Pertanika 13(2), 283·288 (1990)

Composite Models for Short Term Forecasting forNatural Rubber Prices

MAD lASIR SHAMSUDlN and FATIMAH MOHO. ARSHAD

Depanment of Agricultural EconomicsUniversiti Pe1tanian NJalaysia

43400 UPM Serdang, Selnngor Darnl Elzsan, Malnysia

Key words: composite models, short term forecasting, natural rubber prices

ABSTRAK

Teknik ekonomet1ik dan kaedah satu pembolehubah Box-jenkins Lelah digunakan untuk meramal hargagetah asli. Kajian ini membentuk satu model pemmalan jangka jJendek dengan menggabungkan modelekonometrik dan model sattt pembolehubah alau dikenali sebagai model komposit. Penemuan kajian11lenunjukkan model komposit dapat meramal harga getah asli /ebilt cekap daripada model ekonometrik danmodel satu pembolehttbah.

ABSTRACT

The economet1ic technique and Box-jenhins univariate method have been applied in forecasting naturalrubber p,;ces. This study developed a sharl term JOT.casting model/mown as the composite model, by combiningthe econometric and univariate models. The results indicate that the composite rrwdel produces more efficientforecasts than the econometric and univariate models.

INTRODUCTION

Despite the dire need for better forecast ofnatural rubber prices. few attempts have beenmade to do so. Some of the pioneer work onMalaysia natural rubber price forecasting werecarried out by A. Halim (1978), Mohd. Napi(1984) and Mohd. Napi and Mohd. Yusof(1988). They have adopted both econometricand Box:Jenkins univariate methods and foundthal the t\vo techniques were capable ofproducing reliable forecasts. The studies,however, did not compare the relativeperformance of each model. Previous studiesthat compared the relative efficiency of variousforecasting techniques indicate inconclusiveresults. Studies by Cooper (1972), Bourke(1979) and Brandt and Bessler (1981) indicatethe superiority of the Box:Jenkins univariatemodel over econometric models, while studiesby Leuthold et al. (1970) and Gellatly (1979)indicate otherwise.

In view of the inconclusive findings of therelative ability of the forecasting techniquesand the fact that each method has its ownstrengths, the composite model may well be abetter alternative to reconcile the incon­sistencies. This model combines the [\'''0

approaches into a composite one; it retains· thestructural relationship extracted from theeconometric medlod while using the time­series model to explain for the residual. Studiesby Bates and Granger (1969), Granger andNewbold (1977) and Brandt and Bessler(1981) illustrated that the composite forecastsoutperformed the individual forecasts. Theyfurther demonstrated that the compositeforecasts have an enor variance not greaterthan the smallest error variance of theindividual forecasts. This paper attempts toforecasts monthly rubber prices (RSSI) inKuala Lumpur market using the compositemodel and to compare its forecasting

~-1AD NASIR SHA\lSUDlN AND FATIMAH MOHD. ARSHAD

METHODOLOGY

Econometric ModelThe price equation of natural rubber can bespecified as follows:

PR, = f (SCC,_I' WCN,_I' PR,_I' E) (I)

predictability with each individual model (i.e.,econometric and univariatt). The followingparagraphs provide briefly the description ofthe methodology. This is followed by theanalysis of the results and conclusion.

(2)

a random disturbance assumed tobe distributed as N (0, cr')a backward shift operator such thatBZ = Z and B' Z = Z

t I-I I I-I.

The regular autogregressive opera-tor of order p, i.e., ",(B) = (I - "IB- ".,B' - . . .. .. BP)

• pThe seasonal Hutogregressive ope-rator of order pnumber of regular differencesnumber of seasonal dilTerencesthe regular moving average ope­rator of order q, i.e., 0 (B) = (I ­0IB - O"B' - ..... - 0 'EO)• 0the seasonal moving average ope-rator of order Qthe order of the seasonal differences

<l> (B)p

B

" (B) <l> (B') (I - B)' (1 - B')" ZP jl I

= 0 (B) e (B') eII Q I

wheree,

Composite ModelThe composite model refers to thecombination of forecasts from econometricand Box:]enkins models. The advantage of thecomposite model is that its forecasts, in mostcases, outperforms any of the individualforecasts.

Equation (2) is an ARlMA model of order (p,d, q) (P, D, ill,.

forecasting. The identification step involves thecompal;sons of estimated autocorrelation andpartial autocorrelation functions of the seriesof interest with the theoretical autocorrelationand partial autocorrelation fUllctions of knownARlwLA processes. Given a class of ARI Mi\models from the first step, their paI-ametcrvalues can be estimated from the historicalseries using nonlinear least squares. Diagnosticchecks arc then applied to determine anypossible inadequacies in the model, and theprocess is repeated if any afC found. Finally,having arrived at an adequate model, "optimal"forecasts are generated by recursive calculmion.

The general multiplicative seasonalau toregressive-in tergrated-movi ng-avcrage(ARlMA) model for a seasonal Z" t = I, 2,......T with a k.nown period S can be written as

price of natural rubber (RSSl) inKuala Lumpurstock of natural rubber in the con­suming countries lagged one periodtotal consumption of natural rubberlagged one periodprice of natural rubber (RSSI) inKuala Lumpur lagged one periodrandom error term

WCN,-I

E

The price is expected to have poslUverelationships with consumption and one periodlagged price and negative relationship withstock.

