Philippine Agricultural Crop Production Cycle: Evidence and Determinants

A paper submitted to the University of the Philippines School of Economics

in partial fulfillment of the requirements for Economics 199

Submitted by

Christopher John F. Cruz Frances Margaret Canlas

October 2004

Abstract This paper is an attempt to prove the existence of a cycle in agricultural crop production in the

Philippines using time series data from 1970-2003 of the four major crops in the country – rice, corn,

coconut and sugarcane. The extraction of a relevant cycle in time series is implemented using the

Hodrick-Prescott Filter. The study conducts a deviation from trend analysis to test the relevance of

the resulting cycle.

The paper takes the analysis a step further by trying to identify the significant determinants of the

cycle. The authors hypothesize that the major determinant of the cycle is the El Niño Southern

Oscillation (ENSO) extreme events – the El Niño and La Niña. To this end, four multiple regression

models are constructed with the ENSO, domestic and world prices, hectarage, agricultural loans,

implicit tariff and the domestic capital formation as the relevant variables. Two measures of ENSO

are used – the Sea-Surface Temperature Anomaly and the Southern Oscillation Index.

Results from the statistical extraction show that cyclical fluctuations do manifest in the agricultural

crop production in the Philippines. The turning points of the resulting cycle coincide with the ENSO

years. The time series also shows marked deviations from the trend, which coincide with the turning

points of the cycle as well as with the ENSO years. The graphical analyses show that indeed, ENSO

events play a crucial role in the crop production cycle. This is confirmed by the results of the multiple

regression model. Other explanatory variables in the model prove themselves significant as well.

However, the extent of the effect of the determinants, as the findings in the graphical analyses and

regression results show, is determined by crop-specific factors. It is recommended that further

studies consider these crop-specific factors as well as other variables including specific government

policies. In addition, the use of a multivariate analysis as an alternative model to identify the

determinants is recommended as an area for future research.

ii

Acknowledgments The writers would like to extend their sincere gratitude to their parents and families for everything, to

their IDEA family - Mr. Jade Redoblado, Ms. Ruby Lemence, Marsha, Cathy, Pao, Ma’am Ella,

Ma’am Christine, Melissa, Jonats, Lau and Jamir for all the guidance and support, to Christopher’s

brods in the Pan Xenia Fraternity for the encouragement and counseling in the times of mental

breakdown, to the librarians of the School of Economics and of the Bureau of Agricultural Statistics,

and to Dr. Gwendolyn Tecson for making this paper worthy to be called a UP School of Economics

thesis.

iii

Waiver Relevant portions of this work may be quoted and used for research and other scholarly purposes

provided the proper citation is made.

Christopher John Fernandez Cruz Frances Margaret Canlas

iv

“We must remember that knowledge is only in past tense; learning is only in present tense; and prediction is only in future tense. To have sustainable agriculture, we need to be able to know, to learn, and to predict.”

- Chris Maser, The Redesigned Forest

v

Table of Contents I. INTRODUCTION.......................................................................................................................................1 Working Hypothesis 3 Objectives and Significance of the Study 3 Outline of the Study 4 II. REVIEW OF RELATED LITERATURE..........................................................................................5 On Agricultural Crop Production Cycle 5 On the Causes of Agricultural Crop Production Variability 7 On ENSO and the Agricultural Sector 10 III. CONCEPTUAL FRAMEWORK.......................................................................................................17 Defining Cycles in Time Series 17

Agriculture Cycle in the Context of Business Cycles 18 Sources of Production Variability 19 Climate Variability and Production Fluctuations 20 Output Price Risk & the Cobweb Model 23 Capital Requirement and Risk 24 Hectarage/Area Harvested 25 The Role of the Government 27 IV. METHODOLOGY.................................................................................................................................29 Decomposition Analysis of the Time Series 28 The Hodrick-Prescott Filter 30 Dependent Variable Specification 31 Selection of Explanatory Variables 32 Model Specification 33 Data Description, Sources and Limitations 34 V. PRESENTATION OF RESULTS AND ANALYSIS....................................................................37 Data Presentation 37

Proving the Crop Production Cycle 42 Rice Cycle 42 Corn Cycle 44 Coconut Cycle 45 Sugarcane Cycle 46 Deviation from Time Trend 48 Determinants of the Agricultural Crop Production Cycle 51 The Estimated Multiple Regression Models 51

vi

VI. CONCLUSION........................................................................................................................................64 SELECTED REFERENCES.....................................................................................................................68 APPENDICES.................................................................................................................................................72 Appendix A. Consensus List of El Niño and La Niña Years 72 Appendix B. Estimation of Rice Regression Equations 74 Appendix C. Estimation of Corn Regression Equations 77 Appendix D. Estimation of Coconut Regression Equations 80 Appendix E. Estimation of Sugarcane Regression Equations 83

vii

List of Figures

Figure No. Title Page

5.1 Crop Production Raw Time Series Data 37

5.2 Average SOI and SST Anomalies 38

5.3 Average Domestic Prices 39

5.4 Average World Prices (FOB) 39

5.5 Area Harvested 40

5.6 Gross Domestic Capital Formation in Agiculture, 41 Agricultural Loans and Implicit Tariff

5.7 Rice Cycle 42

5.8 Corn Cycle 44

5.9 Coconut Cycle 46

5.10 Sugarcane Cycle 47

5.11 Rice Trend-Cycle vs. Trend 48

5.12 Corn Trend-Cycle vs. Trend 48

5.13 Coconut Trend-Cycle vs. Trend 49

5.14 Sugarcane Trend-Cycle vs. Trend 50

viii

List of Tables

Table No. Title Page

5.1 Rice Regression Result 54

5.2 Corn Regression Result 57

5.3 Coconut Regression Result 58

5.4 Sugarcane Regression Result 61

5.5 Significant Determinants of the Crop Cycles 63

ix

1 Introduction

Agriculture is the economic lifeline of the Philippines. Total agricultural land area is estimated at 10

million hectares, 45% of which are situated in lowlands and 33% in the uplands. Agricultural crops

and livestock account for about one-fifth of the gross national product. Rice and corn remain to be

the most important crops of the country. Historically, though, Philippine agriculture was also

dominated by other crops including coconut, sugar, banana, tobacco, abaca and pineapple.

In the literature, there is a substantial number of studies about agricultural production in the country.

An area of research, which has been the concentration of studies for a number of agricultural

economists, focuses on the factors which affect the level of production, production trends and

variability. Economists agree that technological innovations in the form of enhanced seed varieties,

farm equipment and irrigation facilities have typically resulted in increases in production in both

lowlands and uplands. The increasing trend in production, in absolute levels, may also be attributed

to the more focused and effective government programs of the government since the end of the

Marcos era. Government expenditures devoted to agricultural research, training, as well as in

financing are crucial factors in crop production. However, beyond these internal dynamics, studies

[David in Balisacan and Hill, 2003; Power and Intal, 1990] showed that world market developments

have also affected domestic production. This, in effect, implies that the domestic market is relatively

open to the dynamics of world markets.

Empirically, the absolute level of production fluctuates, that is, it rises and falls depending on certain

shocks and input variability, among other factors. However, little is known about whether in fact,

there exists a cycle, i.e., regular fluctuations, in Philippine Agriculture. According to Niemira and

Kline [1994], economic cycles are a consensus of cycles in many specific activities, which have a

tendency to peak and trough around the same time. There are five types of cycles and one of which

is the agricultural cycle. The other four are Inventory or Kitchin cycles, Fixed Investment or Juglar

1

cycles, Building or Kuznets cycles and Kondratieff cycles. Probably the best-known sector cycle in

economics, the classical agricultural commodity cycle is what Nicholas Kaldor refers to as the

cobweb pattern. However, this cyclical process is most often associated to Mordecai Ezekiel [1944]

who suggested that regular fluctuations occur in agricultural production because (1) the following

period’s production is determined by current or past prices, and (2) the current price is determined

by current production. Conceptually, the agricultural cycle can be triggered by drought or any other

natural disaster at home or abroad that would reduce supply and raise the price of the commodity.

[Niemira and Kline 1994]

In the Philippines, technical studies show that agriculture is one of the sectors vulnerable to extreme

climate events such as the El Niño and La Niña phenomena, the two extreme events of the El Niño

Southern Oscillation (ENSO). Due to climatic factors, production (in absolute levels) of crops like

rice, corn, coconut and sugarcane have declined sharply to cause a fall in the productivity [Tibig

2002]. In the United States, studies abound on yield sensitivity to various climatic scenarios.

Generally, results show that yield deviations associated with two extreme ENSO phases are spatially

and crop dependent. It should be noted, though, that in the Philippines, studies have not proven

that the fall in production due to extreme climatic events is substantial enough to induce an

agricultural cycle, as it is defined in this study. In fact, the National Statistical Coordination Board

(NSCB) does not seem to recognize the importance of the agricultural cycle as it excludes

Agricultural Gross Value Added (GVA) in its computation of the Composite Leading Economic

Indicators (LEIS)1.

1According to the NSCB, the LEIS involve the study of the behavior of indicator that consistently move upward or downward before the actual expansion or contraction of overall economic activity. The system is based on an empirical observation that the cycles of many economic data series are related to the cycles of total business activity, i.e., they expand in general when business is growing and contract when business is shrinking.

2

This study attempts to test the theory on the existence of the agricultural production cycle against

empirical data, i.e., using Philippine time series data from 1970-2003. To do so, the authors conduct

a time series analysis of the production data of the four major crops namely, rice, corn, coconut,

sugarcane and banana. The particular examination for each major crop also aims to determine those

that are most susceptible to the cycle, assuming at this point that it actually exists. The analysis is

taken a step further by considering the possible determinants of the production cycle. Specifically, it

attempts to analyze how the ENSO, domestic and world output prices, capital, agricultural financing

in terms of loans, area harvested and government interventions, impact on the Philippine agricultural

crop production cycle.

Working Hypothesis

There are reasons to believe that the theory on regular fluctuation of agricultural production may be

supported by Philippine time series data. For one, theorists point to drought, i.e., ENSO, as a major

factor inducing the cycle. The Philippines is no exemption when delineating the countries affected by

this extreme climatic event. Hence, the working hypothesis of the paper is this: that there exists a

cycle in Philippine agricultural production and the El Niño Southern Oscillation is a major

determinant of it. The study also hypothesizes that certain factors affecting production variability

may impact to the extent that they cause major fluctuations in the over-all level of production and

therefore may be called determinants of the production cycle. These include domestic and world

output prices, capital, agricultural loans, area harvested and government interventions.

Objectives and Significance of the Study

This study attempts to test the theory of cobweb pattern or what could be generally called the

agricultural cycle against the Philippine time series data on agricultural production. In order to see the

real cycle of the agricultural production in the country, the study takes out the trend factors of the

3

time series data. The paper also intends to examine factors that could possibly determine the

Philippine agricultural production cycle.

The Philippines is primarily an agricultural country. The agricultural sector contributes to the Gross

Domestic Product (GDP) by employing millions of Filipinos as farm laborers and by being the major

source of farm household incomes. Thus, whatever shocks affecting crop production also affect farm

households. In the same breath, these shocks affect the rest of the country through the aggregate

level of domestic supply and by implication, through the commodity prices as well. It is thus

apparent that this investigation on the existence of the agricultural production cycle and its possible

determinants is of great importance, not only to Juan de la Cruz but also to policy makers, most

especially. The results of this study have significant implications in terms of disaster preparation,

crop specialization, financial assistance, output forecasting and further research.

Outline of the Study

This paper is divided into six sections. The second section is a survey of studies, both in the

international and local literature, about the production cycle and the causes of crop production

variability. The third section discusses the theoretical framework behind the study. The fourth

section delineates the techniques employed to prove the existence of a cycle in agricultural crop

production and to examine the significance of certain variables vis-à-vis the crop production cycle.

The fifth section discusses the results and presents the analysis of the study. The last section includes

the conclusion, policy implication and recommendation for further research.

4

2 Review of Related Literature

Agriculture remains a critical sector of the economy although recent income accounts surveys show

that agriculture has been declining in its share of the Philippine GNP or GDP (Gross National

Product/Gross Domestic Product). It was – and perhaps will always be – a key element of the

country’s food security as well as the major source of household income for over half of the

population. Some economists argue that shocks in the agricultural sector mainly explain its instability

relative to other sectors. The first part of this survey of related studies would thus dwell on the early

efforts to explain production variability in the sector.

On Agricultural Production Cycle

While there was a variation in the way theorists talked about the agricultural cycle, it was taken, not as

the major subject of their studies, but as a mechanism in explaining the business cycle at large. In

fact, in a survey of business cycle theories, the agricultural cycle was central to only one of those

delineated – the Climate Theories of Cycles.

In his 1937 book Prosperity and Depression, Gottfried Haberler tries to have a three-fold classification

of the agricultural theories of the business cycle. He first talks about how the periodicity in

agricultural cycles caused business cycles elsewhere. Among the major proponents of this category

are William Stanley Jevons, H.S. Jevons and Henry Moore [Niemira and Kline 1994]. Meanwhile,

A.C. Pigou and Dennis Robertson are placed in the second category which posits that periodic

fluctuations in agricultural crops could act like an exogenous starter and set off a cumulative cyclical

response in the rest of the economy [Niemira and Klein 1994]. The third one, on the other hand,

claims that agriculture, while a passive element, could respond disproportionately to fluctuations

originating elsewhere in the economy.

5

The main proponents of the climate theories of business cycle are Jevons, Moore and Johan

Akerman [Niemira and Klein 1994]. The first thinkers who put forward the idea of a regular cyclical

pattern in the general level of economic activity provide mostly “exogenous cycle” theories.

Essentially, they relate economic cycles to other exogenous cycles found in nature such as weather,

which in turn might be affected by astral phenomena. They claim that these natural phenomena

caused variations or movements among tangible things such as harvests and/or intangibles such as

people’s moods, and that these in turn were the primary causes of observed economic fluctuations.

Expounding on his sunspot theory of the business cycle, Jevons [1884] read two papers before the

British Association in 1875 and 1878. He argues that the variations in sunspot affect the power of the

sun’s rays, influencing the bountifulness of harvests and thus the price of corn, which in turn, affect

business confidence and give rise to commercial crises. Hence, the chain of causation goes from

changes in meteorological conditions to agricultural crop production to business activity. It should be

noted that his work was the first to point to climatic variability as the major cause of variations in

crop production. Jevons [1884] believes that the periodicity of the solar cycle and commercial crises

is “too coincidental to be dismissed.” He is considered as the first economist to argue that the phases

of business activity had a regular, measurable and predictable periodicity.

Akerman [1928, 1932] also claims that the main causes of business cycle are climatic factors.

However, he had a more ingenious idea connecting longer business cycles to the magnified effects of

a series of small, weather-driven seasonal cycles. Moore [1914] shares the same view that weather

conditions cause crop cycles. In Economic Cycles: Their Law and Cause, published in 1914, he notes that:

…these cycles in crops constitute the natural, material current which drags upon its surface the lagging rhythmically changing values and prices with which the economists more immediately concerned.2

2 Quoted in Wesley C. Mitchell, What happens during Business Cycles, 57.

6

Meanwhile, Mordecai Ezekiel [1944], the well known statistician, tried to incorporate the inelasticity

in the demand for agricultural products and the long gestation period (the period between the

decision to produce and the marketing of the output) in agriculture to develop an explanation for

agricultural instability. [Niemira and Kline 1994]

On the Causes of Agricultural Production Variability

Gommes [1999] of the Food and Agriculture Organization (FAO) identifies some of the factors

which affect the variability of agricultural yields. Depending on their variations over time, these

factors are grouped in three categories: (1) growing smoothly, such as more or less regular technology

and management trends (i.e., mechanization, varieties, irrigation, and the farmers’ know-how), but

growing more abruptly for several years in succession in the case of innovations; a typical innovation

was exemplified by the new introduction of irrigation; (2) discontinuous, like extreme factors of

various origins and policy decisions which affect management (for instance, farmers may decide to

use less fertilizer if it is no longer subsidized) or infrastructure, such as the construction of a road

which provides access to new markets; and (3) pseudo-cyclic like weather. [Gommes 1999] He notes

that while weather variability remains one of the main factors behind the inter-annual variability of

agricultural production, it is hard to estimate how much production is actually lost due to variability

under normal circumstances. Essentially, though, he argues that climate variability impacts

agricultural output (production) through its effects on yields and areas planted. He claims that yields

are affected not only by weather, as the main “random factor,” but also by mostly continuous

technological trends (e.g., new varieties and management), innovations (which include management

innovations), agricultural policies (mostly national policies) and extreme factors of various origins.

On the other hand, he notes that areas planted/harvested depend more on economic factors, e.g.,

labor variability, level of mechanization and expected returns (prices), and environmental conditions,

e.g., poor weather during the cycle, damage to infrastructure due to extreme conditions, or even very

low yields for which it is not economical to harvest at all. He proposes a method to estimate how

7

much production is lost due to climate variability. In a production time series, he takes the maximum

production value Pm in the 7-year interval from Y-3 to Y+3 for each year. Then, he computes the

difference between the production P of year Y and Pm and expresses it as a percentage “loss”: (Pm-

P)/Pm*100%. There is, however, a caveat to this method: it implicitly assumes that no marked

technological progress took place in the seven-year period. Using his case studies on Thailand,

Tanzania, Niger, Mexico, Africa, Italy, France and Canada, he concludes that at regional scale, the

largest fraction of inter-annual variability of crop yields and production can be ascribed to technology

and management in developed countries while roughly 20% of that variability is due to other factors,

of which at least half is weather dependent.

In a fact sheet released by the World Meteorological Organization, in collaboration with FAO, it is

stated that considerable variability occurs around the mean global surface temperature increase of

between 0.3°C and 0.6°C since the late 19th century. The fact sheet asserts that approximately 20-30

percent of this temperature variability from the global trends is linked to ENSO, especially in the

Southern Hemisphere. While it is true that production is adjusted to the average climatic conditions

of the region, climatic variability creates hazards to which agricultural systems are generally not well

adapted. While high temperatures exacerbate the effects of drought, damage crops and their

establishments and reduce yields, low temperatures, frost and heavy snowfall cause sharp drops in

yield and destroy crops, it argues.

In Adaptation of Agricultural Production Systems to Climate Variability and Climate Change, Meinke,

Baethgen, and Gimenez [2001] observe that most of the research on the interactions of climate with

agriculture has been focused on yield or production risks, since agriculture is subjected to

uncontrollable events related to weather. They contend that technology plays a crucial role in this

type of risk since it can lead to both a reduction in the variation of productivity levels (e.g.,

introducing irrigation in a water limited environment) or an increase in yield variability (e.g., rainfed

8

systems based on high-yielding cultivars, well fertilized but receiving very variable rainfall). They also

identify three other types of risk crucial in agricultural production: price or market risks (the change

in expected prices after production process began), institutional risks (changes in policies and

regulations affecting agriculture), asset risks (damage to equipment, livestock, etc.) and financial risks

(fluctuations in interest rates on borrowed capital, cash flow difficulties). Taken altogether, they argue

that they affect production variability significantly.

In the Philippines, Tibig [2002] notes that a documentation and analysis study done by a technical

working group in Philippine Atmospheric Geophysical and Astronomical Administration (PAGASA)

shows that production of rice, corn, coconut and sugarcane has been impacted by extreme climate

events (e.g., droughts/floods due to El Niños/La Niñas and occurrences of tropical cyclones)

causing sharp falls in productivity. She stresses that the vulnerability of the agricultural sector to

climate variability/change has been assessed during the past few years but has been limited to a few

important crops using climate scenarios generated by global circulation models (GCMs) and

processed-based crop yields and extreme climate events. Among the non-climatic factors, she

identifies population growth, land use conversion leading to depletion of agricultural/riceland areas

and irrigation development. The annual population growth rate of 2.3% necessitates an

unprecedented increase in food production so that livestock, fisheries and grain production must

necessarily be projected to increase to maintain food security, she argues. This condition is

aggravated by the continuing rural-to-urban migration due to shrinking opportunities in the rural

area. Based on a 1996 Bureau of Soils and Water Management (BSWM) study, depletion of irrigated

ricelands to non-agricultural uses, e.g., human settlements and economic zones, is estimated at a

conservative rate of 2,000 to 3,000 hectares per year. Irrigation development in the country is

estimated at 42% based on the 1996 statistics.

