Final Manuscript - 9th Revision.docx

[Type a 0

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

A Thesis 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 May 2018

1

De La Salle University – Dasmariñas CHAPTER I INTRODUCTION

According to the Agricultural Management Assistance (AMA) (2008) and the Food and Agriculture Organization Corporate Statistical Database (FAOSTAT) (2012), corn, or Zea mays or maize, belongs to the grass family, which originated in Central America and belongs to the top three most grown cereal crops all over the world, together with rice (Oryza sativa) and wheat (Triticum spp.). Its global commercial production in 2010 reached 844.4 million metric tons, at which the harvested land is 161.9 million hectares. 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. According to Gwirtz and Casal (2014), the two basic categories applied in converting maize into other goods for human consumption are dry and wet milling. In the wet milling process, maize is separated into the classes of starch, protein, oil, and fiber. Once the separation has been done, the products are not sold outright instead further industrial processing is required. After the said industrial processes, the four chemical classes can now be sold as sweeteners that are either solid or liquid. The dry milling process, on the other hand, involves particle size reduction of maize, which maintains some of the maize germ and fiber. Again, the maize cannot be sold right away for human consumption, not until some ingredients are further added and some thermal processes such as boiling, drying, frying, and baking are applied.

2

De La Salle University – Dasmariñas This will enable the finished product, though having its nutritional attributes altered changed, to be sold for human consumption.

Even so, there are various ways of

processing corn across various countries. After all, there are so many finished products that can be extracted from corn more than anyone can imagine. In fact, Sailer (2012) stated that corn husks in Mexico 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. 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 current price. Because of this internationally tough competition, farmers plant their corn

3

De La Salle University – Dasmariñas 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 The importance of forecasting commodity prices, corn included, remains a prevailing issue at present. It is important to take a look at some of the techniques and methods used by some researchers in forcasting corn prices, and how effective and accurate these techniques are. 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

4

De La Salle University – Dasmariñas 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 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.

5

De La Salle University – Dasmariñas 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 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. From the studies mentioned above, it can be observed that there are a lot of methods and techniques done in commodity price forecasting. The important issues are the effectiveness and accuracy of these techniques, which are greatly changing over time. This is the exact reason why the researchers never settle on existing forecasting models, instead they either formulate their own models or they improve the existing models that will be enough to account for the present, changing factors. 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.

6

De La Salle University – Dasmariñas 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) 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

7

De La Salle University – Dasmariñas 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 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 monthly farmgate prices of corn in the Philippines from 2007 to 2017?; 2. How do Autoregressive Integrated Moving Average (ARIMA) and AR models perform in forecasting the price of corn in the Philippines?; and 3. How do the predictive data and actual data differ? Objectives of the Study

8

De La Salle University – Dasmariñas 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 monthly farmgate prices of corn in the Philippines from 2007 to 2017; 2. To investigate the performances of ARIMA and AR models in forecasting the price of corn in the Philippines; and 3. To analyze the differences between the predictive data and actual data. 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 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: There is an existence of either trend or seasonality in the corn prices. Teucrium Trading, LLC (2015), in one of its articles, stated that there is a seasonal price patterns in corn. Some of the findings of the study include, but not limited to, the following, namely: (1) the world’s two largest corn-producing countries,

9

De La Salle University – Dasmariñas the US and China, supplies the biggest quantity of corn for approximately 12 weeks, starting from mid-September to mid-December; (2) December is the month with the greatest number of price decreases; and (3) the investors expect great opportunities from both the US and China given the seasonal pattern of their corn prices. H2: ARIMA model provides a better fit than AR model 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 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

10

De La Salle University – Dasmariñas 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 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 long-run. 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

11

De La Salle University – Dasmariñas 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 2007 to 2017 because the data from these years are still available and accessible through the data sources of this study. Definition of Terms 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 determine the gap between the actual and forecasted values.

12

De La Salle University – Dasmariñas Price refers to the farmgate prices of corn in the Philippines.

CHAPTER II REVIEW OF RELATED LITERATURE

This chapter discussed some past researches conducted that are related to forecasting the price of corn in the Philippines. It started with the discussion of previous literature regarding the impact of commodity prices to the economy, thus indicating the economic impact of this study. This is then followed by the discussion of the factors

13

De La Salle University – Dasmariñas that affect the price increases and decreases of commodities. Finally, presented in this chapter are the methods of commodity price forecasting done by various researchers in the past. The related studies done by the researchers in the past enabled the researcher to assess and analyze the studies that have been conducted before, which created a foundation for this study. Furthermore, 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. 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.

14

De La Salle University – Dasmariñas The case of Pakistan clearly displays the relationship between commodity prices and employment. Pakistan, which has a moderate amount of oil production, is a country that relies heavily on oil imports to supply the oil demand most especially by the industrial sector. Ahmad (2013), in his study about the effect of oil prices on unemployment, stated that there has been a very few existing literature regarding the relationship between unemployment and oil process in developing countries. This proposed a challenge because Pakistan is a developing country which is damaged on the increasing oil prices. The results indicated that there is a significant relationship between oil prices and unemployment, but no significant relationship between real interest rate and unemployment. Furthermore, it is suggested that there is a significant relationship between real oil prices in Pakistan are significantly related to the real interest rate. Finally, the study concluded that in the long-run, the oil prices can be used to forecast unemployment rate and real interest rate. 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 economist 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

15

De La Salle University – Dasmariñas 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) Aside from the industrial sector, an agricultural sector also plays a vital role in determining a country’s economic development, most especially in developing countries. Countries such as Liberia and Somalia account agriculture as more than 50% of GDP. 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 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)

