THES’S

LIBRARY
Michigan State
University

This is to certify that the
thesis entitled

DYNAMICS OF AGRICULTURAL WAGES AND RICE PRICES
IN THE PHILIPPINES

presented by

MARIE-CHRISTINE D. LASCO

has been accepted towards fulfillment
of the requirements for the

MS. degree in AGRICULTURAL
ECONOMICS

 

24an VI .BMTM

Major Professor’s Signature

Dearth“, (3,. 2055

Date

MSU is an Affinnative Action/Equal Opportunity Institution

 

PLACE IN RETURN Box to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE

DATE DUE

DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2105 c:/ClRC/DateDue.indd-p.15

 

DYNAMICS OF AGRICULTURAL WAGES AND RICE PRICES
IN THE PHILIPPINES

By

Marie-Christine D. Lasco

A THESIS
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
MASTER OF SCIENCE

Department of Agricultural Economics

2005

ABSTRACT
DYNAMICS OF AGRICULTURAL WAGES AND RICE PRICES
IN THE PHILIPPINES

By

Marie-Christine D. Lasco

Rice trade liberalization in the Philippines will likely cause domestic rice prices to
decrease. This may have a significant impact on agricultural wages and agricultural wage
earners. This study examines the long-run (LR) and short-run (SR) relationship of
agricultural wages and rice prices in the Philippines using a neoclassical wage
determination model. Two frameworks are used in the analysis: A co-integration/error
correction framework which assumes nonstationarity of model variables and an OLS
framework that assumes that the model variables are stationary. The results show that in
the short run, wages adjust partially with a SR elasticity of 0.33-0.41, while in the long
run the adjustment is greater. An analysis of welfare implications for various
groups of agricultural wage earners suggest that most households will benefit from a
decrease in rice prices, although immediate mitigation measures are needed by adversely

affected households that are heavily reliant on agricultural wages for income.

ACKNOWLEDGEMENTS

This project would not have been possible without the following people: Dr.
Richard Bernsten (Major Professor), Dr. Robert Myers (Thesis Advisor), Dr. Eric
Crawford and Dr. John Giles of Michigan State University; Dr. David Dawe and the staff
of the Social Sciences Division at the International Rice Research Institute and the very
helpful staff of the Bureau of Agricultural Statistics.

I am very grateful for my professors and colleagues at the Department of
Agricultural Economics at Michigan State University, and my friends Cathy, Courtney,
Cristobal, Esteban, Josh, Kyunghyun, Kirimi, Neslie, Ricardo, and Valeria for all their
support and encouragement.

I would also like to thank my family for their constant love and encouragement.
Most importantly, I would like to thank God, who gave me gave me strength and enabled
to finish this project, and blessed me with friends and experiences that I will always

cherish.

iii

TABLE OF CONTENTS

LIST OF TABLES .............................................................................................................. v
LIST OF FIGURES ........................................................................................................... vi
1. Introduction ..................................................................................................................... 1
2. Empirical studies on wage determination and agricultural wage response .................... 3
3. Methodology ................................................................................................................... 6
4. Empirical Results .......................................................................................................... 15
A. Unit Root Tests .................................................................................................... 15
B. Dynamics Under Nonstationarity ......................................................................... 21
Long-run Wage Response ......................................................................................... 21
Error Correction Model ............................................................................................. 23
First Difference Model .............................................................................................. 24
C. Dynamics Under Stationarity ............................................................................... 26
D. Welfare Implications ............................................................................................ 29
5. Summary and Conclusions ........................................................................................... 31
REFERENCES ................................................................................................................. 34

LIST OF TABLES

Table 1. Unit root tests ...................................................................................................... 15
Table 2. Engle-Granger test for co-integration with constant ........................................... 22
Table 3. Wage elasticities and adjustment under different model specifications ............. 28
Table 4. Welfare effects on agricultural wage earners of a reduction in rice price .......... 30

LIST OF FIGURES

Figure 1. Natural log of agricultural wages in levels (1975-2003) ................................... 16
Figure 2. Natural log of nominal rice prices in levels (1975-2003) .................................. 17
Figure 3. Natural log of nominal corn prices in levels (1975-2003) ................................ 17

Figure 4. Natural log of Consumer Price Index (1994:100) in levels (1975-2003) ........ 18

Figure 5. Natural log of agricultural area in levels (1975-2003) ...................................... 18
Figure 6. Natural log of agricultural wages in differences (1975-2003) .......................... 19
Figure 7. Natural log of nominal rice prices in differences (1975-2003) ......................... 19
Figure 8. Natural log of nominal corn prices in differences (1975-2003) ....................... 20

Figure 9. Natural log of Consumer Price Index (1994:100) in differences (1975-2003) 20

Figure 10. Natural log of agricultural area in differences (1975-2003) ............................ 21

vi

1. Introduction

The relationship between agricultural wages and staple food prices is an important
empirical issue. Agricultural wages influence rural welfare, especially the welfare of the
poorest groups who rely heavily on these wages for income. In the Philippines, rice is the
staple food of 80% of the population, and rice farming is the largest agricultural sector.
Thus, it is very likely that changes in rice prices will have a significant impact on
agricultural wages. Rice prices in the Philippines have been largely determined by the
government, so it is important to inform policymakers about the impacts of rice price
movements on agricultural wages. This issue becomes even more relevant because of the
country’s commitment to the WTO to liberalize rice trade. Under the Agreement on
Agriculture, all Quantitative Restrictions» (QRs) on rice imports have to be removed and
tariffied. Reduction in tariffs is also expected to follow. Because domestic rice prices are
significantly higher than world prices, rice trade liberalization is expected to cause a
decrease in the price of rice. The removal of QRs was supposed to have taken effect on
January 1, 1995. However, the Philippines, along with Japan and South Korea has been
granted concessions to extend this deadline for at least 10 years (Cororaton, 2004, David,
1997). As of August 2005, the Philippines has yet to implement the removal of QRs. The
government’s hesitation to open up rice trade may be caused in part by the perceived
negative impact to rice producers and agricultural laborers, although it is clear that urban
consumers will benefit from the decrease in rice prices. In order to inform this policy
decision, it is important to analyze of the impacts of a decrease in rice prices to various

sectors in the economy. This study will focus on the effect of a decrease in rice prices on

agricultural wage earners by looking at how changes in rice prices affect the agricultural
wage rate.