9

The PAGASA [2001] documentation and analysis study has three significant and interesting findings.

First, declines in Gross Value Added (GVA)3 and in volume of production in four principal crops,

i.e., rice, corn, sugarcane and coconut, are observed to generally coincide with El Niño years, while

the increases with La Niña years. Sharpest falls in GVAs and volume of production in the agricultural

sector occurred in 1982-1983 and 1997-1998, the periods which saw the birth, peaks and decay of the

two strongest El Niños in the 20th century. Lastly, increases in GVAs in rice and corn are attributed

to favorable rainfall conditions during the La Niña years.

Tibig [2002] concludes that the vulnerability of the agricultural areas and production in the country to

climate variability is a result of the inter-annual occurrence of droughts mainly due to El Niño events,

high-frequency and variability of the occurrence of tropical cyclones with the attendant floods and

the incidence of salt water intrusion near coastal areas which are due to excessive use of available

ground water storage and storm surges.

On ENSO and the Agricultural Sector

The international literature is rich with studies on the factors determining the level and trend of

agricultural production. A number of these studies identify climatic variations associated with ENSO

as major influences in the agricultural sector. Some of these studies deemed relevant by the

researchers would be briefly tackled. The most that could be deduced from the international as well

as in the local literatures are estimates of the ENSO impact on certain agricultural commodities on

particular countries or states. The researchers find the methodology, data sources and results in these

studies worth noting, nonetheless.

3 GVA is the difference between gross outputs or gross value of the goods and services produced during the accounting period and intermediate outputs or value of goods and services such as raw materials and supplies used in the production process during the accounting period estimated at constant 1985 prices.

10

In their 1998 study called ENSO Influences on Agriculture in the Southeastern United States, Hansen,

Hodges and Jones aim to identify crops that are vulnerable to ENSO-related weather variability and

therefore likely to benefit from application of ENSO-based climate forecasts. Using six crops

(peanut, tomato, cotton, tobacco, corn and soybean) in four states (Alabama, Florida, Georgia and

South Carolina), they analyze the historical (1960-1995) response of total value and its components

(yield, area harvested and price) to ENSO phases (El Niño, Neutral, La Niña) and quarterly sea

surface temperature (SST) anomalies in the eastern equatorial Pacific. The authors work on the

hypothesis that ENSO influences the value of crop production both directly through its influence on

weather patterns that affect yields, and indirectly through producer and market response to previous

and expected yields in the region and in other important production regions. Their study tries to

analyze the responses of crop variables to both categorical and continuous measures of ENSO. The

variable used as a continuous measure of ENSO strength was SST anomalies averaged over the

region of eastern equatorial Pacific known as NIÑO3. On the other hand, the three ENSO phases

were used as categorical measures of ENSO activity. Using the study of Sittel [1994], Hansen, et al.,

also used the COAPS (Center for Ocean-Atmospheric Prediction Studies) classification based on

monthly SST anomalies in the region from 150° to 90°W and 4°N to 4°S, smoothed into five-month

running means. If SST anomalies are at least +0.5°C (≤ -0.5°C for La Niña) for at least six

consecutive months, a year is classified as El Niño (La Niña). They also add another requirement to

this classification: that the six-month period should start before October and should include October

through December. Hence, their paper includes ten El Niño events for the period October 1957 to

September 1995 (1957, 1963, 1965, 1969, 1972, 1976, 1982, 1986, 1987 and 1991) and seven La

Niñas (1964, 1967, 1970, 1971, 1973, 1975 and 1988).

To reflect changes in crop management which, they argued, is a function of price fluctuations and

the availability of new technology, their analyses separate each historical yield series into a trend.

They also apply a harmonic smoothing technique, developed by Press, Flannery, Teukolsky and

11

Vetterling [1989] to separate higher-frequency anomalies from lower-frequency trends. The

technique is used because it removes any linear trend from a data series, transforms the data to the

frequency domain using Fourier analysis, applies a low-pass filter covering a specific period,

transforms back to the time domain, then reinserts the linear trend. Since El Niño events (1525-

1988) recurred with a crucial period of 6.7 years according to Hanson, et al., [1989], the authors use a

filtering period of 7.0 years (frequency of 0.143 yr-1) to smooth all crop data. Using the Analysis of

Variance (ANOVA) [Steel and Torrie 1980], they tested their hypothesis that ENSO phase influences

anomalies about the trend for each crop variable. Meanwhile, Duncan’s multiple range test is applied

to identify which ENSO phases differed in their effect for each crop that showed a significant

response to ENSO phase. Results of the study support the hypothesis that ENSO influences the

total value of crops in the southeastern US but not the hypothesis that ENSO influences crop values

by influencing their prices.

David Legler, et al. [1999] carry out a related study with Hansen et al. [1998], but concentrate on the

impact of ENSO-related climate anomalies on crop yields in the United States. They use the three

phases of ENSO – Warm events (El Niño), Cold events (El Viejo or La Niña) and Neutral – to

simulate yields of 7 field crops through the Erosion Productivity Impact Calculator (EPIC)

biophysical model developed by William, Jones and Dyke [1984]. Climate variables used are daily

precipitation and maximum and minimum temperature data for 54 US station over the period 1948-

1987. The index chosen for the ENSO variable is the Japan Meteorological Agency index, which is a

5-month running mean of spatially averaged SST anomalies over the tropical Pacific. Similar to the

study of Hansen, et al. [1998], if index values exceeded 0.5°C for 6 consecutive months including

October-November-December (OND), the ENSO year of October through the following

September is categorized as Warm event (index values greater than +0.5°C); Cold events (index

values less than –0.5°C); or Neutral, all other values [Legler, et al. 1999]. The authors argue that the

specific choice of ENSO year definition was based on prior analyses of Sittel [1994]. Moreover, they

12

justify the use of EPIC model to simulate crop yields accurately by citing the study of Williams, et al.

[1989]. Rosenberg, et al. [1997] also validate the EPIC model for the study of ENSO effects.

While a plethora of studies have documented the effects of relatively high frequency and low

frequency weather developments on economic activity, there was a dearth of studies directed toward

examining the impact of medium-frequency weather fluctuations such as ENSO events. Some

remarkable exceptions to this are Handler [1983], Adams et al. [1995], Debelle and Stevens [1995],

Solow et al. [1998] and the two studies previously discussed. It should be noted, however, that these

studies are of limited use because (1) they focused on a small number of commodities and certain

geographical areas thought to be significantly affected by ENSO [Brunner 2002] and (2) majority of

these studies used dummy variables to represent years in which there was unusual climactic activity.

A variation in the analysis of the impact of ENSO on the agricultural sector is conducted by Brunner

[2002] of the International Monetary Fund. His study focuses on the historical effects of the ENSO

cycle on world primary commodity prices, as well as on other measures of economic activity. The

paper differs from previous studies of the same nature primarily because it estimates several

econometric models using broad measures of price and economic activity and because the models

include continuous measures of ENSO intensity (SST and sea-level air pressure anomalies in the

Pacific Ocean) instead of dummy variable measures. The same kind of ENSO measure will be used

in this study. Plotting SST anomalies against commodity price changes, Brunner [2002] finds that

there is a close association between the two variables. His analysis concludes that El Niños are

generally associated with subsequent real commodity price increases, whereas La Niñas are associated

with price declines. His first conclusion may not be surprising at all since one could expect that drops

in production levels due to ENSO related weather changes necessarily pump up prices in the market,

ceteris paribus. His second conclusion, however, seems a little problematic. Theory aside, one could

argue that La Niñas may help increase production through augmented water supply. However, it is

13

equally true that typhoons associated with La Niña may be strong enough to destroy crops, bringing

down production levels and assuming a relatively inelastic demand curve, should result in a hike in

agricultural commodity prices. While the issue in his paper remains, the researchers find the

econometric models used helpful in their study. Several vector autoregressive (VAR) models were

estimated to gauge the effects of ENSO events on world prices and growth. A measure of ENSO

intensity, ENSOt, which was either the SST anomaly or the SOI anomaly measure, was included in

each of the four-variable VAR model. Each VAR also contained the average CPI inflation rate ( πtg )

and the average GDP growth rate ( ∆yt ) and a measure of real commodity price inflation ( πtcp - πtg ).

Thus, the models were of the following form:

ENSOt = µs + A11(L) ENSOt-1 + εt

Xt = Φs + A21(L) ENSOt + A22(L) Xt-1 + ηt

where ENSOt represents a measure of intensity; Xt =[ πtcp - πtg πtg ∆yt ]; µs and Φs are seasonally

varying constants; A11(L), A21(L) and A22(L) are polynomials in L, the lag operator; εt is an exogenous

shock to ENSOt and ηt is a 3 x 1 vector of innovations to Xt.

Several results of the analysis are significant in this study. For one, the SOI anomaly measure of

ENSO intensity appears to have a much stronger statistical relationship with the economic variables

than the SST anomaly measure does [Brunner 2002]. The Granger causality tests and the impulse

response functions point to a strong statistical relationship between ENSO and world commodity

prices and to a statistically significant relationship, though to a lesser extent, between ENSO and

world inflation and economic activity. For instance, a one-standard-deviation positive surprise in

ENSO raises commodity price inflation by about 3.5 to 4 percentage points. The influence of ENSO

is stronger in the long run as almost 20% of the variation in real primary commodity prices is

associated with it.

14

In his paper Climate Trends and Agricultural Management, Taylor [2002] notes that in his studies in the

early 1960s, it was concluded that El Niño reversed the tropical rain pattern and had little or no

impact elsewhere. However, subsequent to 1972, El Niño has manifested apparent effect on crop

yields in many states in the US. In fact, he suggests that risk planning in the US Corn Belt might be

made purely on the phase of ENSO events. He also concludes that during La Niña years, the risk of

drought is double the long-term average such that crop yield of above or below the trend should be

expected at a confidence level of 80% in the US Corn Belt. It can be inferred from his analysis that

the ENSO impact on crop yields is strong enough to cause a considerable variability in production.

Supporting this result was an earlier study by Carlson, Todey and Taylor [1996] on the risks

associated with the SOI for the US Corn Belt. They conclude that very large crop yields are most

likely during the negative SOI – indicative of the presence of El Niño.

The ENSO phenomenon is pointed out as the primary cause of year-to-year variations in climate in

the United States. Adams, et al. [1999] claim that climate is the primary determinant of agricultural

productivity. If both statements are valid, then it should follow that the primary determinant of

agricultural productivity is the ENSO phenomenon. Their paper tries to assess the economic

consequences of ENSO events in a two-stage process. In the first, the consequences of the changes

in weather patterns due to ENSO phases on crop yields are measured using estimates from both

biophysical simulation models and historical yield data. On the other hand, the second stage

incorporates those yield differences into an economic model in order to assess the aggregate

economic damages of ENSO events. Using EPIC, estimates of the yield implications of weather

changes from each ENSO phase for eight field crops (corn, wheat, soybeans, cotton, barley,

sorghum, oats and hay) are developed. An alternative approach is also employed to estimate yield

consequences of ENSO phases. It is based on 25 years (1972-1996) of crop yield data for all crops

included in the economic model. The yield data are first detrended to remove effects of technological

change and acreage shifts on yields and then yield estimates are projected for each year. The

15

deviations between the projected and actual yields are recorded as a percentage change from the

projected. To obtain a joint probability distribution across 63 US regions based on the 25 historic

weather events, those deviations are applied to the 1997 yield projection. They claim that the

distribution reflects, among other factors or influences, the variation due to the weather including the

ENSO phase. The results from Economic Consequences of ENSO Events [Adams, et al. 2000] reflect a

range of weather and yield conditions. However, the overall implication of the findings is not

surprising – extreme events whether they be El Niño or La Niña, have adverse effects on agriculture.

Meanwhile, Naylor et al. [2002] measure the connections among SST anomalies, rainfall and

Indonesian rice and corn production from 1971-1998. To model the impacts of climate variability on

grain area and production, they initially examine the two intermediate linkages which are the

associations between ENSO and rainfall and between rainfall and rice production, and then examine

the direct association between ENSO and rice area planted and harvested. Despite the number of

available ENSO indices, SST anomaly is chosen because it is thought to contain less high frequency

random variance associated with atmospheric fluctuations that are distinct from ENSO [Cane, Eshel

and Buckland 1994]. The ENSO, rainfall and crop production data are converted to first differences

since this permitted a direct analysis of year-to-year variability and help to remove statistical problems

of first-order autocorrelation among the residuals. They argue that year-to-year changes in

production are easily interpreted relative to deviations from hypothetical trends and that first-

difference models perform consistently better econometrically. From their findings, they conclude

that year-to-year August SSTA fluctuations explain about half the inter-annual variance in paddy

production during the main (wet) season. They also note that Indonesia’s paddy production varies on

the average by 1.4 million tons for every 1°C change in August SSTAs.

16

3 Conceptual Framework

Since this paper attempts to prove the existence of a cycle in agricultural crop production in the

country, the authors should, first and foremost, provide a discussion of the criteria utilized in judging

the presence of relevant cycles in time series. Thus, the first part of this section deals with how

statistical theorists and business cycle practitioners assess the significance of cyclical fluctuations. The

second part, on the other hand, discusses the theoretical underpinnings behind agricultural cycle as

well as the relationship of certain variables to production variability.

Defining Cycles in Time Series

Mitchell and Burns [1946] define business cycle as a type of fluctuation found in the aggregate

economic activities of nations that organize their work mainly in business enterprise and that it

consists of expansions occurring at about the same time in many economic activities, followed by

similarly general recessions, contractions and revivals which merge into the expansion phase of the

next cycle. They also argue that the sequence of changes is recurrent but not periodic. In the

literature, agriculture cycles are often discussed as a means of explaining business cycles.4

Schenk-Hoppe [2001] asserts that business cycles can be defined as deviations of macroeconomic

data from an underlying trend that is not, however, observable in general. He stresses that in the

literature, the decomposition of a time series into a trend and remaining “cyclic” part is in principle

arbitrary. This is supported by Diebold and Rudebush [1999] who identify two widely acknowledged

key characteristics of the cycle: (1) a large number of macroeconomic variables appear to move

together, i.e., there is co-movement of economic series over the cycle; (2) fluctuations in economic

activity exhibit persistence – deviations from the average or trend level of activity5. Kydland and

4 See Niemira and Kline [1994] for one. 5 In this paper, the first characteristic identified by Diebold and Rudebush [1999] does not apply for this study only concerns itself with the cycle in agricultural crop production.

17

Prescott [1990] maintain that any definition of the trend and cycle is necessarily statistical and that

decomposition is a representation of the data.

A large strand of literature assumes that the trend is smooth and that all fluctuations are driven by

small transient productivity shocks. Given this assumption, the usual practice is that aggregate data

are detrended using Hodrick-Prescott filter or the band-pass filter by Baxter and King [1999]. Reeves,

et al. [2000] note that the cyclical residual obtained is an estimate of the combined cyclical and

irregular component of the series. Schenk-Hoppe [2001] stresses that the well-known stylized facts of

the thus-specified business cycle provide a benchmark for any business cycle model.

These studies show that in general, there are no hard-and-fast rules or clear-cut criteria used in

judging the existence of relevant cycles in any time series. Some statistical methods are already

established and have proved themselves in business cycle analysis. In general, business cycle

practitioners are concerned with cyclical fluctuations in GNP and other variables, where cyclical

fluctuations are measured as deviations from the trend of the process. [Canova 1996] Therefore, one

yardstick in determining the existence of cycles in time series is the significance of the deviations

from the trend which, as Lucas [1977] notes, corresponds statistically to estimating a residual, i.e.,

analyzing a detrended series. Having no well-established criteria in detecting the presence of a

significant cycle in any time series, Schenk-Hoppe [2001] emphasizes that any attempt to identify or

approximate economic cycles has to be based on economic theory, which enables one to break open

the circularity of specifying one unobservable variable with the other.

Agriculture Cycle in the Context of Business Cycles

Agricultural cycle is attributed mainly to price fluctuation. [Niemira & Kline 1994] The classical

agricultural commodity cycle posits that agricultural productions seem to have regular fluctuations

because the following period’s production is determined by current or past prices and the current

18

price is determined by current production6. The major claim of the theory is that natural disasters

could set off the cycle. Intuitively, disasters like drought would reduce market supply and, holding all

other things constant, would therefore raise the price of crop products. Assuming that actors are

rational, production the next period would rise in response to the high prices this period. However,

the increased supply, ceteris paribus, would bring down prices and the process would go on. It should

be noted that the length of the cycle is determined by the time required to produce a new generation

of a crop or livestock [Niemira & Kline 1994].

The agricultural commodity cycle is however not completely validated by empirical results. According

to Niemira and Kline [1994], research works find that the theoretical model does not fully capture

the actual production cycle. This is probably because the theory only ascribed the agricultural cycle to

price fluctuations. This leaves open the possibility that other factors actually determine the cycle.

Thus, the next part of this section discusses the variables that affect production variability because of

their risk characteristics.

Sources of Production Variability

Annual fluctuations in crop production have tremendous implications on food security as well as on

food self-sufficiency7 objectives of the country. In the literature, variability in expected outcome or

yields is attributed to the major sources of production risks. It can be argued, therefore, that as a

starting point of defining and analyzing the possible determinants of agricultural production cycle the

authors could look into the concept of risk – and its sources – in agriculture.

6 Price risks shall be elaborated on the latter part of this section where the discussion on the cobweb model shall be formalized. 7 David differentiates food self-sufficiency from food security. She argues that while the former is a political objective, the latter means ensuring that household incomes, especially those of the poor, are sufficient to avail of adequate food at reasonable prices.

19

Uncertainty is a state of mind in which the individual perceives alternative outcomes to a particular

action while risk has to do with the degree of uncertainty in a given situation. [Roumasset 1979]

According to the United States Department of Agriculture Risk Management Agency [1997],

agricultural production implies an expected outcome or yield. In a practical sense, the degree of

disparity between what is expected and the actual production level greatly affects a person’s or a

firm’s financial goals. The agency identifies five primary sources of risk: Production, Marketing,

Finance, Legal, and Human Resources. For the purpose of this paper, discussion will be limited to

production risks8. Kaan [2000] identifies the major sources of production risks as weather, pests,

diseases, and the interaction of technology with other farm and management characteristics, genetics,

machinery efficiency, and the quality of inputs.

Climate Variability and Production Fluctuations

According to the Food and Agriculture Administration, climate variability causes considerable

fluctuations in crop yields and productivity. In fact, Adams, et al. [2000] claim that climate is the

primary determinant of agricultural productivity. Not only do extreme climate conditions often lead

to mass displacement of population, they also cause enormous damages to food production systems.

Climatic variability causes risks to which agricultural systems are not well adapted. A number of

studies look into the effects of weather variability to food production levels and variability. However,

majority of these studies9 has identified climatic variations associated with the ENSO phenomenon

as major influences in production fluctuations. Adams, et al., [2000] argue that one can trace much of

the year-to-year variations in climate to ENSO in many parts of the world.

The El Niño Southern Oscillation (ENSO) refers to a quasi-period redistribution of heat and

momentum in the Pacific Ocean. Solow [1995] notes that it is the largest source of inter-annual

8 For an extensive discussion of the production of the other four sources of risk, please refer to Introduction to Risk Management: Understanding Agricultural Risks published by the United States Department of Agriculture Risk

20

climatic variability on a global scale10. ENSO has three phases: the Warm (El Niño), the Cold (La

Niña) and the Neutral periods. The El Niño phase of ENSO occurs, on average, every four years.

Translated from Spanish as ‘the boy child,’ El Niño was originally used by Peruvian fisherman to

describe the appearance of a warm ocean current off the South American coast around Christmas

season. According to the Commonwealth Bureau of Technology [2004], the term El Niño technically

refers to the extensive warming of the central and eastern Pacific that leads to a major shift in

weather patterns across the Pacific. During El Niño events, significant changes in the atmosphere

and ocean circulation occur. The bureau outlines these as follows: (1) warmer than normal ocean

temperatures across the central and eastern tropical Pacific Ocean, (2) increased convection or

cloudiness in the central tropical Pacific Ocean, (3) weaker than normal easterly trade winds and (4)

low negative values of the Southern Oscillation Index (SOI). El Niño events come in different

strengths: weak, moderate, strong and extraordinary, depending on how much is the increase in the

sea surface temperature relative to the average. Ocean temperatures can average 2°C – 3.5°C above

normal between the date line and west coast of South America during a strong El Niño event.