16

De La Salle University – Dasmariñas 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. Determinants of commodity prices, which are either short- or long-run in nature, can also be either microeconomic or macroeconomic. Frankel and Rose (2009), explained that agricultural and mineral commodities peaked sharply in 2008. The main causes included the ease in monetary policy due to the low real interest rates, a speculative bubble which arose from expectations, some risks and uncertainties, and strong global growth. This sharp spike which happened in 2008 led the researchers in the study which resulted to the analysis of macroeconomic and microeconomic determinants of 11 individual commodity prices. The results indicated that although the

17

De La Salle University – Dasmariñas macroeconomic determinants: global GDP, and real interest rate both have a positive relationship on real commodity prices, the microeconomic determinants: inventory levels, uncertainty measures, and the spot-futures spread have the strongest effects on real commodity prices. Additionally, there is an existence of bandwagon effect. Similar to the study conducted by Good (2008), in a macroeconomic perspective, the determinants of agricultural commodity price volatility include the following: (1) 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) USDollar 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 Working-T 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

18

De La Salle University – Dasmariñas 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, 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

19

De La Salle University – Dasmariñas 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. 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

20

De La Salle University – Dasmariñas 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 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 long-term forecasting. Finally, the study concluded that regime-switching models

21

De La Salle University – Dasmariñas do not provide real forecasting improvement over ARIMA models despite of statistical significance in favour of the regime-switching models. Aside from ARIMA and VAR models, and any other related models, there is yet another method in forecasting commodity prices that is not commonly used, but according to the literature, this method has been used a couple of times in various fields. The artificial neural network (ANN) is one of the useful tools in machine learning that was once limited to studies pertaining to brain and psychology, but is now used in variety of topics such as business, education, arts, and many more. Kulkarni and Haidar (2009) developed a forecasting model for crude oil price on the basis of ANN and commodity futures prices. Similar to the ARIMA model, there is also an optimal ANN model structure where a researcher needs to be very careful about. The study dealt with pre-processing the spot and futures prices into a couple of months in order to determine the optimal lag. The process resulted to a model with an optimal lag of 13 lags to conduct a short-term forecast of until three days in the future. The forecast accuracy of this model for the first, second, and third days are 78%, 66%, and 53%, conclusively. This model is expected to help in the further understanding of crude oil prices, and to enable the investors to have an effective risk management. 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, this paper was able to contribute additional knowledge in commodity price

22

De La Salle University – Dasmariñas forecasting by formulating a methodology that will forecast the price of corn in the Philippines.

CHAPTER III FRAMEWORKS OF THE STUDY

Theoretical Framework 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

23

De La Salle University – Dasmariñas 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 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

24

De La Salle University – Dasmariñas commonly used model in forecasting commodity prices specified by the three order parameters (p, d, q), 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:

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,

25

De La Salle University – Dasmariñas 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 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

26

De La Salle University – Dasmariñas 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.

Actual Farmgat e Corn Prices

Autore Autore gressiv gressiv ee Integrat Integrat ed ed Moving Moving Averag Averag ee Model Model

Autore Autore gressiv gressiv ee Model Model

Pre dict ed Far mga te Cor n Pric es

27

De La Salle University – Dasmariñas

Figure 1. Overall framework of the research

28

De La Salle University – Dasmariñas

CHAPTER IV METHODOLOGY

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.

29

De La Salle University – Dasmariñas Methods of Data Analysis 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 employed. 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 chosen 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 the case of ARIMA forecasting model, six steps will be followed for a more comprehensive model, namely: (1) examining the data, where patterns and irregularities are checked, outliers and missing values are properly filled in, and converting the prices into logarithmic form to better fir the model; (2) decomposing the data, wherein the seasonal, trend, and cycle components are removed; (3) stationarity testing, where the prices are checked whether they have unit root problem or not; (4) test for autocorrelation and the selection of the best ARIMA model by checking the autocorrelation function (ACF) and partial ACF (PACF); (5) fitting the selected ARIMA model; and (6) evaluation of the performance of ARIMA model. (Dalinina, 2017)

30

De La Salle University – Dasmariñas 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: Ŷ

= µ + Yt-1 or

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: Ŷ

= ϕYt−1 + ut or

Predicted Value of FPRICE = ϕFPRICEt−1 + ut , where: 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. Hyndman (2014), in his study on measuring forecast accuracy, explained three points on why researchers should rely on measures on forecast accuracy

31

De La Salle University – Dasmariñas in forecasting prices, namely: (1) a model that perfectly fits the actual data does not necessarily perform well in forecasting; (2) with sufficient number of parameters, perfect fit can be obtained; and (3) over-fitting the model to a data is not a good idea because it fails to understand the systematic pattern of the data. Furthermore, he enumerated various measures of forecast accuracy which the researches can rely on when forecasting prices. Under scale-dependent errors, the two most commonly used measures are Mean Absolute Error (MAE), and Root Mean Squared Error (RMSE). In percentage errors, there is only Mean Absolute Percentage Error (MAPE). Lastly, under scaled errors, there is only Mean Absolute Squared Error (MASE). There was a study conducted by Hyndman and Koehler (2005) regarding a closer look at the measures of forecast accuracy. It was mentioned in the study that the Theil’s Inequality Coefficient or Theil’s U Statistic is also a good measure of forecast accuracy that is commonly cited in literature reviews. This research primarily focused on using RMSE, Theil’s Inequality Coefficient, and MAPE in checking the forecast errors 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. MAPE was chosen over MAE because in some cases, the exact values don’t clearly tell whether the error has significant difference or not. Thus, percentages are far more trusted than actual values. The formula for RMSE is written as: RMSE = √ mean( e 2i )

