Final Manuscript - 1st Revision.docx

1

De La Salle University – Dasmariñas FORECASTING THE PRICE OF CORN IN THE PHILIPPINES

A Thesis Proposal Presented to the Faculty of the Allied Business Department College of Business Administration and Accountancy De La Salle University-Dasmariñas Dasmariñas City, Cavite

In partial fulfillment of the requirements for the degree of Bachelor of Science in Business Administration (Major in Economics)

BILLY JULIUS M. GESTIADA March 3, 2017

2

De La Salle University – Dasmariñas CHAPTER I INTRODUCTION

Corn is the second most bountiful crop grown all over the world, and many people have been consuming this for everyday living. It is a multifaceted crop, and there is no wasted part on its plant. In Mexico, corn husks are made into their traditional tamale. Kernels are converted into food. Animals feed on the stalks, and the corn silks are made into herbal teas. Some food products like corn oil, corn meal, corn sweetener, corn syrup, and even corn whiskey are made from corn. (Sailer, 2012) In the United States (US), even if the farmers are capable of growing different kinds of grains and crops and bringing them to the market, corn accounts for 90 percent of all the produced grain. In 2015, about 80 million acres of farmland are being planted with corn, and the world is being supplied with 20 percent of the American corn. While it is true that the US is maintaining its current reputation as an international exporter of corn, what remains from these corns is not entirely wasted. Given that corn is the primary crop grown in the US, every man, woman, and child consumes four pounds of corn a day, which amounts to a total of more than 1,500 pounds of corn consumed annually. (NathanF, 2015) Even though the US is considered to be the largest exporter of corn in the world, less than 15 percent share of the demand for the US corn is accounted by the exports, which is actually small. This occurrence has something to do with the demand-andsupply-relationship of corn, resulting to the other markets adjusting to the US market’s

3

De La Salle University – Dasmariñas current price. Because of this internationally tough competition, farmers plant their corn after considering the size of the US crop in order to have a market advantage over the short US crops. In fact, some countries like Brazil, India, and South Africa had significant corn exports when international prices are competitive, or the crops are large. (United States Department of Agriculture Economic Research Service [USDA ERS], 2017) In many countries, particularly the developing ones, commodities still remain a reliable source of export earnings. Moreover, price movements of these commodities play a major role on overall macroeconomic performance. Commodity-price forecasts are essential in formulating and planning macroeconomic policies. (Bowman and Husain, 2004) These studies mentioned above are only a very small portion of numerous studies done on commodity prices. In the field of economics, this kind of study is not something new. The efforts of the previous researchers contributed a lot to the present knowledge of commodity prices. Background of the Study A study was conducted by Halonen (2016) showing that there are few statistical techniques that can outperform models that pertain to supply and demand analysis in forecasting the price of corn in the US. The researcher argued that there are some econometric techniques that are costly to use, none of them of being more costly than the supply and demand analysis. The main reason for this much expense is that supply and demand analysis involves gathering and summarizing a large amount of information

4

De La Salle University – Dasmariñas regarding supply and demand. Furthermore, it also requires extensive surveys to be distributed to a large sample in a particular study. That being the case, this study examined if there are some statistical methodologies that can provide forecasts at least as accurate, or even not as costly as the models incorporating supply and demand analysis. Both the statistical methodologies and the supply and demand models were evaluated at one, three, six, nine, and twelve month horizons, given that these horizons are suitable for analyzing commodities that involve buying, selling, production, and contract negotiations. It was found out that an AR model is the best model to use in forecasting over a short horizon, while VAR model is the best model to use in forecasting over a long horizon, over six months. Another study pertaining to forecasting the price of corn, along with other 14 commodities, has been conducted by Bowman and Husain (2004). The research analysed the performances of three different types of commodity price forecasts namely: judgment-based, historical price-based, and commodity futures-based. Since spot prices tend to move forward future prices for most commodities in the long run, and the future prices showing lower variability, it was found out that commodity futures-based model outperforms both the judgment- and historical price-based models in directional terms, at the very least. Aside from the mentioned three different types of commodity price forecasts above, Jha and Sinha (2013) conducted price forecasting on soybean and rapeseedmustard wholesale prices in India using neural network model. The researchers stated that the innovation of Artificial Neural Network (ANN) proved to be feasible given the data

5

De La Salle University – Dasmariñas provided by developing countries. In this study, ANN indicated more significant number of future price changes as compared to linear model. This means that in the context of commodity price forecasting, where turning points are crucial, ANN model might be preferred because it totally outperforms nonlinear models most especially when the series is linear. Lastly, even if the series is nonlinear, combining linear and nonlinear models was observed to perform better than these two models performing independently. Although there have been many papers done in other countries pertaining to models used in forecasting the price of corn, not much is done in the Philippines. This paper will focus on providing an econometric model in forecasting the price of corn in the Philippines. Statement of the Problem Commodity price forecasting is an essential part of any industry involving trading and price analysis. Commodity prices are often unpredictable that becomes even highly unpredictable when you factor the presence of natural calamities droughts, typhoons, floods, and pests. Because of this, there’s a greater risk and uncertainty in formulating a forecasting methodology. In the case of the Philippines, where rice and corn are the major crops, policy makers should see to it that they make reliable, highly accurate forecasts of rice and corn prices in order to ensure food security, thus somehow alleviating hunger and poverty. Farmers will also benefit from commodity price forecasting because they will definitely want to make their production and marketing decisions wisely so that they will be able reap positive financial outcomes in the future. (Jha and Sinha, 2013)

6

De La Salle University – Dasmariñas Another problem in conducting a commodity price forecast is the volatility of prices over time. A study regarding commodity price forecast resulted in forecast prices increasing rapidly, and in the long-run becoming larger due to a spike in futures prices. This resulted to a lower accuracy of the forecasts. It was also mentioned in the study that in order to improve forecast accuracy, dummy variables may be used to adjust for price spikes. Technically, it can be observed that there is a need to compare forecasting models with the other models to ensure that a proper model is used in a proper scenario. (Bowman and Husein, 2004) Another study dealt with the problem of short-term market price forecasting. Time series analysis is usually used in dealing with this problem. Furthermore, ANN, a new technique, has been discovered as a tool in price forecasting. In this study, ANN model has been compared with the time series autoregressive integrated moving average (ARIMA) in forecasting the price of tomato from years 1996 to 2010. The results showed that ANN model performed better than ARIMA model in terms of their relative errors. (Li, Xu and Li, 2010) Corn is second to rice as the most important crop in the Philippines, and yet the studies done regarding forecasting the price of corn in the Philippines are very few. We can only see studies done about the pricing behavior of Philippine corn, relationship between trade liberalization and Philippine corn prices, relationship between the prices of Philippine rice and corn, socio-economic impact of corn in the Philippines, etc. Basically, these studies only present behaviors, relationships, performances, impacts, etc. Like in the other countries, it is important to emphasize methodologies for the improvement of

7

De La Salle University – Dasmariñas forecasting of the price of corn in the Philippines in order to aid both the producers and consumers in making sound decisions. Specifically, this study answered the following questions: 1. What is the trend of the price of corn in the Philippines over time? 2. How well does Autoregressive Integrated Moving Average (ARIMA) model perform in forecasting the price of corn in the Philippines? and 3. How well does autoregressive (AR) model perform in forecasting the price of corn in the Philippines? Objectives of the Study Generally, this study aimed to provide a forecast on the price of corn in the Philippines. In order to carry out the general objective in a more organized and systematic way, the following specific objectives were made: 1. To describe the trend of the price of corn in the Philippines over time; 2. To analyze the price of corn in the Philippines using ARIMA model; and 3. To investigate the price of corn in the Philippines using AR model; Hypotheses of the Study Dash, Solanki and S. (2012) conducted a study in India regarding commodity market behavior, price and its factors. Included in these commodities are the three agroproducts, namely: channa, wheat and pepper. The main factor that affects the prices of these crops, in terms of supply and production, is the monsoons. These crops are also affected by storage constraints that are temporary. Other factors include inflation, supply