The Cold phase of the ENSO, also called La Niña, appears when surface water in the eastern Pacific

is abnormally cold. Typically lasting for approximately 9-12 months, La Niña episodes represent

periods of below-average sea-surface temperatures across the eastern tropical Pacific. Temperatures

average 1°C - 3°C below normal. Interestingly, there have been fewer cold than warm events in the

past two decades. As a result, scientific interest has been focused on El Niño.

Over the equatorial Pacific Ocean, tropical rainfall, wind and air pressure patterns associated with

both El Niño and La Niña are linked to the underlying sea-surface temperatures. They usually begin

to form during June-August, reach peak strength during December-April and then get weak during

Management Agency in December 1997. 9 See the Review of Related Literature section of this paper for a discussion of some of these studies. 10 For a comprehensive review of this phenomenon, see Philander (1990).

21

May-July next year. The two episodes likewise occur every 3-5 years on average though some

prolonged episodes have lasted 2 years and even as long as 3-4 years.

Two measures have been frequently used in studies to quantify and indicate ENSO: the Sea-surface

Temperature Anomaly (SSTA) and the Southern Oscillation Index (SOI). SSTA measures the

deviation in degrees Celsius from a long-term monthly mean sea-surface temperature in the Pacific,

calculated from a base periods of 1950-79. Naylor, et al. [2002] observe that it almost always lies

between plus and minus 4°C. El Niño periods are associated with high and positive SSTAs in the

central Pacific, while La Niña periods are linked to low and negative SSTAs. Meanwhile, SOI is a

measure of the large-scale fluctuations in air pressure occurring between the western and eastern

tropical Pacific. The index has been traditionally calculated based on the difference in air pressure

anomaly between Tahiti and Darwin, Australia. While prolonged periods of negative SOI values

coincide with abnormally warm ocean waters across the eastern tropical Pacific typical of El Niño

episodes, prolonged periods of positive SOI values correspond to the unusually cold ocean waters.

[National Weather Service 2004]

Varied and far-reaching effects of ENSO typically include drought in a number of countries

including the Philippines. Its impact, however, goes beyond alteration of typical weather patterns; it

can also disturb ecosystems and put many species in danger. Consequently, the literature is rich with

studies attempting to quantify the economic impact of ENSO events. For instance, the NOAA

predicts that on average, El Niños result in agricultural losses approaching $2 billion, or nearly 1-2

percent of the total crop output in the United States. Adams, et al., [2000] find that both phases of

ENSO result in economic damages amounting to $1.5 to $1.7 billion loss for the El Niño and $2.2 to

$6.5 billion for La Niña. Generally, the conclusion of these and many other studies is that ENSO

events do induce production variability in agriculture, and consequently impose costs on producers

and consumers.

22

Output Price Risk & the Cobweb Model

Because crop prices tend to fluctuate, farmers are faced with what is generally called output price

risk. This fluctuation may be demonstrated by the cobweb model. Sicat [2003] puts it simply: if

farmers use this year’s market price in determining what and how much to plant in the following

year, then next year’s supply would depend on this year’s market price. Notably, there is a one-year

time lag in the decision of farmers to plant certain crops.

Enders [1995] presents a stochastic version of the traditional cobweb model to explain the volatility

in agricultural prices.

Dt = a – γpt γ > 0

St = b + βpt* + εt β > 0

St = Dt,

where Dt = demand for an agricultural product in period t

St = supply of the agricultural product in t

pt = market price of the agricultural product in t

pt* = price that farmers expect to prevail at t

εt = a zero mean stochastic supply shock

The model11 assumes that consumers would buy as much product as desired at the market clearing

price p. Essentially, farmers form their expectations in a naïve way because they base their supply

decision on their expected price pt*. Actual supply is equal to the planned quantity b + βpt* plus a

random supply shock εt. It should be noted that pt* = pt-1 as the cobweb model goes. The presence of

the stochastic term in the model implies that stochastic equilibrium is such that ever-present supply

shocks prevent the system from remaining at equilibrium. Thus, the cobweb model is a basic

representation of the agricultural cycle.

11 This part is largely taken from Enders [1955].

23

Meanwhile, Duncan [1990] notes that developing countries, which are heavily dependent on

commodity trade, find themselves susceptible to the fluctuations in world market. Fluctuations

through price and exchange rate mechanisms pose difficulties to both producers and consumers and

even to the government. Evidence to this kind of susceptibility abounds in the literature. In Balisacan

and Hill [2003], David argues that the Philippine agriculture sector performed well relative to other

developing Asian countries in the 1970s because of the early advent of the Green Revolution in rice

and the boom in world commodity prices. However, since the 1980s, agricultural growth has been

slow, as marked slowdown in the Gross Value Added growth rates has been recorded. The sharp

drop in world commodity prices in the 1980s undoubtedly contributed to the fall in growth rates.

[David in Balisacan and Hill 2003]

Capital Requirement and Risk

In their paper entitled The Determinants of Agricultural Production, Mudlak, et al., [1997] claim that

agriculture is more cost-capital-intensive than non-agriculture and that capital is all the more

important as a factor of production in that land varies little over time. The level of agricultural

production is highly related to the utilization of technological capital. In the State of Food and

Agriculture press release, the Food and Agriculture Administration [2000] notes that extraordinary

but uneven gains in agricultural production and productivity have been achieved largely as a result of

different approaches to augmenting countries’ technological capital. Underlying major yield increases are

major technological forces, including improved infrastructure. FAO [2000] asserts that such advances

were made possible by public and private investments. It should be noted, however, that increased

inputs in the form of physical investments bring diminishing returns. In that sense, therefore, the

authors argue that fluctuations in agricultural production could be attributed to the efficiency of

physical capital utilized as well as to the level of general domestic capital formation. This argument is

being supported by Gommes [2000] of the FAO. In his theoretical illustration of some of the factors

which affect variability of agricultural yields, he identifies regular technology and management trends,

24

i.e., mechanization, which grow smoothly but more abruptly for several years in succession in case of

innovation.

Pinstrup-Andersen [1982] observes that the influence of technological change in production risk and

capital requirement is another important characteristic of technology that may influence the

distribution of benefits. He adds that the introduction of modern technology requires additional

capital and other factors necessary to ensure optimal production conditions. It is said, though, that

many farmers are reluctant to seek credit, especially if the perceived risks are high. Thus, another

element of capital requirement in the form of financial risk relates to production risk. Lipton12 [1979]

observes that poor farmers are subjected to downward fluctuations in output and income.

Essentially, credit finances much current input and investments, especially for small farmers with few

savings. Rural credit in poor countries is highly localized, aggravating the risks associated with bad

harvests. [Lipton 1979] By implication therefore, empirical researches conclude that credit availability,

as a representation of the capital input, significantly affect production variability.

Hectarage/Area Harvested

Intuitively, area harvested is directly related to the level of production, ceteris paribus. To a great extent,

absolute levels of production have positive relationship with the area planted. However, certain

factors affect the amount of area available and actually used for crop production. It could be argued

that these shocks could result to significant variability in production.

In the PIDS Annual Macroeconometric Model, Yap [2000] outlines the various models of production,

expenditure, and of the financial and fiscal sectors in the Philippines. Hectarage is one of the

significant variables affecting supply of crop products.

25

VAR = β0 + β1 PVARR + β2 HCTRG + β3 GOV AGRI/(PGNP/100) +

β4 Dum80 + β5 VAR(-1) + β6 TIME2 –

β7 PFEEDS*ER/(PGNP/100),

where VAR is the Value Added in Agriculture (Supply at constant 1985 prices)

PVARR is the implicit price of agriculture in real terms

HCTRG is the hectarage of agricultural loan

GOV AGRI is the government expenditures on agricultural sector

Dum 80 is Dummy 1980=1, 0 otherwise

VAR (-1) is the lagged Value Added in Agriculture (supply at constant 1985 prices)

TIME2 is the variable for time squared

PFEEDS is the imported price of feeds

According to Gommes [2000], areas harvested depend on economic factors. For instance, planted

areas vary according to labor availability and expected returns. He further notes that areas harvested

are usually strongly linked to infrastructure damage or even simply to low yields for which it is not

very economical to harvest [Gommes 2000]. In the Philippine historiography, several factors have

significantly affected the available amount of land actually utilized for agricultural production. To a

certain extent, the Comprehensive Agrarian Reform Program of former president Corazon Aquino

has enhanced the incentive of tenant farmers to cultivate their farms. However, owners of large

landholdings have taken advantage of the loopholes of the law by converting their agricultural lands

to industrial uses. Glossing upon the situation, one can conclude that, to a certain extent, this has

reduced the amount of land actually used in crop production. Moreover, land use conversion has

been estimated at a conservative rate of 2,000 to 3,000 hectares per year13. This estimate is based only

on depletion of irrigated ricelands to non-agricultural uses such as human settlement. From this

12 His article is in Roumaset, Boussard and Singh [1979].

26

evidence, the authors find that hectarage is in fact, a significant factor causing agricultural production

variability.

The Role of the Government

In a market economy, the government is supposed to devise and administer institutions that provide

incentives for efficient farm production while investing in the provision of public goods where

appropriate. [FAO 2000] Technological advancements have contributed much to the increase in yield

for the past 50 years. It should be noted that these advancements were rendered possible by research

and investment efforts by the government and by support from national and international

agricultural research centers. Factors from the input side, including availability and adoption of

modern seed varieties and irrigation development, are highly dependent on the government support

of agriculture. Government expenditures on agriculture may be broken down into various parts:

natural resource and environmental management, land acquisition and distribution, price

stabilization, production support, credit/insurance, irrigation, extension, research and development,

road and others.

In Balisacan and Hill [2003], David argues that [some of the] government policies distorted economic

incentives and the choice of policy instruments promoted rent seeking and raised the economic cost

of government interventions. In fact, she notes that past studies have concluded that until the early

1980s, price intervention policies created an incentive structure that was significantly biased against

agriculture [David 1983; Bautista 1987; Power and Intal 1991]. Commodity-specific policies affecting

input prices include import tariffs levied on agricultural inputs. Manasan and Querubin [1997] argue

that by altering prices faced by consumers and producers, tariffs and quotas result in market

distortions that, in turn, lead to inefficiencies in resource allocation. They add that the lowering of

tariffs and the lifting of QRs are expected to result in improved allocative and technical efficiency,

13 This was based on the Bureau of Soils and Water Management study done in 1996.

27

greater access to cheaper-priced, better-quality inputs and superior technologies, greater domestic

competition through a more rational market structure. Essentially, tariffs make imports more

expensive thus suppressing the demand for them. This, in turn, artificially cheapen the foreign

exchange which penalizes the country’s exports though the government initially protects domestic

producers of import substitutes [Manasan and Querubin 1997]

28

4 Methodology

This section describes the two stages involved to prove the existence of a cycle in the agricultural

crop production over the period 1970-2003 and to show how certain variables relate to production

variability and to the production cycle at large. The first stage requires statistical techniques to

detrend the time series. This leaves the cyclical and irregular components of the data. The second

stage, meanwhile, necessitates the use of regression models to determine how certain factors affect

the time series for each crop.

Decomposition Analysis of the Time Series

A time series is an ordered sequence of values of a variable at equally spaced time intervals. Time

series analysis aims not only to describe the various mechanisms that generated the data but also to

forecast values for the short to the long term. A time series could be decomposed into four

components: trend (T), seasonal (S), cyclical movement (C) and an irregular or residual (I)

component. Trend represents a general systematic linear or (most often) nonlinear component that

changes over time and does not repeat within the time series. It is therefore the long-term movement

of the time series. The seasonality component, meanwhile, is the regular periodic pattern that repeats

from year to year. Cyclical factors usually have a longer duration that varies from cycle to cycle.

Typically, a time series contains stochastic elements in the trend, seasonal and irregular components.

However, the irregular component, while not having a regular pattern, is somewhat predictable.

[Enders 1995] Decomposition models for time series include the additive ( T + C + S + I ) and

multiplicative (T x C x S x I) models. According to Bersales [2004] a time series may also have other

components like Easter effect/moving holidays, trading day variation, and extreme values/outliers

aside from those mentioned.

Following Reeves, et al. [2000] and Kydland and Prescott [1990], it is necessary that the seasonality

and trend components be taken out to come up with the cyclical component of the production time

29

series Note, however, that deseasonalizing time series is only applicable to monthly and/or quarterly

data. Since this study uses annual data values, it is unnecessary for the time series to actually be

deseasonalized. The resulting residual, which is presumably composed of the cyclical and irregular

components of the time series, is then matched with the developments in the agricultural crop

production over the period under review. This step tests the validity of the cycles given the historical

events in the country.

To test whether the production cycles are not only regular fluctuations in the time series, it is

necessary that the business cycle definition as used in the literature be focused on as the starting

point of analysis. As discussed in the conceptual framework, there are two defining characteristics of

a cycle as delineated by Diebold and Rudebusch [1999]. The first one does not apply in this study

because it requires the movement of a number of macroeconomic variables, possibly encompassing a

business cycle. The second, however, does apply. Essentially, the second cyclical attribute means that

an economic time series must exhibit persistent deviation from the average of trend level of activity.

In this regard, the study conducts a simple components analysis of the time series. This step utilizes

the trend-cycle components of the time series compared to the average trend level. Note that the

trend-cycle component is a product of the trend and the residual obtained through the Hodrick-

Prescott Filter, presumbaly consisting of the cyclical and the irregular components.

The Hodrick-Prescott Filter14

The Hodrick-Prescott (H-P) Filter is one of the established methods used to remove the trend

component of the time series. It is an algorithm for choosing smoothed values for any time series

14 The discussion on this part is heavily taken from the economics glossary of Mike Moffatt, available on the internet on http://economics.about.com/library/glossary/bldef-hodrick-prescott-filter.htm

30

data. By choosing smooth values {st} for the series {xt} of T elements (t=1 to T), it tries to solve the

following problem:

minimize {( xt - st )2 ... etc. }

Technically, the parameter λ>0 is the penalty on variation, where variation is measured by the

average squared second difference. As such, a larger value of λ makes the resulting {st} series

smoother; less high-frequency noise. Note that the commonly applied value of λ is 1600. It should

also be noted that H-P filtered data show less fluctuation than first-differenced data, since the H-P

filter pays less attention to high frequency movements. Further, H-P filtered data show more serial

correlation than first-differenced data.

Dependent Variable Specification

Since one of the objectives of the study is to capture the effects of irregular factors on the time series

data, the detrended time series is recommended by statisticians to represent the dependent variable.

As Reeves, et al. [2000] stress, the detrended time series or the cyclical residual obtained is an

estimate of the combined cyclical and irregular component of the series. In this manner, fluctuations

in production could be traced primarily to the impact of these irregular factors. Note that whatever

the resulting coefficients would be, one thing is certain: the trend effects have been filtered out in the

process.

To correspond to the agricultural production as a whole, four major crops are used in this study: rice,

corn, coconut and sugarcane. Historically, Philippine agriculture is dominated by these crops [Power

and Intal 1990]. Rice and corn are the main food grains while coconut and sugar are the main

traditional exports. In terms of area used in production, rice farms account for less than half of the

total number of farms. Corn and coconut farms were the next most numerous, together accounting

for almost 45 percent of the farm area.

31

Selection of Explanatory Variables

Since the authors hypothesize that the major determinant of the agricultural production cycle is the

El Niño Southern Oscillation (ENSO), a fair representation of the phenomenon is necessary. Both

SOI and SSTA have gained respect in climate-related studies. Brunner [2002] finds that SOI anomaly

measure of ENSO intensity appears to have a much stronger statistical relationship with the

economic variables than the SST anomaly measure does. However, a number of studies, like those of

Naylor, et al. [2001] use SSTA in their analysis and find the ENSO measure a significant variable in

their analysis as well. This study uses both ENSO measures to test which of them captures better the

effect of the ENSO phenomenon on the Philippine time series. Lagged values for each ENSO

measure are utilized. This is necessary as the effect of ENSO may be reflected in the data one year of

even two years after it hit the country. Since ENSO highly determines variability in rainfall and other

climatic factors [Naylor, et al. 2002; Legler 1999], the authors find it unnecessary that they be

included in the model. This is to avoid muticollinearity problems as well.

In accordance with the Cobweb Model, prices at first and second lags for each of the four crops are

included in the model. There are two variables of output prices included. The obvious one would be

the domestic output prices. However, given the relative openness of the country in terms of trade

and the historical effects of trade prices on domestic production15, the authors find it necessary to

include world commodity prices as well. To the extent that both prices are not perfectly correlated,

then the model does not have multicollinearity problems.

Given the two ways in which capital16 affects agricultural production in the country, two capital

variables are used in this study. The first one is the gross domestic capital formation in agriculture

which accounts for agricultural machineries like tractors, threshers, etc. The second is the agricultural

15 See David in Balisacan & Hill [2003] for a comprehensive discussion. 16 See the Conceptual Framework section of this study.

32

loans which essentially affect the capability of farmers to purchase agricultural inputs and their

expected rate of return. Similar to Yap [2000], the amount of area harvested is included as an

explanatory variable.

Studies show that the government plays a pivotal role in determining not only production levels per

se, but also economic incentives17 of farm and market players. Effective rates of protection affect

resource allocation and intermediate input prices [David in Balisacan & Hill 2003]. To account for

this government factor, the implicit tariff on agricultural inputs is used. These inputs include

fertilizer, pesticide and water pumps. Meanwhile, following Yap [2000] the study includes time and its

square as relevant variables in the model.

Model Specification

Since in this study, production levels of the four major crops are used in lieu of the aggregate

agricultural production, four separate regression models are constructed. In each model, two ENSO

measure will be used. Logically, the domestic price, world price and area harvested vary for each

crop.

Yit = β0 + β3 ENSOt + β1 DOMPRICEi(t-1) + β2 WORLDPRICEi(t-1) +

β4 GDKFt + β5 AGRILOANSt + β6 IMPTARIFFt + β7 GOVTEXPt +

β8 AREAit + TIME + TIME2 + εt

Where Yit is the detrended time series data for each crop;

i =1 for rice, 2 for corn, 3 for coconut, and 4 for sugarcane

ENSOt is the ENSO measure which could be ENSO1t for SOI or ENSO2t for SST

DOMPRICEit -1 is the lagged output farmgate prices (in pesos per kilo)

WORLDPRICEit-1 is the lagged world output prices (in dollar per metric ton)

33

GDKFt is the Gross Domestic Capital Formation in Agriculture (in million pesos)

AGRILOANSt is the amount of Agricultural Loans (in million pesos)

IMPTARIFFt is the Implicit Tariffs on agricultural inputs (%)

GOVTEXPt is the Government Expenditure on Agricultural Research (in million pesos)

AREAit is the harvest area in hectares

TIME & TIME2 are time variables

εt is the error term

Data Description, Sources and Limitations

Data on Southern Oscillation Index (SOI) anomalies – the deviations between air pressure

differentials in the South Pacific and their historical averages – were secured from the Bureau of

Meteorology of the Australian National Climate Center. Meanwhile the Sea-surface Temperature

Anomaly (SSTA) – deviations between sea surface temperatures in a given region and the region’s

historical average – indices were downloaded from the website of the Climate Prediction Center at

the National Oceanic and Atmospheric Administration18.

Domestic crop prices are categorized into three: farmgate, retail and wholesale prices. The study uses

the farmgate price since it is the most appropriate following the cobweb model. Note that it is the

farmgate prices that really affect the behavior of farmers and not the other two prices. Data for

farmgate prices were obtained from Power and Intal [1990] and from the Bureau of Agricultural

Statistics (BAS). Meanwhile, world output prices are freight on board (FOB) prices and were

obtained from the World Bank.

17 See David in Balisacan & Hill [2003]. 18 The data could be accessed at http://www.cpc.cep.noaa.gov/data/indices.

34

In the agricultural statistics website publication of the Philippine Institute for Development Studies

(PIDS)19, gross domestic capital formation in agriculture consists of breeding stock and orchards,

agricultural machineries, and tractors other than steam. Meanwhile, based on the National Statistical

Coordination Board (NSCB)20 publications, capital formation is comprised of fixed capital, including

construction, durable equipment and breeding stock and orchard development. In this study, data on

gross domestic capital formation in agriculture comprise of breeding stock and orchards, agricultural

machineries and tractors other than steam. However, the time series data on these variables were

measured differently over the 30-year period under study. Hence, the authors choose to limit the

components of the capital formation variable to tractors other than steam and agricultural

machineries, which have been measured consistently in the time series data available. The 1981-2003

figures were obtained from the National Statistical Coordination Board while the 1970-1980 figures

came from Power and Intal [1990].