,

32

De La Salle University – Dasmariñas where ei is the forecast error expressed as the difference between the actual and predicted values in the forecast sample at period i. The general formula of Theil’s Inequality Coefficient is expressed as:

TH =

√

1 n

n

√

1 n

n

∑ e2 i =1

√∑

∑ y + 1n i =1

where n is the sample size of the study,

2

ŷ

,

n

ŷ

2

i =1

is the predicted value of y, and e is the

equivalence factor, denoting economies of scale. The general formula for MAPE is written as: MAPE =

1 n

n

∑ || Actual Actual |

Forecast |

i =1

CHAPTER V

x 100

33

De La Salle University – Dasmariñas RESULTS AND DISCUSSION

The first part of this section provided the monthly farmgate prices of corn in the Philippines from years 2007 to 2017, which were obtained primarily from PSA. 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. Furthermore, this chapter provided a decision criteria on which model performed better in terms of forecasting the farmgate prices of corn in the Philippines. The last part of this chapter compared the predictive and actual data using both the AR and ARIMA models. 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

34

De La Salle University – Dasmariñas 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. 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. Trend of Farmgate Prices of Corn in the Philippines Tiffany (2009) conducted a study in the US pertaining to the environmental and economic impacts of the usage and production of US corn ethanol. In this paper, the objectives include, but not limited to, the impacts of corn ethanol production to the farmers’ decision-making, and the relationship of corn ethanol production and the prices of corn. Wisner (2014) showed the positive relationship of crude oil and corn oil prices from 2003 to 2014. It led to a conclusion that corn ethanol, being a substitute of crude oil, also led to an increase in the corn prices. In the Philippines, various factors are considered when corn prices. Table 1 shows the monthly farmgate prices of corn in the Philippines from 2007 to 2017 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, coupled with evident price fluctuations. The upward trend, with some price

35

De La Salle University – Dasmariñas fluctuations, can be attributed to a lot of factors. Padin (2016) proved that natural weather, no matter how unlikely the result is, can affect the corn prices in the Philippines. Juliano and Gonzales (n.d.) mentioned technological advancement and policies as important factors in the movement of corn prices. With technological advancements, it becomes much easier for the farmers to produce more corn, and for them to decide the farmgate price of corn. Furthermore, the study also mentioned that policies regarding low taxes on fertilizers, and credit at reasonable rates will definitely enhance corn production.

Table 1

36

De La Salle University – Dasmariñas Monthly Farmgate Prices of Corn, Philippines, 2007-2017 Year

Month

2007

January February March April May June July August September October November December January February March April May June July August September October November December January February March April May

2008

2009

Continued …

Price (in PhP/kg.) 9.28 9.62 10.02 10.31 10.65 10.51 10.26 10.07 9.81 9.93 9.90 9.98 10.24 10.60 10.84 11.04 10.74 11.04 10.79 10.51 10.45 10.84 11.28 11.71 12.91 13.55 12.38 12.07 11.51

Percent change 3.66% 4.16% 2.89% 3.30% -1.31% -2.38% -1.85% -2.58% 1.22% -0.30% 0.81% 2.61% 3.52% 2.26% 1.85% -2.72% 2.79% -2.26% -2.59% -0.57% 3.73% 4.06% 3.81% 10.25% 4.96% -8.63% -2.50% -4.64%

37

De La Salle University – Dasmariñas Year

2010

2011

Continued …

Month June July August September October November December January February March April May June July August September October November December January February March April May June July August September October November

Price (in PhP/kg.) 10.35 10.05 8.99 8.49 8.68 9.49 9.71 10.13 10.82 11.40 11.30 11.02 11.40 11.31 10.96 10.38 10.55 11.19 11.92 12.55 12.85 12.59 11.74 11.66 11.55 11.32 10.89 10.73 11.65 12.69

Percent change -10.08% -2.90% -10.55% -5.56% 2.24% 9.33% 2.32% 4.33% 6.81% 5.36% -0.88% -2.48% 3.45% -0.79% -3.09% -5.29% 1.64% 6.07% 6.52% 5.29% 2.39% -2.02% -6.75% -0.68% -0.94% -1.99% -3.80% -1.47% 8.57% 8.93%

38

De La Salle University – Dasmariñas Year

2012

2013

2014

Continued …

Month

Price (in PhP/kg.)

Percent change

December January February March April May June July August September October November December January February March April May June July August September October November December January February March April

13.01 13.67 13.58 12.63 12.46 12.72 12.46 12.32 12.14 12.01 12.28 12.56 12.58 12.48 11.97 11.87 11.89 12.02 12.08 11.83 11.58 11.74 11.81 11.75 11.83 11.97 11.94 12.21 12.54

2.52% 5.07% -0.66% -7.00% -1.35% 2.09% -2.04% -1.12% -1.46% -1.07% 2.25% 2.28% 0.16% -0.79% -4.09% -0.84% 0.17% 1.09% 0.50% -2.07% -2.11% 1.38% 0.60% -0.51% 0.68% 1.18% -0.25% 2.26% 2.70%

39

De La Salle University – Dasmariñas Year

2015

2016

Continued …

Month

Price (in PhP/kg.)