Equation (1) was estimated using ordinaryleast squares (OLS). The sample period wasfrom January 1978 to March 1988. The datawas obtained from the MRRDB "MalaysianRubber Review" and lRSG "Rubber StatisticalBulletin".

Box:Jenkins ModelThe Box:Jenkins technique, or autoregressive­integrated-moving average (ARIMA) model, iswell documented, in particular. by Box andJenkins (1976), Nelson (1973) and O'Dono­van (1983). Briefly, this technique is aunivariate approach which is built on thepremise that knowledge of past values of a timeseries is sufficient to make forecasts of the­variable in question. In other words, themodels are void of economic theory and thischaracteristic has been critisized as an inherentweakness of the time series application.

Box and Jenkins (1976) set forth four stepsfor this approach: model indentification,parameter estimation, diagnostic checking and

wherePR,

2H4 PERTANIKA VOL. 13 NO.2, 1990

CO:'\1J>OSITE MODELS FOR SHORT TER:'\'! FOREC-\!)"TINC FOR NATUR..\.L RUBBER PRICES

A basic problem underlying the generationof composite forecasts is what weight to applyto each individual forecast". Bates and Granger(1969) and Granger and Newbold (1977)discuss several procedures for determiningthese weigh ts. One of the methods which hasbeen proven sllperiOl' to other methods is theminimum variance (Brandt and Bessler(1981» which can be expressed as follows:

Actual

Turn

No tutTI

TABLE ITypes of turning point

Prediction

Turn No turn

a b

b d

\I..·here K is the weight assigned to forecastmethod \, cr.:? is the sample period forecasterror varian~c associated with method i, andp.. is the correlation coefficient between the

" . d'errors of forecasts 1 an J.

Forecast EvaluationEvaluation of the forecasts is based on theabsolute accuracy and turning points. Threemeasures of absolute accurancy are used, Theyare root mean square error (RMSE) , rootmean square percent error (RJ\lfSPE) and Theilinequality coefficient.

Turning points evaluation can becategorised"" into statistical and cyclicalturning points. The statistical turning pointsrefer to an enor which relates to anyforecast direction of change that does notagree with the actual direction of movem~nt.The cyclical turning points refer to a turmngpoint in the economic sense of a reversal ofcurrent trend.

Tn evaluating cyclical turning points 1:\\"0

types of errors are involved: (i) a turning pointmay be incorrectly predicted (Type 1 error)or (ii) none predicted \\Then one actuallyoccurs (Type II error).

Quantitative measures of these errors canbe expressed as:

£, = b/(a+b}; f, = c(c+d)

where f1

and f2

refer to type I and type 11errors, respectively while a, b, c and d are typesof turning point (Table I).

Generally, the lower fj

the belter, but bothmust be considered ill association since Type1 errors may be avoided by never prediciting

K =J

RESULTS AND DISCUSSION

(5)

+ 0.040 WCN,.,( 1.289)

R' = 0.935 SER = 10.687 h ~ 0.191

where PR, is the price of natural rubber (RSSl)in Kuala Lumpur, SeCt_I is the stock of naturalrubber in the consuming countries lagged one

Period and WCN is total consumption of'·1

natural rubber lagged one period. The figuresin parentheses are the t-values of thecoefficients.

The estimated equation appears to fit thedata well, as evidenced by the R2 and t-values.All the estimated coefficients have theexpected signs, although the estimate forlagged total consumption is insignificant. Theresults suggest that the price of natural rubberis determined by the stock levels and the pricein the previous periods where they aresignificant at five percent level.

The estimated coefficients in equation (5)are used to estimate ex-post forecasts fromApril 1988 to March 1989 (Table 2). The valueof the root mean square percent error whichmeasure the deviation of the forecasted valuefrom its actual value in percentage terms issmall (7.210 percent). The Theil inequalitycoefficient is less than one. These figuresindicate that the forecasting pelformance of

PR, = 70.048 - 0.0698 SCC'_I(-3.337)

+ 0.893 PR,.,(24.667)

Econometric jWodelThe estimated equation of the price of naturalrubber is as follows:

a turning point, and similarly for Type IIerrors.

(4)

(3)., cr

°2~_-PI\!°1 :!0l~ + 0/ - 2p!:! oJ o~

K.,~I-K,

PERTAi'l"IKA. VOl.. 13 NO.2, 1990 285

MAD NASIR SHAt\JISUDIN Ai~D FATIMAH MOHO. ARSHAD

TABLE 2Natural rubber price forecasts, RSSI (CenLS/kg.)

April 1988 305.5 285.3 292.4 286.9May 335.6 302.7 290.1 299.8

June 331.5 331.5 291.4 322.4

July 348.2 369.9 287.6 351.1August 342.8 342.6 285.7 329.6

September 315.9 341.4 279.5 327.2

October 284.6 321.0 277.1 310.9November 275.1 281.4 275.2 279.9December 287.9 298.9 275.3 293.0

January 1989 303.0 289.9 274.9 286.6

February 307.0 307.2 278.8 300.7March 299.7 313.0 276.2 304.8

RMSE 23.367 40.815 22.466RMSPE 7.210 11.819 6.557

U .369 0.682 .350Turning Point

Cyclical errorsf, 66.7 50.0 66.7

f, 28.6 16.7 28.6Statistical

correct (%) 33.3 66.7 33.3

Forecasted price

Period Actualprice

£cono- Box-metric Jenkins

Compo­site

forecast than the one identified in this study.The chosen model is then fined into the

data from January 1978 to March 1988 whichyield the following estimates (with estimatedstandard error in parenthesis):

°1 = -.20707 (.25672)

<1>'0 -.07969 (.10271)8, .43172 (.22600)8, = .51455 (.20986)8 10 .82500 (.05891)

The diagnostic checks shown in Table 3,do not detect any model inadequacies. Thereis no lag that is significant at either five percentor ten percent leveL

TABLE 3Diagnostic Chi-Square statistics for residual

series of mbber prices

Lag Chi-square Degree of freedom Probability

6 .37 I .541212 5.59 7 .588718 15.81 13 .259624 24.45 19 .179630 28.59 25 .281436 35.21 31 .2756

the estimated equation is satisfactory. Themodel however is not able to simulateaccurately the turning points of the actualseries. It only manages to predict correctly 33.3percent of the turning points.