Meanwhile, implicit tariff is defined as the tariff revenue on a good or group of goods, divided by the

corresponding value of imports. It is also defined as the difference between the price just inside a

border and the price just outside it, especially in the case of a good protected by an import quota. It

is often lower than the official or statutory tariff, due to failures in customs collection. Estimates of

the implicit tariff from 1970 until 2002 were lifted from David in Balisacan and Hill [2003].

Agricultural Loans consist of loans granted by commercial, rural, development and agricultural

banks. Data (in million pesos) came from the Central Bank of the Philippines and were published in

the Philippine Statistical Yearbook. On the other hand, data on the area harvested for each of the

crops under study were published by the Bureau of Agricultural Statistics and downloaded from the

PIDS website.

19 See www.pids.gov.ph for more statistical figures and analysis related to agriculture.

35

Since data pertaining to most of the variables in this study are officially provided by the

aforementioned government institutions, the authors take them as objective, true and accurate. There

might be a problem though with the data on implicit tariffs as these are estimates. However, since the

values were estimated by reputable economists, the authors take them accurate as well.

20 Refer to the website of the agency: www.nscb.gov.ph for more information.

36

5 Presentation of Results and Analysis

This section is divided into four parts: the first includes a preliminary analysis of the data used in the

study. Meanwhile, the second part presents and discusses the evidence to prove the existence of a

cycle in the Philippine agricultural crop production. This entails matching the developments in the

Philippine agriculture, with particular focus on rice, corn, coconut and sugarcane, and the

characteristics of the cyclical-irregularity residual obtained through the Hodrick-Prescott Filter. This

step is necessary, as it tests whether the graphical presentation of the production cycle actually makes

sense given the agricultural developments over the period under review. The third part includes a

deviation from trend analysis to test whether the cycles obtained differ from regular fluctuations in a

time series. Meanwhile, the last part attempts to evaluate the possible determinants of the production

cycle through a discussion and interpretation of the regression results.

Data Presentation

Raw time series data for the four crops under study - rice, corn, coconut and sugarcane - reflect a

general upward trend. In fact, using Hodrick-Prescott Filter, one can verify this upward trend in total

production over time. Except for sugarcane production which has considerable fluctuations, the

agricultural crops show a generally smooth time series.

Figure 5.1. Crop production raw time series data (1970-2003)

0

5000000

10000000

15000000

20000000

25000000

30000000

1970

1972

1974

1976

1978

1980

1982

1984

1986

1988

1990

1992

1994

1996

1998

2000

2002

met

ric to

ns

rice corn coconut sugarcane

37

Typically, time series data have the four components mentioned in the early part of this paper -

trend, seasonality, cycle and irregular factors. Since this study uses annual data, the seasonality factor

is presumably absent in the analysis. In order to detrend the data, the Hodrick-Prescott Filter is used.

The smoothing parameter λ=100 is utilized given the frequency of the data.

Figure 5.2. Average SOI and SST Anomalies

(1970-2003)

-13

-8

-3

2

7

12

1970

1972

1974

1976

1978

1980

1982

1984

1986

1988

1990

1992

1994

1996

1998

2000

2002

sst

soi

As shown in Figure 5.2, average SOI and SST values generally coincide with each other and show

cyclical fluctuations. This in part is due to the fact that the values of both measure range from

positive to negative numbers. The figure is consistent with the findings of Brunner [2002] that there

is a close association between the two variables. However, the average values of the SOI show more

fluctuations than the SST’s. Note that the SOI is measure of air pressure while the SST is a measure

of temperature. It could thus be expected that the SST experiences less variability than the SOI.

Having an ocular inspection of the figure alone, one can verify that 1972-73, 77-78, 82-83, 87-88, 91-

93, 94-95 and 02-03 are El Niño years while 1971-72, 73-76, 88-89, 98-99 and 00-01 are La Niña

years.

38

Figure 5.3. Average Domestic Prices (1970-2003)

0

2

4

6

8

10

12

14

16

18

1970

1972

1974

1976

1978

1980

1982

1984

1986

1988

1990

1992

1994

1996

1998

2000

2002

peso

s

rice corn coconut sugarcane

The average domestic prices of rice, corn, coconut and sugarcane show considerable fluctuations

over time. However, an upward trend is apparent in the four prices. While from 1970 until the start

of the 80s, all the four prices seem to converge at the P0.3 to the P2 level, the mid 1980s marked

their continuous upward trend and fluctuation. This could be partly attributed to the political

instability at the end of the Marcos regime and restructuring from the administration of Aquino. Of

all the four crops, only the domestic price of sugarcane rose to unprecedented levels by 1986.

Average sugarcane prices seem to manifest a relatively stable trend over time, compared to the other

three crop prices. Meanwhile, cyclical fluctuations are apparent in average world prices of the

products of the four crops. Only corn prices seem to have a generally smooth trend. Interestingly,

Figure 5.4. Average World Prices (FOB) (1970-2003)

0

100

200

300

400

500

600

700

800

1970

1972

1974

1976

1978

1980

1982

1984

1986

1988

1990

1992

1994

1996

1998

2000

2002

US

$

corn copra rice sugar

39

the average world prices of rice, corn and coconut seem to manifest cyclical fluctuations. This only

shows how volatile the world market prices can be.

Figure 5.5. Area Harvested

(1970-2003)

0

500000

1000000

1500000

2000000

2500000

3000000

3500000

4000000

4500000

rice corn coconut sugarcane

Average area utilized for coconut plantation soared in the 1970s from almost 1.9 million hectares at

the start of the decade to a little less than 3.5 million hectares at the end of the decade. However,

beginning early 1980s until 2003, there was no marked increase in the area used in coconut

production. Harvested areas for sugarcane show a relatively stagnant graph as well. This is in contrast

to the fluctuations characteristic of both the rice and corn graphs. Area used in rice production rose

at the early 1970s, but started to decline by the late 70s until the early 90s when it reached an

unprecedented level of 4 million hectares. Meanwhile, corn area experienced an uptick from the early

70s until the mid 80s. From then until 2003, harvested area for corn has been declining.

Gross Domestic Capital Formation in Agriculture shows considerable fluctuation over the period

under study. In general, though, it manifests an upward trend from the start of the 70s. This uptrend

was interrupted in the mid 80s, when it dropped to levels similar to the 1970 value. By the end of the

80s, it started its climb though it still shows fluctuation in the 90s. Meanwhile, agricultural loans show

an uptrend from the start of the period under study until the end. They experienced a rapid uptrend

40

in the 90s, reaching P126 billion in 2003. On the average, implicit tariff has been considerably stable

except for an increase in the late 70s. From the start of the early, 80s, it started to fall until it reached

the year 2000 value which is 50% lower than the 1970’s.

Figure 5.6. Gross Domestic Capital Formation in Agiculture, Agricultural Loans and Implicit Tariff21

(1970-2003)

0

20

40

60

80

100

120

140

1970

1972

1974

1976

1978

1980

1982

1984

1986

1988

1990

1992

1994

1996

1998

2000

2002

agriloans implicit tariff gdkf in agri

21 In this graphical presentation, gross domestic capital formation, which includes agricultural machineries, and tractors other than steam, is expressed in ten million pesos at constant 1985 prices; agricultural loans, taken as the sum of loans granted by commercial banks, development banks, rural banks & agricultural banks for each year, are expressed in billion pesos, at constant 1985 prices.

41

Rice Cycle

The extraction of the cyclical component from a time series using the Hodrick-Prescott filter resulted

in a production cycle as shown in the subsequent figures. The cycle of rice production displays

significant fluctuations over the period 1970-2003. It only has two major troughs, occurring on the

years 1972-1974 and 1997-1999. Coincidentally, strong El Niños hit the country on 1972-1973 and

1997-199822. These extreme climate events may have factored in the plummeting of total production

during those periods.

Figure 5.7. Rice Cycle

0.99

0.995

1

1.005

1.01

1970

1972

1974

1976

1978

1980

1982

1984

1986

1988

1990

1992

1994

1996

1998

2000

2002

From the early 1970s, rice productivity increased due to the introduction and increasing cultivation of

high-yielding varieties developed in the mid-1960s at the International Rice Research Institute. Green

Revolution, as it was popularly known, was accompanied by an expanded use of chemical inputs. In

fact, total fertilizer consumption rose by 80 percent, from 668 tons in 1976 to 1,222 tons in 1988.

The government also undertook a major expansion in the irrigation system to further stimulate

productivity. In sum, these developments bolstered production, which reached its peak around the

early 80s. A strong El Niño disrupted the momentum of the increase in 1982-1983, causing a slight

decline in production as seen in Figure 5.7. Over the period 1980-1985, average annual growth dived

to a mere 0.9 percent, in contrast to 4.6 percent for the preceding fifteen years. The 1982-83 drought

22 For a complete list of El Niño years, see the appendices.

42

was accompanied by the general economic downturn of the 1980s and the 1983-85 economic crisis,

all of which resulted in a sharp slide in the growth of value added in the rice industry in the 1980s.

Prices of agricultural inputs jacked up, crop loans staved off and palay prices dipped in this period.

As a result, plant nutrient consumption was estimated to have dropped by almost 15 percent and

increasing debts and shrinking income troubled the farmers. Hectarage used in palay production also

fell by an average of 2.4 percent per annum during the first half of the 1980s. By 1985, the country

imported 538,000 tons of rice.

The conditions improved in the late 1980s, resulting in a kink in the production cycle as seen in

Figure 5.7. However, the improvement in the market and farm conditions was not enough to pump

up production as the cycle marks a continuous downtrend until the late 1990s. To a large extent, this

was probably due to bad weather conditions. An El Niño struck the country in 1987-88, which was

followed by a strong La Niña in 1988-89. Over the period 1991-93, the country was hit by a twin El

Niño, a stronger one in 1991-92 followed by another one in 1992-93. Output levels fell by 1.5

percent, forcing the importation of an estimated 400,000 tons of rice during this period. As if the

occurrence of either the El Niño or La Niña was not enough for the century, the country

experienced a severe drought in 1997-98 causing output levels to plunge and the rice cycle to carry

on with its downswing. However, a La Niña immediately followed this drought in 1998-99. To a

certain extent, this somewhat reversed the situation or farm conditions. In fact, it may have been

beneficial to the rice industry at large, as the rice cycle shifted gears and started to climb up once

again. This uptrend in the rice cycle continues until the time of this writing. This is despite the

occurrence of a La Niña in 2000-01 and an El Niño in 2002-03. This upswing may therefore be

attributed to the increases in yield (due to the widespread use of modern and enhanced modern

varieties of rice) and hectarage in the past years. Improvement in the coverage and in the facilities of

the irrigation system may have likewise helped in the growth in yield and in easing the effect of the

long dry spell in 2002.

43

Corn Cycle

More pronounced fluctuations characterize the corn cycle. The launching of the Masagana '99 in the

early 1970s revolutionized the corn industry. This caused the slight uptrend in the corn cycle and

made the country self-sufficient in white corn. However, this was dampened by the occurrence of a

strong El Niño in 1972-73, which was followed by a strong La Niña in 1973-74, a weak La Niña in

1974-75 and again, a strong La Niña in 1975-76.

Figure 5.8. Corn Cycle

0.99

0.995

1

1.005

1.01

1970

1972

1974

1976

1978

1980

1982

1984

1986

1988

1990

1992

1994

1996

1998

2000

2002

From then on, the downtrend in the corn cycle went through. Ironically, around 1982-83, the corn

cycle went up despite the occurrence of a severe drought during the period. This suggests that other

factors may have played in the process that altered the negative impact of the ENSO. These include

developments in the corn seeds, which increased the yield of corn grains, and the sudden uptrend in

area used in corn production starting early 1980s. The upswing in the corn cycle proceeded

uninterrupted, despite the occurrence of an El Niño in 1987-88 and a La Niña in 1988-89, until 1991-

92 when a strong El Niño stopped the sustained increases in corn production. The downtrend, this

time, was exacerbated by a series of ENSO events of the century: 1992-93, 1994-95, and 1997-98 El

Niño. The rains brought about by the La Niñas of 1998-99 and 2000-01 may have been favorable to

corn production as the uptrend in the corn cycle strengthened from 1997 up to the time of this

writing.

44

Coconut Cycle

The Philippines is one of the largest coconut producers in the world and in fact, the biggest supplier

of coconut oil in the world. After Indonesia, it is the world's second largest producer of coconut

products. One can therefore immediately see how open the coconut industry is to the dynamics of

world coconut products' market.

Compared to the rice and corn cycles, there are no huge fluctuations in the coconut cycle. From the

very start of the 1970s, the coconut industry already experienced an upsurge in total production. Part

of the reason is that land devoted to coconut cultivation inched up by about 6 percent per year

during the 1960s and 1970s. This expansion was in response to the devaluation of the peso in 1962

which meant that local prices of coconut products became cheaper relative to their foreign

counterparts and therefore made them more competitive in the world market. To a large extent,

intensified world demand (which could also have been the effect of lower Philippine coconut

products' prices) served as an impetus for hectarage growth. In the local scene, the continued

upswing in local coconut production was also the result of the Philippine government encouraging

the processing of copra domestically. The government also provided investment incentives to

increase the construction of coconut oil mills. Note that as perennial crop, coconut is hardly affected

or destroyed by natural calamities such as typhoons and floods, relative to rice and corn. This does

not mean though that it is totally unaffected by natural calamities. For the whole 1970s up to the

early 80s, the El Niños and La Niñas seem to have no effect in the local production of coconut.

However, in the early 1980s, the coconut cycle started to tumble as a strong El Niño hit the country

in 1982-83. To a large extent, it could be argued that the downswing in the coconut cycle was caused

by the declining coconut yields due to the aging of coconut trees in a number of regions. The

downtrend in the coconut cycle continued as the number of aging coconuts spiraled. In 1983, 25 to

30 percent of coconut trees were estimated to be at least sixty years old; but by 1988, the proportion

had inched up to between 35 and 40 percent.

45

Figure 5.9. Coconut Cycle

0.8

0.85

0.9

0.95

1

1.05

1970

1972

1974

1976

1978

1980

1982

1984

1986

1988

1990

1992

1994

1996

1998

2000

2002

Sugarcane Cycle Sugarcane grows luxuriantly and well in the middle islands of the Visayas and Mindanao. It suffers

little from typhoon damage and is therefore a crop preference for a number of provinces which are

vulnerable to drastic weather disturbances. In fact, the southernmost island of Mindanao is typhoon

free and henceforth, sugar competes well with other tropical crops in investment opportunities.

High world market prices in the 1970s caused a boom in the sugar industry. This is despite the US

quota on sugar imports from the Philippines in the 50s. Production levels continued to soar, the

sugar industry prospered and old mills expanded capacity. In the midst of rising world market prices

in the 70s, the Marcos dictatorship felt the potential of the sugar industry as a financial source and

therefore, established a monopoly to handle all sales of sugar and promote further development in

the industry. With the expiration of the Laurel-Langley Agreement which provided limitless

preferential access to Philippine sugar in the US market, the Philippines stood to lose its export

market. The monopoly itself suffered as world market prices dropped below production costs. The

pervasive policy of the Marcos regime of price controls and government monopolies caused market

distortions which to a large extent, caused the collapse of the sugar industry. To exacerbate the

situation, twin ENSO events hit the country: an El Niño in 1987-88 and a strong La Niña in 1988-

46

89. As a result, production levels contracted by over 100% from a high of 2.7 million tons to a low of

1.3 million toms in 1987 alone. In fact, this is the only trough of the sugarcane cycle as shown in

Figure 5.10. The sugarcane industry was called a sunset industry during this time. However, the

industry was able to overcome the crisis of the 1980s as the sugarcane cycle started to uptick in the

early 1990s. Market forces were restored as the sugar monopoly was dismantled during the

presidency of Corazon Aquino. The Sugar Regulatory Administration was established to rebuild the

industry but its seeming contradictory policies have resulted in pricing problems which affected the

producers gravely. To make matters worse, the Philippine government opened the market to trade

liberalization, and by implication, to efficient exporters and to the volatility of the world market. The

local sugar industry attempted to reform itself and cope with the changing domestic and international

environment. However, during this time, a series of ENSO events struck the country one after the

other, dampening the gains of the reform to a certain extent. As a consequence, the sugarcane cycle

practically stagnated from the early 1990s until the time of this writing.

Figure 5.10. Sugarcane Cycle

0.6

0.7

0.8

0.9

1

1.1

1970

1972

1974

1976

1978

1980

1982

1984

1986

1988

1990

1992

1994

1996

1998

2000

2002

47

Deviation from Time Trend

Given the second key characteristic defining a cycle (fluctuations must be persistent in the sense that

there are deviations from the average or trend level of the activity) put forth by Diebold and

Rudebusch [1999], this paper conducts a simple comparative analysis of the trend and trend-cycle

components of the time series for each crop. This analysis follows Schenk-Hoppe [2001] that in the

business cycle literature, researchers generally assume that the trend of most time series is smooth

and that all fluctuations are driven by transient productivity shocks.

Figure 5.11. Rice Trend-Cycle vs. Trend

(1970-2003)

400000050000006000000700000080000009000000

1000000011000000120000001300000014000000

1970

1972

1974

1976

1978

1980

1982

1984

1986

1988

1990

1992

1994

1996

1998

2000

2002

trend-cycle trend

Figure 5.12. Corn Trend-Cycle vs. Trend

(1970-2003)

1500000

2000000

2500000

3000000

3500000

4000000

4500000

5000000

1970

1972

1974

1976

1978

1980

1982

1984

1986

1988

1990

1992

1994

1996

1998

2000

2002

trend-cycle trend

Rice production shows a generally smooth uptrend over the period under study. This result,

generated through the Hodrick-Prescott Filter, supports the assumption by most business cycle

practitioners that most time series display a smooth trend over time. From Figure 5.11, it is apparent

48

that the trend-cycle components of the time series do manifest deviations from the trend level. In

fact, the biggest deviations from the trend occurred in 1972-76, 1992-95 and 1999-01. Interestingly, it

was the period of 1972-76 when three consecutive ENSO events occurred: a strong El Niño in 1972-

73 and three consecutive La Niñas from 1973 to 1976. The country was likewise hit by a strong El

Niño in 1992-93 and in 1994-95. The 1999-01 period was marked by a twin La Niñas: one in 1998-

99 and the other in 2000-01. Meanwhile, corn production is characterized by an upward trend from

the 1970s until the early 1990s. However, by mid 90s, the trend started to fall, albeit at a marginal

rate. Figure 5.12 shows marked deviations from the trend line during the years 1984-86 and 1996-00.

Similar to the case of rice, these were also the periods when the country suffered from ENSO events.

Figure 5.13. Coconut Trend-Cycle vs. Trend (1970-2003)

0

2000000

4000000

6000000

8000000

10000000

12000000

14000000

16000000

1970

1972

1974

1976

1978

1980

1982

1984

1986

1988

1990

1992

1994

1996

1998

2000

2002

trend-cycle trend

Both coconut and sugarcane production seem to show the same trend and deviation attributes.

Marked deviations from the trend line characterize the production of the two crops. In general, there

was an upward trend in the production, though there was a slight downtrend by the late 80s. The

years 1971-74, 1976-80, 1984-86 and 1990-94 seem to register the most significant fluctuations. Akin

to the findings in the cases of rice and corn, these years were marked by strong El Niño and La Niña

events.