Percent change

May June July August September October November December January February March April May June July August September October November December January February March April May June July August September October

13.02 13.42 13.22 13.94 12.95 12.84 12.51 12.50 12.28 12.40 12.66 12.67 12.79 12.68 12.65 12.40 11.82 11.60 11.49 11.43 11.67 12.25 12.60 12.51 12.87 13.00 12.98 12.48 11.54 10.96

3.83% 3.07% -1.49% 5.45% -7.10% -0.85% -2.57% -0.08% -1.76% 0.98% 2.10% 0.08% 0.95% -0.86% -0.24% -1.98% -4.68% -1.86% -0.95% -0.52% 2.10% 4.97% 2.86% -0.71% 2.88% 1.01% -0.15% -3.85% -7.53% -5.03%

40

De La Salle University – Dasmariñas Year

Month

November December 2017 January February March April May June July August September October November December Mean Standard Deviation Source: Philippine Statistics Authority

Price (in PhP/kg.)

Percent change

10.86 10.96 11.25 11.45 11.41 11.41 11.53 11.33 11.20 11.21 11.43 11.58 12.07 12.34 11.60 1.08

-0.91% 0.92% 2.65% 1.78% -0.35% 0.00% 1.05% -1.73% -1.15% 0.09% 1.96% 1.31% 4.23% 2.24% 0.28% 3.65%

41

De La Salle University – Dasmariñas Figure 2 shows the graphical representation of the monthly farmgate prices of corn in the Philippines. The highest price, which is 13.94 PhP/kg., occurred on August 2014. The PSA has attributed this to the high corn output on that same year. This good corn production, which led to high farmgate prices, was caused by the increased use of good quality seeds. Despite of the high farmgate prices in 2014, there was a downward trend of prices in 2015 due to the lingering effects of El Niño, as well as the shortage of water supply due to drought. The lowest price, 8.49 PhP/kg., occurred on September 2009. This is primarily because 2009 was considered the deadliest season in the Philippines after decades. According to Corpuz (2010), typhoons Ketsana (Ondoy) and Parma (Pepeng), which both occurred on this year, have devastated a lot of Filipino homes as well as the entire Philippine agriculture, with its effects remaining until 2011. Consequently, some price drops in 2010 can be attributed to the occurrence of El Niño. Though there are a few rare instances where natural calamities don’t necessarily negatively affect agriculture, this is usually an expected outcome especially to the farmers.

42

Ye ar

2011

2012

2013

Farmgate Prices of Corn

2014

2015

2016

2017

De La Salle University – Dasmariñas

8.00

9.00

10.00

11.00

12.00

13.00

14.00

2007

2008

2009

2010

Figure 2. Farmgate prices of corn in the Philippines

Farmgate Price (in PhP/kg.) Theories have stated that commodity prices are bound to increase over time due to inflation. Regarding the price fluctuations, it is not just the frequency of them that

43

De La Salle University – Dasmariñas matters, it should also be important to consider how high or low the price increases and decreases are. In a study conducted by Bäckman and Sumelius (2009), there are numerous factors affecting the price fluctuations of food products according to various literature reviews. The researchers enumerated some common and a few uncommon factors considering both the supply and demand factors. Under demand factors, there are three, namely: (1) energy price, which has a positive relationship with the demand of agricultural products; (2) population growth, which exhibits a positive relationship on the demand for agricultural commodities, and is the least emphasized one among the demand factors; and (3) consumer habits, which positively affects the demand for meat products, and is defined by the study as the increased protein intake of the consumers. There are six enumerated supply factors in the study, namely: (1) input factors, which directly affect the production of agricultural products; (2) weather, which refers to the natural occurrence that humans cannot control such as rain, or even floods, typhoons, and the presence of insects in the farm; (3) climate, which is capable of changing the agricultural production from one place to another; (4) technological development, which exhibits a positive relationship on the supply of goods, that drives increased production; (5) policies and institutions, which means the unstable policies, rules, and regulations regarding the commodity production, that negatively affects the supply of agricultural foods; and (6) prices, which have a direct relationship on the supply of commodities. Performance of the Forecasting Models

44

De La Salle University – Dasmariñas This is the second 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, there will be six steps to be followed under the ARIMA model. Furthermore, mentioned in the Methodology section, are the bases of this study in determining the best model, which are the values of MAPE, RMSE, and Theil’s Inequality Coefficient. Autoregressive Integrated Moving Average Model Step 1: Examining the Data Table 1 and Figure 2 showed the monthly farmgate prices of corn in the Philippines in tabular and graphical forms, respectively. It can be seen that there are complete data from 2007 to 2017, without any missing values. Due to the volatility of prices of corn, outliers are expected to occur, like what is shown above. It is worth noting some comparisons between the logarithmic form of farmgate prices compared to its original form. First, the logarithmic prices show that they have been deflated as compared to the original prices. Second, from the original trend of volatile price prices with heavy price spikes and drops, it has now become a steady, an almost horizontal trend with some weaker price spikes and drops. Lastly, the logarithmic form of farmgate prices has helped in stabilizing the price increases and decreases. Step 2: Decomposing the Data Since the prices of corn are gathered monthly, it is now possible to decompose the logarithmic corn prices in order to eliminate the presence of seasonality, trend, or cycle. The researcher used the additive model, which is done by subtracting the

45

De La Salle University – Dasmariñas seasonal component from the original series, in decomposing the entire series. This resulted in a series having lower prices compared to the original series. Having a deseasonalized series means that the seasonal, cycle, and trend components of the series have been eliminated. The next step required is to determine whether there is a stationarity problem or not. Step 3: Stationarity Testing It is commonly assumed that a time series data is stationary. Stationarity refers to the statistical properties, such as mean, variance and autocorrelation, which do not change over time. Table 2 shows the Augmented Dickey-Fuller unit root test on deseasonalized logarithmic form of farmgate prices of corn (LFPRICESA) at level. The results showed a Philips-Perron test statistics probability value of 0.9%, which is less than 5%. This only means that the null hypothesis can be rejected, and that there is no stationarity problem in the series.