There is some contention regarding the direction and magnitude of wage response
to changes in rice prices in the Philippines. Several studies in Bangladesh have found a
positive relationship between agricultural wages and rice prices (Boyce and Ravallion,
1991, Ravallion, 1994, Palmer-Jones, 1993, Palmer-Jones and Parikh, 1998). The general
hypothesis of these studies is that the relationship is positive with partial adjustment in
the short run and full adjustment in the long run. The reason for this is that an increase in
price will generally increase incentives for farmers to engage in rice production. This
could lead to a greater demand for factor inputs, including labor, which causes wages to
increase. Similarly, if prices decrease then this may lead to a decrease in factor demand
and wages. However, there are also some studies which show that agricultural wages and
rice prices have no relationship or a negative relationship. A recent study in Bangladesh
by Rashid (2002) found that agricultural wages and rice prices have no long run
relationship. In the Philippines, Dawe (2003) argued that a decrease in the price of rice
will exert upward, rather than downward pressure on agricultural wages. He said that a
decrease in rice prices will not decrease demand for labor in the agricultural sector
because farmers will diversify to other crops such as vegetables. Since vegetables are
more labor intensive than rice, there may even be a higher demand for labor and which
may lead to increased wages. Moreover, he contends that the effect of a decrease in rice
prices will primarily affect land rents (as opposed to wages and prices of tradable inputs)
received by landowners of rice farmlands. This is because farmers have little control over

prices of tradable inputs like fertilizer and pesticides which are imported. Thus, the

adjustments must be absorbed by either land rents or agricultural wages. Since land is the
more inelastic input, it will be affected more than wages.

This study informs this debate by explicitly looking at the relationship between
agricultural wages and rice prices, both in the long-run and the short-run. The developing
country literature on the subject is limited and no previous empirical study has been
conducted to measure agricultural wage response to changes in rice prices in the
Philippines.

This study addresses the general research question of how agricultural wages are
affected by changes in the price of rice. Specifically, this study 1) tests whether there is a
long-run relationship between the agricultural wage rate and the price of rice in the
Philippines and 2) estimates the long-run and short-run elasticities of the wage rate with
respect to the price of rice. Knowledge of short-run and long-run agricultural wage
responses will provide insight into the direction, magnitude and persistence of the wage
adjustment. This will aid in the analysis of the impacts of rice trade liberalization on the
Philippines’ rural sector. In this paper, the empirical findings are used to assess welfare
implications of a decrease in rice price to agricultural wage earners in the Philippines

who are net suppliers of agricultural labor.

2. Empirical studies on wage determination and agricultural wage response

Although the literature on wage response to changes in rice prices in the
Philippines is limited, several studies have sought to measure wage response to changes
in rice or other food prices in other developing countries. Boyce and Ravallion (1991)

(hereafter BR), conducted a study on wage determination in Bangladesh using annual

data from 1949/50 to 1980/81. BR assumed a long-run equilibrium relationship among
the nominal agricultural wage and the nominal prices of rice, jute and com, the
manufacturing wage, and a productivity index. From this they estimated an error
correction model (ECM) to analyze the short-run (SR) wage adjustment. They found
that wages responded significantly to rice price in the SR with an elasticity of 0.22. Based
on the SR error correction model, BR derived a model for the long-run (LR) wage
adjustment mechanism and reported a LR elasticity of 0.47 for the nominal agricultural
wage and 0.53 for wage deflated by the price of rice.

Palmer-Jones (1993), however, argued that the previous BR model failed
prediction and stability tests. Extending the BR data set to 1949/50 to 1989/90, he
proposed a different model wherein nominal agricultural wage is a function of its own
lagged value, nominal prices of rice and (lagged) jute and (lagged) manufacturing wages.
Moreover, he added a dummy variable for the period 1972-19741 and an ad-hoc time
trend which takes on the value of time from 1949-1950 to 1964, and zero onwards. In a
rebuttal, Ravallion (1994) criticized Palmer-Jones’ modeling technique, claiming that the
model is not homogenous, i.e., the implicit long run real wage rate depends on real and
nominal variables. Furthermore, he questioned the use of a sub-period dummy and a half-
time trend because this imposes a model structure that is not appropriate for the case of
Bangladesh. Interestingly, Palmer-Jones arrived at a similar result (0.22) for the short-
run elasticity of agricultural wages with respect to rice price. Moreover, the long-run
elasticity reported (0.46) was also close to that of BR’s.

New developments in econometric time series analysis have shown that many

 

' According to Palmer Jones, this was a period of great disruption and inflation following independence and
the associated national and political disturbances in Bangladesh.

macroeconomic data, such as income and prices, appear to be nonstationary. If variables
are nonstationary and not co-integrated, classical regression, as used by Palmer-Jones,
leads to spurious relationships among variables and inconsistency of parameter estimates
(Pindyck and Rubinfeld, 1998). In the case of the BR study, a co-integrating relationship
must first be established among the variables before an ECM can be used. Techniques
based on a co-integration framework have been developed to analyze the relationships
among nonstationary time series variables. More recent studies have adopted these
methodologies.

Rashid (2002) re-analyzed the first two studies by BR and Palmer-Jones, using a
co-integration framework. Rashid pointed out that tests on the variables used in the
previous studies showed that all variables were nonstationary, and therefore the use of
classical regression was inappropriate. He also pointed out that the models used in both
studies face an identification problem. Although these studies assumed a priori that all
right-hand side variables were exogenous, a weak exogeneity test showed that rice price
cannot be exogenous. Furthermore, Rashid argued that the inclusion of the agricultural
productivity variable (measured as an index of per acre production) in the BR model was
inappropriate because prices and wages are inflationary while productivity is not.
Analyzing the same data used in the two studies using co-integration techniques, Rashid
reported higher estimates of SR and LR elasticities for both models. The short-run
elasticities of wage to rice price were 0.32 and 0.25 for the BR and Palmer-Jones models
respectively. Moreover the corresponding long-run elasticities were 0.72 and 0.69.

In the same study, Rashid also analyzed data from 1976/77-1998/99, which he

called 'post-famine' data. He hypothesized a long-run relationship among nominal

agricultural wage, rice price, and the urban wage rate. However, he found that in the post-
famine period, rice prices and wages were no longer co-integrated (i.e. there is no stable
LR relationship between them).

Palmer-Jones and Parikh (1998) also applied co-integration techniques to the data
set from the 1993 Palmer-J ones study. From a wage determination model, they estimated
the nominal agricultural wage-rice price SR elasticity to be 0.11, while the LR elasticity
was 0.48.

Datt and Olmsted (2004) used a Generalized Method of Moments framework to
measure short and long-run elasticities of nominal agricultural wages to food prices in
Egypt. Using panel data from 18 governorates from 1976-1993, they found that over a
16-month period, the SR elasticity of wages to food prices was only 0.27. Moreover, it
took up 5-7 years for wages to adjust to 90-95% of the food price increase.

Although there are differences in the literature about the appropriate model and
estimation procedure, almost all studies point to a significant but sluggish response of
agricultural wages to changes in rice price in the short-run. In the long-run, most studies

reported a greater wage adjustment, although the LR elasticity is still less than one.

3. Methodology

The empirical model is derived based on the neoclassical utility maximizing
framework for a representative farm household. In a study of wage determinants and
labor supply in rural India, Rosenzweig (1980,1984) argued that the neoclassical
competitive framework is appropriate, even for developing countries. Using district and

household level data, he found rural wage rates to be endogenously determined and

highly responsive to changes in labor supply. A similar study in the Muda River Valley,
a farming community in Northwest Malaysia also found evidence for competitive labor
markets in rural areas (Barnum and Squire, 1979).