49

Figure 5.14. Sugarcane Trend-Cycle vs. Trend (1970-2003)

0

5000000

10000000

15000000

20000000

25000000

30000000

1970

1972

1974

1976

1978

1980

1982

1984

1986

1988

1990

1992

1994

1996

1998

2000

2002

trend-cycle trend

Summary

In general, the graphical analyses of the production cycles show that both the turning points of the

cycles as well as the years when crop production experienced marked deviations from the trend line

coincide with the occurrence of ENSO events – either El Niño or La Niña. The peaks of the four

crop cycles are as follows: 1980-82 and 2001-02 (rice); 1974-76 and 1989-91 (corn); 1979-80

(coconut) and 1978-81 (sugarcane). Meanwhile, the trough years are: 1973-74 and 1996-98 (rice);

1982-84 and 1997-99 corn; 1990-92 (coconut) and 1989-91) sugarcane. In general, the turning points

of the cycles are not synchronous. This is despite the fact that these turning points coincide with

ENSO event years. This finding suggests that other factors may have likewise played crucial roles in

the process. It also implies that the extent of the effect of ENSO to the four crops in the study is

subjected to some crop-specific factors, which possibly include government policies. Meanwhile, to

distinguish the production cycles from ordinary fluctuations, the deviation from trend analysis

following Diebold and Rudebusch [1999] and Schenk-Hoppe [2001] is implemented. Interestingly,

the years when there are marked deviations from the trend line are, in general, the years of the peaks

and troughs of the production cycles as well. The finding that there are indeed significant deviations

from the average or trend level of the activity corroborates the earlier analysis of the production

cycles. The analyses conducted in this study suggest that the crop production cycles are not mere

50

fluctuations in the graphical presentation. In fact, the historical events are in general, consistent with

the cycles presented.

Determinants of the Agricultural Crop Production Cycle Essentially, running the regression models aims to determine which factors significantly affect the

production variability to the extent that they can be called determinants of the crop production cycle.

In other words, the study would like to capture the effects of the “irregular” component of the cycle.

To this end, therefore, removing the trend of the dependent variable will help make the effect of

these factors clearer.

The Estimated Multiple Regression Models

Four multiple regression models, corresponding to the four crops under review, are estimated. This

analysis assumes that there is no significant inter-crop relationship though certain factors may

simultaneously affect all the four crops. For each crop, two sets of regression models are constructed:

one containing the SST variable and the other, the SOI variable, both of which represent the ENSO

phenomenon. This separate estimation aims to determine which ENSO measure captures more the

impact of the phenomenon on crop production time series in the Philippines. Each model is also

composed of crop-specific variables – domestic price, world price and hectarage – and of the gross

domestic capital formation in agriculture, agricultural loans, implicit tariff and time variables.

Rice

For rice production, the model with the SST variable performed well with R-squared equal to 90.27%

and adjusted R-squared equal to 85.4%. The SST measure of ENSO significantly affects rice

production in the country. As expected the sign of the coefficient is negative. This finding implies

that on the average both El Niño and La Niña depress local production because of the weather-

related disturbances they impose. This finding is consistent with that of Hansen, et al. [1998], Legler

51

[1999], Adams, et al. [1999], among others who claim that climate-related events are the primary

factors affecting agricultural productivity. Note that in the graphical analysis in the first part of this

paper, it is found that the turning points of the crop cycles coincide with the ENSO years. The

regression results confirm how the ENSO events actually affect local rice production.

In accordance with the cobweb model, the domestic price of rice at first lag turned significant at the

5% level. However, domestic prices at second lag were insignificant even at the 10% level. This

confirms the cobweb theory that farmers’ behaviors are in fact, affected by the previous period’s

prices. However, the sign of the domestic prices variable was not expected. This could mean that

while the levels of farmgate prices the previous period affect the profit prospects of farmers,

significant fluctuations, i.e., a sudden spike in prices, do not necessarily have the effects predicted by

the cobweb theory. This scenario could be likened to the expectations theory of interest rate

determination in monetary economics. If short-term interest rates are high, people usually expect

them to come back down. In effect, they have a particular belief that interest rates follow a certain

trend, and farmers adjust their responses accordingly. Applying this logic to the farmgate price

effects, one could argue that farmers also believe that significant fluctuation in farmgate prices will

not last for long as eventually, prices will have to go back to their long-term trend. In effect,

production adjustments do not correspond to the same extent as the movement of prices. Another

of way of looking at it is that the negative sign in the coefficient of the lagged domestic prices is a

reflection of the distortions in the farmgate prices determination. In the Philippines, most of the

farmers are beholden to their creditors, landlords or to a group of local harvest buyers who distort

the price determination process of market forces, in one way or another.

World rice prices at any lag were insignificant even at the 10% level. This implies that over the period

under review, it is the domestic price, largely determined by the conditions of the domestic market,

which really affects the behavior of farmers and therefore the total production. The finding suggests

52

that the dynamics of world rice market may not significantly affect local rice production, save the

effect of rice importation to meet rising local demands for certain years. Historically, the domestic

rice market is largely protected because of trade barriers including tariffs. However, the insignificance

of the implicit tariff variable implies that the effective protection rates do not significantly affect rice

production as a whole. That is, the market distortion imposed by the implicit tariffs to imported

inputs, including fertilizer, pesticide, water pumps, tractors and threshers, are not significant enough

to result in variability in rice production.

The amount of area used in palay production is also a significant determinant of the production cycle

at the 5% level. The sign of the coefficient is positive as expected. This result may not be so

surprising as intuitively, one can say that if the hectarage used in production increases, then, ceteris

paribus, the production level should also step up.

Both time variables are significant at the 5% level. At first glance, this result is not expected, as the

detrended values of rice production represent the dependent variable. Note that detrending is one

way of taking out the effect of time. The statistical significance of the time variables, therefore,

suggests that to a certain extent, they actually capture the effects of missing variables of the model. In

fact, the time variables are crucial in determining the actual fit of the model. Removing the time

variables and running the same kind of regression, the authors find that the new model registered a

lower R-squared value of 56.4%. In order to test specification error for the two models, the Ramsey-

rest Test is utilized. Testing the hypothesis that there is no specification error, the authors find that

both models are correctly specified, at least from a statistical viewpoint. The model that includes the

time variables has p-value of 0.001252 while the other one has 0.00018 – both of which are less than

1%, 5% and 10%.

53

Meanwhile, a second set of regression models is estimated, now replacing the SST measure with the

SOI measure of the ENSO phenomenon. Like the firsts set of regression runs, the first model

includes the time variables. Lagged domestic rice prices, hectarage, and the time variables are all

significant at the 5% level. However, the ENSO measure of SOI is not significant for both first and

second lags, even at the 10% level. Further, the SOI model has a lower R-squared value of 87.3%.

This suggests that the SST anomaly is the better measure of the ENSO when studying the agriculture

variables in the country. It should be noted that Brunner [2002] finds that the SOI anomaly measure

of ENSO intensity appears to have a much stronger statistical relationship with economic variables

than the SST anomaly measure does. The results of the regression runs in this paper prove otherwise,

at least in the case of rice production variables.

Table 5.1. Rice Regression Result23

Dependent Variable: RICECYCLE Method: Least Squares Sample(adjusted): 1972 2002 Included observations: 31 after adjusting endpoints

Variable Coefficient Std. Error t-Statistic Prob. C 0.055982 0.111753 0.500946 0.6219

SST(-1) -0.024746* 0.007828 -3.161226 0.0049 SST(-2) -0.004889 0.006377 -0.766667 0.4522

DPRICE(-1) -0.096246* 0.031695 -3.036625 0.0065 WPRICE(-1) -0.007002 0.029751 -0.235354 0.8163

GDKF -0.021628 0.016773 -1.289437 0.2120 AGRILOANS 0.007220 0.017697 0.407953 0.6876 IMPTARIFF -0.034225 0.030858 -1.109125 0.2805 AREARICE 2.79E-07* 3.28E-08 8.513759 0.0000

TIME 0.022928* 0.002931 7.823728 0.0000 TIMESQ -0.000769* 9.32E-05 -8.248365 0.0000

R-squared 0.902702 Mean dependent var 0.996687 Adjusted R-squared 0.854053 S.D. dependent var 0.062801 S.E. of regression 0.023992 Akaike info criterion -4.350774 Sum squared resid 0.011512 Schwarz criterion -3.841940 Log likelihood 78.43700 F-statistic 18.55538 Durbin-Watson stat 2.073104 Prob(F-statistic) 0.000000

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

23 The regression results presented here are those which have the highest R-aquared values. Other regression results are presented in the appendices.

54

In the two sets of regression models, agricultural loans and the gross domestic capital formation in

agriculture variables turned insignificant even at the 10% level. These findings suggest that the efforts

of the government to provide assistance to the farmers through an increase in the available loans for

agricultural purposes as well as through the augmentation of capital in the form of tractors and

threshers do not translate to a significant change in the over-all level of rice production, at least over

the period under review.

Corn

The corn model did not perform as good as the rice production regression model, in terms of the

over-all fit, as it registered a lower R-squared value. The first set which utilizes the SST anomaly as

ENSO measure has an R-squared value of 84.5% percent for the equation which includes the time

variables and 67.4% for the one which does not. The SST anomaly measure of ENSO is only

significant at the second lag at the 5% level. This implies that either the El Niño or La Niña takes its

toll, on the average, two periods after its occurrence. In contrast to rice production, the coefficient of

the SST variable is positive. This implies that, on the average, the occurrence of ENSO events

actually has a positive impact on the production of corn. This finding supports the earlier analysis of

the corn cycle. The upsurge of corn production starting in 1997 may be attributed, to a certain extent,

to the rains brought about by the 1998-99 and 2000-01 La Niñas.

The domestic price of corn at fist lag is significant at the 5% level. The coefficient also bears the

expected sign, confirming the cobweb theory: corn prices at period t-1 affect the behavior of farmers

at period t. Meanwhile, the world price variable is not significant at any lag at the usual levels,

suggesting the relative closeness of the local market, i.e., that the dynamics of local corn production

are hardly affected by movements in the foreign market.

55

The gross domestic capital formation (GDKF) in agriculture variable turned out to be a significant

determinant of the corn cycle. The positive sign attached on the coefficient suggests that the GDKF,

e.g., agricultural machineries and tractors other than steam, is positively related to corn production.

The same is true for the area harvested for corn production. The results for both GDKF and

hectarage actually confirm intuitive reasoning and do not need much explanation. What is interesting

though, is that agricultural loans at any lag do not significantly affect corn production. This finding

implies that the government’s attempt to help corn farmers in propping up production, through

financial means, is actually in vain, to a large extent. Government market interventions in the form of

implicit tariff do not significantly affect corn production via the distortion of prices of agricultural

inputs. This finding, however, could be explained by the fact that only a few of the enumerated

inputs in Balisacan and Hill [2003] are relevant in the production of corn. The estimated effective

protection rates thus, turned out to be insignificant in the corn regression equations. However, it

should be noted that the nominal protection rate for corn has risen steadily over time, from about

25% in the late 1970s and early 1980s to nearly 90% by the late 1990s [David in Balisacan and Hill

2003]. This suggests that the imposition of protection rates does not affect the corn subsector

through the input mechanism.

Meanwhile, as in the case of rice, the time variables are also significant factors, implying that they

somehow capture the effects of other variables. R-squared value dropped from over 84% to 67%

when the time variables are excluded from the model. Interestingly, whether or not the model

includes the time variables, the results of the Ramsey-reset test suggest that there is no model

specification error.

Replacing the SST measure with SOI, the authors find that the model performed just as well as the

first set. R-squared value of the one which includes time and the SOI measure is almost 83%,

compared that to the 84% of the one with SST measure. Further, SOI turned significant only at the

56

second lag, confirming the ENSO effect the regression results showed at the preceding paragraphs.

Similar to the first set of regression runs for corn, other significant variables include lagged domestic

prices, GDKF, hectarage and the time variables.

Table 5.2. Corn Regression Result

Dependent Variable: CORNCYCLE Method: Least Squares Sample(adjusted): 1972 2002 Included observations: 31 after adjusting endpoints Variable Coefficient Std. Error t-Statistic Prob.

C 0.373608 0.094860 3.938528 0.0008 SST(-1) -0.009015 0.009096 -0.991120 0.3335 SST(-2) 0.018316* 0.008036 2.279346 0.0338

DPCORN(-1) 0.161575* 0.052249 3.092398 0.0057 WPCORN(-1) -0.050203 0.040431 -1.241695 0.2287

GDKF 0.034885* 0.014408 2.421215 0.0251 AGRILOANS -0.031551 0.019066 -1.654799 0.1136 IMPTARIFF 0.021625 0.028565 0.757042 0.4579 AREACORN 2.00E-07* 3.31E-08 6.038514 0.0000

TIME -0.026104* 0.006300 -4.143270 0.0005 TIMESQ 0.000849* 0.000196 4.330842 0.0003

R-squared 0.844583 Mean dependent var 0.998615 Adjusted R-squared 0.766875 S.D. dependent var 0.056677 S.E. of regression 0.027365 Akaike info criterion -4.087665 Sum squared resid 0.014977 Schwarz criterion -3.578831 Log likelihood 74.35881 F-statistic 10.86862 Durbin-Watson stat 1.696650 Prob(F-statistic) 0.000004

Coconut

The exclusion of the time variables in the regression equations which included the SST measure

instead of the SOI measure did not matter much as R-squared value only differed by just almost 3

percentage points: 91% for the time-including equation and almost 88% for the time-excluding. To a

certain extent, this implies that the model does not suffer from the exclusion of other variables, i.e.,

from model specification. In fact, using Ramsey-reset to test the hypothesis that there is no

specification error, one could verify that model is in fact correctly specified, statistically speaking.

Domestic prices at first and second lags turned significant in the model. This proves that the cobweb

model is also true in coconut production. The significance of the second lag means that price effects

last until after the second period. This is probably because of the nature of coconut production – it

57

takes time for farmers to be able to harvest their produce. This result may also be partly attributed to

the fact that for most of the years under review, the main contention in the industry was the aging of

coconut trees. To reiterate, as of 1988, 35 to 40 percent of the coconut trees are already aging.

However, similar to the case of rice, there is a major twist in the story: the coefficients bear negative

signs, in contrast to what is expected. While puzzling at first, answers could be found in the historical

events in the coconut industry. Government interventions in the industry for the period under study

have caused market distortions, which to a certain extent reversed the effect of price changes to

farmers' incentives.

Table 5.3. Coconut Regression Result

Dependent Variable: COCONUTCYCLE Method: Least Squares Sample(adjusted): 1972 2002 Included observations: 31 after adjusting endpoints

Variable Coefficient Std. Error t-Statistic Prob. C 0.674665 0.039563 17.05284 0.0000

SST -0.000135 0.000510 -0.265390 0.7936 SST(-1) 0.002944 0.004469 0.658684 0.5180 SST(-2) -0.000267 0.004915 -0.054242 0.9573

DPCOCO(-1) -0.011115** 0.005830 -1.906583 0.0718 DPCOCO(-2) -0.009049** 0.004604 -1.965305 0.0642

WPCOPRA(-1) 9.37E-05* 4.33E-05 2.165292 0.0433 WPCOPRA(-2) 0.000124* 3.94E-05 3.151580 0.0053

GDKF -0.014147 0.011975 -1.181340 0.2520 AGRILOANS -0.016364 0.009829 -1.664964 0.1123 IMPTARIFF 0.048739* 0.017455 2.792282 0.0116 AREACOCO 8.57E-08* 1.58E-08 5.420840 0.0000

R-squared 0.877895 Mean dependent var 0.992164 Adjusted R-squared 0.807202 S.D. dependent var 0.035249 S.E. of regression 0.015477 Akaike info criterion -5.214254 Sum squared resid 0.004551 Schwarz criterion -4.659162 Log likelihood 92.82093 F-statistic 12.41849 Durbin-Watson stat 1.586012 Prob(F-statistic) 0.000002

The martial law regime merged all coconut-related government operation within the Philippine

Coconut Authority (PCA), which was mandated to collect a levy of P0.55 per 100 kilograms on the

sale of copra. This was aimed at stabilizing the domestic price of coconut-based consumer goods like

cooking oil. The levy was increased to P20 in 1974 when the Coconut Industry Development Fund

was created. The PCA acquire the United Coconut Planters Bank in 1975 and Eduardo Cojuangco, a

58

business associate of then President Marcos, chaired the bank. Levies collected by the PCA were

placed in the bank. Starting 1978, the UCPB purchased coconut mills as a measure to cope with the

excess capacity in the industry. To compensate for the price controls on coconut-based consumer

products, mills not owned by Cocofed members were denied subsidy. By early 1980, an industry

composed of some 0.5 million farmers and 14,000 traders was highly monopolized as power remains

in the hands of PCA President Cojuangco and Defense Minister Juan Ponce Enrile. In sum, the price

control policy of the government as well as the levy collection imposed on coconut farmers, have

reversed the incentive mechanism of crop prices as posited by the cobweb model. The negative

impact of the government monopoly on coconut price determination may be presumed to be

continuing until the time of this writing, as the alleged misuse of the coconut levy funds remains a

hot issue.

World coconut prices are also significant at first and second lags. This result only demonstrates the

relative openness of the coconut industry to the international market. It also reflects the export

market dependency of the coconut industry at large. Note that in the graphical analysis presented at

the early part of this paper, it is noted that much of the expansion in coconut production in the

1970s is attributed to the devaluation of the peso, rendering local products more attractive to the

world market. Meanwhile, for the first time in the analysis, implicit tariff became a significant

determinant of the production cycle. This finding confirms that of David in Balisacan and Hill

[2003]. Tariff protection on exportable crops including coconut products has greatly affected the

production of the agricultural crop. For instance, to protect consumers and agroprocessors from

high domestic prices caused by the devaluation of the peso and subsequent boom in world

commodity prices in the 1970s, David [2003] notes that coconut products were penalized by the

negative protection rates ranging from –4% to –28%. She adds that the rates are rising despite the

fact the fluctuations in the protection rates have to some extent reflected government attempts to

stabilize domestic prices.

59

Interestingly, in all the equations estimated for the coconut cycle, neither the SST measure nor the

SOI measure turned significant in the regression results. If one takes a look at the nature of coconut

as a perennial crop, this result may not be totally puzzling. Coconuts are hardly affected or destroyed

by natural calamities such as typhoon or flood. However, the idea that the El Niño Southern

Oscillation in the form of El Niño and La Niña affects the level of coconut production can not be

dismissed. For instance, agronomists point to the improved rainfall brought by the 2000-01 La Niña

for the uptrend in production by as much as 45%. It can, therefore, be concluded that while in the

case of coconut crop, ENSO may not be a statistically significant determinant of the production

cycle, it remains a factor affecting production variability in general.

As in the first two crops, the hectarage variable turned to be significant. Historically, some of the

uptrend in the levels of coconut production was ascribed to the expansionary effect of exchange rate

declines to the area used by the coconut industry. Similar to the case of rice, both the gross domestic

capital formation and agricultural loans variables turned insignificant even at the 10% level. This

finding suggests that capital in the form of tractors and other machineries does not significantly

affect coconut production in the country. Intuitively, this is not totally surprising since coconut

production does not require much farm machineries as in the case of other crops. Despite the

increasing trend in agricultural loans, the insignificance of the loans variable implies that on the

average, the pumped up loans in the agriculture sector do not translate to a corresponding changes in

the level of coconut production.

Sugarcane

Among the estimated equations for sugarcane production, those time-and-SOI-including equations

performed better than those SOI-including and time-excluding. R-squared values for the former

range from 85% to almost 89%. The SOI measure of ENSO turned significant at lag 0. This implies

that ENSO affects sugarcane production contemporaneously, which is really not surprising given the

60

time period it takes for sugarcane production. Interestingly, the SOI measure was also significant at

the second lag for all the estimated equations. Statistically, this means that the ENSO actually affects

production even two periods after its occurrence. While there is no theoretical justification for this

effect, the result is not puzzling if one looks at what exactly is SOI. SOI is a measure of the large-

scale fluctuations in air pressure occurring between the western and eastern tropical Pacific. The

index has been traditionally calculated based on the difference in air pressure anomaly between Tahiti

and Darwin, Australia. It may, therefore, be the case that seemingly, the actual effect of ENSO takes

its toll on sugarcane production only after two periods because of the kind of measure used in this

analysis. In any case, however, the regression results prove that ENSO is in fact a significant

determinant of the sugarcane cycle.