Table 2 Augmented Dickey-Fuller Testing at Level

Augmented Dickey-Fuller test statistic Test critical values: 1% level

t-Statistic

Prob.*

-3.541718 -3.481217

0.0083

46

De La Salle University – Dasmariñas 5% level 10% level

-2.883753 -2.578694

*MacKinnon (1996) one-sided p-values. Null Hypothesis: LFPRICESA has a unit root Exogenous: Constant Lag Length: 1 (Automatic - based on SIC, maxlag=12)

Step 4: Autocorrelation Testing and ARIMA Model Selection Table 3 shows the information on correlogram of LFPRICESA, with 36 lags included. Given below are the values of autocorrelation (AC), partial autocorrelation (PAC), Q-statistics and its p-values. AC does three important things, namely: (1) it shows the correlation of the LFPRICESA values and lags; (2) it determines the appropriate differencing of the best ARIMA model; and (3) it helps to determine the order of the MA(q) component of the ARIMA model. Furthermore, the two lines in the AC plot determine whether the autocorrelation is statistically different from 0 at 5% significance level. If the AC is within the range of those two lines, then it is not

47

De La Salle University – Dasmariñas statistically different from 0. On that note, all AC values at orders 1 to 36 are display high autocorrelation that decays over time. On the other hand, PAC shows the correlation of LFPRICESA values and its lags that is not explained by the past lags. It also determines the order of the AR(p) component of the ARIMA model. Additionally, there are also two lines in the PAC plot that function the same as the two lines in the AC plot. It can be seen that the PAC value at order 1 displays a high autocorrelation that cuts down to the next order. The Q-statistics values help determine whether the values of LFPRICESA have correlation with each other or not. However, it is not easy to manually choose the best ARIMA model to use given the combination of orders to be tested.

Table 3 Correlogram of LFPRICESA

48

De La Salle University – Dasmariñas Date: 05/10/18 Time: 11:45 Sample: 2007M01 2017M12 Included observations: 132 Autocorrelation

Partial Correlation 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36

Step 5: Fitting the ARIMA Model

AC

PAC

0.916 0.790 0.658 0.530 0.426 0.355 0.317 0.309 0.326 0.348 0.348 0.315 0.269 0.224 0.194 0.189 0.195 0.221 0.257 0.288 0.309 0.309 0.282 0.230 0.187 0.164 0.146 0.153 0.177 0.202 0.224 0.224 0.208 0.164 0.115 0.064

0.916 -0.301 -0.055 -0.046 0.068 0.076 0.085 0.088 0.091 0.002 -0.114 -0.111 0.029 0.055 0.108 0.095 -0.019 0.082 0.000 -0.028 0.025 -0.003 -0.051 -0.074 0.102 0.059 -0.065 0.103 0.013 -0.024 -0.002 -0.100 0.046 -0.102 0.048 -0.081

Q-Stat 113.26 198.25 257.62 296.44 321.76 339.45 353.67 367.31 382.64 400.24 417.92 432.56 443.35 450.86 456.57 462.02 467.86 475.46 485.83 498.91 514.10 529.50 542.40 551.06 556.85 561.32 564.91 568.88 574.28 581.36 590.11 598.96 606.70 611.58 614.01 614.76

Prob 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

49

De La Salle University – Dasmariñas Table 4 shows the ARIMA model summary and criteria table for LFPRICESA. In the criteria table, there are values of log-likelihood (LogL), Akaike Information Criterion (AIC), Schwarz Criterion (BIC), and the Hannan-Quinn Criterion (HQ). These three criteria are only used to determine the appropriate number of AR(p) and MA(q) terms. The criteria table below chose ARIMA Model (3,3)(0,0) as the best model given the lowest value of AIC, as well as the remaining criteria. This suggests that the ARIMA Model, with the terms AR(3) and MA(3) without differencing, is the best model to forecast the price of corn in the Philippines. However, this model will still undergo further testing to see whether it satisfies the certain conditions to be met in forecasting.

Table 4 ARIMA Criteria Table and Summary

50

De La Salle University – Dasmariñas Model

LogL

AIC*

BIC

HQ

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

292.026579 291.828095 292.459581 292.287075 289.608300 292.463764 287.438595 283.022405 283.297938 283.286081 285.083540 283.999839 283.352419 283.305837 284.282916 284.093237 283.057686 281.812034 280.017613 269.112200 267.913519 254.651094 235.022074 196.542821 126.106301

-4.303433 -4.300426 -4.294842 -4.292228 -4.281944 -4.279754 -4.233918 -4.227612 -4.216635 -4.216456 -4.213387 -4.212119 -4.202309 -4.201604 -4.201256 -4.198382 -4.197844 -4.194122 -4.182085 -4.032003 -3.968387 -3.782592 -3.500334 -2.932467 -1.880398

-4.128718 -4.125710 -4.098287 -4.095674 -4.129068 -4.061360 -4.059203 -4.140255 -4.107438 -4.107259 -4.060511 -4.081082 -4.071273 -4.070567 -4.048380 -4.045507 -4.066807 -4.084925 -4.094727 -3.966485 -3.837350 -3.673395 -3.412977 -2.866949 -1.836720

-4.232437 -4.229429 -4.214971 -4.212358 -4.219822 -4.191009 -4.162922 -4.192114 -4.172263 -4.172083 -4.151265 -4.158872 -4.149062 -4.148356 -4.139135 -4.136261 -4.144597 -4.149749 -4.146587 -4.005379 -3.915139 -3.738220 -3.464836 -2.905843 -1.862649

Model Selection Criteria Table Dependent Variable: LFPRICESA Date: 05/10/18 Time: 12:05 Sample: 2007M01 2017M12 Included observations: 132

51

De La Salle University – Dasmariñas

Table 5 shows the correlogram table of LFPRICE, after differencing once. As opposed to the results of Table 3, this table shows that there is no significant autocorrelation on both the AC and PAC plots, as shown by the values being within the limit of the two lines in their respective plots. This means that there is no presence of cuts or decays, and that the values under AC and PAC plots are all statistically the same with 0 at 5% significance level.