8

De La Salle University – Dasmariñas constraints, costs of production, foreign exchange holdings, and some international policies pertaining to imports and exports. Thus, in order to carry out the study more properly and systematically, the researcher hypothesized that: H1: Commodity prices are generally nonstationary. In a study conducted by Wang and Tomek (2004) regarding commodity prices and unit root tests, it was found out that commodity prices are generally treated as stationary. However, unit root tests prove that commodity prices are generally nonstationary, most especially when the test specification does not account for structural changes. H2: There is an upward trend in corn prices. The European Central Bank (2014) reported on its July Monthly Bulletin that commodity prices (oil and food) had an upward trend despite of being interrupted by a financial crisis in 2008. H3: Farmgate prices, rather than wholesale or retail prices, should be the primary concern of forecasting the price of corn in the Philippines. Okunmadewa (n.d.) explained that despite of the farmers giving their best effort in producing crops or livestock, they tend to get the least out of it when it comes to selling the products in the market. This is the same with forecasting the price of the corn in the Philippines. This paper would like to focus on the farmer’s side, whose efforts are much more rigorous compared to the consumers, rather than the consumer’s side. H4: ARIMA Model is an efficient tool in forecasting the price of corn. Jadhav, Reddy and Gaddi (2017) conducted a study on the application of ARIMA Model for forecasting the prices of paddy, ragi, and maize (corn) in India. The results

9

De La Salle University – Dasmariñas showed that ARIMA Model is a powerful tool in forecasting commodity prices. Furthermore, the research checked the validity of the model using the values of MSE, MAPE, and Theil’s U, and these values indicated that the forecasted values are almost similar to the actual values. Lastly, one of the limitations of the ARIMA Model is that the time series should be long, which makes the said model really suitable in forecasting the price of corn in the Philippines. Significance of the Study This study compared the performances of both AR model and ARIMA model in order to determine the model that is flexible enough to the volatility of corn’s prices in the Philippines. The government, most especially the policy makers, this impacts their decision as to how they are going to forecast the price of corn in the Philippines. Given the uncontrollable circumstances that could negatively affect the commodity prices, it is better to have many alternative models that could fit the scenario given certain factors. The farmers are guaranteed to benefit on this study as they will be guided on what decisions should be made in the future in order to be financially stable. Having a reliable commodity price forecasting method to account for yields will be very helpful. Though farmers are considered starving and dying in the Philippines, the opportunity to receive financial incentives in the future is always there for as long as they are willing to grab it. The students should be able to learn the value of food security in the long-run as early as possible. In response to this, through this study, they will learn that commodity

10

De La Salle University – Dasmariñas price forecasting is not simply about being able to understand numbers and figures, but by those figures and numbers, policies can be derived in order to secure food in the longrun. This study could be further improved by the future researchers who will be conducting a research similar to this. The fact that this study only has one variable, it might be better for the other researchers to come up with models, aside from the commonly used ARIMA and AR models, which could easily deal with univariate analysis while also looking into the effectiveness of their performances as well. Scope and Limitations This study covered the prices of corn from 80 provinces/cities including Metro Manila, the same with the provinces/cities covered by the Philippine Statistics Authority (PSA). This study is limited only to the data available at PSA as the said organization has the wholesale, retail, and farmgate prices of corn in the Philippines. This follows the assumption that the data provided by PSA are all accurate. This study is limited only to the use of two models, AR and ARIMA. This paper’s main model will be ARIMA while AR will only be a model for comparison. The data that used in forecasting the price of the corn in the Philippines is only from 1960 to 2016 because the data from these years are still available and accessible through the data sources of this study. Definition of Terms

11

De La Salle University – Dasmariñas Commodity Price refers to the wholesale, retail, or farmgate price of crops such as rice, corn, sugar, cassava, vegetables, fruits, and rubber, which could be either wholesale or retail. Corn or yellow corn specifically is the second most important crop in the Philippines, and is the main subject of this study. Farmgate Price means the price of corn set by the producer itself. It is also termed as the producer price. Forecasting is the method used in this study that uses historical prices of corn in order to generate policies and recommendations pertaining to food security and valuation of corn prices in the Philippines. Price refers to the farmgate prices of corn in the Philippines, and is one of the main variables used in this study.

CHAPTER II REVIEW OF RELATED LITERATURE

This chapter discussed some past researches conducted that are related to this study. This chapter also critically evaluated and analyzed the studies that have been

12

De La Salle University – Dasmariñas conducted before, which enabled the researcher to create a foundation for the study. Through the help of review of related literature, the researcher determined what has been discussed by the previous studies so far, and what has not yet been discussed that can serve as a research gap. This chapter focused on the previous researches done on forecasting the price of commodities. This chapter, review of related literature, discussed previous studies conducted relating to impact of commodity prices to the economy, determinants of commodity prices, and commodity price forecasting. Impact of Commodity Prices to the Economy Sands (2015) stated that fluctuations in commodity prices affect the entire economy in terms of employment, public and private expenditures, and capital accumulation. When the prices fluctuate down, the rate of return of commodity sectors exceeds that of the non-commodity sectors. In addition, a lot of economic problems arise whenever economies rely on commodities as the main component of their Gross Domestic Product (GDP). Because of this, we see a shift from commodity sectors into productive non-commodity sectors. Brazil is said to be one of the major commodity exporters all over the world, and it has its own major stocks as well. However, a commodity deflation has been experienced at around April 2015, which forced Brazil’s majors stocks to give negative returns. The researcher then concluded that in order to adjust to lower commodity prices, two steps under fiscal policy can be undertaken. First is for the government to reduce taxes to increase household spending. Last is to handle both unemployment and the investment cycle by investing in other productive assets.

13

De La Salle University – Dasmariñas An Australian economist said that it can be challenging on the part of a researcher to analyze how commodity products are likely to impact both the customers and the whole economy. The researcher further explained that one of the pressing issues concerning commodity markets is the dramatic declines in the industrial commodity prices such as iron-ore and oil. Basically, the study showed how this scenario would impact both the global and Australian economies. For the global economy, a fall in oil prices will have significant implications for oil importers and exporters, consumers and governments. In this case, Russia and Organization of the Petroleum Exporting Countries (OPEC) countries, which rely heavily on oil revenues to fund their government expenditures, will lose a lot during heavy price decreases of oil. Although affiliated companies such as energy-mining companies and the like will be negatively affected, a lot of countries will still benefit. In fact, industries that have higher input costs on oil will have free cash flows, and will be able to operate at higher margins. As for the Australian economy, the results showed that the impacts will most likely be seen in inflation and interest rates in the short-run. (Oster, 2015) An agricultural sector becomes successful provided that it supports economic growth. The US has a strong economy in terms of agriculture. American farmers are capable of producing vegetables, fruits, grains, meat, and dairy products at a low cost. As a result of this, domestic food supply becomes safe and secured. Furthermore, through modern technology, the American agriculture sector is capable of producing biofuels and other sources of alternative energy in order to minimize dependence on foreign oil. This helps to reduce the costs incurred by the businesspeople and consumers in purchasing gas

14

De La Salle University – Dasmariñas or oil. Finally, it is truly important for rural areas and small towns to have a strong agricultural economy. In fact, farmers and ranchers give full support to farm industries, and they purchase local goods and services, which results to an increased production. This high level of production has contributed a lot to the businesses given that a strong agricultural economy exists. (United States Congress Joint Economic Committee [JEC], 2013) Determinants of Commodity Prices There has been a vast study regarding both short- and long-term determinants of commodity prices. Over the years, studies pertaining to this topic become more prevalent. Good (2008) conducted a study on the factors affecting corn and soybean prices. The researcher stated that the agricultural commodities have been influenced by the change of value of US-Dollar which has a negative relationship for both the corn and soybean prices. Changes in crude oil prices are considered to affect both the corn and soybean prices negatively. News pertaining to exports also affects both corn and soybean prices. Weather is an important factor to every agricultural commodity, corn and soybean included. Another important factor is production as it is highly related with weather. Finally, the developments in the financial markets have positive effect on corn and soybean prices. Similarly, any weakening of those markets will have a negative effect on both commodities. Similar to the study conducted by Good (2008), in a macroeconomic perspective, the determinants of agricultural commodity price volatility include the following: (1)

15

De La Salle University – Dasmariñas stocks that has a negative relationship with price volatility ; (2) Southern Oscillation Index (SOI) that has a positive relationship with price volatility; (3) world market structure that has a negative relationship with price volatility; (4) biofuel production that has a positive relationship with price volatility; (5) Kilian index; (6) crude oil price behavior that has a positive relationship with price volatility; (7) US-Dollar exchange rate volatility that has a positive relationship with price volatility; (8) US interest rate that has a negative relationship with price volatility; (9) the Scalping index; and (10) the WorkingT

index.