Table 5.4.Sugarcane Regression Result

Dependent Variable: SUGARCANECYCLE Method: Least Squares Sample(adjusted): 1972 2002 Included observations: 31 after adjusting endpoints

Variable Coefficient Std. Error t-Statistic Prob. C 0.687196 0.084026 8.178418 0.0000

SOI 0.002774* 0.000875 3.169136 0.0056 SOI(-1) 0.001095 0.000757 1.446759 0.1662 SOI(-2) 0.002485* 0.000923 2.692111 0.0154

DPSUGAR(-1) -0.101892* 0.048773 -2.089118 0.0520 DPSUGAR(-2) -0.046144 0.040798 -1.131045 0.2737 WPSUGAR(-1) 0.016543 0.022889 0.722735 0.4797 WPSUGAR(-2) 0.037349* 0.017256 2.164390 0.0450

GDKF 0.019491 0.018621 1.046703 0.3099 AGRILOANS -0.066804* 0.026171 -2.552536 0.0206 IMPTARIFF 0.064567* 0.031081 2.077402 0.0532

AREASUGAR 4.39E-07* 1.18E-07 3.712967 0.0017 TIME 0.027990* 0.004960 5.642795 0.0000

TIMESQ -0.000686* 0.000125 -5.487544 0.0000 R-squared 0.887842 Mean dependent var 0.988415 Adjusted R-squared 0.802073 S.D. dependent var 0.052503 S.E. of regression 0.023358 Akaike info criterion -4.373304 Sum squared resid 0.009275 Schwarz criterion -3.725697 Log likelihood 81.78622 F-statistic 10.35164 Durbin-Watson stat 2.052910 Prob(F-statistic) 0.000012

The domestic price variable is significant only at first lag at the 5% level. However, similar to the

cases of rice and coconut, the sign of the coefficient is negative and not what is expected. This

61

suggests that, there may be factors which altered the incentive mechanism of competitive price

determination. This is, in fact, the case given the historical developments in the government's

intervention in the sugar industry. As discussed in the graphical analysis presented at the first part of

this chapter, the Marcos regime's policy on price interventions caused market distortions which since

not sustained, caused the collapse of the whole industry in 1987. This is compounded by the

government monopoly, and the establishment of SRA whose contradicting policies proved to be a

major distortion in the sugarcane market. The eventual restructuring in the sugarcane industry may

have fixed the price distortion caused by the Marcos regime’s intervention but not to the extent that

would completely restore the incentive mechanism of the competitive price determination. This

could partly explain why the coefficient of the domestic price variable is not what is expected.

Meanwhile, the world price variable is significant at the 5% level only at the second lag. This only

reflects the period of adjustment of farmers to the changes in the world prices of sugarcane products.

This lag in the adjustment period may have resulted from the rising nominal protection rates for

sugar products. Farmers can not immediately react to the fluctuations in world prices because in the

first place, the industry is highly regulated.

The statistical significance of the implicit tariff reflects the trade distortions imposed in the industry

by government policies, including quota restrictions imposed by the United States and by the SRA

after the EDSA 1. This finding corroborates the claim of David in Balisacan and Hill [2003] that the

sugar industry has historically been the most highly protected crop, even in the 1970s. She argues that

the protection was due to the economic rents conferred by the premium prices obtained under the

US sugar quota, and the concomitant implicit tax on consumers due to import restrictions.

Meanwhile, only in the case of sugarcane did the agricultural loans variable turn to be significant.

This significance is partly explained by the fact that the industry is highly regulated for most of the

62

years included in the analysis. Similar to the three other crops, the amount of area used in the

sugarcane production as well as the time variables are all significant at the 5% level. Again, the

significance of the time variables suggests that there may be other variables which are excluded in the

model, whose effects are being captured by the time variables. Interestingly, Ramsey reset test shows

that the model does not suffer from specification error.

Summary

Table 5.5. Significant Determinants of the Crop Cycles*

Rice Corn Coconut Sugarcane

ENSO a a a Domestic Price a a a a World Price a a GDKF in Agriculture a Agricultural loans a Implicit tariff a a Area harvested a a a a

*The checked variables mean that they are significant factors affecting the crop production cycles.

Regression results in this study show that the ENSO is a major factor affecting the crop cycles,

except for the case of coconut. The insignificance of the ENSO variable in coconut production

could be explained by the perennial nature of the crop. In any case, however, the importance of the

ENSO events in the crop production is supported by the graphical analyses presented at the second

and third parts of this chapter. Not all of the explanatory variables turned significant if each crop

case is examined. However, ENSO, domestic and world prices, GDKF in agriculture, agricultural

loans and implicit tariff are significant in at least one of the crop cases. This means that there are

crop-specific factors which affect the average level of production as well as the crop production

cycles. The graphical presentation at the early part of this section combined with this finding imply

that even the extent of the effect of each of the explanatory variables could be affected by the very

nature of the crop as well as other intervening variables not included in the study.

63

6 Conclusion

This paper attempts to prove the existence of a cycle in agricultural crop production in the

Philippines. It uses time series data from 1970-2003 of the four major crops in the country – rice,

corn, coconut and sugarcane. Taken altogether, these crops account for almost 90% of total

production value of agricultural crops. Essentially, time series analysis and a couple of statistical

techniques are utilized to determine whether or not crop production in the Philippines is

characterized by regular fluctuations, to the extent that one can fairly conclude that a cycle in fact,

exists.

The paper takes the analysis a step further by looking at how certain variables affect this cycle. In

other words, the paper tests whether these variables are significant determinants of the cycle. To this

end, four multiple regression models are constructed. Explanatory variables of the models include

domestic and world prices and hectarage for each crop, and agricultural loans, implicit tariff,

domestic capital formation and time variables.

Two major questions are posted in this paper: (1) Is there an agricultural crop production cycle in the

Philippines? and (2) Are the extreme ENSO events a major determinant of the cycle? Statistical

analysis shows that the production of the four crops under review do manifest a cyclical behavior.

The cycles presented in this study are the results of the method of extraction of time series

components, which is being used in economic cycle analysis by practitioners. In the analysis of the

production’s cyclical attributes, it is found that historical events are in fact reflected in the crop

cycles. The second defining characteristic of an economic cycle put forth by business cycle

practitioners is confirmed in the deviation from trend analysis. Significant deviations from the trend

line are present in all the four crops under study. The years of marked deviations coincide with the

peaks and troughs of the production cycles. Interestingly, both the turning points of the cycles as

well as the years of marked deviations coincide with the ENSO years. This is one piece of evidence

64

that clearly answers the second question of the study: extreme ENSO events are in fact major factors

determining the production cycle in the Philippines. To verify this finding, the authors conduct a

simple multiple regression analysis.

The multiple regression estimates confirm the initial findings in the cyclical analysis conducted in the

early part of the study. The El Niño Southern Oscillation – which hits the country in the form of El

Niño or La Niña – is found to be a significant factor affecting the crop cycle. The regression results

indicate that the ENSO affects production the first until the second year, at the most, after it hits the

country. The extent of the effect could be seen in the graphical presentation of the cycles. It is quite

apparent indeed that the turning points of the cycles as well as the years of the deviation from the

trend are marked by the ENSO years. However, two issues still need to be addressed in this part: (1)

the graphical analysis of the turning points of the cycles shows that the effects of the ENSO events

are not simultaneously manifested by the four crops in the period 1970-2003, i.e., the peaks and

troughs are not synchronous; and that (2) the ENSO variable is not significant in the coconut

equation. It could be argued that while ENSO events do in fact affect production variability to the

extent that they could be called determinants of the cycles, there are crop-specific factors which

intervene in the process. Likewise, the extent of the effects are themselves a function of other

factors, e.g., government policies, market conditions and perhaps, even the degree of disaster

preparedness in the part of the farmers. In any case, however, the bottomline is that clearly, ENSO

events play a crucial role in the crop production cycles.

As the cobweb model predicts, domestic and world prices are significant determinants of the cycle.

However, the responses of crop production to price variables are mixed. Specifically, the signs of the

coefficients of the domestic prices of rice, coconut and sugarcane are not expected. This result

suggests that market conditions also play a crucial role in determining how exactly will price variables

affect production. Normally, if there are no price distortions, price determined in the market provide

65

incentives for farmers to increase production, the next period. This works as the cobweb model

predicts. However, government interventions in the market usually impinge on this mechanism,

resulting in the reversal of the expected sign of the coefficients as the results show.

Domestic Capital Formation was insignificant in all the equations estimated. This only suggests that

while capital variables affect the trend of production levels, the effect is not much for capital to be

considered determinant of the production cycle. Meanwhile, agricultural loans and implicit tariff

turned significant only in the coconut and sugarcane models. This suggests that government policies

in the form of loans and trade distortions affect crops, e.g., sugarcane and coconut, whose markets

are relatively open to foreign trade. Hectarage also turned as a significant variable explaining the

behavior of the cycle. Intuitively, this makes sense as area harvested is directly related to the level of

production, ceteris paribus.

Policy Implications and Recommendation for Further Research

Given the finding that crop production in the Philippines experiences regular fluctuation and

persistent trend deviation characteristic of an economic cycle, it is necessary that policymakers look

into the variables which determine the cycle. This becomes crucial when they formulate and

implement policies that will avert the negative impact of certain variables, i.e., sudden downfall and

continuous downtrend in the levels of production.

Perhaps, the most important contribution of this paper is that it showed that weather-related

disturbances, e.g., El Niño and La Niña, significantly affect production levels, resulting in their

sudden downfall, on the average. The finding that for most of the crops, the effect of ENSO lingers

even after two years suggests that the government should implement a long-term disaster-

preparedness strategy for the agricultural sector. Equally important is the need for the government to

invest in crop research and development that will be able to counter the negative effects of the said

66

weather disturbances. Meanwhile, the conversion of agricultural lands to industrial and residential

uses apparently reduces the available land for agricultural production. This immediately impacts on

the level of production, holding all other things constant. In this case, the response of the

government should be founded on economic grounds. If the crop being affected by the continuous

decline in the available land for production is in fact, losing its comparative advantage, it is probably

better not to be so much reactive to the case through a deliberate attempt to maintain or increase the

amount of area devoted to the crop. Indeed, it is recommended that the government weigh its

options and carefully analyze the status of the crop subsector relative to the developments in the

domestic and world economy at large.

As in the case of rice, coconut and sugarcane, government interventions in the form of policy

controls and trade distortions negatively impact on the incentive mechanism of the market. On the

average, these cause a downfall in the absolute levels of production. Therefore, if the government’s

main concern is to prop up crop production in the country, it should explicitly and implicitly

minimize policies which distort the market for crop products.

For further research, the authors recommend that a multivariate analysis be utilized in the time series

analysis. This would test the joint impact of the explanatory variables on the crop cycle. Note that

this paper assumes that there is no significant inter-crop relationship that will distort the impact of

the explanatory variables on the dependent variable. It is also recommended that further research be

focused on the particular government policies which affect production at large, whether in a positive

or negative manner.

67

Selected References

Adams, R., C. Chen, B. McCarl, and R. Weiher. [1999] “The Economic Consequences of ENSO

events,” Climate Research. 13(3): 165-172.

Adams, R., et al. [1995] “Value of Improved Long-Range Weather information,” Contemporary Economic Policy 13:10-19.

Akerman, J. [1928] “Weather driven Seasonal Cycle,” http://cepa.newschool.edu. Balisacan, A. and H. Hill. [2003] The Philippine Economy: Development, Policies and Challenges.

Quezon City: Ateneo de Manila University Press. Bautista, G. [1987] Impact of Agriculture on the Labour Market in the Philippines. India: KO-

ARTEP. Baxter, M. and R. G. King. [1999] Lectures on Macroeconomics. Cambridge: MIT Press. Bersales, L. [2004] “Time Series Analysis and Forecasting.” Lecture delivered during the IDEA

Forecasting Workshop Series. Brunner, A. [2002] “El Niño and World Primary Commodity Prices: Warm Water or Hot Air?,” The

Review of Economics and Statistics, 84(1):176-183. Burns, A. and W. Mitchell. [1946] Measuring Business Cycles. New York: National Bureau of

Economic Research.

Cane, M., G. Eshel, and R.W. Buckland. [1994] “Forecasting Zimbabwean maize yield using eastern equatorial Pacific sea surface temperature,” Nature, 370:204-205.

Canova, F. [1996] “Three tests for the existence of cycles in time series,” Ricerche Economiche,

50:135-162.

Chen, C. and B. McCarl. [2002] “Agricultural Values of ENSO Information Under Alternative Phase Definition,” Climatic Change, 54(3):305-325.

Commonwealth Bureau of Meteorology website. http://www.austehc.unimelb.edu.au. Accessed July

2004. Debelle, G. and G. Stevens. [1995] “Monetary Policy Goals for Inflation in Australia,” Reserve Bank

of Australia Research Discussion Paper No. 9503. Diebold, F. and G. Rudebusch. [1999] Business Cycles: Durations, Dynamics and Forecasting.

Princeton: Princeton University Press. Duncan, B. [1990] World Trade. Pennsylvania: Tab Books. Enders, W. [1995] Applied Econometric Time Series. New York: Wiley.

68

Ezekiel, M. [1944] “The Cobweb Effect,” Readings in Business Cycle Theory. Philadelphia: Blakiston.

Food and Agriculture Administration (FAO) website. http://www.fao.org. Accessed July 2004. Gommes, R. [2000] Production Variability and Losses. Food and Agriculture Organization of the

United Nations. Haberler, G. [1937] Prosperity and Depression. Geneva: League of Nations. Hall, A.D., J. Skalin, and T. Terasvirta. [2001] “A nonlinear time series model of El Niño,”

Environmental Modelling and Software, 16:139-146. Handler, P. [1983] “Climatic Anomalies in the Tropical Pacific Ocean and Corn Yields in the US,”

Science 20:1155-1156. Hansen, J. W., A.W. Hodges, and J.W. Jones. [1998] “ENSO influences on agriculture in the

Southeastern US,” Journal of Climate, 11(3):404-411. Hanson, K. and G.A. Maul. [1989] “Evidence of Significant nonrandom behavior in the recurrence

of Strong El Nino between 1525 and 1988,” Geophys. Res. Let. 16, 1181-1184.

Holden D., et al. [1991] Risk in Agriculture. Washington, D.C.:The World Bank. Jevons, H.S. [1884] Investigations in Currency and Finance. London: Macmillan. Jones, J. [2003] Agricultural Responses to Climate Variability and Climate Change. Background paper for

NOAA Workshop entitled insights and Tools for Adaptation: Learning from Climate Variability.

Kaan, D. [2000] Risk and Resilience in Agriculture. Colorado State University. Kydland, F. and E. Prescott. [1990] “Business cycles; real facts and a monetary myth,” Federal

Reserve Bank of Minneapolis, Quarterly Review, Spring, 3-18. Legler, D.M., K.J. Bryant, and J.J. O’Brien. [1999] “Impact of ENSO-related Climate Anomalies on

Crop Yields in the US,” Climatic Change, 42:351-375. Manasan R. and R. Querubin. [1997] “Assessment of Tariff Reforms in the 1990s,” PIDS Discussion

Paper Series No. 97-10. Mendelsohn, R., W. Nordhaus, and D. Shaw. [1994] “The Impact of Global Warming on

Agriculture: A Ricardian Analysis,” American Economic Review, 84(4):753-771. Meinke, H., W. Baethgen, A. Gimenez. [2001] Adaptation of Agricultural Production Systems to

Climate Variability and Climate Change: Lessons Learned and Proposed Research Approach. MGAP website www.mgap.gub.uy.

Mitchell. W. [1959] Business Cycles and their Causes. Berkeley: University of California. Moore, H.L. [1914] Economic Cycles: Their Law and Causes.

69

Mudlak, Y., D. Larson, and R. Butzer. [1997] “The determinants of agricultural production: a cross- country analysis,” Policy Research Working Papers, World Bank.

National Oceanic and Atmospheric Administration (NOAA) website. http://www.noaa.gov.

Accessed July 2004. National Statistical Coordination Board (NSCB) website. http://www.nscb.gov.ph. Accessed July

2004. National Weather Service website. http://www.nws.noaa.gov. Accessed July 2004. Naylor, R., W. Falcon, N. Wada, and D. Rochberg. [2002] “Using El Niño Southern Oscillation

Climate Data to Improve Food Policy Planning in Indonesia,” Bulletin of Indonesian Economic Studies, 38(1):75-91.

_______________. [2001] “Using El Niño/Southern Oscillation Climate Data to Predict Rice

Production in Indonesia,” Climatic Change, 50:255-265. Niemira, M. and P. Klein. [1994] Forecasting Financial and Economic Cycles. New York. John Wiley &

Sons, Inc. PAGASA. [2001] Documentation and Analysis of Impacts of and Responses to Extreme Climate

Events: Agriculture Sector, Technical Working Group for Agriculture. Philippine Atmospheric and Astronomical Services Administration.

Philippine Institute for Development Studies (PIDS) website. http://www.pids.gov.ph. Accessed

July 2004. Pigou, A.C. [1927] Industrial Fluctuations. London: Macmillan. Pinstrup-Andersen, P. [1982] Agricultural Research and Technology in Economic Development.

London: Longman.

Power, J. and P. Intal, Jr. [1990] Trade, Exchange Rate, and Agricultural Pricing Policies in the Philippines. Washington, D.C.:The World Bank.

Press, W.H., B.P. Flannery, S.A. Teukolsky and W.T. Vetterling. [1989] Numerical Recipes: The Art of

Scientific Computing. Cambridge University Press. Reeves, J., et al. [2000] “The Hodrick-Prescott Filter, a Generalization, and a New Procedure for

Extracting an Empirical Cycle from a series,” Studies in Nonlinear Dynamics and Econometrics, 4(1):1-16.

Risk Management Agency. [1997] Introduction to Risk Management. United States Department of

Agriculture. Rosenberg, et al. [1997] “Sensitivity of North American Agriculture to ENSO-based Climate

Scenarios.”PNL-116699/UC-01000,Pacific Northwest National Laboratory, Washington, D.C.

Roumasset, J. et al. [1979] Risk, Uncertainty and Agricultural Development. Southeast Asian Regional

70

Center for Graduate Study and Research in Agriculture and the Agricultural Development Council.

Schenck-Hoppe, K. [2001] “Economic Growth and Business Cycles: A Critical Comment on

Detrending Time Series,” Studies in Nonlinear Dynamics and Econometrics, 5(1):75-86. Sicat, G. [2003] Economics, Volume 3: Philippine Economic and Development Issues. Pasig City:

Anvil Publishing. Sittel, M. [1994] “Differences in the means of ENSO extremes for maximum temperature and

precipitation in the United States.” COAPS Report 94-2, Florida State University, COAPS, Tallahassee, FL 32306-2840.

Solow, A. [1995] “An Exploratory Analysis of a Record of El Niño Events, 1800-1987,” Journal of

the American Statistical Association, 90(429):72-77. __________. [1998] “The Value of Improved ENSO Prediction to U.S. Agriculture,” Climatic

Change, 39(1):47-60. Steel, R.G.D. and J.H. Torrie. [1980] Principles and Procedures of Statistics: a Biometrical Approach, Second

Edition. McGraw-Hill. Taylor, S. [2003] Climate Trends and Agricultural Management. Iowa State University. Taylor, S., Carlson and Todey.[1996]. Risk Associated with SOI. Tibig, L. [2002] “A Survey of Climate Variability/Change Impacts in the Agriculture Sector.”

Philippine Atmospheric and Astronomical Services Administration. Terasvirta, T. [1994] “Specification, Estimation, and Evaluation of Smooth Transition Autoregressive

Models,” Journal of the American Statistical Association, 89(425):208-218. Williams, J. R., C. A. Jones, and P. T. Dyke. [1984] “A modeling approach to determining the relationship between erosion and soil productivity,” Trans. ASAE, 27, 129-144. Williams, J.R., C.A. Jones, and D.A. Spanel. [1989] “The EPIC crop growth model.” Trans. ASAE,

32:497-511.

Yap, J. [2000] “PIDS Annual Macroeconometric Model 2000.” Accessed at http://www.pids.gov.ph.