BoxJenkins ModelAfter following the Box:Jenkins procedure ofidentification, estimation and diagnostic chec­king. the following model was chosen:

(1-B') (1-BlO) (1-0,B) (1-<I>lOB'O) Z,

= (1-ll,B - 8,B') (I-GlOB'O) e, (6)

This model involves a first order auLO­regressive element, a second order moving ave­rage element, a seasonal autoregressive ele­ment and a seasonal moving average element.The model is different from the one developedby Mohd. Napi and Mohd. Yusoff (1988). Theapplication of the model adopted by the twoauthors to an updated data provides inferior

The estimated model is then used toobtain ex-post forecasts for the twelve monthsbeginning from April 1988. Tbe forecastedvalues are shown in Table 2. The values ofRMSE, RMSPE and U are larger than thecorresponding values in the econometricmethod. Unlike most studies on comparisonof forecasting methods, this study indicates thatthe econometric technique outperforms theBox:Jenkins model in predicting monthlyrubber p1ices.

The relative efficient performance of theeconometric method could be attributed to itsability to capture the dynamics of the srrueturalchanges in the market due to variation in thefundamentals which is pertinent in the naturalrubber market. For instance, there was an"abnormal" upswing in natural rubber pricesin the second and third quarters of 1988. Theprices in fact exceeded INRO's "must sell"level. Prices hit an all-time high in June whenRSSI was traded at $3.76 per kg. This firmness

286 PERTANIKA VOL. 13 NO.2, 1990

COMPOSITE MODELS FOR SHORT TERM FORECASTING FOR NATURAL RUBBER PRICES

in prices was partly due to increased buyingby the centrally planned countries, particularlyChina, strong global demand for the tyreindustry and uprecedented boom in theproduction of condoms and surgical gloves asresult of worldwide concern about AIDS. Theprevalence of wintering and adverse weatherconditions in Indonesia further aggravated thepnces. Since the Box:Jenkins model isessentially an extrapolation technique which isbased on past values and not on economictheory. it could not capture this "abnormal"condition of supply and demand. Theeconometric model incorporated fundamentalfactors of supply and demand and inventorylevel in its specification, hence it has the abilityto translate the upswing effects on its forecasts.The track.ing performances of the Box:Jenkinsmodel were relatively better. It correctlypredicted 66.7 percent of the turning pointsof the actual series.

Composite ModelFollowing equations (3) and (4), theeconometric and Box:Jenkins models could heoptimally combined by assigning a weight ofK, = 0.7713 to the former and K., = 0.2287 tothe latter. The forecasted values generatedfrom the composite model are presented inTable 2 which clearly indicates that it providessuperior forecasts. The RMSE of the compositeforecasts, 22.466, is about half as large as thevalue reponed for the Box:Jenkins model, butit is only marginally less than that for theeconometric model. The RMSPE and U of thecomposite forecasts is almost equivalent to theRMSE. The results obtained are consistent withthe theoretical expectation in that if there aretvvo (or more) forecasting techniques (\vhichare based on different sets of information), thecombination of the methods would improvethe forecast quality.

CONCLUSION

This study seeks to develop a short termforecasting model for natural rubber prices(RSS1) using a composite approach. Aminimum variance criterion was used tocombine the forecasts generated by theeconometric and ARlMA models. Despite the

fact that the econometric model outperformsthe Box:Jenkins model, it is possible to use theminimum variance criterion and combine thetwo approaches to produce even more efficientforecasts. The results show that forecast errorsare indeed reduced by combining theindividual predictions. This implies thatforecasts taken from individual models are notlikely to provide the user with the mostaccurate information. The user might be betteroff dle combine the forecasts from alternativemodels to remove the likelihood of makinggross mistakes based on forecasts of individualmodels.

REFERENCES

ABDUL HAuM HAsSAl'\!. 1978. Meramal Harga GeLahAsli. Sahagian I - Satu Model Ekonometrik. JRubb. Res. Insl. Malaysia 2; 23-32.

ABDUL HALIM HASSAN. 1978. Meramal Harga GetahAsh. Bahagian II - Harga Gab 2, 3 dan PuraLaBulanan. J Rubb. Res. hut. Malaysia 2: 53-60.

BRANDT, J.A. and D.A. BESSLER. 1981. CompositeForecasting: An Application with U.S. HogPrices, Amer. j. Agr. £Cou. 63; 135-140.

BATES, ].M. and C.W.]. GRANGER. 1969. TheCombination of Forecasts. Operational Res.Quart. 20; 451-68.

BOURKE, I]. 1979. Comparing The Box:Jenkins andEconometric Techniques for Forecasting BeefPrices. Rro. oj Mktg. aud Agr. Econs. 47(2); 95­106.