The

performances

of

Generalized

Autoregressive

Conditional

Heteroskedasticity-Mixed-data Sampling (GARCH-MIDAS) and GARCH(1,1) were compared, which GARCH-MIDAS always performed better than the other model. This analysis was applied and tested for wheat, corn, and soybean. (Dönmez and Magrini, 2013) Adeyanju (2014) argued that corn has been an important food to the entire human race. However, more than just a food source, corn has also become an important fuel source. Thus, the researcher enumerated the top factors that either increase or decrease the price of corn. First is the effect of Ethanol, which comes from corn. Given that an increase in the demand for ethanol would increase the demand for corn, which will surely increase the price of corn. However, when the demand for ethanol decreases, decreasing the demand of corn, it is not necessarily equal to the effect of increasing demand for corn given that only 40 percent of corn becomes ethanol. Another factor is the crude oil prices which has a positive relationship with corn prices most of the time. This is because even corn has been functional as an energy commodity as well. Next is the speculator effect,

16

De La Salle University – Dasmariñas which is considered to be the biggest driver of corn prices. Naturally, it will be smart for investors to observe how corn is being valued before taking any actions. Climate is a very important factor of corn included. Another important factor, though not as significant as the other factors, is the Chinese effect. China is said to be taking efforts to have a cleaner energy, therefore there will be an increase in demand for ethanol, which will most likely contribute to an increase in demand for corn. Finally, geopolitical issues play an important role in the corn since corn production is unevenly distributed worldwide. Technically, a change in economy affects corn industries. Commodity Price Forecasting There were a lot of researches done on the forecasting of commodity prices using different econometric methods. Most researchers generally use either ARIMA model, or VAR (Vector Autoregressive) model, for multivariate studies, or AR, for univariate studies, in commodity price forecasting. In fact, Tripathi et al. (2014) conducted a study in India regarding rice productivity and production using ARIMA models. The paper focused on the analysis of trend of rice area, production, and productivity of Odisha as compared to India using data from years 1950 to 2009. It also focused on forecasting the rice area, production, and productivity using ARIMA models. It was found out that there is an increasing trend in productivity and production for both India and Odisha, with Odisha having a lesser rate of increase than India. The researchers believed that it is because of the low input in agricultural operations and other biotic and abiotic factors.

17

De La Salle University – Dasmariñas Overall, it was proved that ARIMA model can be successfully used to forecast rice area, productivity, and production for both Odisha and India in the coming years. In a study pertaining to forecasting major fruit crops productions in Bangladesh, Box-Jenkins ARIMA model was used. The study aimed to fit the Box-Jenkins ARIMA model in forecasting three of the major fruit crops in Bangladesh namely: Mango, Banana, and Guava. It was found out that for Mango, the best chosen Box-Jenkins ARIMA model, accounting for more than 5% level of significance, is ARIMA(2,1,3); for Banana, it is ARIMA(3,1,2); and for Guava, it is ARIMA(1,1,2). The researcher concluded that given that these three models are capable of practically explaining the situation, they are the best model to use in forecasting. The researcher further recommended that these models can be used for decision-making by the researchers, policymakers, businessmen, etc. Finally, this study concluded that Box-Jenkins ARIMA model performs good in short-term forecasting. (Hamjah, 2014) In addition to the usage of ARIMA model in forecasting commodity prices of various places and periods, other researchers have forecasted commodity prices using regime-switching models, which this paper will also use in forecasting the price of corn in the Philippines. Ubilava and Helmers (2011) conducted a study regarding the impact of El Niño Southern Oscillation (ENSO) – a natural phenomenon characterized by wind variations and changes in sea surface temperature – on predicting world Cocoa prices. The researchers contributed to the previous knowledge of commodity price forecasting by considering that a nonlinear causal relationship between ENSO and world Cocoa prices would be possible to compare the performances between linear and nonlinear

18

De La Salle University – Dasmariñas models. The smooth transition autoregressive framework (STAR) model, the model used by the researchers, and is under the regime-switching models, proved that nonlinear models are more reliable in out-of-sample forecasting compared to linear models. Furthermore, the study concluded that there exists a Granger causality between ENSO and world Cocoa prices. The STAR model was also used in forecasting Corn and Soybean basis using regime-switching models, a study conducted by Sanders and Baker (2012). In this study, it was stated that producers of corn and soybean in the core production areas in the US have noticed a great increase in the volatility of prices in their recent years, which resulted to an increase of price risk of producers in decision-making. This paper aimed to apply regime-switching models to formulate a model that could adjust to the prices’ changing volatilities, and to provide more accurate forecasts especially in periods of changing volatilities. The researchers found out over the course of their study that time series econometrics perform better at short-term forecasting, but difficult to use in longterm forecasting. Finally, the study concluded that regime-switching models do not provide real forecasting improvement over ARIMA models despite of statistical significance in favour of the regime-switching models. This paper focused on forecasting the price of corn in the Philippines. The previous studies that have been presented in this section clearly explained the need to forecast commodity prices in various places, as well as how these commodity prices will have an impact on the economy. Through these past studies done by different researchers,

19

De La Salle University – Dasmariñas this paper was able to contribute additional knowledge in commodity price forecasting by formulating a methodology that will forecast the price of corn in the Philippines.

CHAPTER III FRAMEWORKS OF THE STUDY

Theoretical Framework

20

De La Salle University – Dasmariñas The previous chapters of this study have mentioned some among the numerous studies done on commodity price forecasting models. As mentioned in the Chapter II of this study, among the most used models by the researchers when dealing with commodity price forecasting are ARIMA and AR/VAR models. Though not as common as the previous mentioned two models, there are still a lot of models that can be used in forecasting commodity prices as they will have their own importance depending on the scenario. Judgmental Forecasting. This a forecasting method which relies on a person’s own judgment of a particular situation. It is naturally expected to be subjective because it does not rely on historical and other statistical data, which means that it can only be used on qualitative researches. Hillier, F. S. and Hillier, M. S. (2001) enumerated the commonly used judgmental forecasting methods, namely: (1) manager’s opinion which relies on a single manager’s best judgment in forecasting; (2) jury of executive opinion that is similar to the first one, except now that there is a small group of managers who combine their best judgments; (3) sales force composite which is often used by the companies when they want to generate higher sales by hiring sales forces; (4) consumer market survey that relies on surveying actual or potential customers in order to determine their responsiveness to the new products or new features of the existing products; and (5) Delphi method which involves a group of experts from various locations independently filling out a series of questionnaires. Unit root model. The unit root problem is demonstrated when the presence of unit root in a time series affects statistical inferences due to some vague, unpredictable

21

De La Salle University – Dasmariñas patterns. The solution provided to this problem is the unit root testing which ensures that the time series is stationary, that is the statistical properties do not change over time. Some commonly used unit root tests include, but not limited to, the Dickey Fuller Test, Augmented Dickey-Fuller (ADF) Test, and Phillips-Perron (PP) Test. The unit root model with trend and drift is the simplest form of forecasting model, and it can be written as: yt = µ + yt-1 + ut, where yt is the natural logarithm of the commodity price at period t, and the error term, ut is assumed to be a white noise. ARIMA model. The ARIMA model was first introduced by the statisticians George E.P. Box and Gwilym M. Jenkins and thus being commonly known as BoxJenkins model which is used as a forecasting model. This is probably the most commonly used model in forecasting commodity prices, and is also the most commonly used model in forecasting other prices given that it can convert non-stationary time series data in to stationary time series data using differentiation.