71

Appendix A. Consensus List of El Niño and La Niña Years

Winter WRCC CDC CPC MEI Consensus

1950-51 C+ C C C La Niña 1951-52 W+ W- 1952-53 1953-54 W W-

1954-55 C C-

1955-56 C+ C+ C Strong La Niña

1956-57 C C- C- Weak La Niña

1957-58 W W W+ W El Niño 1958-59 W+ W-

1959-60 1960-61 1961-62 C- 1962-63 C- 1963-64 W W- 1964-65 C C C- La Niña 1965-66 W+ W W W El Niño 1966-67 C- 1967-68 C- 1968-69 W W- 1969-70 W W 1970-71 C C C La Niña 1971-72 C C- C- Weak La Niña

1972-73 W+ W W+ W Strong El Niño

1973-74 C+ C C+ C+ Strong La Niña

1974-75 C C- C- Weak La Niña

1975-76 C+ C C+ C Strong La Niña

1976-77 W W- 1977-78 W+ W- W- El Niño 1978-79 1979-80 W- W- 1980-81 1981-82 1982-83 W+ W W+ W+ Strong El Niño

72

1983-84 C- 1984-85 C- C- 1985-86 1986-87 W W 1987-88 W+ W- W W- El Niño 1988-89 C+ C- C+ C Strong La Niña

1989-90 1990-91 W+ 1991-92 W W W+ W+ Strong El Niño

1992-93 W W+ W- El Niño 1993-94 W+ W 1994-95 W+ W W- El Niño 1995-96 C- C- 1996-97 1997-98 W+ W W+ W+ Strong El Niño

1998-99 C+ C C- La Niña 1999-00 C C 2000-01 C C C- C- La Niña 2001-02 2002-03 W W W W El Niño

Source: Golden gate Weather Services http://ggweather.com/enso/years.htm

73

Appendix B. Estimation of Rice Regression Equations Dependent Variable: RICECYCLE Method: Least Squares Sample(adjusted): 1972 2002 Included observations: 31 after adjusting endpoints

Variable Coefficient Std. Error t-Statistic Prob.

C 0.055982 0.111753 0.500946 0.6219DPRICE(-1) -0.096246 0.031695 -3.036625 0.0065WPRICE(-1) -0.007002 0.029751 -0.235354 0.8163

SST(-1) -0.024746 0.007828 -3.161226 0.0049SST(-2) -0.004889 0.006377 -0.766667 0.4522GDKF -0.021628 0.016773 -1.289437 0.2120

AGRILOANS 0.007220 0.017697 0.407953 0.6876IMPTARIFF -0.034225 0.030858 -1.109125 0.2805AREARICE 2.79E-07 3.28E-08 8.513759 0.0000

TIME 0.022928 0.002931 7.823728 0.0000TIMESQ -0.000769 9.32E-05 -8.248365 0.0000

R-squared 0.902702 Mean dependent var 0.996687Adjusted R-squared 0.854053 S.D. dependent var 0.062801S.E. of regression 0.023992 Akaike info criterion -4.350774Sum squared resid 0.011512 Schwarz criterion -3.841940Log likelihood 78.43700 F-statistic 18.55538Durbin-Watson stat 2.073104 Prob(F-statistic) 0.000000

Dependent Variable: RICECYCLE Method: Least Squares Sample(adjusted): 1972 2002 Included observations: 31 after adjusting endpoints

Variable Coefficient Std. Error t-Statistic Prob.

C 0.780399 0.140029 5.573134 0.0000DPRICE(-1) -0.059494 0.062500 -0.951905 0.3515WPRICE(-1) -0.024881 0.059694 -0.416813 0.6809

SST(-1) -0.045630 0.014529 -3.140525 0.0048SST(-2) 0.010841 0.012121 0.894356 0.3808GDKF -0.003659 0.033423 -0.109469 0.9138

AGRILOANS -0.007812 0.035534 -0.219844 0.8280IMPTARIFF 0.070999 0.056651 1.253274 0.2233AREARICE 6.85E-08 3.83E-08 1.788559 0.0875

R-squared 0.563834 Mean dependent var 0.996687Adjusted R-squared 0.405228 S.D. dependent var 0.062801S.E. of regression 0.048433 Akaike info criterion -2.979564Sum squared resid 0.051607 Schwarz criterion -2.563245Log likelihood 55.18323 F-statistic 3.554942Durbin-Watson stat 1.888565 Prob(F-statistic) 0.008635

Dependent Variable: RICECYCLE Method: Least Squares Sample(adjusted): 1972 2002 Included observations: 31 after adjusting endpoints

Variable Coefficient Std. Error t-Statistic Prob.

C -0.052914 0.120786 -0.438077 0.6660DPRICE(-1) -0.083450 0.035899 -2.324547 0.0307WPRICE(-1) -0.010140 0.035643 -0.284480 0.7790

SOI(-1) 0.001589 0.000914 1.738163 0.0976SOI(-2) 0.000455 0.000798 0.569966 0.5750GDKF -0.018827 0.019444 -0.968302 0.3445

AGRILOANS 0.000193 0.020335 0.009502 0.9925IMPTARIFF -0.036258 0.036460 -0.994473 0.3319AREARICE 3.05E-07 3.67E-08 8.326088 0.0000

TIME 0.025894 0.003569 7.255954 0.0000

74

TIMESQ -0.000863 0.000110 -7.821543 0.0000

R-squared 0.872655 Mean dependent var 0.996687Adjusted R-squared 0.808983 S.D. dependent var 0.062801S.E. of regression 0.027448 Akaike info criterion -4.081657Sum squared resid 0.015067 Schwarz criterion -3.572823Log likelihood 74.26569 F-statistic 13.70542Durbin-Watson stat 1.746732 Prob(F-statistic) 0.000001

Dependent Variable: RICECYCLE Method: Least Squares Sample(adjusted): 1972 2002 Included observations: 31 after adjusting endpoints

Variable Coefficient Std. Error t-Statistic Prob.

C 0.674897 0.155873 4.329785 0.0003DPRICE(-1) -0.017391 0.067892 -0.256155 0.8002WPRICE(-1) -0.040155 0.069231 -0.580018 0.5678

SOI(-1) 0.002714 0.001654 1.641143 0.1150SOI(-2) -0.002202 0.001395 -1.578633 0.1287GDKF 0.004155 0.037521 0.110739 0.9128

AGRILOANS -0.013583 0.039855 -0.340795 0.7365IMPTARIFF 0.095718 0.063923 1.497403 0.1485AREARICE 8.37E-08 4.35E-08 1.923016 0.0675

R-squared 0.457132 Mean dependent var 0.996687Adjusted R-squared 0.259725 S.D. dependent var 0.062801S.E. of regression 0.054034 Akaike info criterion -2.760720Sum squared resid 0.064232 Schwarz criterion -2.344401Log likelihood 51.79116 F-statistic 2.315688Durbin-Watson stat 1.610682 Prob(F-statistic) 0.056890

Dependent Variable: RICECYCLE Method: Least Squares Sample(adjusted): 1972 2003 Included observations: 32 after adjusting endpoints

Variable Coefficient Std. Error t-Statistic Prob.

C 1.092175 0.073663 14.82656 0.0000DPRICE(-1) -0.094814 0.052356 -1.810954 0.0827

DPRICEDETREND(-2) 0.031320 0.055026 0.569189 0.5745WPRICE(-1) -0.021775 0.048121 -0.452510 0.6550

WPRICEDETREND(-2) -0.009014 0.046142 -0.195355 0.8468SST -0.013956 0.016018 -0.871295 0.3922

SST(-1) -0.060088 0.014130 -4.252557 0.0003SST(-2) 0.007928 0.016788 0.472246 0.6410

R-squared 0.452337 Mean dependent var 0.997174Adjusted R-squared 0.292601 S.D. dependent var 0.061841S.E. of regression 0.052013 Akaike info criterion -2.862337Sum squared resid 0.064928 Schwarz criterion -2.495903Log likelihood 53.79739 F-statistic 2.831791Durbin-Watson stat 1.954907 Prob(F-statistic) 0.026642

Dependent Variable: RICECYCLE Method: Least Squares Sample(adjusted): 1972 2003 Included observations: 32 after adjusting endpoints

Variable Coefficient Std. Error t-Statistic Prob.

C 1.049830 0.094448 11.11546 0.0000DPRICE(-1) -0.061087 0.061008 -1.001307 0.3267

DPRICEDETREND(-2) 0.057668 0.062487 0.922885 0.3653WPRICE(-1) -0.017613 0.059934 -0.293868 0.7714

WPRICEDETREND(-2) -0.031273 0.057203 -0.546704 0.5896SOI -0.001490 0.001742 -0.854963 0.4010

SOI(-1) 0.004444 0.001732 2.565870 0.0170

75

SOI(-2) -0.002843 0.001778 -1.599101 0.1229

R-squared 0.243400 Mean dependent var 0.997174Adjusted R-squared 0.022725 S.D. dependent var 0.061841S.E. of regression 0.061135 Akaike info criterion -2.539163Sum squared resid 0.089698 Schwarz criterion -2.172729Log likelihood 48.62660 F-statistic 1.102978Durbin-Watson stat 1.738426 Prob(F-statistic) 0.392836

Dependent Variable: RICECYCLE Method: Least Squares Sample(adjusted): 1972 2002 Included observations: 31 after adjusting endpoints

Variable Coefficient Std. Error t-Statistic Prob.

C 0.107335 0.158426 0.677510 0.5091DPRICE(-1) -0.046044 0.042360 -1.086952 0.2954WPRICE(-1) -0.050193 0.048954 -1.025320 0.3226

SST(1) 0.027851 0.011187 2.489548 0.0260SST(-2) 0.002671 0.008252 0.323640 0.7510GDKF -0.034666 0.025396 -1.364992 0.1938

GDKF(-1) 0.016815 0.027359 0.614604 0.5487GDKF(-2) -0.009049 0.020179 -0.448432 0.6607

AGRILOANS 0.009487 0.031093 0.305121 0.7648AGRILOANS(-1) -0.011768 0.044516 -0.264361 0.7954AGRILOANS(-2) -0.008628 0.036864 -0.234042 0.8183

IMPTARIFF 0.031385 0.046207 0.679225 0.5081IMPTARIFF(-1) -0.055443 0.043983 -1.260541 0.2281IMPTARIFF(-2) 0.019835 0.057119 0.347261 0.7336

AREARICE 2.70E-07 4.68E-08 5.767356 0.0000TIME 0.018507 0.005389 3.434118 0.0040

TIMESQ -0.000651 0.000171 -3.811106 0.0019

R-squared 0.915877 Mean dependent var 0.996687Adjusted R-squared 0.819737 S.D. dependent var 0.062801S.E. of regression 0.026664 Akaike info criterion -4.109181Sum squared resid 0.009953 Schwarz criterion -3.322801Log likelihood 80.69230 F-statistic 9.526476Durbin-Watson stat 2.490951 Prob(F-statistic) 0.000060

76

Appendix C. Estimation of Corn Regression Equations Dependent Variable: CORNCYCLE Method: Least Squares Sample(adjusted): 1972 2002 Included observations: 31 after adjusting endpoints

Variable Coefficient Std. Error t-Statistic Prob.

C 0.345919 0.121050 2.857655 0.0097

DPCORN(-1) 0.149214 0.052073 2.865480 0.0096WPCORN(-1) -0.043712 0.049170 -0.889002 0.3846

SOI(-1) 0.000570 0.001348 0.423180 0.6767SOI(-2) -0.001805 0.000897 -2.012279 0.0579GDKF 0.033338 0.014946 2.230637 0.0373

AGRILOANS -0.028847 0.019796 -1.457200 0.1606IMPTARIFF 0.025391 0.030298 0.838063 0.4119AREACORN 2.16E-07 4.66E-08 4.623200 0.0002

TIME -0.029792 0.010089 -2.952960 0.0079TIMESQ 0.000961 0.000309 3.111567 0.0055

R-squared 0.829753 Mean dependent var 0.998615Adjusted R-squared 0.744630 S.D. dependent var 0.056677S.E. of regression 0.028641 Akaike info criterion -3.996525Sum squared resid 0.016406 Schwarz criterion -3.487691Log likelihood 72.94614 F-statistic 9.747639Durbin-Watson stat 1.551812 Prob(F-statistic) 0.000010

Dependent Variable: CORNCYCLE Method: Least Squares Sample(adjusted): 1972 2002 Included observations: 31 after adjusting endpoints

Variable Coefficient Std. Error t-Statistic Prob.

C 0.701225 0.088460 7.927069 0.0000DPCORN(-1) 0.185230 0.064164 2.886847 0.0086WPCORN(-1) -0.129048 0.050967 -2.532015 0.0190

SOI(-1) 0.003751 0.000991 3.786724 0.0010SOI(-2) -0.000989 0.001011 -0.978865 0.3383GDKF 0.026743 0.018810 1.421726 0.1691

AGRILOANS -0.029002 0.025117 -1.154673 0.2606IMPTARIFF 0.045562 0.037895 1.202338 0.2420AREACORN 6.48E-08 1.59E-08 4.071900 0.0005

R-squared 0.698380 Mean dependent var 0.998615Adjusted R-squared 0.588700 S.D. dependent var 0.056677S.E. of regression 0.036348 Akaike info criterion -3.553639Sum squared resid 0.029066 Schwarz criterion -3.137321Log likelihood 64.08141 F-statistic 6.367429Durbin-Watson stat 1.584812 Prob(F-statistic) 0.000260

Dependent Variable: CORNCYCLE Method: Least Squares Sample(adjusted): 1972 2002 Included observations: 31 after adjusting endpoints

Variable Coefficient Std. Error t-Statistic Prob.

C 0.373608 0.094860 3.938528 0.0008DPCORN(-1) 0.161575 0.052249 3.092398 0.0057WPCORN(-1) -0.050203 0.040431 -1.241695 0.2287

SST(-1) -0.009015 0.009096 -0.991120 0.3335SST(-2) 0.018316 0.008036 2.279346 0.0338GDKF 0.034885 0.014408 2.421215 0.0251

AGRILOANS -0.031551 0.019066 -1.654799 0.1136IMPTARIFF 0.021625 0.028565 0.757042 0.4579AREACORN 2.00E-07 3.31E-08 6.038514 0.0000

TIME -0.026104 0.006300 -4.143270 0.0005

77

TIMESQ 0.000849 0.000196 4.330842 0.0003

R-squared 0.844583 Mean dependent var 0.998615Adjusted R-squared 0.766875 S.D. dependent var 0.056677S.E. of regression 0.027365 Akaike info criterion -4.087665Sum squared resid 0.014977 Schwarz criterion -3.578831Log likelihood 74.35881 F-statistic 10.86862Durbin-Watson stat 1.696650 Prob(F-statistic) 0.000004

Dependent Variable: CORNCYCLE Method: Least Squares Sample(adjusted): 1972 2002 Included observations: 31 after adjusting endpoints

Variable Coefficient Std. Error t-Statistic Prob.

C 0.678524 0.095156 7.130654 0.0000DPCORN(-1) 0.209114 0.069577 3.005525 0.0065WPCORN(-1) -0.116917 0.051288 -2.279620 0.0327

SST(-1) -0.031487 0.010279 -3.063296 0.0057SST(-2) 0.012732 0.010814 1.177291 0.2517GDKF 0.027710 0.019769 1.401661 0.1750

AGRILOANS -0.020517 0.025964 -0.790222 0.4378IMPTARIFF 0.033952 0.039138 0.867482 0.3950AREACORN 5.99E-08 1.67E-08 3.598222 0.0016

R-squared 0.674337 Mean dependent var 0.998615Adjusted R-squared 0.555915 S.D. dependent var 0.056677S.E. of regression 0.037769 Akaike info criterion -3.476946Sum squared resid 0.031383 Schwarz criterion -3.060627Log likelihood 62.89266 F-statistic 5.694323Durbin-Watson stat 1.606614 Prob(F-statistic) 0.000551

Dependent Variable: CORNCYCLE Method: Least Squares Sample(adjusted): 1972 2003 Included observations: 32 after adjusting endpoints

Variable Coefficient Std. Error t-Statistic Prob.

C 0.335954 0.107599 3.122288 0.0052DPCORN(-1) 0.207560 0.060888 3.408901 0.0026WPCORN(-1) -0.072293 0.040047 -1.805189 0.0854

SST(-1) -0.011028 0.008785 -1.255276 0.2232SST(-2) 0.021049 0.008989 2.341667 0.0291

GDKF(-1) 0.036249 0.018312 1.979508 0.0610AGRILOANS(-1) -0.032260 0.019107 -1.688380 0.1061IMPTARIFF(-1) 0.005402 0.030058 0.179719 0.8591

AREACORN 2.09E-07 3.13E-08 6.678328 0.0000TIME -0.026219 0.005463 -4.799431 0.0001

TIMESQ 0.000858 0.000167 5.132538 0.0000

R-squared 0.820909 Mean dependent var 0.999808Adjusted R-squared 0.735628 S.D. dependent var 0.056162S.E. of regression 0.028877 Akaike info criterion -3.985265Sum squared resid 0.017511 Schwarz criterion -3.481418Log likelihood 74.76424 F-statistic 9.625896Durbin-Watson stat 1.806966 Prob(F-statistic) 0.000008

Dependent Variable: CORNCYCLE Method: Least Squares Sample(adjusted): 1972 2003 Included observations: 32 after adjusting endpoints

Variable Coefficient Std. Error t-Statistic Prob.

C 0.942694 0.088190 10.68939 0.0000DPCORN(-1) 0.185550 0.082854 2.239475 0.0339WPCORN(-1) -0.129816 0.058011 -2.237780 0.0340

SOI -0.001648 0.001260 -1.308085 0.2023SOI(-1) 0.003767 0.001336 2.818607 0.0091SOI(-2) -0.001483 0.001367 -1.084586 0.2881

78

R-squared 0.385799 Mean dependent var 0.999808Adjusted R-squared 0.267683 S.D. dependent var 0.056162S.E. of regression 0.048061 Akaike info criterion -3.065335Sum squared resid 0.060056 Schwarz criterion -2.790510Log likelihood 55.04536 F-statistic 3.266279Durbin-Watson stat 1.204983 Prob(F-statistic) 0.020235

Dependent Variable: CORNCYCLE Method: Least Squares Sample(adjusted): 1972 2003 Included observations: 32 after adjusting endpoints

Variable Coefficient Std. Error t-Statistic Prob.

C 0.948157 0.089156 10.63483 0.0000DPCORN(-1) 0.154561 0.093361 1.655519 0.1098WPCORN(-1) -0.104512 0.058291 -1.792925 0.0846

SST 0.000666 0.015308 0.043490 0.9656SST(-1) -0.035548 0.012874 -2.761255 0.0104SST(-2) 0.007435 0.017143 0.433689 0.6681

R-squared 0.385255 Mean dependent var 0.999808Adjusted R-squared 0.267035 S.D. dependent var 0.056162S.E. of regression 0.048082 Akaike info criterion -3.064451Sum squared resid 0.060109 Schwarz criterion -2.789626Log likelihood 55.03122 F-statistic 3.258798Durbin-Watson stat 1.285962 Prob(F-statistic) 0.020433

79

Appendix D. Estimation of Coconut Regression Equations Dependent Variable: COCONUTCYCLE Method: Least Squares Sample(adjusted): 1972 2002 Included observations: 31 after adjusting endpoints

Variable Coefficient Std. Error t-Statistic Prob.

C 0.240691 0.332605 0.723653 0.4777DPCOCO(-1) 0.022955 0.095781 0.239663 0.8130

WPCOPRA(-1) -0.030233 0.138170 -0.218810 0.8290SST(-1) -0.028250 0.028879 -0.978204 0.3397SST(-2) -0.014555 0.032076 -0.453760 0.6549GDKF 0.026362 0.066743 0.394983 0.6970

AGRILOANS -0.033501 0.066625 -0.502821 0.6206IMPTARIFF 0.246319 0.118150 2.084798 0.0501AREACOCO 2.08E-07 1.53E-07 1.358214 0.1895

TIME -0.009928 0.020187 -0.491810 0.6282TIMESQ 0.000186 0.000504 0.368487 0.7164

R-squared 0.366956 Mean dependent var 0.992404Adjusted R-squared 0.050435 S.D. dependent var 0.112567S.E. of regression 0.109692 Akaike info criterion -1.310867Sum squared resid 0.240645 Schwarz criterion -0.802033Log likelihood 31.31844 F-statistic 1.159340Durbin-Watson stat 1.269075 Prob(F-statistic) 0.371166

Dependent Variable: COCONUTCYCLE Method: Least Squares Sample(adjusted): 1972 2002 Included observations: 31 after adjusting endpoints

Variable Coefficient Std. Error t-Statistic Prob.