Box, G.E.O. and G.M. ]ENKlNS. 1976. Time SeriesAnalysis, Revised Edition, San Francisco:Holder-Day.

CoOPtR, R.L. 1972. The Predictive Perfonnance ofQuarterly Econometric Models of the U.S.Econometric Models of Cyclical Behaviour1972. ed. Hickman. New York: Columbia Univ.Press.

GELL-\.TLY, C. 1972. Forecasting N.s.W. Beef Pro­duction: An Evaluation of AlternativeTechnique. Rev. 1.WRtg. and AgJ·. Ecan. 47.

GRANGER, C.WJ. and P. NEWBOLD. 1977. ForecastingEconomics Time Series. New York: Academic Press.

HWA, E.C. 1979. Price Determination in SeveralIntenlational Primary Commodity Markets: AStructural Analysis, IMF Staff Papers. 26; 157­188.

LEUTHOLD, R., A. MacCORMICK, A. SCHMITZ and D.

PERTANIKA VOL. 13 NO.2. 1990 287

AHMAD ZUBAlDI BAHARUMSHAH

growing consumer preference for high qualitycakes, bread, and rolls that require high pro­teiD wheat.

Information on production, consumptionand import behavior of the various marketagents of rice in Malaysia is important as thegoverment continually evaluates rice program­mes in Malaysia. The purpose of this paper isto identify empirically the factors affecting thesupply and demand for rice in Malaysia. Inaddition, this paper attempts to explain thechange over time in the level of protectionusing the price linkage equation or policyreaction function. Thus, the behaviour agentSconsidered in the model includes consumers,farmers and the government who cono-olprices with quota/tariffs.

Polity EnvironmentAfter independence, three major policy goalsin Malaysia were formulated. These were to (1)reduce dependency on world markets (2) saveforeign exchange, and (3) improve the welfareof rice producers. These objectives are to beattained through the adoption of modernagriculture inputs, large investments ininfrastructure and a producer- orien ted ricepolicy. The imports of rice are restl;cled andrice prices are maintained above world prices(Goldman, 1975).

1n the 1959's Peninsular Malaysiaproduced about 50-60% of the domestic ricerequirements and by the middle of the 70'sthis percentage had increased to 90%. Theincreased paddy production was the result ofimproving water control, the opening of newrice land and a high producer price supportprogramme. Despite the rapid adoption ofdouble-cropping and widespread planting ofnew paddy varieties, Malaysia's yield per­formance has been disappointing. Goldmanreported that from 1962, prior to theintroduction of new seed varieties, to 1974, dry­season yields in irrigated areas increased at anaverage compound rate of only 1.86%.

Malaysia nearly achieved self-sufficiency inrice in the 70's but was unable to maintainthis policy objective in the 80's because of thechanging economic environment. The cost ofpursuing self-sufficiency is too high and with

the large budget. deficit, the administrationthought that it would be advantageous to aimto a lower degree of self-sufficiency, relyingmore on imports to make up for the shortfall.This change in policy is facilitated by thedeclining international rice price. Further, thecounu-y's imparl requirement is not largerelative to the world markeL Malaysia, in its1984 Malaysian National Agriculture Policy,lowered the self-sufficient target to 80-85% andsubsequently this was lowered to 60-65%.

The decision to give up the objective ofself-sufficiency in rice and rely on imports fora significan t share of consumption wasimplemented without drastic change in u-adepolicies. Malaysia still maintains a restrictivetrade policy in rice although tradeliberalisation is under consideration currentlyby the administration. Imports are taxed toprovide revenue to the government andprovide protection to domestic producers.Malaysia still maintains a certain level of stockas food security reselve. The government seeksa gradual adjustment of production throughthe measure of freezing price support ratherthan cutting it suddenly.

The extent of gevernment intervention inthe rice sector is significant. Both theproducers and consumer prices for rice areconsiderably above world level, sometimes asmuch as 1.5 to 2.0 times the prevailing worldmarket. However, Malaysia is not the onlycountry that provides incentives to the riceproducers. For example, Japanese rice priceswere about 200% above the internationalprices in early 1980. Countries like Korea andTaiwan maintained domestic prices 150%above the international price in the same timeperiod.

.Relevant LiteratureWong (1978) developed a simple dynamicsimultaneous equation model which allows forinteraction between rice production,consumption and export sectors to evaluate theeffect of taxation on rice exports in Thailand.The model consist of four equations - adomestic supply equation, a domesticconsumption equation, export price equationand price transmission equation. The results

290 PERTANIKA VOL. 13 NO.2, 1990

INCORPORATING GOVERNMENT INTERVENTION IN RICE MARKET, CASE OF RICE POLICY

obtained were consistent with theoreticalexpectation.

Nik Fuad (1985) evaluated the Malaysianrice policy using an ecornometric model. Hefound that the estimated elasticity of areaplanted \'fith rice with respect to the gua­ranteed minimun price ranges from 0.17 to1.13 for the various regions in Malaysia. Overallresults suggested a positively sloped supplycurve althought the respond to support pricewas small. Nik Fuad reported an estimatedprice elasticity of demand -0.5 and found thatthis was comparable to other studies. Theincome elasticity was small but negetive (-0.13).The small negative income elasticity reflects achanging consumption pattern away fromstable food, as is usually expected as thecountry reaches a higher average income.

