institute for Development Studies.

International Research Institute for Climate and Society (IRI). 2008. IRI seasonal climate forecasts [online] http://portal.iri.columbia.edu/portal/server.pt?open=512&objID=944&PageID=0&cached=true&mode=2&userID=2. [Accessed August 2008].

ITSMF-NL. 2006. Metrics for IT service management. Zaltbommel, Netherlands: Van Haren Publishing.

Kajisa, K. and T. Akiyama. 2003. Evolution of rice price policies over four decades: Thailand, Indonesia, and the Philippines. Tokyo, Japan: Foundation for Advanced Studies on International Development.

Kaul, S. (n.d.). Bioeconomic modelling of climate change on crop production in India. New Delhi, India: Indian Agricultural Statistics Research Institute.

Kutner, M.H., C.J. Nachtsheim, and J. Neter. 2004. Applied linear regression models (4th ed.). New York: McGraw-Hill/Irwin.


Lansigan, F.P. 2004. Coping with climate variability and change in rice production systems in the Philippines. Rice is life: scientific perspectives for the 21st Century. Proceedings of the World Rice Research Conference held in Tsukuba, Japan, 4–7 Novermber.

Lawrence, D.B. 1999. The economic value of information. New York: Springer-Verlag, Inc.

Lewis, C.D. 1982. Industrial and Business Forecasting Methods (2nd ed.). London: Butterworths.

LINDO Systems. 2008. “What’s Best!” [online] http://www.lindo.com. [Accessed 25 July 2008].

McCarl, B.A. and T.H. Spreen. 2007. Applied mathematical programming using algebraic systems. [online]. http://agecon2.tamu.edu/people/faculty/mccarl-bruce/mccspr/newb.pdf. [Accessed May 2008].

Naylor, R.L., W.P. Falcon, D.Rochberg, and N.Wada. 2001. Using El Niño/Southern Oscillation climate data to predict rice production in Indonesia. Climate Change 50:255–265.

Ramos, C.G. 2000. State intervention and private sector participation in Philippine rice marketing. Quezon City, Philippines: Management and Organizational Development for Empowerment, Inc.

Reburiano, M.P. 2005. NFA rice importation. Presented during the First Consultation Meeting ACIAR-funded project Bridging the gap between seasonal climate forecasts and decisionmakers in agriculture, 8 July, NEDA Building, Makati City, Philippines.

Recide, R.S. 2005. Inter-Agency Committee on Rice and Corn (IACRC). Presented during the first consultation meeting on ACIAR-funded project Bridging the gap between seasonal climate forecasts and decisionmakers in agriculture, 8 July, NEDA Building, Makati City, Philippines.

Reyes, C.M., S.N. Domingo, C.D. Mina, and K.G. Gonzales. 2009. Policy options for rice and corn farmers in the face of seasonal climate variability. PIDS Discussion Paper Series No. 2009-11. Makati City, Philippines: Philippine Institute for Development Studies.

Reyes, C.M. and C.D. Mina. 2009. Profitable use of SCF in a policy context: the case of rice stockholding in the Philippines. PIDS Discussion Paper Series No. 2009-09. Makati City, Philippines: Philippine Institute for Development Studies.

Rockafellar, R.T. 2001. Optimization under uncertainty. Lecture Notes. University of Washington.

Sarma, A.A.L.N., T.V. Lakshmi Kumar, and K. Koteswararao. 2008. Development of an agroclimatic model for the estimation of rice yield. Journal of the Indian Geophysical Union 12(2):89–96 (April).


Satake, T. and S. Yoshida. 1978. High temperature induced sterility in Indica rices at flowering. Japanese Journal of Crop Science 47:6–17.

SCF Project Team. 2005. Understanding the ENSO phenomenon and its implications. Economic Issue of the Day 5(1). Makati City, Philippines: Philippine Institute for Development Studies.

Tao, F., M. Yokozawa, J. Liu, and Z. Zhang. 2008. Climate-crop yield relationships at provincial scales in China and the impacts of recent climate trends. Climate Research 38:83–94.

Tolentino, B., B. de la Peña, E. Noveno, B. Rayco, and I.L. Villapando. 2002. 101 facts about the NFA. DA-DOF-ADB TA 3429. 10 May.

Unnevehr, L.J. 1985. The costs of squeezing marketing margins: Philippine government intervention in rice markets. The Developing Economies 23(2):159–172.

Verceles, E.R. 2005. Basics of El Niño Southern Oscillation (ENSO). Presented during the Pulong Saliksikan, 21 April, NEDA Building, Makati City, Philippines.

Zubair, L. 2002. El Niño Southern Oscillation influences on rice production in Sri Lanka. International Journal of Climatology 22:249–260.

Date post:	24-Sep-2020
Category:	Documents
Upload:	others
View:	0 times
Download:	0 times

Incorporating Regional Rice Production Models in a ... · and relative price of rice. Because 1991...

Documents