The equation for ARIMA model in a stationary time series analysis is a linear equation, which can be expressed as:

22

De La Salle University – Dasmariñas

Futures Forecast Model. The futures price is one way of forecasting commodity spot prices. Mckenzie and Holt (1998) and Chinn and Coibion (2010) stated that the futures price is an unbiased predictor of future spot prices, and there is a little evidence that it is also the best forecast according to Alquist and Kilian (2010) and Alquist et al. (2011). Despite of a large literature proving that the capacity of futures price to forecast exceeds that of the random walk model, the model concerning futures prices performs differently depending on the commodity, whether it is consumed daily, weekly, monthly, or even yearly. The general futures forecast model is expressed as: St = α + βFt|t-k + et, where Ft|t-k is the price for period t with future markets in period t-k. Vector Autoregressive Model. The VAR Model, which is a simple, yet flexible model that deals with multivariate time series data, is just a natural extension of the AR Model, which deals with univariate time series data. Being one of the most commonly used model in forecasting commodity prices, VAR Model is often compared to ARIMA Model alongside Error Correction Model (ECM) in terms of their effectiveness given various situations. However, it was also found out that there are times when VAR Model is preferred over ARIMA Model because there are more theoretical backgrounds on the former model than the latter. This model was popularized by the American econometrician and macroeconomist Christopher A. Sims (1980) on his journal article

23

De La Salle University – Dasmariñas entitled Macroeconomics and Reality. In that article, Sims demonstrated that VAR model is able to provide a flexible, better framework in analyzing economic time series data. Assuming there are three different time series variables, denoted by xt,1, xt,2, and xt,3, the VAR model of order 1 is expressed as: xt,1 = α1 + ϕ11xt-1,1 + ϕ12xt-1,2 + ϕ13xt-1,3 + wt,1 xt,2 = α2 + ϕ21xt-1,1 + ϕ22xt-1,2 + ϕ23xt-1,3 + wt,2 xt,3 = α3 + ϕ31xt-1,1 + ϕ32xt-1,2 + ϕ33xt-1,3 + wt,3 , where α is constant, ϕ is the phi coefficient, and wt is the error term. Conceptual Framework Mentioned in the hypotheses of the study are the characteristic and trend of the commodity prices, most especially price of corn. Furthermore, it has been mentioned the superiority of farmgate prices over wholesale and retail prices, and the importance of focusing more on ARIMA model than the other models. Figure 1 represents specifically the model which this study used in forecasting the price of corn in the Philippines.

Historic al Prices

essive essive Integrat Integrat ed ed Moving Moving Averag Averag ee Model Model

Autore Autoregr gr

Autore Autore gressiv gressiv ee Model Model

Fut ure Pric es

24

De La Salle University – Dasmariñas

Figure 1. Overall framework of the research

CHAPTER IV

METHODOLOGY

25

De La Salle University – Dasmariñas

Research Design This study dealt with the quantitative aspect of research. Specifically, this paper aimed to assess whether what model performs the best in forecasting the price of corn in the Philippines. The models included ARIMA model and AR model. This study used the historical design of research. The historical design of research enabled the researches to gather and synthesize past data in order to accept or reject a hypothesis – to prove whether corn prices in the Philippines have an upward or downward trend. Furthermore, this study is also an evaluative research. This paper also provided an evaluation and assessment on what model performs the best in forecasting the price of corn in the Philippines. Since food security is a very serious matter not only in the Philippines, but in the other countries as well, the forecasting method should be ensured that it provides the best, most accurate forecasts as possible. Sources of Data This study gathered data from the secondary sources that are available and accessible to the public online. These data came from government agencies, specifically the Philippine Statistics Authority (PSA), whose scope includes the gathering price of the agricultural crops in the Philippines. Aside from PSA, the data also utilized the study conducted by Power and Intal, Jr. (1990), under the World Bank Comparative Studies, entitled Trade, Exchange Rate, and Agricultural Pricing Policies in the Philippines. Methods of Data Analysis

26

De La Salle University – Dasmariñas This study used some statistical techniques depending on the requirements presented on the objectives of the study. This section mentioned the different statistical techniques that this study will employ. In the case of historical design, tables and graphs are used in order to clearly see the trend of prices of corn in the Philippines. Using these tools enabled the researcher to analyze the patterns displayed in the historical data gathered, which will led to an intelligent conclusion as to why such pattern/s occurred. As to the evaluative design of this study, both the ARIMA and AR models are utilized for comparison as to what model performs best in forecasting the price of corn in the Philippines. These two models are suitable to use when a particular study concerning forecasting has only one variable available. In order to determine the significance of the overall models of the study, both the coefficient of determination (R2) and the F-statistic are checked as well. Eviews is used in the estimation procedure. The ARIMA model, which satisfied the second objective of the study is: Predicted Value of FPRICE = µ + FPRICEt-1 , where: FPRICE = Farmgate Price of Corn in the Philippines (in PhP/kg.) t = Time µ = Constant term, average change over time The AR model, on the other hand, which satisfied the third objective of the study is: Predicted Value of FPRICE = ϕFPRICEt−1 + ut , where:

27

De La Salle University – Dasmariñas c = Constant ϕ = Phi coefficient (should not be less than 1) u = Random error at period t There are various ways of measuring the effectiveness of forecast performances of different models such as mean absolute relative pricing error (MARPE), mean absolute error (MAE), or root mean squared error (RMSE). This research primarily focused on using RMSE and Theil’s Inequality Coefficient in checking the average forecast error of both AR and ARIMA. RMSE is also a suitable measure to use given that it can only be used for a specific commodity and not for comparison across various commodities. The formula for RMSE is written as:

√

1 RMSE = n

n

∑ ( Si - FCi )2

,

i =1

where Si is the spot price of the commodity, and FCi is the commodity forecast price. And the general formula of Theil’s Inequality Coefficient is expressed as:

TH =

√

1 n

n

√

1 n

n

∑ e2 i =1

√

∑ y 2 + 1n i =1

where n is the sample size of the study, ŷ

n

,

∑ ŷ2 i =1

is the predicted value of y, and e is the

equivalence factor, denoting economies of scale.

28

De La Salle University – Dasmariñas

CHAPTER V RESULTS AND DISCUSSION

29

De La Salle University – Dasmariñas The first part of this section provided the prices of corn in the Philippines (farmgate, retail, and wholesale) annually from years 1960 to 2016, which were obtained primarily on the Philippine Statistics Authority. It is followed by the results of ARIMA and AR models pertaining to their respective forecast performances in the farmgate prices of corn in the Philippines, as well as the detailed discussions of those results. This chapter ended with a decision criteria on which model performed better in terms of forecasting the farmgate prices of corn in the Philippines. Timmer (2008) conducted a study concerning the causes of high food prices. It is because of these high food prices that poor consumers are experiencing grave consequences concerning food security. The study concluded some factors that affect the food prices depending on the year. In 2004, at least three main factors are found to be dominant, namely: (1) China’s rapid economic growth and the excess of demand over supply in India; (2) a constant decline in the value of US dollar; and (3) the combined high and still rising prices of fuel that were found out to be related to the other commodity prices. In the Philippines, one of the most common agricultural problems is the climate or weather. During typhoons, the usual scenario is that people expect a price spike in the agricultural prices due to the damage dealt to the farmlands and its farmers. Contrary to this belief, the Bureau of Agricultural Statistics (BAS) (2013) stated that prices of rice and corn remained stable in Visayas region during the week when typhoon Yolanda, one of the strongest typhoons recorded in the world, devastated the said region.

30

De La Salle University – Dasmariñas This is a scenario which is not commonly seen among different countries, and therefore should not be expected to frequently occur. Padin (2016) reported that the average farmgate prices of local corn have risen during the recent weeks as El Niño continues to pester the areas in the Philippines where corn is thriving. Here, the farmers had a difficult time earning due to the harsh climate, which forced the prices of corn to increase. The Status of Prices of Corn in the Philippines Farmgate Prices. Table 1 shows the farmgate prices of corn in the Philippines from 1960 to 2016 while Figure 2 is the presentation of these tabulated prices in a graphical form. Generally, from the graph, the trend is found to be upward, with some evident price fluctuations starting 1984 up to 2016.

Table 1

31

De La Salle University – Dasmariñas Farmgate Prices of Corn, Philippines, 1960-2016 Year 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 Continued …

Price (in PhP/kg.) 0.17 0.19 0.18 0.23 0.25 0.26 0.28 0.26 0.26 0.27 0.33 0.48 0.54 0.56 0.91 0.93 0.94 0.99 0.97 1.00 1.16 1.29 1.34 1.39 2.36 2.91 2.70 2.98 2.96

Percentage change (%) 11.765 -5.263 27.778 8.696 4.000 7.692 -7.143 0.000 3.846 22.222 45.455 12.500 3.704 62.500 2.198 1.075 5.319 -2.020 3.093 16.000 11.207 3.876 3.731 69.784 23.305 -7.216 10.370 -0.671

32

De La Salle University – Dasmariñas

Year

Price (in PhP/kg.)