This model assumes that agricultural households derive utility from the

consumption of three goods: an agricultural commodity produced by the household (Xa ),

a market purchased good (X m ), and leisure (XI ). The household chooses a consumption
level that maximizes overall utility subject to income, time and production constraints
(Equation 1):

Max U=U(Xa,Xm,X,) s.t. Fax, +Pme = W(T-x.)+n
Xa,Xm,Xl,L

where:

Pm — nominal retail price of market goods
Pa — nominal farm price of household-produced agricultural commodity

W —- nominal agricultural wage rate

T —- total time available to household

['1 = Pa Q(L,A,Z) - WL

Q(L,A,Z) —— production function of agricultural commodity
L — total labor input allocated to farm production

A — area devoted to agricultural production

Z — other non-labor, non-land inputs

The household is a price taker in the markets for Xa and Xm, Price-taking
behavior is plausible even for X8 because most agricultural commodities are also

imported and/or subject to price controls. Family labor and hired labor are assumed to be
perfect substitutes and all labor is valued at W, the agricultural wage rate. It is assumed
that there is separability of production and consumption decisions. That is, households
make production decisions to maximize profit and then choose consumption based on

realized income from their production decision. Furthermore, it is assumed that urban

and rural markets are separate, i.e. non agricultural wages do not affect the decision to
supply agricultural labor. This is reasonable in the case of the Philippines where there is
significant unemployment in the urban and rural areas. In 2002, the urban unemployment
rate was 13.2% while the rural unemployment rate is 7.3% (National Statistics
Coordination Board, 2002). This means that all jobs in the urban areas are taken by
urban laborers before they become available to rural laborers. Moreover, in many rural
areas in the Philippines, labor is not mobile from rural to urban areas.

To maximize profit, a household chooses a level of labor and inputs such that the
marginal revenue product of labor and other inputs are equal to the wage rate and other
input prices respectively. In this model only labor is considered as variable. In the
Philippines, this stylization is reasonable because the major input to agricultural
production is labor. For example, in rice farming, 50% of production cost is for labor.
The second major input is fertilizer which is 16% of production cost (2004, International
Rice Research Institute- Social Sciences Division).

Taking the derivative of (l) with respect to L, the labor demand from the
agricultural household is derived:

L* = L*(W, Pa , A, Z) (2)

Substituting optimal profit [I * into (1) and taking the derivative with respect to
X., the demand for leisure is obtained:

Xl* =X.*(Pm, Pa, W, T, IT“) (3)

In equilibrium under a single representative household, L = T- XI. Imposing this
restriction and using (2) and (3) gives:

F(W,Pa,A,Z,Pm)=O (4)

which defines the long-run equilibrium relationship among the variables in the systemz.

Note that I'I* = II(W, Pa, A, Z) so (4) completely characterizes the equilibrium

relationship between wages and prices.

In the econometric model, the Consumer Price Index is used to reflect Pm . Pa is

represented by the nominal prices of rice (Pr) and corn (PC ), Which are the primary

agricultural crops in the Philippines. Since there are no data to quantify other non-labor,
non-land inputs, a time trend (t) is used to capture changes in Z, as well as other time-
varying omitted variables, such as labor productivity. Define X as the vector of variables
in the equilibrium relationship, that is:

X=(W,Pr,PC,Pm,A,t) (5)
The empirical relationship among these variables is examined in both the long run and in
the short run.

Annual time series data from 1975-2003 are used in this analysis. Data on
agricultural wages, rice and corn prices and agricultural area are from various surveys
conducted by the Bureau of Agricultural Statistics (BAS) in the Philippines. Wages are
obtained by taking the weighted average of wages received by rice, corn, coconut and
sugar laborers who do not receive meals. Weights are determined by the size of
production of the abovementioned commodities in the province. Since there are
differences in wages across provinces, the provinces which produce more of a given
commodity are given more weight. Rice and corn prices are national annual averages
computed from daily and monthly prices for different provinces in the Philippines. Area

is computed by summing the land areas planted to rice and corn. The data on Consumer

 

2 T is assumed to be a constant and was therefore excluded from the relationship.

Price Indices are from various volumes of the Philippine Statistical Yearbook, published
by the National Statistics Coordination Board (N SCB).

If the variables in (5) are covariance stationary then the short-run and long-run
relationship between wages and prices can easily be estimated using OLS, assuming that
the following reduced form relationship exists:

W=f(Pr,PC,Pm,A,t). (6)
The assumption of exogeneity of the right-hand side variables (6) is a fairly strong
assumption. Due to the lack of available instrumental variables, the endogeneity of each
of the explanatory variables was not tested. This limitation should be noted when
interpreting the results of this model under the assumption of stationarity.

It is also possible that some of the variables in X may have unit roots (i.e., are
integrated of order one or 1(1)) and that there exist a long-run equilibrium among these
1(1) variables. To evaluate this hypothesis, the Augmented Dickey-Fuller (ADF) and
Phillips-Perron tests are used. The ADF test is the most commonly used procedure for
testing for unit roots. The Phillips-Perron test is similar to the ADF test, but allows for
more relaxed assumptions on the error term. (i.e., it does not assume that errors are
uncorrelated with constant variance.) The following equation is used to implement the
ADF:

m
Ayt=a0+}y,_] +all+z 6Ay,_,-+e, (7)
i=1

where y, is the variable being tested for unit root, A is the first difference operator, t is a

time trend and e, is the residual term.

10

The null hypothesis of the test is 7 =0 (nonstationarity). OLS is used to estimate
(7) and the 1 statistic associated with the y parameter is compared to the critical value in

the Dickey-Fuller tables to determine if the null hypothesis can be rejected. The lag
length m is chosen by adding additional lags until no more autocorrelation is found in the
residuals. The procedure outlined by Enders (2004) for determining whether to include a
constant and a time trend in (7) is used. This method involves estimating the most general
form which includes a constant and a time trend and sequentially imposing restrictions
that test whether a regressor can be excluded from the equation.

The classical approach to dealing with nonstationarity is to difference the data and
then use OLS estimation. However, differencing removes important information about
the long-run relationship among the levels of the variables. In order to capture this
information, a co-integration framework can be used to analyze the relationship among
these variables. If the variables are indeed co-integrated then, consistent with the Granger
Representation Theorem, an error correction model (ECM) can be used to analyze the
short-run adjustment.

Co-integration implies a long-run equilibrium relationship among I(l) variables,
although in the short-run there may be substantial deviations from this long-run
equilibrium. Engle and Granger (1987) give a formal definition of co-integration: “The

components of the vector x, are said to be co-integrated of order d,b, denoted xt~CI(d, b)
if (i) all components of x, are I(d); (ii) there exists a vector a(¢0) so that zt = a'xt ~ I(d—b),

b>0. The vector is called the co-integrating vector.” The co-integrating vector (CV) is not

unique. A scalar combination of the CV is also a co-integrating vector. Moreover there

11

may be other equilibrium relationships among a subset of the variables in the co-
integrating regression.