Agricultural economists typically treatpolicy variables exogenously in econometricmodels. There are at least four reasons for this.These are (1) the analyst believes that policyvariables are essen tially random variablesunrelated to the system being modeled; (2)the analyst believes that policy variables areinfluenced by the economic environment butthe relationship is weak; (3) the purpose ofthe analysis is to calculate policy multipliers,so the policy variable must be exogenous; and/or (4) the major variables are difficult if notimpossible to measure.

Meilke and Griffith (1983), however,argued that the above reasons are invalid andshowed that policy reaction function can besuccessfully estimated and included in astructural econometric model. They used atrade model of the international market forsoybeans, rapeseed and other related productsto illustrate the idea. By simulationexperiments, they showed that the model ,vithendogenous policy variables performs well incomparison 'With a model which treats policyexogenously. Examples of other research thathave attempted to account for governmentintervention include Reed and Ladd (1980)and Sarris and Freebairn (1983).

Model Specification and EstimationThis paper employs a similar model to thatformulated by Wong to identify the economic

factors affecting the rice industry in Malaysia.The model used in this study, however,incorporates the guaranteed minimum priceequation which was suggested by Meilke andGriffith. The model, formulated in aggregateterms for this paper, consists of 3 equations.It includes a domestic supply equation (Q.), adomestic consumption equation (C.), a priceformulation (GMP,), and an identity (themarket clearing identity). The specification ofthe model is a follow:

1. The Domestic Supply EquationThe quantity available for consumption frOIDdomestic production (Q.) in a particular yearis primarily the result of production decisions.The supply equation is the familiar Nerloviantype where the quantity of rice produced isregarded as a function of the expected pricesof rice and of competitive crops. Quantities areadjusted each period by a fraction of thediscrepancy between the last period's observedvalue and the desired value. This hypothesis isconsistent with an economy where there arerigidities which prevent complete adjustmentin each period. I In Malaysia, paddy farmers areguaranteed a minimum price (GMP

t) and the

GMP is usually maintained above the worldprice.

Paddy farmers depend on other source (s)of income to supplement their income frompaddy. The variable PR,_1' the price of rubber,is the expected opportunity cost of resources.Based on the optimization behaviour of theproducers, it is hypothesized that both Q.-l andGMP, will be positively related to Q. and thevariable PR

t_

1will have a negative sign.

2. The Domestic Consumption EquationEconomic theory states that the quantitydemanded is a funtion of the price of thecommodity itself, the price of related

lThe partial adjustment (PA) and adaptive expectationmodel has long been used to investigate the dynamics of thesupply of agriculture products. Recently, these models havebeen nested in the more general PAAE model by Doran(1988) to diagnose for the appropriate specification usingthe likelihood principles. To simplify the analysis, it isassumed here that the PA r....lodel is sufficient lO capture theunderlying dynamiCS.

PERTANIK;\ VOL. 13 NO.2. 1990 291

AH~IAD ZL:BAJDl BAHARU~-tSH.AH

commodity and income level. During eachmarketing year, the National Paddy and RiceAuthority (LPN) announce the wholesale andretail prices for each grade of rice. Thedomestic consumption equation is simplyspecified with the retail price of rice (RPRJ,world price of wheat (WPWI), income (\';) andpopulation (POP.) serving as the dettrminants.

Bases on the utility optinlization behaviourof the individual, it is hypothesized that theretail price will be negatively correlated toconsumption. Income as mcaslu"ed by grossnational products and price of wheat, asubtitutc for rice arc bolli expected to havepositive signs. Population growth has beenidentified by policy analysIs to be the mostimportant factor affecting the domestic as wellas the international demand for food. Thedomestic demand for food products willprobably increase in accordance withpopuL:Hjon growth. Thus, population (inthollsand persons) is included to capture thispositive effect on aggregate domesticconsumption.

3. The Guaranteed Minimum Price EquationMalaysian rice policy emphasizes protection ofthe domesiic indusu-y and domestic price sta­hility. Malaysia's trade policy in rice is subjectedto import quotas, and a minimum pricemaintained above the world market price. Theobjective of the pricing policy is to provide a"fair" farm income to rice producers. The pricesupport programme is often rationalized ondistt-ibutional grounds by the decision makersand the guaranteed price is sometimes viewedas the relevant domestic supply inducingfactor.

To estimatt the behavioral equations forthe minimum price variable, a simple spe­cification employed by Meilke and Griffith(1983) is used in this study. Three variablesare expected to capture the major factorsinfluencing guaranteed price and they are theworld market price (vVPRt), the volume ofimported rice (M,) and the lagged dependentvariable (GMP,_I)' The lagged dependentvariable is included to allow for partialadjustments which reflects a government'sstrong tendency to be cautious in adjusting

rninimllm price. In Malaysia, the minimumguaranteed producer price is generally setabove the world prices. As the world marketprice is increased, the subsidy or protectionprovided to the producers declines. Hence, thisshould result in increased pressure fromproducers to increase price supports. Thus, thewodd price for rice (P,",'.) is expected to bepositively related LO GMP

t• The volume of

imports (M,) is a policy variable and ishypothesized to be negatively correlated toGMP.,

The complete model for the rice sector inMalaysia is as follows:The domestic supply equation is