52

De La Salle University – Dasmariñas

Table 5 Correlogram of D(LFPRICESA)

53

De La Salle University – Dasmariñas Date: 05/10/18 Time: 11:51 Sample: 2007M01 2017M12 Included observations: 131 Autocorrelation

Partial Correlation

AC 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36

Table 6 ARIMA Criteria Table and Summary

0.397 0.153 0.029 -0.125 -0.282 -0.322 -0.261 -0.195 -0.037 0.151 0.265 0.106 0.022 -0.112 -0.203 -0.179 -0.127 -0.108 0.061 0.156 0.225 0.299 0.276 0.039 -0.129 -0.225 -0.273 -0.251 -0.187 -0.007 0.027 0.101 0.269 0.178 0.025 -0.101

PAC 0.397 -0.005 -0.036 -0.147 -0.217 -0.157 -0.080 -0.077 0.039 0.114 0.111 -0.173 -0.124 -0.201 -0.130 0.019 0.049 -0.029 0.101 -0.028 -0.016 0.087 0.131 -0.092 -0.056 -0.108 -0.100 -0.024 -0.008 0.074 -0.046 -0.089 0.065 -0.078 -0.036 -0.085

Q-Stat 21.124 24.304 24.420 26.551 37.556 52.008 61.571 66.954 67.144 70.424 80.630 82.278 82.348 84.213 90.420 95.248 97.726 99.520 100.11 103.94 111.96 126.28 138.53 138.77 141.50 149.91 162.36 173.00 178.95 178.96 179.09 180.87 193.74 199.45 199.57 201.45

Prob 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

54

De La Salle University – Dasmariñas Model

LogL

AIC*

BIC

HQ

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

289.712029 291.260057 289.787562 289.777055 289.775437 290.940821 289.796609 284.959874 283.656726 285.496100 284.467090 284.961370 283.396888 282.002414 278.842711 279.995169 278.843181 278.843111 278.548822 277.415747 280.357794 279.078541 278.923165 279.408440 267.671571

-4.283516 -4.276668 -4.269509 -4.269349 -4.269325 -4.256679 -4.254494 -4.226665 -4.222072 -4.219638 -4.219198 -4.211536 -4.187832 -4.181855 -4.179435 -4.166593 -4.164291 -4.164290 -4.159831 -4.157814 -4.156936 -4.152705 -4.150351 -4.142552 -4.025327

-4.130640 -4.080113 -4.094793 -4.094634 -4.094610 -4.038285 -4.057939 -4.095628 -4.112875 -4.066762 -4.088162 -4.058660 -4.034956 -4.050818 -4.113917 -4.057396 -4.076933 -4.076932 -4.072473 -4.092296 -4.025900 -4.043508 -4.041154 -4.011516 -3.981648

-4.221394 -4.196797 -4.198512 -4.198353 -4.198329 -4.167934 -4.174623 -4.173418 -4.177699 -4.157516 -4.165951 -4.149414 -4.125710 -4.128608 -4.152811 -4.122221 -4.128792 -4.128791 -4.124332 -4.131191 -4.103689 -4.108333 -4.105978 -4.089305 -4.007578

Model Selection Criteria Table Dependent Variable: D(LFPRICESA) Date: 05/10/18 Time: 11:58 Sample: 2007M01 2017M12 Included observations: 131

55

De La Salle University – Dasmariñas

Table 6 showed the ARIMA criteria table for D(LFPRICESA). As previously shown in Table 5, there is no significant autocorrelation in both the AC and PAC plots. The ARIMA terms, AR(p) and MA(q), are given the values 2 and 3, respectively. This means that given the AIC, BIC, and HQ values, the ARIMA model chosen was (2,3) (0,0). It can be concluded that the ARIMA Model (2,3)(0,0) is the best model in forecasting the price of corn in the Philippines at first difference. Step 6: Evaluation of the ARIMA Model The ARIMA model is expressed as the form ARIMA (p,d,q), where p is the autoregression order, d is the differencing level, and q is the moving average order. There are four models that are chosen for comparison based on steps 4 to 5, satisfying the following conditions, namely: (1) the ARMA model (3,3) using LFPRICESA at level; (2) the ARMA model (3,3) at first difference; (3) the ARMA model (2,3) using LFPRICESA at level; and (4) the ARMA model (2,3) using LFPRICESA at first difference. Figures 3 to 6 show the evaluation of the ARIMA models (3,0,3), (3,1,3), (2,0,3), and (2,1,3), respectively. The values of RMSE, MAE, MAPE, and AIC were checked to compare the performances of the four models. The model that displays the lowest value of the four criteria will be the best ARIMA model. After estimating the models individually, the next step is proceed directly to forecasting, where the values of

56

De La Salle University – Dasmariñas RMSE, MAE, and MAPE are shown. Tables 4 and 6 show the AIC values of the four models. The results show that ARIMA model (3,1,3) is the best ARIMA model for comparison against AR model in forecasting the price of corn in the Philippines on the bases of the four mentioned values.