Engle and Granger have developed a simple two-step procedure for estimating
the co-integrating vector and formulating the corresponding ECM. This involves
estimation of the co-integrating regression by OLS and incorporating lagged errors of this
regression into a differenced equation. However, there are some limitations to this
method. First, the results of the co-integration test may be sensitive to the choice of
dependent variable, although asymptotically the results should be consistent. Secondly,
the method does not allow for determining if there is more than one co-integrating vector.
Lastly, because a two-step procedure is used, it is possible that errors in the first step will
be carried over to the next step. For example, an estimation bias in the parameters of the
co-integrating regression could be transmitted to the regression that tests for co-
integration (Davidson and MacKinnon, 1993, Enders). Johansen (1988) developed a
method based on the estimation of a multivariate Vector Autoregression (VAR) by
maximum likelihood which overcomes these limitations. However, an undesirable
characteristic of the VAR is that it requires a lot of parameters to be estimated. If the
VAR were to be used is this analysis, the very low power of the test will not yield useful
results. Due to the fact that only a limited number of observations are available, the two-
step procedure proposed by Engle and Granger is used in this study. Some of the
limitations mentioned above are addressed by testing for co-integration using each of the
model variables as the dependent variable. This evaluates whether the results are robust

to the choice of normalization variable.

12

The first step in the Engle-Granger methodology is to use OLS to estimate a
model containing the variables that are hypothesized to be co-integrated. If the variables
are not co-integrated, then OLS leads to ‘spurious regressions’, where the regression
results are characterized by a high R-squared and significant t-values, but the relationship
has no economic meaning. However, an interesting result of OLS estimation is that if the
variables are co-integrated, then the parameters are super consistent estimators of the
long-run equilibrium relationship. This means that the parameters converge to their true
value faster than OLS estimates involving stationary regressors. However, R-squared and
standard errors are still misleading, and drawing any statistical inference about the
parameters from conventionally estimated standard errors is inappropriate.

A log-log functional form is used in the model. Since wage response is the focus
of this paper, wage is assigned to be the dependent or normalizing variable. The equation

to be estimated is:
W! = .30 +1811?” +182Pcr +:63.l7mt +540: +fl5t+51 (8)
where lower case italicized letters represent natural logarithms of the model variables.

If the variables in (8) are co-integrated, then the ,Bs can be interpreted as long-run

elasticites.

To determine if the variables in (8) have a co-integrating relationship, the
residuals from the regression are tested for stationarity using the Engle-Granger test for
co-integration. The following equation is used for this test:

k
Aé, =boé,_1+Zb,-Aé,_,~+v, (9)

i=1

where .53, is the estimated residual from estimation of (8).

13

The null hypothesis is b0 = 0 (nonstationarity). If autocorrelation is found in v, ,
then successive lags of Aé,_1 are included in the equation. The test above is similar to

the ADF test, but in this case, standard Dickey-Fuller tables cannot be used for
hypothesis testing. This is because the test is based on estimated residuals, and this makes
the procedure biased towards stationarity. Several tables of critical values, which correct
for this bias, have been developed. Here, the values presented by Davidson and
MacKinnon are used. The critical values are determined by the number of 1(1) variables
in the co-integrating relationship and the nature (trend, constant) of the nonstochastic
regressors.

Once it is established that there is a co-integrating relationship among the
variables in the model, the co-integrating vector can be used to formulate the ECM. The
intuition behind the ECM is that deviations from the long-run equilibrium in the previous
period are corrected in the current period. Thus, it is necessary to establish a LR
equilibrium first before using an ECM. The ECM is formulated as follows:

Aw, = a0 + alApr, + azApC, +a3Apm, +a4Aa, + /l[wt_1 —fl'xt_1]+u, (10)
where x is a vector consisting of the variables pr, pc, pm,a and t and [I is composed of

the parameters (not including the coefficient of the normalizing variable) from the co-
integrating regression. Thea parameters can be interpreted as short run elasticities. The
term [wt_1 —- B'xt_l ] is the error correction (EC) term and it is the speed of adjustment
parameter which measures how past period deviations of the wage rate from its long-run
equilibrium are ‘corrected’ or adjusted in the current period. The larger the value of it the

greater the adjustment to deviations from the previous period’s disequilibrium. The EC

term can be conveniently replaced by the lagged residuals from the co-integrating

14

regression because the magnitude of the residuals reflects the deviations from the long-

run equilibrium in the past period.

4. Empirical Results
A. Unit Root Tests

The results of the ADF and Phillips-Perron tests are presented in Table 1. Both
levels and differences were tested. The second column shows the form of the equation
used in the test (Random walk with drift (RWD), Random Walk (RW)). The inclusion of
the drift was determined by testing for its significance in the regression equation for each
of the variables. The levels of all the variables were found to be nonstationary using a 5%
significance level, while the differences of all the variables were found to be stationary at

the same significance level. These results suggest that all variables are I(1).

 

Table 1. Unit root tests

 

 

 

 

 

 

 

 

 

 

 

 

 

VARIABLE Null ADF Phillips- 2’ Critical 2' Critical Order of

Hypothesi r Stat Perron Value Value Integra-
s r Stat (5%) (10%) tion

LEVELS

w RWD -1.696 -1.696 -2.99 -2.62 I(1)

Pr RWD -1.218 -0.982 -2.99 -2.62 1(1)

1% RWD -1.287 4.129 -2.99 -2.62 I(1)

Pm RWD -1.938 -2.296 -2.99 -2.62 I(1)

a RW -0.331 -0.265 -1.95 -1.60 I(1)

DIFFERENCE

S

d.w RW -2.358 3.800 4.95 4.60 I(O)

d. Pr RW -2.974 1.551 4.95 4.60 I(O)

d PC RW -4.697 1.306 4.95 4.60 I(O)

d pm RW -2.009 4.080 -1.95 -1.60 I(O)

d.a RW -7.389 -0.265 -1.95 -1.60 I(O)

 

All variables are in natural logarithms.

15

Figures 1 to 10 show the levels and differences of the natural logs of the variables
in the model. While it is apparent from Figures 5-10 that the differences are stationary,
the graphs of the levels suggest that the variables could be trend stationary, as opposed to
having unit roots. It should be noted that due to the small sample size, the ADF and
Phillips-Perron tests have very low power (Enders). Thus it is possible that the variables
are stationary, but the null hypothesis in the ADF and Phillips-Perron tests cannot be
rejected due to the low power of the test.

In order to allow for the possibility that the variables are stationary and the
possibility that the variables have unit roots, both cases are analyzed. In the first part of
the next section, it is assumed that the variables are nonstationary and a co-integration
framework is used to derive the long run and short run elasticities. In the second part, it is
assumed that the variables are stationary and OLS estimation is used to obtain long run
and short run elasticities. These alternative specifications allow the results to be evaluated

under alternative assumptions regarding the stationary properties of the variables.

