Study1: Supply and Demand for Cereals in Nepal, 2010-2030

Reference: Prasad, Pullabolta and Ganesh-Kumar. Supply and Demand for Cereals in Nepal, 2010-2030. IFPRI. September 2012. Link

Motivation: The study attempts to estimate supply and demand models for the 3 most important  cereals: rice, wheat, and maize.

Methods Overview:

1) Supply projection - based on a single-crop production function (using data from 1995-2008)

2) Demand function/projection - based on the data from NLSS II (2003-04)

Methods in Detail:

1) SUPPLY PROJECTION

A. Building a model:

• Model: Three single-crop production functions
  
• Model specification:
  
• Y: crop production amount (tons)
• Xs:
• area under  crop (A)
• improved seed quantity (S)
• gross irrigated area (I)
• total fertilizer supply (F)
• annual average rainfall (R)
• Functional form: Double-log Cobb-Douglas production function models is prepared for each of the 3 crops
• Multicollinearity & Heteroskedasticity:
• For each, Durbin-Watson d-statistic was used to test for autocorrelation, and if AC present, the regression model was estimated through a Prais-Winstein regression
• Final production function:
• Only significant variables were retained for each model and the model was re-estimated

B. Use the model for projection:

• Build auxiliary models to first build the future values of X-variables; then use these to predict Y
• Identify policy variables
• Variables that the government can manipulate - such as  irrigation level, fertilizer supply and seed supply)
• Since these are manipulated by the government, we can assume several scenarios and determine future X-values accordingly.
• Business as usual: We can examine the growth rate and assume this growth rate will continue (Business as Usual); for this we need to regress each of the three policy variables with time variables (years) and calculate growth rate (b-coeff in double log auxiliary regression - b/c log allows us to interpret in % terms)
• Other growth rate assumptions: We can also assume certain growth rate and use that in the model.
• Identify behavioral variables
• Variables that depend on farmer choices and preferences, such as crop acreage
• Here, we can regress crop acreage with other X-variables and see which ones have what kind of influence on farmer's preference of how many acres of crop to plant (X-vars include - irrigation, rainfall, fertilizer, seed supply)
• Now predict Y
• Now, determine which years you want to forecast for: 2010, 2015, 2020, 2025 and 2030
• Now use the time trends of policy variables --> predict crop acreages (behavioral variable)
• Then use crop acreages & policy variables --> predict Ys
• This can be done for several scenarios:
• Business as Usual
• Pessimistic
• Optimistic

2) Demand Projection

• Theory/background:
• Total domestic demand for 3 cereals consists of:
• Direct demand by households
• Indirect demand for processing (ex: seed, feeds, wastage)

• To determine direct demand by households
• Use Household demand model - Linear Approximate - Almost Ideal Demand System (AIDS) Model
• Formulation:
• Start with AIDS model 
• Use linear price index (Stone price index) to use LA-AIDS
• Parameters
• Y: wi = % of HH expenditure on food items
• pi = price of i good
• y = total expenditure on all goods
• P = price index
• Estimate the model
• Use seemingly unrelated regression (SUR) procedure
• B/c of possible cross-equation correlations
• Compute elasticities 
• Expenditure elasticity
• Price elasticity
• To determine indirect demand for processing
• Use Food budget model 
• B/c AIDS model captures % expenditure on food items only and not the budget itself
• Estimate seed, feed and wastage 
• Data:
• Nepal Living Standards Survey (Household survey)