Percentage change (%)

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016

4.15 4.26 3.68 4.99 4.62 4.98 6.28 6.16 5.97 5.60 5.39 6.37 6.50 6.42 6.67 8.49 7.54 9.11 10.09 10.79 10.44 11.26 11.95 12.43 11.62 12.73 12.01 11.79

40.203 2.651 -13.615 35.598 -7.415 7.792 26.104 -1.911 -3.084 -6.198 -3.750 18.182 2.041 -1.231 3.894 27.286 -11.190 20.822 10.757 6.938 -3.244 7.854 6.128 4.017 -6.516 9.552 -5.656 -1.832

33

De La Salle University – Dasmariñas

Figure 2. Farmgate prices of corn in the Philippines

34

De La Salle University – Dasmariñas Retail Prices. Table 2 shows the retail prices of corn in the Philippines from 1960 to 2016 while Figure 3 is the presentation of these prices graphically. As shown in the graph, the prices display an upward trend, with strong price spikes from around 1983 until 2016.

35

De La Salle University – Dasmariñas Table 2 Retail Prices of Corn, Philippines, 1960-2016 Year 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987

Price (in PhP/kg.) 0.29 0.30 0.30 0.37 0.40 0.45 0.50 0.46 0.46 0.46 0.50 0.81 0.82 0.91 1.39 1.53 1.46 1.60 1.60 1.67 1.90 2.20 2.24 2.34 3.71 5.11 4.95 5.12

Percentage change (%) 3.448 0.000 23.333 8.108 12.500 11.111 -8.000 0.000 0.000 8.696 62.000 1.235 10.976 52.747 10.072 -4.575 9.589 0.000 4.375 13.772 15.789 1.818 4.464 58.547 37.736 -3.131 3.434

36

De La Salle University – Dasmariñas 1988 Continued …

Year 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016

5.19

1.367

Price (in PhP/kg.) 5.93 7.05 6.80 8.10 8.07 8.53 9.79 10.97 11.10 11.66 11.73 12.71 13.41 13.45 12.98 14.40 14.30 14.65 15.79 18.18 19.90 19.26 19.80 21.51 22.04 20.76 20.70 20.36

Percentage change (%) 14.258 18.887 -3.546 19.118 -0.370 5.700 14.771 12.053 1.185 5.045 0.600 8.355 5.507 0.298 -3.494 10.940 -0.694 2.448 7.782 15.136 9.461 -3.216 2.804 8.636 2.464 -5.808 -0.289 -1.643

37

De La Salle University – Dasmariñas

38

De La Salle University – Dasmariñas

Figure 3. Retail prices of corn in the Philippines

39

De La Salle University – Dasmariñas Wholesale Prices. Table 3 shows the wholesale prices of corn in the Philippines from 1960 to 2016 while Figure 5 presents the graphical form of these prices. It can be seen from the graph that there is an upward trend in the prices, with rapid increase from around 1983 up to 2016.

40

De La Salle University – Dasmariñas Table 3 Wholesale Prices of Corn, Philippines, 1960-2016 Year 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987

Price (in PhP/kg.) 0.22 0.25 0.20 0.27 0.28 0.36 0.36 0.33 0.33 0.33 0.38 0.64 0.63 0.67 1.07 1.16 1.19 1.22 1.23 1.26 1.41 1.59 1.59 1.78 2.92 3.57 3.48 3.63

Percentage change (%) 13.636 -20.000 35.000 3.704 28.571 0.000 -8.333 0.000 0.000 15.152 68.421 -1.563 6.349 59.701 8.411 2.586 2.521 0.820 2.439 11.905 12.766 0.000 11.950 64.045 22.260 -2.521 4.310

41

De La Salle University – Dasmariñas 1988 Continued …

Year 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016

3.67

1.102

Price (in PhP/kg.) 4.47 4.80 4.40 6.00 5.60 6.20 7.40 7.71 7.63 8.32 8.47 9.20 9.43 8.91 8.56 10.14 9.48 10.85 11.44 13.14 13.84 14.41 15.13 15.78 15.93 14.31 15.52 15.63

Percentage change (%) 21.798 7.383 -8.333 36.364 -6.667 10.714 19.355 4.189 -1.038 9.043 1.803 8.619 2.500 -5.514 -3.928 18.458 -6.509 14.451 5.438 14.860 5.327 4.118 4.997 4.296 0.951 -10.169 8.456 0.709

42

De La Salle University – Dasmariñas

43

De La Salle University – Dasmariñas

Figure 4. Wholesale prices of corn in the Philippines

44

De La Salle University – Dasmariñas Performance of the Forecasting Models This part of the chapter presents the models, ARIMA and AR, used in forecasting the farmgate prices of corn in the Philippines. In the case of ARIMA Model, there is no need to check whether the variable is stationary or not because the Automatic ARIMA Forecasting option in Eviews. This option is capable of detecting right away whether or not the model is stationary or not. If it turns out to be stationary, then the software will directly proceed to forecasting. On the contrary, if it is found out to be non-stationary, then the software will differentiate the variable as many times as possible just to make sure that the variable becomes stationary. For the AR Model, given that this study has only one dependent variable, FPRICE, it’ll be much easier to determine whether the independent variables are significant enough to explain the changes in FPRICE. This will be done through estimating an AR Model using FPRICE, then estimating an Ordinary Least Squares (OLS) Model using the estimated AR Model in order to check the significance of the independent variables to changes in FPRICE. If the independent variables are found to be significant, then the AR Model can proceed directly to forecasting, without the need of differencing. Autoregressive Integrated Moving Average Model. The ARIMA Model in this study covers Tables 4 to 6, and Figures 5 to 7. Here the readers are presented with tables and graphs to inform the readers regarding the significance, and the forecast performance of the said model.

45

De La Salle University – Dasmariñas Table 4 Equation Output Dependent Variable: D(FPRICE,2) Method: ARMA Maximum Likelihood (BFGS) Date: 12/07/17 Time: 14:47 Sample: 1962 1996 Included observations: 35 Convergence achieved after 40 iterations Coefficient covariance computed using outer product of gradients Variable

Coefficient Std. Error

t-Statistic

Prob.

C AR(1) AR(2) MA(1) MA(2) SIGMASQ

0.013853 -1.149650 -0.674533 -0.057885 -0.942108 0.069103

4.161666 -9.648197 -5.369211 -0.000229 -0.000187 0.009928

0.0003 0.0000 0.0000 0.9998 0.9999 0.9921

R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood F-statistic Prob(F-statistic)

0.840283 0.812745 0.288791 2.418608 -6.404218 30.51412 0.000000

Inverted AR Roots -.57+.59i Inverted MA Roots 1.00

0.003329 0.119157 0.125630 252.4355 5035.699 6.960627

Mean dependent var S.D. dependent var Akaike info criterion Schwarz criterion Hannan-Quinn criter. Durbin-Watson stat

-0.004000 0.667371 0.708812 0.975444 0.800853 2.025409

-.57-.59i -.94

This equation output shows the overall performance of the model. The independent variables AR(1) and AR(2) have p-statistics values of less than 5%, which make them significant, while the independent variables MA(1) and MA(2) are insignificant to explain the changes in the dependent variable D(FPRICE,2). Overall, the model is good given the R-squared and F-statistic values. Also, the model is not spurious

46

De La Salle University – Dasmariñas because

the

R-squared

is

less

than

the

Durbin-Watson

Table 5 Summary of the ARIMA Forecasting Model Automatic ARIMA Forecasting Selected dependent variable: D(FPRICE, 2) Date: 11/30/17 Time: 18:35 Sample: 1960 1996 Included observations: 35 Forecast length: 20 Number of estimated ARMA models: 25 Number of non-converged estimations: 0 Selected ARMA model: (2,2)(0,0)

statistic.