C 0.417845 0.226866 1.841816 0.0790DPCOCO(-1) 0.032118 0.092400 0.347601 0.7314

WPCOPRA(-1) -0.030951 0.131124 -0.236041 0.8156SST(-1) -0.026181 0.028018 -0.934440 0.3602SST(-2) -0.016302 0.030545 -0.533698 0.5989GDKF 0.032718 0.064644 0.506122 0.6178

AGRILOANS -0.035918 0.064799 -0.554301 0.5850IMPTARIFF 0.247075 0.113880 2.169611 0.0411AREACOCO 1.11E-07 6.69E-08 1.659853 0.1111

R-squared 0.338672 Mean dependent var 0.992404Adjusted R-squared 0.098190 S.D. dependent var 0.112567S.E. of regression 0.106898 Akaike info criterion -1.396189Sum squared resid 0.251396 Schwarz criterion -0.979870Log likelihood 30.64093 F-statistic 1.408303Durbin-Watson stat 1.280660 Prob(F-statistic) 0.247684

Dependent Variable: COCONUTCYCLE Method: Least Squares Sample(adjusted): 1972 2002 Included observations: 31 after adjusting endpoints

Variable Coefficient Std. Error t-Statistic Prob.

C 8.914433 7.851490 1.135381 0.2684DPCOCO(-1) 0.032118 0.092400 0.347601 0.7314

WPCOPRA(-1) -0.030951 0.131124 -0.236041 0.8156SST(-1) -0.026181 0.028018 -0.934440 0.3602SST(-2) -0.016302 0.030545 -0.533698 0.5989GDKF 0.032718 0.064644 0.506122 0.6178

AGRILOANS -0.035918 0.064799 -0.554301 0.5850IMPTARIFF 0.247075 0.113880 2.169611 0.0411AREACOCO 1.11E-07 6.69E-08 1.659853 0.1111

R-squared 0.338672 Mean dependent var 0.992404

80

Adjusted R-squared 0.098190 S.D. dependent var 0.112567S.E. of regression 0.106898 Akaike info criterion -1.396189Sum squared resid 0.251396 Schwarz criterion -0.979870Log likelihood 30.64093 F-statistic 1.408303Durbin-Watson stat 1.280660 Prob(F-statistic) 0.247684

Dependent Variable: COCONUTCYCLE Method: Least Squares Sample(adjusted): 1972 2002 Included observations: 31 after adjusting endpoints

Variable Coefficient Std. Error t-Statistic Prob.

C 0.581648 0.049638 11.71774 0.0000DPCOCO(-1) -0.004970 0.005573 -0.891845 0.3842DPCOCO(-2) -0.006603 0.004117 -1.603818 0.1262

WPCOPRA(-1) 3.56E-05 4.06E-05 0.876077 0.3925WPCOPRA(-2) 8.00E-05 3.91E-05 2.043504 0.0559

SST(-1) 0.001678 0.003960 0.423671 0.6768SST(-2) 0.000434 0.004188 0.103725 0.9185GDKF -0.004736 0.011324 -0.418257 0.6807

AGRILOANS -0.011981 0.008770 -1.366152 0.1887IMPTARIFF 0.060747 0.015827 3.838221 0.0012AREACOCO 1.34E-07 2.38E-08 5.608891 0.0000

TIME -0.005965 0.002854 -2.090416 0.0510TIMESQ 0.000128 6.90E-05 1.850636 0.0807

R-squared 0.910137 Mean dependent var 0.992164Adjusted R-squared 0.850229 S.D. dependent var 0.035249S.E. of regression 0.013641 Akaike info criterion -5.456336Sum squared resid 0.003350 Schwarz criterion -4.854987Log likelihood 97.57321 F-statistic 15.19210Durbin-Watson stat 1.466034 Prob(F-statistic) 0.000001

Dependent Variable: COCONUTCYCLE Method: Least Squares Sample(adjusted): 1972 2002 Included observations: 31 after adjusting endpoints

Variable Coefficient Std. Error t-Statistic Prob.

C 0.547546 0.043983 12.44891 0.0000DPCOCO(-1) 0.001401 0.012666 0.110581 0.9131

WPCOPRA(-1) -0.000574 0.018271 -0.031428 0.9752SST(-1) 0.000190 0.003819 0.049718 0.9608SST(-2) 0.000333 0.004242 0.078455 0.9382GDKF 0.008903 0.008826 1.008758 0.3251

AGRILOANS -0.006393 0.008810 -0.725570 0.4765IMPTARIFF 0.071705 0.015624 4.589413 0.0002AREACOCO 1.49E-07 2.02E-08 7.357772 0.0000

TIME -0.008251 0.002670 -3.090715 0.0058TIMESQ 0.000175 6.67E-05 2.616858 0.0165

R-squared 0.887100 Mean dependent var 0.992164Adjusted R-squared 0.830650 S.D. dependent var 0.035249S.E. of regression 0.014506 Akaike info criterion -5.357153Sum squared resid 0.004208 Schwarz criterion -4.848319Log likelihood 94.03587 F-statistic 15.71483Durbin-Watson stat 1.156766 Prob(F-statistic) 0.000000

Dependent Variable: COCONUTCYCLE Method: Least Squares Date: 09/13/04 Time: 21:13 Sample(adjusted): 1972 2002 Included observations: 31 after adjusting endpoints

Variable Coefficient Std. Error t-Statistic Prob.

C 0.580208 0.057776 10.04245 0.0000

81

DPCOCO(-1) -0.003523 0.007665 -0.459640 0.6520DPCOCO(-2) -0.007338 0.005023 -1.460961 0.1634

WPCOPRA(-1) 4.40E-05 8.11E-05 0.542196 0.5952WPCOPRA(-2) 5.72E-05 8.98E-05 0.637516 0.5328

WPCOCOOIL(-1) -1.20E-05 6.42E-05 -0.187531 0.8536WPCOCOOIL(-2) 1.81E-05 6.46E-05 0.280145 0.7830

SST(-1) 0.001326 0.004282 0.309669 0.7608SST(-2) 0.000517 0.004793 0.107940 0.9154GDKF -0.003199 0.013099 -0.244237 0.8102

AGRILOANS -0.011688 0.009409 -1.242234 0.2321IMPTARIFF 0.061769 0.016980 3.637777 0.0022AREACOCO 1.33E-07 2.71E-08 4.914881 0.0002

TIME -0.005934 0.003169 -1.872703 0.0795TIMESQ 0.000126 8.04E-05 1.572402 0.1354

R-squared 0.910976 Mean dependent var 0.992164Adjusted R-squared 0.833081 S.D. dependent var 0.035249S.E. of regression 0.014401 Akaike info criterion -5.336688Sum squared resid 0.003318 Schwarz criterion -4.642823Log likelihood 97.71867 F-statistic 11.69484Durbin-Watson stat 1.438057 Prob(F-statistic) 0.000007

Dependent Variable: COCONUTCYCLE Method: Least Squares Sample(adjusted): 1972 2002 Included observations: 31 after adjusting endpoints

Variable Coefficient Std. Error t-Statistic Prob.

C 0.547546 0.043983 12.44891 0.0000DPCOCO(-1) 0.001401 0.012666 0.110581 0.9131

WPCOPRA(-1) -0.000574 0.018271 -0.031428 0.9752SST(-1) 0.000190 0.003819 0.049718 0.9608SST(-2) 0.000333 0.004242 0.078455 0.9382GDKF 0.008903 0.008826 1.008758 0.3251

AGRILOANS -0.006393 0.008810 -0.725570 0.4765IMPTARIFF 0.071705 0.015624 4.589413 0.0002AREACOCO 1.49E-07 2.02E-08 7.357772 0.0000

TIME -0.008251 0.002670 -3.090715 0.0058TIMESQ 0.000175 6.67E-05 2.616858 0.0165

R-squared 0.887100 Mean dependent var 0.992164Adjusted R-squared 0.830650 S.D. dependent var 0.035249S.E. of regression 0.014506 Akaike info criterion -5.357153Sum squared resid 0.004208 Schwarz criterion -4.848319Log likelihood 94.03587 F-statistic 15.71483Durbin-Watson stat 1.156766 Prob(F-statistic) 0.000000

Dependent Variable: COCONUTCYCLE Method: Least Squares Sample(adjusted): 1972 2003 Included observations: 32 after adjusting endpoints

Variable Coefficient Std. Error t-Statistic Prob.

C 0.582871 0.172340 3.382091 0.0027DPCOCO -0.111818 0.076672 -1.458396 0.1589

DPCOCO(-1) -0.062485 0.075727 -0.825133 0.4182DPCOCO(-2) 0.045734 0.078153 0.585191 0.5644WPCOPRA 0.265744 0.120970 2.196768 0.0389

WPCOPRA(-1) 0.169514 0.122385 1.385087 0.1799WPCOPRA(-2) 0.094909 0.116278 0.816223 0.4231

SOI -0.005727 0.003622 -1.581236 0.1281SOI(-1) -0.000678 0.003303 -0.205360 0.8392SOI(-2) 0.001966 0.003276 0.600085 0.5546

R-squared 0.258578 Mean dependent var 0.993568Adjusted R-squared -0.044731 S.D. dependent var 0.110932S.E. of regression 0.113386 Akaike info criterion -1.265734Sum squared resid 0.282840 Schwarz criterion -0.807691Log likelihood 30.25174 F-statistic 0.852522Durbin-Watson stat 0.799482 Prob(F-statistic) 0.578477

82

Appendix E. Estimation of Sugarcane Regression Equations Dependent Variable: SUGARCANECYCLE Method: Least Squares Sample(adjusted): 1972 2002 Included observations: 31 after adjusting endpoints

Variable Coefficient Std. Error t-Statistic Prob.

C -0.389540 0.388365 -1.003026 0.3278DPSUGAR(-1) 0.390271 0.193206 2.019970 0.0570WPSUGAR(-1) 0.028492 0.134938 0.211149 0.8349

SST(-1) 0.019615 0.038558 0.508704 0.6165SST(-2) 0.058570 0.039750 1.473468 0.1562GDKF 0.060042 0.107736 0.557310 0.5835

AGRILOANS 0.050070 0.137338 0.364576 0.7193IMPTARIFF 0.074278 0.181581 0.409063 0.6868

AREASUGAR 1.36E-06 5.84E-07 2.330408 0.0304TIME 0.033738 0.022158 1.522607 0.1435

TIMESQ -0.000820 0.000562 -1.458138 0.1603

R-squared 0.587808 Mean dependent var 1.002586Adjusted R-squared 0.381712 S.D. dependent var 0.182800S.E. of regression 0.143738 Akaike info criterion -0.770232Sum squared resid 0.413210 Schwarz criterion -0.261397Log likelihood 22.93859 F-statistic 2.852110Durbin-Watson stat 1.557784 Prob(F-statistic) 0.022075

Dependent Variable: SUGARCANECYCLE Method: Least Squares Sample(adjusted): 1972 2002 Included observations: 31 after adjusting endpoints

Variable Coefficient Std. Error t-Statistic Prob.

C 0.513381 1.384262 0.370870 0.7146DPSUGAR(-1) 0.421585 0.214497 1.965461 0.0634WPSUGAR(-1) 0.005988 0.138426 0.043254 0.9659

SOI(-1) -0.001323 0.004627 -0.286048 0.7778SOI(-2) -0.003729 0.004283 -0.870643 0.3943GDKF 0.066993 0.111764 0.599414 0.5556

AGRILOANS 0.083543 0.140458 0.594788 0.5587IMPTARIFF 0.055559 0.188565 0.294641 0.7713

AREASUGAR 1.48E-06 5.98E-07 2.479570 0.0222TIME 0.036120 0.024439 1.477970 0.1550

TIMESQ -0.000869 0.000624 -1.393008 0.1789

R-squared 0.557188 Mean dependent var 1.002586Adjusted R-squared 0.335781 S.D. dependent var 0.182800S.E. of regression 0.148981 Akaike info criterion -0.698574Sum squared resid 0.443906 Schwarz criterion -0.189740Log likelihood 21.82790 F-statistic 2.516586Durbin-Watson stat 1.495970 Prob(F-statistic) 0.037862

Dependent Variable: SUGARCANECYCLE Method: Least Squares Sample(adjusted): 1972 2002 Included observations: 31 after adjusting endpoints

Variable Coefficient Std. Error t-Statistic Prob.

C -0.497153 0.390113 -1.274380 0.2171DPSUGAR(-1) 0.421585 0.214497 1.965461 0.0634WPSUGAR(-1) 0.005988 0.138426 0.043254 0.9659

SOI(-1) -0.001323 0.004627 -0.286048 0.7778SOI(-2) -0.003729 0.004283 -0.870643 0.3943GDKF 0.066993 0.111764 0.599414 0.5556

AGRILOANS 0.083543 0.140458 0.594788 0.5587IMPTARIFF 0.055559 0.188565 0.294641 0.7713

83

AREASUGAR 1.48E-06 5.98E-07 2.479570 0.0222TIME 0.036120 0.024439 1.477970 0.1550

TIMESQ -0.000869 0.000624 -1.393008 0.1789

R-squared 0.557188 Mean dependent var 1.002586Adjusted R-squared 0.335781 S.D. dependent var 0.182800S.E. of regression 0.148981 Akaike info criterion -0.698574Sum squared resid 0.443906 Schwarz criterion -0.189740Log likelihood 21.82790 F-statistic 2.516586Durbin-Watson stat 1.495970 Prob(F-statistic) 0.037862

Dependent Variable: SUGARCANECYCLE Method: Least Squares Sample(adjusted): 1972 2002 Included observations: 31 after adjusting endpoints

Variable Coefficient Std. Error t-Statistic Prob.

C 0.562667 0.067213 8.371419 0.0000DPSUGAR(-1) -0.062107 0.038529 -1.611970 0.1235WPSUGAR(-1) 0.005846 0.024254 0.241036 0.8121

SOI 0.002059 0.000829 2.484266 0.0225SOI(-1) 0.000711 0.000797 0.891241 0.3839SOI(-2) 0.002329 0.000848 2.746542 0.0128GDKF 0.029432 0.019515 1.508121 0.1480

AGRILOANS -0.034189 0.023950 -1.427532 0.1697IMPTARIFF 0.063033 0.032206 1.957156 0.0652

AREASUGAR 4.98E-07 1.04E-07 4.806183 0.0001TIME 0.030302 0.004632 6.541897 0.0000

TIMESQ -0.000738 0.000119 -6.229426 0.0000

R-squared 0.851986 Mean dependent var 0.988415Adjusted R-squared 0.766294 S.D. dependent var 0.052503S.E. of regression 0.025382 Akaike info criterion -4.224944Sum squared resid 0.012240 Schwarz criterion -3.669853Log likelihood 77.48664 F-statistic 9.942408Durbin-Watson stat 1.986020 Prob(F-statistic) 0.000010

Dependent Variable: SUGARCANECYCLE Method: Least Squares Sample(adjusted): 1972 2002 Included observations: 31 after adjusting endpoints

Variable Coefficient Std. Error t-Statistic Prob.

C 0.687196 0.084026 8.178418 0.0000DPSUGAR(-1) -0.101892 0.048773 -2.089118 0.0520DPSUGAR(-2) -0.046144 0.040798 -1.131045 0.2737WPSUGAR(-1) 0.016543 0.022889 0.722735 0.4797WPSUGAR(-2) 0.037349 0.017256 2.164390 0.0450

SOI 0.002774 0.000875 3.169136 0.0056SOI(-1) 0.001095 0.000757 1.446759 0.1662SOI(-2) 0.002485 0.000923 2.692111 0.0154GDKF 0.019491 0.018621 1.046703 0.3099

AGRILOANS -0.066804 0.026171 -2.552536 0.0206IMPTARIFF 0.064567 0.031081 2.077402 0.0532

AREASUGAR 4.39E-07 1.18E-07 3.712967 0.0017TIME 0.027990 0.004960 5.642795 0.0000

TIMESQ -0.000686 0.000125 -5.487544 0.0000

R-squared 0.887842 Mean dependent var 0.988415Adjusted R-squared 0.802073 S.D. dependent var 0.052503S.E. of regression 0.023358 Akaike info criterion -4.373304Sum squared resid 0.009275 Schwarz criterion -3.725697Log likelihood 81.78622 F-statistic 10.35164Durbin-Watson stat 2.052910 Prob(F-statistic) 0.000012

84

Dependent Variable: SUGARCANECYCLE Method: Least Squares Sample(adjusted): 1972 2002 Included observations: 31 after adjusting endpoints

Variable Coefficient Std. Error t-Statistic Prob.

C 0.703412 0.105691 6.655389 0.0000DPSUGAR(-1) -0.075487 0.054819 -1.377020 0.1864DPSUGAR(-2) -0.019462 0.043552 -0.446869 0.6606WPSUGAR(-1) 0.016607 0.028913 0.574390 0.5732WPSUGAR(-2) 0.038296 0.021109 1.814199 0.0873

SST -0.022934 0.011104 -2.065287 0.0545SST(-1) -0.005172 0.008011 -0.645643 0.5271SST(-2) -0.014256 0.010735 -1.327975 0.2017GDKF 0.022401 0.023412 0.956801 0.3521

AGRILOANS -0.049074 0.031842 -1.541196 0.1417IMPTARIFF 0.060464 0.036745 1.645504 0.1182

AREASUGAR 3.42E-07 1.40E-07 2.450345 0.0254TIME 0.020079 0.005302 3.786969 0.0015

TIMESQ -0.000482 0.000132 -3.640162 0.0020

R-squared 0.833983 Mean dependent var 0.988415Adjusted R-squared 0.707028 S.D. dependent var 0.052503S.E. of regression 0.028418 Akaike info criterion -3.981124Sum squared resid 0.013729 Schwarz criterion -3.333517Log likelihood 75.70742 F-statistic 6.569143Durbin-Watson stat 1.234208 Prob(F-statistic) 0.000248

Dependent Variable: SUGARCANECYCLE Method: Least Squares Sample(adjusted): 1972 2002 Included observations: 31 after adjusting endpoints

Variable Coefficient Std. Error t-Statistic Prob.

C 0.670082 0.083166 8.057120 0.0000DPSUGAR(-1) -0.007466 0.005855 -1.275178 0.2194DPSUGAR(-2) -0.004480 0.005489 -0.816125 0.4257WPSUGAR(-1) 0.000103 9.79E-05 1.056243 0.3056WPSUGAR(-2) 0.000112 5.89E-05 1.905063 0.0738

SST -0.015451 0.010181 -1.517727 0.1475SST(-1) -0.003210 0.007444 -0.431210 0.6717SST(-2) -0.008412 0.010146 -0.829138 0.4185GDKF 0.017594 0.022572 0.779457 0.4464

AGRILOANS -0.064259 0.029432 -2.183304 0.0433IMPTARIFF 0.024917 0.036355 0.685376 0.5023

AREASUGAR 3.12E-07 1.31E-07 2.380552 0.0293TIME 0.021237 0.004718 4.501576 0.0003

TIMESQ -0.000309 0.000171 -1.801269 0.0894

R-squared 0.847094 Mean dependent var 0.988415Adjusted R-squared 0.730166 S.D. dependent var 0.052503S.E. of regression 0.027273 Akaike info criterion -4.063393Sum squared resid 0.012645 Schwarz criterion -3.415786Log likelihood 76.98260 F-statistic 7.244567Durbin-Watson stat 1.246688 Prob(F-statistic) 0.000133

Dependent Variable: SUGARCANECYCLE Method: Least Squares Sample(adjusted): 1972 2002 Included observations: 31 after adjusting endpoints

Variable Coefficient Std. Error t-Statistic Prob.

C 0.607084 0.073296 8.282691 0.0000DPSUGAR(-1) -0.008778 0.005209 -1.685176 0.1102DPSUGAR(-2) -0.007955 0.005110 -1.556576 0.1380WPSUGAR(-1) 8.14E-05 8.02E-05 1.014994 0.3243WPSUGAR(-2) 8.99E-05 5.37E-05 1.673846 0.1125

SOI 0.002009 0.000816 2.463256 0.0247SOI(-1) 0.000765 0.000736 1.039931 0.3129

85

SOI(-2) 0.001812 0.000886 2.045261 0.0566GDKF 0.022884 0.018637 1.227869 0.2362

AGRILOANS -0.072360 0.024694 -2.930257 0.0093IMPTARIFF 0.019296 0.031595 0.610717 0.5495

AREASUGAR 4.05E-07 1.16E-07 3.487232 0.0028TIME 0.028752 0.005002 5.747823 0.0000

TIMESQ -0.000430 0.000153 -2.807954 0.0121

R-squared 0.886763 Mean dependent var 0.988415Adjusted R-squared 0.800169 S.D. dependent var 0.052503S.E. of regression 0.023470 Akaike info criterion -4.363731Sum squared resid 0.009364 Schwarz criterion -3.716124Log likelihood 81.63783 F-statistic 10.24055Durbin-Watson stat 1.883389 Prob(F-statistic) 0.000013

86