Figure 1. Natural log of agricultural wages in levels (1975-2003)

 

 

 

 

 

 

 

6T —--~- -- -
5 “ ’l/fl/Jv—f—r—e
4 .. , v—d/MA-e/“T’ . . -i,
E 3 " * /‘/‘ T' Trad! A
5 ._.. ”a w-/’(
2 a .V.”" [H ._ _ _ . . . 7 _ l
1 _, _ 7
O I T ' TI I I I I F I I I I I I T I I TIT T I l I I I I fir r
l 1975 1978 1981 1984 1987 1990 1993 1996 1999 2002
l Year

 

 

 

 

 

16

Figure 2. Natural log of nominal rice prices in levels (1975-2003)

 

l

 

 

 

 

 

2.5 - - - ~ ~ , - , flue—-m—
2 _.
1:15- ~ - ~ ~ ,2 , i e iiiii v #—
3 [J/
E 1- f,/ “*5" , L _ L, A
0.5 - -, 9- e m L e
0 "ii—"T“‘T'fi' 7-4-1!" I I T I 1 T . I - T if f I I I I I I I
1975 1978 1981 1984 1987 1990 1993 1996 1999
Year

 

 

Figure 3. Natural log of nominal corn prices in levels (1975-2003)

 

[i

2 _- .- - ,, -L . - . . ..-.__ 2W _.

 

,JfiMT-A
1.51 , - r\/ it ~ -~ — — -‘

fl

 

 

 

 

Year

 

A 1 _ ,. .,,,.x~"“‘--,-~-""" .‘ 7 .. ,v
E /
E 0.5 9 7W / W 9- . r - a» ,_
#r,r/
O‘hmly’w’;r I I I | I“ I I I T I—. I I I I fT I I a
_05 _l__. -_..z, ._.. , ..__. ____._--. u- .. .. ..__.__. _-..___ __-__-._.-_-.--
197 1978 1981 1984 1987 1990 1993 1996 1999

 

17

 

 

 

Figure 4. Natural log of Consumer Price Index (1994==100) in levels (1975-2003)

 

 

"upnv

 

 

 

I I I I I

1996 1999

l IWI I

1978

I I I

1993

I I

1990
Year

1975 1981 1984 1987

 

I

T I I

2002

 

 

Figure 5. Natural log of agricultural area in levels (1975-2003)

 

 

15i85
1513
1575
151'
‘H165
15£5
15155 ~*
1515.
15A51r
15AI‘

nua)

 

 

 

l
'1
~4
l

 

1EL35

1984

I T I T —I

1990 1998
Year

I I

1975 1978 1981 1987 1996 199

 

9

I I I

2002

 

18

 

 

Figure 6. Natural log of agricultural wages in differences (1975-2003)

 

I

(135
(13

if (12
015
01
005

0

mm

 

L__z_-

(125 -

 

 

 

 

 

1975

1978

I

1981

1984

I

198

7 ' 1990
Year

I T I I I I

1993 1996

I

1999

I

2002

 

Figure 7. Natural log of nominal rice prices in differences (1975-2003)

 

C16
(15
(14
(13

I
l
l

In(d.pr)

01

-01
412

 

02 .4

0-.

—4

.4

—4

 

j -

1975 1978

1981

1984

{Fr
/ ‘w
/ x‘
I/(
I

1 0 “ V\,
1 ,r' \ .4
' 'i”’T,“*"T—T—_TfiTTTLI

'I.’

1987 1990 1993 1996

Year

 

. . ,\
l/\/ .
/
I I I

I IVI

N/I

5 Add

IV

I

l
4.
7,

l

 

1999 2002

 

19

 

 

Figure 8.

Natural log of nominal corn prices in differences (1975-2003)

 

 

In(d.pc)

0.6
0.5
0.4
0.3
0.2
0.1

-0.1
-0.2

 

— — a m—* — gm “—1
r ‘1 W
I l,

—1 f q‘ A 7 ‘ ‘ vi ——

‘1

.1

m

’1

_4 __ _._. ._... ._..... _-.J

 

1987 1990 1993 1996
Year

1975 1978 1981 1984

 

 

 

1999 2002

 

Figure 9. Natural log of Consumer Price Index (1994:100) in differences (1975-2003)

I
l
1

 

 

In(d.pm)

0.5
0.4
0.3
0.2
0.1
0
-0.1

 

 

 

 

4 A , - _ ~— _
s / i L _ I
/ \
4 . fl, "/ ~— 11 \ - ,, g, or" K! - ~ —~ M— I —~
\M/ W
1975 1978 1981 1984 1987 1990 1993 1996 1999 2002

Year

 

20

 

 

 

Figure 10. Natural log of agricultural area in differences (1975-2003)

 

 

 

 

0.251 ~ , ~ ,- ~ .- ”222m -
0.2 - , - z - A _ -_--
0.151 ~ ~- ~ *-~ .
0.1 1 . L i
0.05 . ”x 9- ,5.
0 db“% “vr‘“‘”i+i—:fi“:**r—~rr‘lfi I 7871/ I mam, I I 7’I Is I
005 ~ ' ‘ 3/
-0.1 ~ , ' - e e a.“ , -
-015 ~ ~ , ,- . w e m
-02 l _ - _ _ - _.
1975 1978 1981 1984 1987 1990 1993 1996 1999 2002

’ Year

In(d.a)

 

 

 

 

 

 

 

 

 

B. Dynamics Under Nonstationarity
Long-run Wage Response

Following the methodology developed by Engle and Granger, OLS was used to
estimate the following equation which is hypothesized to represent a co-integrating
regression:

w, = 1.2553 + 0.7807 p,, — 0.1234 pa + 0.0547 pm, + 0.0385a, + 0.05011+ s, (11)

SE. (5.5685) (.2396) (.2399) (.1966) (.3504) (.0120)
t-stat (0.23) (3.26) (-0.51) (0.28) (0.11) (4.16).
F(5, 22) = 862.96 R-squared = 0.9947

Prob > F = 0.0000 Adj R-squared = 0.9935

If the variables in (11) are co-integrated, then the coefficients can be interpreted
as long run elasticities so the long run elasticity of wage with respect to the price of rice

is 0.78.

21

Using error terms from equation (1 1) to estimate (9) yields:

Aé, = -.4835.€~,_l +.2929A.€,_l + v. (12)
SE. (.1628) (.1528)
T -Stat (-2.97) (1.92)

where a lag of A 5, was added to correct for autocorrelation.

The critical value for five variables in the co-integrating regression (with a trend
and a constant) is -4.72 at the 5% significance level, and -4.43 at the 10% significance
level. Since the I—Stat of the lagged error term (-2.97) is not greater (in absolute terms)
than the critical values, the null hypothesis of nonstationarity of the residuals cannot be
rejected. This implies that the variables in the equation (11) are not co-integrated, and no
stable long-run relationship exists among them.