47

De La Salle University – Dasmariñas AIC value: 0.670498281469

The summary shows that out of the 25 estimated models of the software, the ARMA model (2,2)(0,0) came out to be the best model. The forecast length of this model is 20 years, from 1997 to 2016. The dependent variable has been differenced twice, suggesting that it requires two differencing processes in order to make the variable stationary. The main basis of the software for choosing the best model is the Akaike information criterion (AIC), an estimator comparing the qualities of a certain model with the other models. The lower the value of AIC, the better. In ARMA model (2,2)(0,0), the AIC value proved to be at the minimum level.

48

De La Salle University – Dasmariñas Actual and Forecast 18 16 14 12 10 8 6 4 2 88

90

92

94

96

98

00

02

Forecast

04

06

08

10

12

14

16

Actual

Figure 5. Forecast graph

It can be seen from the graph the comparison between the actual values of the farmgate corn prices and forecasted values of the farmgate corn prices over the 20-year period. While both of the values exhibit upward trends, it is evident that more fluctuations can be seen in the actual values, while there exists a steady increase, with

49

De La Salle University – Dasmariñas minimal fluctuations in the forecasted values. The upward trend of the forecasted values assumes that in the long-run, farmgate prices of corn will continue to rise steadily, with almost no fluctuations at all. Given that assumption, the producers of corn will most likely earn profit in the future from the high prices, probably because of a good weather, an advanced technology, positive expectations, and many more.

50

De La Salle University – Dasmariñas Forecast Comparison Graph 24 20 16 12 8 4 0 1998

2000

2002

ARMA(3,3)(0,0) ARMA(1,0)(0,0) ARMA(3,0)(0,0) ARMA(1,2)(0,0) ARMA(2,4)(0,0) ARMA(4,1)(0,0) ARMA(3,4)(0,0) ARMA(3,2)(0,0) ARMA(2,2)(0,0)

2004

2006

2008

2010

ARMA(1,4)(0,0) ARMA(1,3)(0,0) ARMA(2,0)(0,0) ARMA(0,3)(0,0) ARMA(0,2)(0,0) ARMA(4,0)(0,0) ARMA(0,4)(0,0) ARMA(2,3)(0,0)

2012

2014

2016

ARMA(0,0)(0,0) ARMA(0,1)(0,0) ARMA(1,1)(0,0) ARMA(3,1)(0,0) ARMA(2,1)(0,0) ARMA(4,4)(0,0) ARMA(4,3)(0,0) ARMA(4,2)(0,0)

Figure 6. Forecast comparison graph

This is the graphical representation of the selection of the best model among the estimated 25 models. Most of these models look identical graphically given minimal

51

De La Salle University – Dasmariñas price fluctuations, but AIC proved to be the most important factor in choosing the best model.

Table 6 ARIMA Criteria Table Model Selection Criteria Table Dependent Variable: D(FPRICE, 2) Date: 11/30/17 Time: 18:35 Sample: 1960 1996 Included observations: 35 Model

LogL

AIC*

BIC

HQ

(2,2) (0,0)

-6.404218

0.670498 0.931728 0.762594

(4,2) (0,0)

-4.482015

0.674704 1.023010 0.797498

(2,3) (0,0)

-6.383138

0.723413 1.028181 0.830858

(3,2) (0,0)

-6.404214

0.724552 1.029320 0.831997

(4,3) (0,0)

-4.417846

0.725289 1.117134 0.863433

(0,4) (0,0)

-7.600198

0.735146 0.996376 0.827242

(3,4) (0,0)

-4.690908

0.740049 1.131894 0.878193

52

De La Salle University – Dasmariñas (4,4) (0,0)

-4.416331

0.779261 1.214644 0.932754

(4,0) (0,0)

-8.517299

0.784719 1.045949 0.876815

(4,1) (0,0)

-8.017047

0.811732 1.116501 0.919177

(2,1) (0,0)

-10.785248

0.853257 1.070948 0.930003

(0,2) (0,0)

-12.404468

0.886728 1.060881 0.948125

(2,4) (0,0)

-8.538437

0.893970 1.242276 1.016764

(3,1) (0,0)

-10.564258

0.895365 1.156595 0.987461

(0,3) (0,0)

-12.397380

0.940399 1.158091 1.017145

(1,2) (0,0)

-12.400924

0.940590 1.158282 1.017337

(1,1) (0,0)

-15.569217

1.057796 1.231949 1.119193

(2,0) (0,0)

-16.152033

1.089299 1.263452 1.150696

(3,0) (0,0)

-16.031148

1.136819 1.354510 1.213565

(0,1) (0,0)

-18.795438

1.178132 1.308747 1.224180

(1,3) (0,0)

-18.600630

1.329764 1.590994 1.421860

(1,0) (0,0)

-28.424926

1.698645 1.829260 1.744693

53

De La Salle University – Dasmariñas (0,0) (0,0)

-35.001225

2.000066 2.087143 2.030765

(1,4) (0,0)

-34.230442

2.228673 2.533441 2.336118

(3,3) (0,0)

-35.043330

2.326667 2.674973 2.449461

54

De La Salle University – Dasmariñas Akaike Information Criteria (top 20 models) 1.2 1.1 1.0 0.9 0.8 0.7

(0,1)(0,0)

(3,0)(0,0)

(2,0)(0,0)

(1,1)(0,0)

(1,2)(0,0)

(0,3)(0,0)

(3,1)(0,0)

(2,4)(0,0)

(0,2)(0,0)

(2,1)(0,0)

(4,1)(0,0)

(4,0)(0,0)

(4,4)(0,0)

(3,4)(0,0)

(0,4)(0,0)

(4,3)(0,0)

(3,2)(0,0)

(2,3)(0,0)

(4,2)(0,0)

(2,2)(0,0)

0.6

Figure 7. ARIMA Criteria Graph

Autoregressive Model. Tables 7 to 9, and Figure 8 fall under the AR Model. This portion presents with some evidences via tables and graph that show the forecast performance of the AR Model.

Table 7

55

De La Salle University – Dasmariñas Summary of the AR Model Vector Autoregression Estimates Date: 11/30/17 Time: 11:55 Sample (adjusted): 1962 2016 Included observations: 55 after Adjustments Standard errors in ( ) & t-statistics in [ ] FPRICE FPRICE(-1)

0.698025 (0.13270) [ 5.26031]

FPRICE(-2)

0.323340 (0.13613) [ 2.37523]

C

0.187513 (0.11130) [ 1.68468]

R-squared

0.982361

Adj. R-squared

0.981683

Sum sq. resids

16.09511

S.E. equation

0.556346

56

De La Salle University – Dasmariñas F-statistic

1448.032

Log likelihood

-44.24913

Akaike AIC

1.718150

Schwarz SC

1.827641

Mean dependent

4.564000

S.D. dependent

4.110710

With 2 lags set as the optimum number of lags, an AR equation has been estimated using FPRICE. There are 2 independent variables, 1 independent variable, with 3 coefficients in this model. So far, judging from the R-squared and F-statistic,

the

model’s

overall

Table 8 Ordinary Least Squares

System: UNTITLED Estimation Method: Least Squares Date: 12/07/17 Time: 11:17 Sample: 1962 2016 Included observations: 55

performance

is

good.

57

De La Salle University – Dasmariñas Total system (balanced) observations 55 Coefficient Std. Error

t-Statistic

Prob.

C(1)

0.698025

0.132696

5.260315

0.0000

C(2)

0.323340

0.136130

2.375234

0.0213

C(3)

0.187513

0.111305

1.684677

0.0980

Determinant residual covariance0.292638

Equation: FPRICE = C(1)*FPRICE(-1) + C(2)*FPRICE(-2) + C(3) Observations: 55 R-squared

0.982361

Mean dependent var 4.564000

Adjusted Rsquared

0.981683

S.D. dependent var

4.110710

S.E. of regression 0.556346

Sum squared resid

16.09511

Durbin-Watson stat 2.046628

Table 8 showed the OLS estimation of FPRICE after its AR estimation, as shown by the 2 lags set. Even if Table 7 has shown that the overall performance is good, it will still not be enough is the independent variables are insignificant. To check whether the independent variables are significant or not, we need to determine the pstatistics first. If the value of p-statistics is less than 5%, then the variable is significant, otherwise it is insignificant. The table above shows that the value of p-statistics of the independent variables, FPRICE(-1) and FPRICE(-2), are 0.00% and 2.13%,

58

De La Salle University – Dasmariñas respectively. Therefore, both FPRICE(-1) and FPRICE(-2) are significant enough, individually,

to

explain

the

changes

in

FPRICE.