I. Q, :: fJ. u + fJlIQ.~t + f3t~GMP, + fJl~l'R'_l + UI<

The domestic consumption equation is

2. C, '" fJ~n + f3~,RPR, + PttY, + fJ~.~\\rpW, + fJt,POP, + U1 ,

The guaranteed minimum price equation is

and the market clearing identity is

4. S, - S,_I == Q, - C, + \1,

The model, formulated in aggregatedterms, consists of 3 equations (3 endogenousvariables) and a market clealing identity. Theendogenous variables are ~=domestic

production of rice; Ct=domestic consumptionof rice, GMPI=guaranteed minimum price andSt=stocks.:! There are ten exogenous vmiables.The exogenous variables are PR._,=price ofnatural rubber; Yt=gross national product,GMPt_l=guaranteed minimum price lagged oneperiod; RPRt=retail price of rice in Malaysia;WPRI=world price of rice; ~_l=domestic

production lagged one pel-iod; vVPW,=worlrlprice of wheat; POP,=population; Ml=volumeof rice imported into Malaysia and S,_,=stockslagged olle period. The U;, (where i=I,2,3)are the random disturbances_ The small lettert appealing a., subcript denotes t1le time period

tSlOtks nlay he an importam policy variahlc in thepresent comex!. Howevcr, given that data Oil stocks arcsubjected to large errors and the specification of the stockequaunn usuall~' invokes a complex dynamic relationship.stock an: treated as an identity"

PERTA"NlKA VOL. l~ NO.2, 1990

I:\,CORPORATTNG GOVERN;"'IET\Cf lNTERVENTION IN RICE MARKET: CASE OF RICE POLICY

- crop year for agri<.:ultllrc production statisticsand calender year for all other statistics. Pricesand income have been deflated by theconsumer price index with 1980 as the baseyear. The order and rank conditions ofidentification are both satisfied and equation(I) to (3) are overidentified. The 2SLS methodwill privide consistent estimates of the struc­tural coeffiecient in equations (1) to (3). The2SLS is a limited-information estimation tech­nique. The resulL~ of this estimation techniquearc compared with the full-informatiunmethod and for this purpose, the 3SLS is used.

DataThe annual data used in fitting the model areobtained mainly from the Malaysian Ministryof Agriculture and Rural Development (PaddyStatistics), the Statistical Year Book for Asia andthe Pacific, the International Financial StatisticYearbook dan World Rice Statistics 198.'5. Theperiod of analysis is from 1960 to 1985, thusproviding twenty-six observations in eachequation.

Thl' Regrt}.t~i()n ResultsThe estimates of the structural coefficienL'" arepresented in Table J. The figures in paren­thesis arc the standard errors. Most of the signson all the estimated coefficients are consistentwith theoretical expectations. The supply equa­tion fits the data fairly well bu t the estimatedcoefficients were found to be statisticall}'insignificant. The coefficient of adjustment is0.1738, implying thal about 17 percent of thediscrepancy between the desired and actualproduction level is eliminated in a year. TheresulL~ of the regression support the hypothesisthe producers respond positively to the pricethey received and with some lagged response.the coefficients for Q-l is statistically significantat the one percent level.

The explanatory v·.ariables appering in theconsumption equation have the expccted signs.In genend, the resulLli from fitting the con­sumption equation are better than the supplyequation. In should be noted that theconsumpt.ion figures used in this study includeprivate stock. It is not possible to separatethem in the analysia given that data on private

stocks arc poorly document for the periodunder investigation.

The explanatOl)' power of the estimatedguaranteed price equation is high, suggestingthat the lagged GMP, world market price(\NPR) and imporL~ are imponant policy varia­bles. Positive feedback from world marketprices was found in the Malaysian market forrice. This suggests that Malaysia is not willingto alter its domestic agricultural policies, atleast in the short-run, in respond to fluctuatingworld market conditions. This is consistentwith the nation expressed by developmenteconomists that stable domestic price is animportant policy target which cannot beallowed to fluctuate which external conditions.Further, it can be concluded that the strengthof the world market price feedback is weakcompared to the lagged dependent val;able.This implies that the domestic price policy hasa strong partial a~jllstment component in theMalaysian rice market.

The elasticities at the mean correspond tothe 2SLS results are presented in Table 1. Theyshow that production adjusts rather slowly toprice changes. This is consistent with thefindings by Wong (1978) for Thailand and byNik Fuad (1984) for Malaysia. The estimatedshort-run price elasticity of demand is -0.14.Nik Fuad, however, reported -0.5 [or his studyand concluded that the estilnate was probablybiased upward because of the inclusion ofprivate S[ock change in the demand equation.The negative income elasticity implies that riceis an inferior good. Nik Fuad also foundnegative income elasticities for Malaysia. Therelatively low elasticity of price transmissionsupports the theory that domestic pricingpolicies are effective in insulating domesticmarket from a change in world price at leastin the shon nm.

The same model was estimated by 3SLSmethod and the reSulL'i are also reported inTable 1 so that a comparison can be madebetween the procedure of the two estimations.The values of the coefficeints bet\\1cen the twoestimating procedures do not appear to bedifferent. The similarities are revealing whenthe elasticities are compared. For examples, the2SLS estimates imply a shon-run price elasticity

PERTANIK:\ VOL U ;\'0. 2, 1990 293

AHMAD ZUBAJDI BAHARUMSHAH

TABLE 1The results of the regression analysis

No. Estimated equations

A. Two-Stage Least Square:

1. <4 = 241.07 + 0.8262 <4-1 + 0.9299 GMP, - 3.3560 PR'_1 R'=.84 h=-1.l5(137.32) (0.1249)** (1.0798) (0.5460)