Tests for co-integration are also done under alternative normalizations by varying
which variable in (11) is specified as the dependent variable. The results of the Engle-

Granger test under these alternative normalizations are presented in Table 2.

 

Table 2. Engle-Granger test for co-integration with constant

 

 

 

 

 

 

NORMALIZING Lags r Stat 1 Critical Value (5%)
VARIABLE

w 1 -2.97 -4.72

Pr 0 -5.14* -4.72

Pc 0 -6.63* -4.72

Pm O -3.14 -4.72

a O -4.91* -4.72

 

The null of no co-integration (5, s are nonstationary) can be rejected in three of

the five cases. Since the sample size is limited, it is expected that the Engle-Granger test

has low power. Thus, it is possible that the regressions having w and pm as dependent

22

variables have stationary residuals, and a co-integrating relationship does exist among the
variables. The analysis of the short-run dynamics is dependent on the assumption about
the long-run relationship among the variables. Since there is evidence for both co-
integration and no co-integration, we present two models of the short-run dynamics in the
following sections. The error correction model assumes that the variables in the model
are co-integrated while the first difference model assumes that the variables are not co-
integrated.

Error Correction Mode!

If the variables in (1 1) are co-integrated, then an ECM is appropriate in analyzing
short run wage adjustments. The ECM assumes that there is a significant error correction
(EC) term [Wt—l —B'xt_1], which corrects past-period deviations from the long run

equilibrium in the current period. Estimation of (10) yields the following (Equation 13):

Aw, = 0.0290 + 0.3345Apr, — 0.0387Apc, + 0.1356Apm, —0.0364Aa, + 0.3389Aw,_1 -O.2617EC +0,

SE. (.0141) (.1131) (.0666) (.1418) (.1138) (.0933) (.0991)
t-Stat (2.06) (2.96) (~O.58) (0.96) (-0.32) (3.63) (-2.64)
F(6, 20) = 14.21 R-squared = 0.8100

Prob > F = 0.0000 Adj R-squared = 0.7531

where a lag of Am was added to correct for autocorrelation. The residuals of (13) passed

tests for white noise using Portmanteau’s Q test (Pval> [(21) = 0.6508) and Durbin H

(Pval> 13m: = 0.5649) test for first order serial autocorrelation. Because each term in the

equation above is stationary, standard asymptotic theories apply, and inferences can be
made about the model based on t and F statistics. The individual coefficients of Ap,, ,

Aw“ and the error correction term are significant at the 5% level. The coefficient of the

23

error correction term is negative (-0.26). This means that if the EC term is positive, or the
wage in period H is above the long run equilibrium, then the error correction mechanism
will drive the current wage lower. Under the assumptions that the variables in (11) are
co-integrated and have an ECM representation, the SR contemporaneous elasticity of
wages with respect to rice prices is 0.33. This means that in the very short run, a 1%
decrease in nominal rice prices will cause a .33% decrease in the agricultural wage.

First Difference Model

The ECM assumes that there is a co—integrating relationship among the variables
in (11). However, the results of the Engle-Granger tests for co-integration also gave some
evidence that there is no co-integrating relationship among the variables in the model. To
allow for this possibility, the short run relationship among the variables is analyzed using
a F DM. This model assumes that the variables in (11) are not co-integrated. Since it is
also assumed that the variables have unit roots, first differencing is needed to make the
variables stationary. Note that since the data have been differenced, inferences about the
LR relationship among the levels of the variables cannot be made. However, short~run
elasticities can still be derived. OLS is used to estimate the following model:

Aw, = a0 + alApr, + azApC, + a3ApW + a4Aa, + a5t + a), (14)

This model is similar to the ECM model in (10) but the error correction term is excluded
because there are no adjustments from a long equilibrium that need to be captured. A
time trend is included because there is no underlying long run relationship that would
have captured the effects of time varying omitted variables. Estimation of (14) yields the

following results (Equation 15):

24

Aw, = 0.0110+ 030926,)” —0.0339Ap,., + 0.1764Apm, —0.0|08Aa, + 0.00041 + 0.3957Aw,_1 + a),

8.13. (.0308) (.1341) (.0774) (.1777) (.1314) (.0011) (.1104)
t-Stat (0.36) (2.31) (044) (0.99) (-0.08) (0.38) (3.59)
F( 6, 20)=9.77 R—squared = 0.7457
Prob > F = 0.0000 Adj R-squared = 0.6694

where a lag of Aw, was added to correct for autocorrelation.

Portmanteau’s Q (Pval> ,5) -- 0.6603) and Durbin H (Pval>z(21) = 0.6112) tests

showed that the residuals of (15) are uncorrelated of order one and follow a white noise
process. The coefficients of Aw“ and Apr, are significant at the 5% significance level
with t-statistics of 3.59 and 2.31, respectively. These results are similar to the results of
the ECM. Moreover, the elasticities of wages with respect to rice prices are close, at 0.33

for the ECM and 0.31 for the FDM.

A F -test was performed on the non-significant variables to see if they are jointly

zero. The null that 612 = a3 = a4 = a5 = 0 was not rejected at 5% significance level
(Pval > F = 0.8604). By removing the variables Apc,,Apmt,Aa, andt, a more

parsimonious model is derived:

Aw, = 0.0279 + 0.3608Apr, + 0.4048Aw,_l + a), (16)
SE. (.0131) (.0611) (.0975)

t-Stat (2.14) (5.91) (4.15)

F(2, 25) = 32.34 R-squared = 0.7294

Prob > F = 0.0000 Adj R-squared = 0.7068

25

The residuals of this model were found to be white noise using Portmanteau’s Q

(Pva1> 15) = 0.125) and Durbin H (Pval>z(21) = 0.7238) tests. Under this model, which

is the preferred model under the nonstationary and not co-integrated assumptions, the
short run elasticity of wages with respect to rice prices is 0.36. This means that for a 1%
change in rice prices, agricultural wages will change by .36% in the same year.
C. Dynamics Under Stationarity

If the variables in (6) are 1(0) then the short run elasticities can be obtained by
using OLS estimation. To make the stationary model (SM) fully consistent (that is, terms
in both equations are similar) with (16), the model estimated is:

W! = ,30 + ’51P” + ,32Prt—1 + ,53Wt—1 + [3410—2 + Z: (17)

The corresponding long run elasticities can be derived by using the formula

 

LR Elasticity = fl' (18)
1— ,33

where ,6] is the coefficient of p,., while [33 is the coefficient of w,_].