59

De La Salle University – Dasmariñas 14 12 10 8 6 4 2 0 60

65

70

75

80 FPRICE

85

90

95

00

05

10

15

FPRICE (VARSCEN)

Figure 8. Forecast graph

FPRICE is the farmgate prices, and is the actual values, while FPRICE (VARSCEN) is the forecasted value using the AR Model. It suggests that the price of corn will continue to rise in the future, followed by some fluctuations. With proper

60

De La Salle University – Dasmariñas decision-making and with proper timing, the farmers will still be able to enjoy themselves in the future. Effectiveness of the Forecasting Models This is the last part of this section where the researcher conducts an evaluation as to what model is better in forecasting the price of corn in the Philippines. As mentioned in Chapter IV, the basis of this study in determining the best model is the RMSE and Theil’s Inequality Coefficient. Table 9, and Figures 9 and 10 both fall under this category.

61

De La Salle University – Dasmariñas

20

16

12

8

4

0 60

65

70

75

80

85

90

95

00

05

10

15

FPRICE FPRICE_ARIMA FPRICE (VARSCEN)

Figure 9. Forecast graph

Figure 9 presents the graphical comparison summary of the performance models of ARIMA and AR models. Even though it seems that the AR Model is quite closer to the actual price than the ARIMA Model, it is not enough to conclude that the former model is better than the latter overall.

62

De La Salle University – Dasmariñas

Table 9 Evaluation of the Forecasting Models Forecast Evaluation Date: 12/01/17 Time: 11:44 Sample: 1960 2016 Included observations: 57 Evaluation sample: 1960 2016 Training sample: 1997 2016 Number of forecasts: 7 Combination tests Null hypothesis: Forecast i includes all information contained in others Forecast

F-stat

F-prob

FPRICE_ARIMA FPRICE_VAR

170.2885 0.0000 1.150590 0.2881

Evaluation statistics Forecast

RMSE

MAE

MAPE

Theil

FPRICE_ARIMA FPRICE_VAR Simple mean Simple median Least-squares Mean square error MSE ranks

1.697087 0.680486 0.949273 0.949273 0.708649 1.475662 0.764124

0.923498 0.348393 0.499466 0.499466 0.372900 0.806033 0.371899

10.33805 4.509829 6.259343 6.259343 4.678298 9.208170 5.014489

0.125223 0.056632 0.074331 0.074331 0.059326 0.110671 0.061053

*Trimmed mean could not be calculated due to insufficient data The table above shows the evaluation of the 2 forecasting models used in this study. When using the Root Mean Squared Error, the lower the value, the better the performance of the model. In the case of Theil’s Inequality Coefficient, if the value is 1,

63

De La Salle University – Dasmariñas the forecasting model is called perfectly fit, which means that the actual and forecasted values are the same. If the value is 0, then the predictive power of the forecasting model is at its worst. Given that those 2 values are almost never seen in real life situation, there exist values in between 0 and 1. The closer the value is to 0, the better the performance of the forecasting model. In the table above, the shaded are is noticeable, stating that the AR Model has lesser RMSE value compared to the ARIMA Model, and has a Theil’s Inequality Coefficient Value of closer to 0. Therefore, in this study, the best model to use in forecasting the farmgate prices of corn in the Philippines is the AR Model.

64

De La Salle University – Dasmariñas

Forecast Comparison Graph 18 17 16 15 14 13 12 11 10 9 06

07

08

09

10

11

FPRICE FPRICE (VARSCEN) Simple median Mean square error

12

13

14

FPRICE_ARIMA Simple mean Least-squares MSE ranks

15

16

65

De La Salle University – Dasmariñas Figure 10. Forecast comparison graph

CHAPTER VI SUMMARY, CONCLUSIONS, AND RECOMMENDATIONS

Summary The research was primarily concerned in determining the model that would best forecast the farmgate prices of corn, for 20 years, in the Philippines. Given the wide range of forecasting models available, the researcher decided to compare the performances of ARIMA and AR models in this univariate analysis. The main motivation for this study is the fact that the corn is second most important crop in the Philippines and almost no studies were made regarding forecasting the price of the said commodity. Without the imported goods from abroad, the Filipinos would turn to the appetizers such as Filipino bread, corn, and mashed potatoes. Of course, the decrease in demand for corns will strongly affect the producers. Thus, this study was made to somehow help the producers in their future decisions. Both historical and evaluative methods were used in this study, which covered the years 1960 to 2015.

66

De La Salle University – Dasmariñas Tables 1 to 3 showed the trend of the prices of corn in the Philippines (farmgate, retail, and wholesale) from years 1960 to 2015. It was found out that regardless of the varying price fluctuations, all these 3 kinds of prices exhibit a continuous upward trend, which led to an assumption that in the long-run, prices of corn will further increase given various factors. Before estimating the equations, it is necessary first to check whether the overall model performs good, and whether the variables are significant or not. There was no problem with estimating the ARIMA model because Eviews is capable of directly proceeding with forecasting whether or not the variable is stationary or not. If the variable is non-stationary, the software automatically differences it – it can be once or twice depending on the situation. All in all, the ARIMA model forecasting went smoothly because the variables and the overall model were thoroughly checked. The ease, however, of estimating an ARIMA model is not the same with estimating a AR model. First, an AR model was estimated, and looking at the R-squared and F-statistic, the overall model is good. Next, an OLS regression is estimated to find out whether the independent variables, FPRICE(-1) and FPRICE(-2) are significant enough individually to explain the changes in the dependent variable, FPRICE, and it turned out that the p-statistics of the 2 independent variables are both less than 5%, which means that the independent variables are significant enough. From there, a forecasting model based on AR can be estimated. Conclusions

67

De La Salle University – Dasmariñas The univariate study was conducted in order to compare the performances of ARIMA and VAR models, and determine whether which of these models performs better in forecasting the farmgate prices of corn in the Philippines. The results show that given the values of R-squared and F-statistic, both models are good overall, and that both models have significant variables with the exception of ARIMA model, which has 2 significant independent variables out of all the 4 independent variables. This is to be expected given that the researcher hypothesized that there is an upward trend in commodity prices, including corn, due to inflation which is experienced by any country. Both models are also non-spurious because their R-squared values exceed their DurbinWatson statistic values. The deciding factor of these two models is when their RMSE and Theil’s Inequality Coefficient Values are checked. When looking into the RMSE, when the value is lower, it means that the predictive capacity of a forecasting model is better. On the other hand, Theil’s Inequality Coefficient might exhibit 3 kinds of values, namely: (1) 0, where the forecast model’s predictive power is at its worst; (2) 1 (perfectly fit), where the actual and forecasted values are the same; and (3) in between 0 and 1, wherein the predictive power of the forecasting model becomes better as the value approaches near 0. The results showed that the RMSE value of AR Model is lower than that of ARIMA Model’s, and the former model’s Theil’s Inequality Coefficient value is closer to 0 than the latter model, which means that the AR model is the best model in forecasting the farmgate prices of corn in the Philippines. Recommendations

68

De La Salle University – Dasmariñas The research was able to show clearly that the AR Model is better than the ARIMA Model. However, some things have to be taken into consideration. First, there is only one variable used in this study, and that is the price of corn itself. This means that this study assumed that corn prices can be assumed, ceteris paribus. For the future researchers, they are recommended to find other variables that might be useful in forecasting corn prices. That will make this study much more realistic when other factors will be included that will greatly make or break the performance of the forecasting model. For the Philippine government, it is recommended that they pay attention to the importance of forecasting the corn prices in the Philippines. This will be very helpful in ensuring the security of corn farmers in the future. The preservation of the farmers is the preservation of the commodity itself. The majority of the Filipinos, especially the adults, know the nutritional benefits of corn, as it even surpasses the health benefits of rice. Given that the Philippine government have pointed out the importance of ensuring the agricultural stability of a nation, it is recommended for them to put more emphasis on the major crops in the Philippines, such as corn. This study is only limited to the corn prices. The future researchers, and the government as well are recommended to gather the data of corn prices to the neighboring ASEAN countries for a more comprehensive study. That study will not only benefit the Filipino people, but also those neighboring ASEAN countries. This study forecasted the annual corn prices in the Philippines from 1960 to 2016. The future researchers are recommended to improve this study by setting the time frame

69

De La Salle University – Dasmariñas to quarterly, semi-annually, or even monthly. It will make this study more comprehensive, and much closer to real-time events, as the researchers will be able to carefully analyze what causes the price fluctuations given the present factors.