57

De La Salle University – Dasmariñas

2.7

Forecast: A Actual: LFPRICESA Forecast sample: 2007M01 2017M12 Adjusted sample: 2007M03 2017M12 Included observations: 130 Root Mean Squared Error 0.084588 Mean Absolute Error 0.067504 Mean Abs. Percent Error 2.757657 Theil Inequality Coefficient 0.017317 Bias Proportion 0.042072 Variance Proportion 0.522860 Covariance Proportion 0.435068

2.6 2.5 2.4 2.3 2.2 2.1 07

08

09

10

11

12 A

13

14

15

16

17

± 2 S.E.

Figure 3. Forecast evaluation graph ARIMA model (2,0,3)

58

De La Salle University – Dasmariñas

.08

Forecast: B Actual: D(LFPRICESA) Forecast sample: 2007M01 2017M12 Adjusted sample: 2007M04 2017M12 Included observations: 129 Root Mean Squared Error 0.029642 Mean Absolute Error 0.022025 Mean Abs. Percent Error 140.2943 Theil Inequality Coefficient 0.806206 Bias Proportion 0.000151 Variance Proportion 0.797569 Covariance Proportion 0.202280

.06 .04 .02 .00 -.02 -.04 -.06 -.08 07

08

09

10

11

12 B

13

14

15

16

17

± 2 S.E.

Figure 4. Forecast evaluation graph ARIMA model (2,1,3)

59

De La Salle University – Dasmariñas

2.7

Forecast: CC Actual: LFPRICESA Forecast sample: 2007M01 2017M12 Adjusted sample: 2007M04 2017M12 Included observations: 129 Root Mean Squared Error 0.080513 Mean Absolute Error 0.064124 Mean Abs. Percent Error 2.598020 Theil Inequality Coefficient 0.016524 Bias Proportion 0.161954 Variance Proportion 0.537861 Covariance Proportion 0.300185

2.6 2.5 2.4 2.3 2.2 2.1 07

08

09

10

11

12 CC

13

14

15

16

17

± 2 S.E.

Figure 5. Forecast evaluation graph ARIMA model (3,0,3)

60

De La Salle University – Dasmariñas

.08

Forecast: DD Actual: D(LFPRICESA) Forecast sample: 2007M01 2017M12 Adjusted sample: 2007M05 2017M12 Included observations: 128 Root Mean Squared Error 0.029114 Mean Absolute Error 0.021480 Mean Abs. Percent Error 148.0214 Theil Inequality Coefficient 0.772748 Bias Proportion 0.000330 Variance Proportion 0.757820 Covariance Proportion 0.241850

.06 .04 .02 .00 -.02 -.04 -.06 -.08 07

08

09

10

11

12 DD

13

14

15

16

17

± 2 S.E.

Figure 6. Forecast evaluation graph ARIMA model (3,1,3)

61

De La Salle University – Dasmariñas

Autoregressive Model Table 7 Autoregression Estimates

LFPRICESA LFPRICESA(-1)

1.329899 (0.07990) [ 16.6437]

LFPRICESA(-2)

-0.425318 (0.07803) [-5.45061]

C

0.234847 (0.06597) [ 3.55966]

R-squared Adj. R-squared Sum sq. resids S.E. equation F-statistic Log likelihood Akaike AIC Schwarz SC Mean dependent S.D. dependent

0.904805 0.903306 0.098570 0.027859 603.5534 282.5321 -4.300495 -4.234321 2.450104 0.089592

Vector Autoregression Estimates Date: 05/10/18 Time: 12:51 Sample (adjusted): 2007M03 2017M12 Included observations: 130 after adjustments Standard errors in ( ) & t-statistics in [ ]

62

De La Salle University – Dasmariñas The Autoregression estimates above showed the summary of the estimated AR model. The right-hand side of the variable, LFPRICESA, shows the values of coefficient, standard errors, and t-statistics. From here, there are four things that can be concluded, namely: (1) the coefficient value of LFPRICESA at lag one of 1.329899 means that an increase of one percent in LFPRICESA(-1) will lead to an increase in the value of LFPRICESA by 1.329899%; (2) likewise, the coefficient value of LFPRICESA at lag two of -0.425318 means that as LFPRICESA(-2) increases by one percent, LFPRICESA decreases by 0.425318%; (3) the standard errors calculate the statistical reliability of the estimated coefficients, which means that the larger the value of the standard errors, the greater the problem is observed in the estimates; and (4) to interpret the t-statistic values, which is the coefficient value divided by the standard error value, it is important to take a look at the value of probability of t-statistics. With 2 lags set as the optimum number of lags, an AR equation has been estimated using LFPRICESA. There are 2 independent variables, 1 independent variable, with 3 coefficients in this model. So far, judging from the R-squared and Fstatistic, the model’s overall performance is good. However, as mentioned above, not enough information is shown whether the independent values are significant enough to explain

the

dependent

variable.

63

De La Salle University – Dasmariñas

Table 8 Ordinary Least Squares Method

Coefficient

Std. Error

t-Statistic

Prob.