Estimation of (17) yields:

w, = 0.5829 + 0.4128p,, — 0.1208p,,_| + 0.8839w,_l — 0.1261w,_2 + z, (19)
SE: (.1579) (.0536) (.0847) (.1689) (.1143)

t-Stat: (3.69) (7.70) (-l.43) (5.23) (-l.10)

F( 4, 22) = 6435.17 R-squared = 0.9991

Prob > F = 0.0000 Adj. R-squared = 0.9990

The coefficients of p,.,_] and w,_2 are individually and jointly non-significant
(Pval>F = 0.3397). Removing p,,_1 and w,_2 from (19) leads to a more

parsimonious model:

26

w, = 0.8498 + 0.4126 p,, + 0.6494w,_1 + z, (20)

SE. (.0828) (.0506) (.0402)
t-Stat: (10.26) (8.14) (16.16)
F( 2, 25) = 862.96 R-squared = 0.9986
Prob > F = 0.0000 Adj R-squared = 0.9985

The residuals of (20) passed Durbin H (Pval> [(21): 0.2426) and Portmanteau’s Q

(Pval> [(21) = 0.1969) tests for first order autocorrelation and white noise.

To verify if the results of the full model estimation are robust to underlying
stationary assumptions, the full model is also estimated assuming that all variables are
stationary and all right hand side variables are exogenous:

w, = 3.7348+0.5722p,, —0.0669pc, -0.0948p,,,, -0.1617a, +0.0066t + 0.596lw,_1 + z, (21)

3.13. (2.5923)(.1268) (.1106) (.1021) (.1627) (.0082) (.0742)
t-Stat:(l.44) (4.51) (-0.61) (-0.93) (-0.99) (0.80) (8.03)
F( 6, 21) = 3042.70 R-squared = 0.9989

Prob > F = 0.0000 Adj R-squared = 0.9985

A lag of w, is added to remove autocorrelation. The residuals of (21) passed Durbin H

test (Pval> 151): 0.5157) and Portmanteau’s Q test (Pva1> 45(21): 0.4783) for first order

autocorrelation and white noise.

In Equation 21, the signs of ApcraAPm, and Au, are not expected. The t—statistics

also show that these variables and t are non-significant. A joint F-test indicated that the

coefficients of Apc,,Ap m t , Aa, and! are jointly equal to zero at the 5% significance level

27

(Fm, = 0.97). Removing the non-significant variables in (21) we derive a model that is
identical to (20) which is the preferred model under this specification.

Under the assumption that the variables are stationary, the short-run elasticity is
0.41 while the long run elasticity is 1.17. This implies that in the SR, a 1% change in rice
price will lead to a 0.41% change in the agricultural wage. Under this model, the LR
elasticity is greater than one. As mentioned in the previous chapter, results from this
model should be interpreted with caution because of the potential endogeneity problem
associated with this specification.

Table 3 summarizes the short run and long run elasticities obtained from the error
correction model and the stationary model. The first difference model is excluded

because long run elasticities cannot be obtained from this model. It is of interest to know

 

Table 3. Wage elasticities and adjustment under different model specifications

 

 

 

 

 

 

 

 

 

 

 

 

MODEL

ECM SM
Short-Run 0.33 0.41
ElasticitL
Long-Run 0.78 1.18
Elasticity
YEAR % Adjustment to the LR Elasticity
Year 1 0.57 0.58
Year 2 0.62 0.73
Year 3 0.64 0.82
Year 4 0.64 0.88
Year 5 0.65 0.92
Year 10 0.65 0.99
Year 15 0.65 1.00

 

how long it takes before the long run adjustment is reached. By summing the annual
marginal effect of a percentage change in the price of rice in time t to the agricultural

wage, we can obtain the annual percentage adjustment to the long run elasticity.

28

The ECM model suggests that even after 15 years, long run adjustment has not
been reached. In the case of a price decrease, this means that agricultural wages will not
decrease as much as rice prices, even in the long run. The SM suggests that 90% of the
adjustment to the LR elasticity takes place after 5 years.

D. Welfare Implications

So far the effect of changes in rice price on the agricultural wage has been
discussed, but nothing has been said about its effect on agricultural wage earners. Note
that agricultural wage earners refer to net demanders of rice and net suppliers of labor
(i.e. those households that supply more labor than they hire). The short-run and long-run
elasticities provide information about how quickly wages respond to changes in rice
prices, and enable the identification of groups who will be adversely affected by the
decrease in rice prices. Using a household model similar to Equation 1 to analyze the
effect of changes in wages to food prices, Ravallion (1990) showed that the necessary
and sufficient condition for households that are net demanders of rice and net suppliers of
agricultural labor to benefit from a small increase in food price is that the elasticity of
agricultural wages with respect to food prices should be greater than the share of food
expenditure in the household’s labor earnings. Based on the same reasoning, the
condition for households to benefit from a decrease in rice price is if the elasticity of

wages to rice prices (77) is less that the share of rice expenditures in labor earnings (17*).

Conversely, households that are net demanders of rice and net suppliers of labor will
incur a loss if the elasticity of wages with respect to rice price is greater than the share of

rice expenditures in labor income, i.e., 17 > 77 *. To illustrate this concept, Table 4 presents

the different welfare implications of decreasing the price of rice for different types of

29

agricultural wage earners, given the SR and LR elasticity. Since rice is a staple food in
the Philippines, a substantial proportion of household expenditure is devoted to
purchasing rice. The monetary amount is more or less fixed for a household of a given
size, but the share in the total budget changes depending on the income. The poorest
households in the Philippines typically spend 30% of their income on rice expenditures
(Cororaton). Case 1 represents the poorest agricultural household that relies solely on

agricultural wages where,77 * = 0.30. Since the SR and LR 77 values in the ECM, FDM
and SM models are not greater than 17 * , these households are expected to incur welfare

losses both in the SR and in the LR if rice prices were to decrease.

 

Table 4. Welfare effects on agricultural wage earners of a reducton in rice price

 

 

 

 

 

 

 

ECM F DM SM
Case % of % of % of SR gain LR gain SR gain SR gain LR gain
rice income rice or loss or loss or loss or loss or loss
exp. in from ag. exp. in = = = = =
total HH labor ag. (”SR (”LR (”SR (”SR (”LR
budget income 0-33) 0-78) 0.36) 041) 1-18)
(77*)

1 30 1 00 0.30 Loss Loss Loss Loss Loss

2 30 50 0.60 Gain Loss Gain Gain Loss

3 20 50 0.40 Gain Loss Gain Gain Loss

4 10 20 0.50 Gain Loss Gain Gain Loss

5 10 10 1 Gain Gain Gain Gain Loss

 

 

 

 

The value of 77* increases as households rely less heavily on agricultural wages, or
incomes increase such that rice expenditures become less relative to the household
budget. In general, as in Cases 2-4, households benefit from the decrease in rice price in
the SR but eventually lose in the LR, as wages are further driven down in response to the

decrease in rice prices. Case 5 shows that a household that is a net supplier of labor can

30

only unambiguously gain from the decrease in rice price if the income is high enough and
if the income from agricultural wages is only small portion of total income.