REFERENCES

Adeyanju, C. (2014). The Top Factors that Move the Price of Corn. Retrieved from https://www.futuresknowledge.com/news-and-analysis/grains/the-top-factors-thatmove-the-price-of-corn/. Alquist, R. and Kilian, L. (2010). What do we Learn from the Price of Crude Oil Futures? Retrieved from https://deepblue.lib.umich.edu/bitstream/handle/2027.42/75776/1159_ftp.pdf? sequence=1. Alquist, R., Kilian, L. and Vigfusson, R. J. (2011). Forecasting the Price of Oil. Retrieved from https://www.federalreserve.gov/pubs/ifdp/2011/1022/ifdp1022.pdf. Bowman, C. and Husain, A. M. (2004). Forecasting Commodity Prices: Futures Versus Judgment. Retrieved from https://www.imf.org/external/pubs/ft/wp/2004/wp0441.pdf. Chinn, M. D. and Coibion, O. (2010). The Predictive Content of Commodity Futures. Retrieved from http://www.nber.org/papers/w15830.pdf. Dash, M., Solanki, A. and Shobana, T. (2012). A Study on Commodity Market Behaviour, Price Discovery and Its Factors. Retrieved from https://www.researchgate.net/publication/228150427_A_Study_on_Commodity_ Market_Behaviour_Price_Discovery_and_Its_Factors. Dönmez, A. and Magrini, E. (2013). Agricultural Commodity Price Volatility and Its Macroeconomic Determinants. Retrieved from http://ftp.jrc.es/EURdoc/JRC84138.pdf.

70

De La Salle University – Dasmariñas European Central Bank. (2014). Monthly Bulletin July. https://www.ecb.europa.eu/pub/pdf/mobu/mb201407en.pdf.

Retrieved

from

GMA News Online. (2013). Yolanda Deals Central PHL P3.7B in Agri Damage, Supplies Stable – DA. Retrieved from http://www.gmanetwork.com/news/money/content/334930/yolanda-deals-centralphl-p3-7b-in-agri-damage-supplies-stable-da/story/. Good, D. (2008). Many Factors Influencing Corn and Soybean Prices. Retrieved from http://www.farmdoc.illinois.edu/marketing/weekly/html/092208.html. Halonen, D. G. (2016). Forecasting the Spot Price of Corn: Methods and Assessment. Retrieved from http://repository.stcloudstate.edu/cgi/viewcontent.cgi? article=1006&context=econ_etds. Hamjah, M. A. (2014). Forecasting Major Fruit Crops Productions in Bangladesh using Box-Jenkins ARIMA Model. Retrieved from http://iiste.org/Journals/index.php/JEDS/article/viewFile/12578/12897. Hillier, F. S. and Hillier, M. S. (2001). Introduction to Operations Research. Retrieved from https://notendur.hi.is/kth93/3.20.pdf. Jha, G. K. and Sinha, K. (2013). Agricultural Price Forecasting Using Neural Network Model: An Innovative Information Delivery System. Retrieved from http://ageconsearch.umn.edu/bitstream/162150/2/8-GK-Jha.pdf. Jadhav, V., Reddy, B. V. and Gaddi, G. M. (2017). Application of ARIMA Model for Forecasting Agricultural Prices. Retrieved from http://jast.modares.ac.ir/article_16988_5dcee1eff8387db448a1ee6725b3aa72.pdf. Jorgensen, N. (2014). A Comparison of Corn Yield Forecasting Models. Retrieved from http://openprairie.sdstate.edu/cgi/viewcontent.cgi?article=1339&context=etd. Latham Hi-Tech Seeds. (2012). The Importance of Corn. Retrieved http://www.thefieldposition.com/2012/06/the-importance-of-corn/.

from

Li, G., Xu, S. and Li, Z. (2010). Short-Term Price Forecasting For Agro-products Using Artificial Neural Networks. Retrieved from http://ac.elscdn.com/S2210784310000367/1-s2.0-S2210784310000367-main.pdf? _tid=2aecff02-19c1-11e7-b21b00000aacb360&acdnat=1491370489_9c6a3b54b1b21f46a5e3fdd2d4427431.

71

De La Salle University – Dasmariñas Mckenzie, A. M. and Holt, M. T. (1998). Market Efficiency in Agricultural Futures Markets. Retrieved from https://ageconsearch.umn.edu/bitstream/20933/1/spmcke01.pdf. Off The Grid News. (2015). The Importance of Corn in The American Economy. Retrieved from http://www.offthegridnews.com/off-grid-foods/the-importance-ofcorn-in-the-american-economy/.

Okunmadewa, F. (n.d.). Enhancing Farm Gate Prices: The Role of Agricultural Marketing Research. Retrieved from http://ageconsearch.umn.edu/bitstream/147523/2/Prof. %20Okumadewa.pdf. Oster, A. (2015). Commodity Price Declines and Their Economic Impact. Retrieved from https://internationalbanker.com/banking/commodity-price-declines-and-theireconomic-impact/. Padin, M. G. (2016). El Niño Causes Spike in Local Corn Prices. Retrieved from https://businessmirror.com.ph/el-nino-causes-spike-in-local-corn-prices/. Ribeiro, C. O. and Oliveira, S. M. (2011). A Hybrid Commodity Price-forecasting Model Applied to the Sugar–alcohol Sector. Retrieved from http://onlinelibrary.wiley.com/doi/10.1111/j.1467-8489.2011.00534.x/full. Sanders, D. J. and Baker, T. G. (2012). Forecasting Corn and Soybean Basis Using Regime-Switching Models. Retrieved from http://www.farmdoc.illinois.edu/nccc134/conf_2012/pdf/confp16-12.pdf. Sands, S. (2015). How the World Economy Adjusts to Lower Commodity Prices. Retrieved from http://marketrealist.com/2015/10/world-adjusts-lower-commodityprices/. Sims,

C. A. (1980). Macroeconomics and Reality. Retrieved from http://www.ekonometria.wne.uw.edu.pl/uploads/Main/macroeconomics_and_reali ty.pdf.

72

De La Salle University – Dasmariñas Taleghani, S., Aslani, S. and Shiry, S. (2008). Robust Moving Object Detection from a Moving Video Camera Using Neural Network and Kalman Filter. Retrieved from https://www.researchgate.net/publication/220797913_Robust_Moving_Object_De tection_from_a_Moving_Video_Camera_Using_Neural_Network_and_Kalman_ Filter.

Timmer, C. P. (2008). Causes of High Food Prices. Retrieved from https://www.adb.org/sites/default/files/publication/28375/economics-wp128.pdf

Tripathi, R., Nayak, A. K., Raja, R., Shahid, M., Kumar, A., Mohanty, S. et al. (2014). Forecasting Rice Productivity and Production of Odisha, India, Using Autoregressive Integrated Moving Average Models. Retrieved from https://www.hindawi.com/journals/aag/2014/621313/.

Ubilava, D. and Helmers, C. G. (2011). The ENSO Impact on Predicting World Cocoa Prices. Retrieved from http://ageconsearch.umn.edu/bitstream/103528/2/Ubilava_Helmers_2011.pdf.

United States Department of Agriculture Economic Research Service. (2017). Trade. Retrieved from https://www.ers.usda.gov/topics/crops/corn/trade.aspx#world. United States Congress Joint Economic Committee. (2013). The Economic Contribution of America’s Farmers and the Importance of Agricultural Exports. Retrieved from https://www.jec.senate.gov/public/_cache/files/266a0bf3-5142-4545-b806ef9fd78b9c2f/jec-agriculture-report.pdf. Wang, D. and Tomek, W. G. (2004). Commodity Prices and Unit Root Tests. Retrieved from http://www.farmdoc.illinois.edu/nccc134/conf_2004/pdf/confp16-04.pdf.