2. C, = -772.0 - 2.566311.1'11., - 0.0213 ~ + 4.7231 WPW, + 0.2311 POP, 11.'=.93 DW=1.37(336.13)* (2.7586) (0.0063)** (2.3656) (0.0361)**

3. GMP, = 0.7793 + 0.9685 GMP,_, + 0.1158 WPW, - 0.0183 M, R,=.99 h~0.63

(7.8707) (0.0359)** (0.0239)** (0.0147)

B. Three Stage Least Squre:

1. <4 = 178.53 + 0.8886 <4-1 + 0.2267 GMP, - 0.1933 PR,_, 11.' = .84 h~2.51

(134.95) (0.1210)** (1.0315) (0.5181)

2. C, = -709.5 - 1.9661 RPR, - 0.225 Y, + 4.2794 WPW, + 0.2244 POP, R,=.93 DW=1.35(298.5)* (1.9707) (0.0055)** (1.8329)* (0.0321)**

3. GMP, = -7.812 + 0.9730 GMP'_I + 0.1156 "VPW, - 0.0003 M, R,= .99 h=2.39(7.7260) (0.0354)'* (0.0229)*' (0.0143)

Estimates of Elasticities at Means:

1. Price elasticity of supply2. Cross price elasticity3. Price elasticity of demand4. Cross price elasticity of demand5. Income elasticity6. Elasticity of price transmission

Short-run0.0830

-0.0569-0.1387

0.2161-0.1223

0.1093

Long-run0.4856

-0.3273

4.4790

Note: The estimated par.:uneters of 2StS and 3SLS were obtained using the SHAZAM package. The usual test of signific..'lllcemay not be applicable here and the single equations statistics reponed here must be interpreted wilh care. The estimatedoflhe elasticities were calculated using the resullS from the 2SLS method, Figures in the parentheses are the standard errors.** significant at one percent level.* significant at five percent level

to demand of -0.14 while the 3SLS results inan e1asliciIy of -0.1 1. Similarly, the computedelasticities for all others do not differ betw'eenthe two estimation techniques.

The 3SLS method is a system method andit utilizes more information than the 2SLS thatis, it takes into account the entire structure ofthe model. Both estimators are biased but theyare consistenL The 3SLS. however, is an asymp­totically efficient estimator. The reason for thelack of asymptotic efficiency in 2SLS is becauseit disregards the correlation of the disturbancesacross equations and over identified restrictionsin other equations. However, the 3SLS requiresa complete knowledge of the entire model. A

single specification error in one of theequations is transmitted during estimation toall Ihe equations in the model. The 2SLS ispreferred over the 3SLS since the accuracy ofthe specification of some of the equations isuncertain. The 2SLS, however may be lesssensitive to specification error in the sense thatthose parts of the system that are correctlyspecified will not be affecled appreciably byerrors in specification in other parts (SeeKlein, 1974). Finally. the Breusch-Pagan LM.test for diagonal covariance matrix indicatedthat the covariance matrix was diagonalimplying that the 3SLS estimalor would reduceto 2SLS estimator.

294 PERTANIKA VOL. 13 NO.2, 1990

INCORPORATING GOVERNMENT INTERVENTION IN RICE MARKET, CASE OF RICE POLICV

2. HeteroskedasticityTo test for heteroskedasticity, a simple testproposed by Breucsh and Pagan (1979) isavailable. Here we assume that

where Zj a' = h(o.o + alZ,1 + (J.2Zi2 + ... +a Z ) and h(.) possesses first and second de-

p 'privatives with respect to a. Our Ho: CJ.'l= ... =a =0, so that under Ho: V(U)~h(al)~constant.The LM test statistic for Ho is 1/2 SSR fromthe regression of e 2 ja'!. on Z_ where, e. areOLS residuals and cr'~E -'E In. Roenker (1981)

Model Diognostic TestOne of the most important components ofeconomic model-building is tests forspecification en-ors. The pupose of this sectionis (0 test the assumptions on the disturbanceterm. The disturbance term in the classicalregression analysis assumed that the dis·turbance are homoskedastic, serial inde­pendent and normally distributed. Thefollowing section discusses the tests.

1. Serial Correlation:Serial correlation of the error term can leadto inefficien t estimators and predictions, andto inconsistent estimates of ~ if a laggeddependent variable is in the set of regressors.Perhaps the most useful test for serialindependence against AR(p) and MA(q) hasbeen developed by Breusch (1978) and God­frey (1978). If U, follows an AR(p) process,then Ut=PIU I _ 1 + P2UI_2 + ... + PpUt _ p + E l ,

where E l is white noise. The LM test of thenull hypothesis H o: PI=P,= .•.. =pp~O isobcained by replacing U . with U . U~1, 2, ..

t-J t-J., p), the lagged values of the OLS residuals.The LM test statistic under the null hypothesisis calculated as TR' and TR' - .«p) under Ho'where T is the samples size and R is the co­efficient of multiple determination from theauxiliary regression. The test is conducted forthe three system of equations using theresiduals from the 2SLS. The computed .<'(3)are 1.07,4.0, and 3.7 for equations 1,2 and 3respectively and the critical .<'(3) equal 37.65.Thus, the null hypothesis H o: PI~P,~P.~O

cannot be rejected at 0.05 level of significance.