The analysis on Table 4 applies to a specific type of household, which is a net
demander of rice and net supplier of agricultural labor. Although other types of
households are not the focus of this paper, it is beneficial to briefly extend the welfare
application to other household types. Households that are net demanders of food and net
demanders of labor will unambiguously benefit from a decrease in the price of rice for all

77>0, because they will have both the benefit of decreased rice expenditures as well as

decrease in the cost of hiring labor. Urban consumers, who are net demanders of food,
but have no agricultural income can also be categorized under this group. Urban
consumers will clearly gain from a decrease in rice price because their food expenditures
will decrease while their income will remain the same.

On the other hand, households that are net suppliers of rice and net suppliers of
labor will be adversely affected by the price decrease for any 77>0 regardless of how
much their own consumption of rice is. This is because their income from selling rice
will decrease and at the same time, their wage from agricultural labor will also decrease.
For households that are net suppliers of rice but are net demanders of labor, the welfare
effect is less predictable. It will depend on the loss of income due to the decrease in price

at which rice is sold, relative to the decrease in the cost of hiring agricultural laborers.

5. Summary and Conclusions

The objective of this paper is to determine whether there is a long-run relationship

between agricultural wages and rice prices. and to measure the short-run and long-run

31

wage rate elasticities with respect to rice price. This study was motivated by the expected
decline of rice prices in the Philippines, as a result of the government’s efforts to
liberalize rice trade in the country. The impact on wages is important because
agricultural wages are the primary source of income of some of the poorest households in
the Philippines’ rural sector.

The relationship of agricultural wages and rice prices is analyzed using a
neoclassical wage determination model where the wage rate is determined by supply and
demand of labor in the agricultural sector. The ADF and Phillips-Perron tests for
nonstationarity showed that variables in the model are 1(1). However, it is recognized that
the tests for nonstationarity may have low power. Thus, the LR relationship was analyzed
under two assumptions: stationarity and nonstationarity. In the first part, it was assumed
that the variables are nonstationary and a co-integration framework was used. In the
second part, it was assumed that the variables are stationary and OLS was used to derive
SR and LR elasticities. Because the test for co-integration revealed mixed results about
the existence of a long run equilibrium relationship among the variables in the model,
two models under the nonstationarity assumption are also specified: an error correction
model and a first difference model.

The results confirm the general hypothesis that rice prices are an important
influence on agricultural wages. Depending on the model specification, the short-run
elasticity of wages with respect to rice prices ranges from 0.33-0.41, while the long-run
elasticity ranges from 0.61-1.18. In all model specifications, wages adjusted to the LR

elasticity by more that 50% after the first year. The SM suggests that at least 90% of the

32

adjustment to the LR elasticity will occur in less than 5 years while the ECM model
suggests that the adjustment is slower, with only 64% adjustment even after 15 years.

An analysis of the welfare implications of a decrease in rice prices for agricultural
households that are net demanders of rice and net suppliers of labor showed that most
households benefit in the SR but lose in the LR. Moreover, households who rely
primarily on agricultural wages and spend a large percentage of their wage income on
rice will be adversely affected by the decrease in rice prices in both the short-run and the

long-run.

33

REFERENCES

Barnum, Howard and Lyn Squire, 1979, A Model of an Agricultural Household.
Washington, DC.

Boyce, J. K., and M. Ravallion, 1991, A dynamic econometric model of agricultural
wage determination in Bangladesh. Oxford Bulletin of Economics and Statistics. Vol.53,
No. 4, pp. 361—376.

Cororaton, Caesar B., 2004, Rice Reform and Poverty in the Philippines: A CGE
Analysis, Philippine Institute for Development Studies Discussion Paper Series 2004-14,
Philippine Institute for Development Studies, Makati City, Manila.

Datt, Gaurav and Jennifer Olmsted, 2004, Induced Wage Effects of Changes in Food
Prices in Egypt, The Journal of Development Studies, Vol. 40, No. 4, pp. 137-166

David, Cristina C., 1997, Agricultural Policy and the WTO Agreement: The Philippine
Case, Philippine Institute for Development Studies Discussion Paper Series [997-13,
Philippine Institute for Development Studies, Makati City, Manila.

Davidson, Russell and James G. MacKinnon, 1993, Estimation and Inference in
Econometrics,1993, Oxford University Press,New York, New York.

Dawe, D., 2003, “Equity effects of rice trade liberalization in the Philippines,” in Rice
Science: Innovations and Impact for Livelihood. Proceedings of the International Rice
Research Conference, 16-19 September 2002, Beijing, China. Mew TW, Brar D, Peng S,
Dawe D, Hardy B (editors). International Rice Research Institute, Chinese Academy of
Engineering, Chinese Academy of Agricultural Sciences, pp. 1007-1022.

Enders, Walter, 2004, Applied Econometric Time Series 2nd Edition, John Wiley and
Sons, Inc., Hoboken, New Jersey.

Engle, Robert F. and C.W.J. Granger, 1987, Co-Integration and Error Correction:
Representation, Estimation and Testing, Econometrica, Vol.55, No.2, pp.251-276.

International Rice Research Institute - Social Sciences Division, 2004, unpublished
survey data

Johansen S., 1998, Statistical Analysis of Co-integration Vectors, Journal of Economic
and Dynamic Control, Vol.12 , pp.23 1-254.

34

National Statistical Coordination Board (Philippines), 1982, 1994, 2002,2004, Philippine
Statistical Yearbook, Makati, Metro Manila, Philippines

Palmer-Jones, R.W. 1993. Agricultural wages in Bangladesh: What do the figures really
show?, Journal of Development Studies, Vol.29, No.2, pp.277-295.

and Parikh, A., 1998, The Determination of Agricultural Wage Rates in
Bangladesh, Journal of Agricultural Economics, Vol.49, No.1, pp.111-133.

 

Pindyck, Robert S. and Rubinfeld, BL, 1998, Econometric Models and Economic
Forecasts 4th Ed., McGraw-Hill, Singapore.

Rashid, Shahidur, 2002, Dynamics of Agricultural Wage and Rice Price in Bangladesh :
A re-examination, Post-doctoral paper for the International Food Policy Research
Institute, Washington DC.

Ravallion, M., 1990, Rural welfare effects of food price changes under induced wage
responses: Theory and evidence for Bangladesh, Oxford Economic Papers, Vol. 42,
No.3, pp.574-585.

Ravallion, M., 1994, Agricultural Wages in Bangladesh: What do the Figures Really
Show?, Journal of Development Studies, Vol. 31, No. 2, pp. 334-345.

Rosenzweig, Mark.R., 1980, Neoclassical theory and the optimizing peasant: An
econometric analysis of market family labor supply in a developing country, Quarterly
Journal of Economics, Vol. 95, pp.31-55.

Rosenzweig, Mark R., 1984, Determinants of Wage Rates and Labor Supply Behavior in
the Rural Sector of a Developing Country, in Contractual Arrangements, Employment,
and Wages in Rural Labor Markets in Asia (Hans P. Binswanger and Mark R.
Rosenzweig, Eds.), Yale University Press, New Haven, New Jersey.

35

S RA
|

'111311111121iiiiiiijiiiiiiiijlii