CONCLUSION

In general, the results of the model satis­factorily explained the behaviour of the Malay­sian rice sector and the residuals are well­behaved despite the weakness of data used inthe analysis. After analyzing the empirical re­sults, one can conclude that rice producers arerelatively unresponsive to guaranteed mini­mum price in the short-run and are also fairlyunresponsive in the long-run. The low priceelasticities suggest that there is little scope forprice manipulation in achieving the rice policyof self-sufficiency. Thus, if policy makers areunwilling to accept the income consequencesof large price changes. they may need 000­

price policies to boost domestic production infuture. Such measures should primarily aim atraising productivity through the introductionof high-yielding seeds and improvement In

production technology and practices.

3. Normality TestNumerous tests have been proposed to test fornormality. The test proposed by Bera andJarque (1981) is used in this paper. The teststatistic under the null hypothesis is given by:

LM = n [b,l6 + (b,-3)'/24j - .<'(2)

where b l = 0,/0,'. b, = 0,/0" 0, = I/n Lei(r~2, 3, 4) and b (i= I, 2) are estimates of f3. s.The normality tlest is carried out for the 3equations using the residual form the 2SLS.The computed x: for equations 1, 2 and 3 are9.31, 9.33 and 7.77, respectively. Thus, thehypothesis, that the errors are normallydistributed, cannot be rejected.

cr'; = V(U;) = h(Z; a)

The test is constructed for the system ofequation and the results show no evidence ofheteroskedasticity.

showed that this test is not robust under oon­normality and suggested a slight modificationof the test statistic. Asymptotically under thenull hypothesis of homoskedasticit)',

LM = TR'2 ~ x(P-11 under Ho

where R is the coefficient of determination inregressing E i on Z;.

(i=I, 2, ... , T)cr', = V(U) = h(Z; a)

PERTANIK.O\ VOL. 13 NO.2, 1990 295

AI-L\1AD ZUBAIDI BA.HARU~ISHAH

The important factors that have beenidentified as affecting consumption arc popu­lation, prices and income. Both the incomeand price elasticities are found to be inelastic.Wheat is a substitute and has acquired animportant position in the Malaysian diel. Thehypothesis that rice is an inferior good isconsistent with recent findings.

The empirical estimates of the guaranteedprice equation suggest that the minimum priceis dctenninecd by the world market price,import volume and the:: lagged guaranteedminimum price. The Malaysian price supportprogramme is sensitive to changes in worldmarket situations. Some form of feedback rulerather than an ad-hoc price support policy isused to induce producers to make necessaryadjusunents.

Finally, the model recognized theimportance of including domestic pricingpolicy in the rice sector in Malaysia but ofcourse it is not without limitations. The modeldoes not consider stock changes to avoid thecomplex dynamic problem. The model, beinga partial equilibrium one, also ignoresinteraction with other commodities.

REFERENCES

BERo\. A.R. and C.M. jARQUE. 1981. An EfficientLarge-Sample Test for Normality of Obser­vations and Regression Residuals. AustralianNatiollal Universiry. Canberra.

BREUCH, T.S. 1978. Testing for AutocondaLioll inDynamic Linear Models. Australian EconomirsPap" 17:~~4-3~5.

BREL1CH, T.S. and AR PA<.;A'\L 1979. A Simple Testfor Heleroskeda.~ticit)'and Random CocfliciemVariation. Ero1101nf'ltira 47:1287-1294.

BREL"CI-I T.S. and A.R. PA{;AK 1980. The Langrange

Multiplier Test and irs Application to ModelSpecification in Econometrics. RelliPlIJ ojEconomir Studies 47:239-253.

DORAN E.D. 1988. Specification Test for the PartialAeljustment and Adaptive Expectations Models.A mer. I Agr. Emil. 70:713-723.

COLOMA.", R.H. 1975. Staple Food Self·Sufficiencyand Distributive Impact or Malaysian RicePolicy. Food Research TnsJilule Studies 14:251-293.

GODFREY, L.G. 1978. Testing ror Higher Order SerialCorrelation in Regression Equations 'WhenRegressors Included Lagged DependentVariable. }:.:rollometrica 46: 1303-131 O.

KlEIN, L.R. 1974. A Textbook ofEconometrics. 2nd edn.,New.Jersey: Prentice~Halllnc., Englewood ClilTs.

KOENKER, R 1981. A Note on Studentizing a Testfor Heteroskedasticity. Junwl of Econolnl'lrirs17:107-112.

NIl\. Fu.w KA.\{IL. 1984. An Economic Analysis of theMala)'"sian Rice Sector: Prospects and PolicyAlternatives. Ph. D. Thesis. vVashington StateUniversit}'.

MHKU:, K.D., and G.K. GRUHTH. 1983. Incor­porating Polic)' Variables in a Model of the"\lorld Soybean/Rapeseed Market. Amer. j. Agr.Eeon. 65:fi5-n

REE.D, M.R. and G.\V. Lo\.DD. 1980. Feed Grain Im­ports and Feed Grain Prices in ImportingCounu·ies. Research Bulletin 588, Iowa Agri~

culture and Home Economics ExperimentalStation" Ames, Iowa.

SARRIS, A.H. and J. FREE&-\IR."l. 1983. EndogenousPrice Policies and International 'Wheat Prices.Amer. j. Agr. Eron. 65:214-224.

WONG, CHUKG Mr:"JG. 1978. A Model for Evalualingthe Effects of Thai Government Taxation onRice ExporLS on Trade and \Vclfare Amer. j. Agl".Ecoll 65:65·73.

(RPcfiued 27 June, }989)

2YG PERTANIKA VOL 13 NO.2, 1990