1.329899 -0.425318 0.234847

0.079904 0.078031 0.065975

16.64375 -5.450610 3.559656

0.0000 0.0000 0.0005

Determinant residual covariance

0.000758

C(1) C(2) C(3)

Equation: LFPRICESA = C(1)*LFPRICESA(-1) + C(2)*LFPRICESA(-2) + C(3) Observations: 130 R-squared 0.904805 Mean dependent var 2.450104 Adjusted R-squared 0.903306 S.D. dependent var 0.089592 S.E. of regression 0.027859 Sum squared resid 0.098570 Durbin-Watson stat 2.049812 System: UNTITLED

64

De La Salle University – Dasmariñas Estimation Method: Least Squares Date: 05/10/18 Time: 12:58 Sample: 2007M03 2017M12 Included observations: 130 Total system (balanced) observations 130 Table 8 showed the OLS estimation of LFPRICESA after its AR estimation, as shown by the 2 lags set. Even if Table 5 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 probability of t-statistics first. If the value is less than 5%, then the variable is significant, otherwise it is insignificant. The table above shows the coefficients of the OLS model as well as their respective t-statistics probability values. The equation written in the lower portion of the OLS estimation summary shows the corresponding independent variables the coefficients are attributed into. The value of t-statistics probability of LFPRICESA(-1) is found to be less than 5%, which implies its statistical significance. In the same manner, the probability of t-statistics value of LFPRICESA(2) of 0.00% shows that it is statistically significant at 5%. Overall, both LFPRICESA(1) and LFPRICESA(-2) are significant enough to explain the changes in LFPRICESA. The model proves that it is not spurious because the R-squared value is less than the value of Durbin-Watson statistics. Therefore, the farmers can rely on this model when making their decisions whether or not to produce more corn in the future.

Table 9

65

De La Salle University – Dasmariñas Forecast Evaluation (AR Model)

Variable LFPRICESA

Inc. obs.

RMSE

MAE

MAPE

Theil

132

0.081858

0.064553

2.628879

0.016690

RMSE: Root Mean Square Error MAE: Mean Absolute Error MAPE: Mean Absolute Percentage Error Forecast Evaluation Date: 05/10/18 Time: 13:00 Sample: 2007M01 2017M12 Included observations: 132 Figures 3 to 6 evaluated the performance of the top 4 ARIMA models, and selected among the best of them, while Table 9 showed the evaluation of the AR model. 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, 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 1, the better the performance of the forecasting model. On the basis of MAPE, the smaller the value is, the lesser the forecast error becomes in percentage terms. The ARIMA model (3,1,3) has an RMSE value of 0.029114, while the AR model has 0.081858. On the basis of RMSE, clearly, the ARIMA model performed better. On Theil’s Inequality Coefficient,

66

De La Salle University – Dasmariñas the ARIMA model (3,1,3) performed better given its value of 0.772748, as opposed to the AR model’s value of 0.016690. Finally, the MAPE value of the ARIMA (3,1,3) model is 14,802.14%, while the AR model’s MAPE value is 262.8879%. The ARIMA model (3,1,3) failed proved to perform better than the AR model on the basis of MAPE. Therefore, on the basis of RMSE, and Theil’s Inequality Coefficient values, the best model to use in forecasting the farmgate prices of corn in the Philippines is the ARIMA (3,1,3)

Model.

67

De La Salle University – Dasmariñas 2.7 2.6 2.5 2.4 2.3 2.2 2.1 07

08

09

10

11

12

13

LFPRICESA_ARIMA

14

15

16

17

18

LFPRICESA

Figure 7. Actual and forecasted prices using ARIMA

Figure 7 showed the graphical representation of the actual monthly prices from 2007 to 2017, as well as the forecasted prices on 2018, using ARIMA (3,1,3) model. This graph shows that in 2018 will start with a few months of increasing prices, which will be followed by some months of decreasing prices. After the months of decreasing prices, the remaining months will be coupled with increasing prices. Overall, it is safe

68

De La Salle University – Dasmariñas to assume that the corn farmers will earn profit if they are to start selling corn at present.

Table 10 Actual and Forecasted Farmgate Prices of Corn Using ARIMA (3,1,3) Model Year

Month

2007

January February March

Actual Price (in PhP/kg.) 2.21 2.22 2.27

Forecasted Price (in PhP/kg.) -

69

De La Salle University – Dasmariñas

2008

2009

April May June July August September October November December January February March April May June July August September October November December January February March April May

2.31 2.34 2.34 2.32 2.33 2.34 2.34 2.32 2.31 2.31 2.32 2.35 2.38 2.35 2.38 2.37 2.37 2.41 2.43 2.45 2.47 2.54 2.57 2.48 2.47 2.42

-

Month

Actual Price (in PhP/kg.) 2.32 2.30 2.22 2.20

Forecasted Price (in PhP/kg.) -

Continued … Year

June July August September

70

De La Salle University – Dasmariñas

2010

2011

October November December January February March April May June July August September October November December January February March April May June July August September October November

2.21 2.28 2.28 2.30 2.34 2.40 2.40 2.38 2.42 2.42 2.42 2.40 2.40 2.44 2.49 2.51 2.51 2.50 2.44 2.43 2.43 2.42 2.41 2.43 2.50 2.57

-

Month

Actual Price (in PhP/kg.)

Forecasted Price (in PhP/kg.)

December January February March

2.57 2.59 2.57 2.50

-

Continued … Year

2012

71

De La Salle University – Dasmariñas

2013

2014

April May June July August September October November December January February March April May June July August September October November December January February March April May

2.50 2.52 2.51 2.51 2.52 2.55 2.56 2.56 2.54 2.50 2.44 2.44 2.45 2.46 2.47 2.46 2.47 2.52 2.52 2.49 2.48 2.46 2.44 2.47 2.50 2.54

-

Month

Actual Price (in PhP/kg.)

Forecasted Price (in PhP/kg.)

June July August September

2.58 2.58 2.66 2.62

-

Continued … Year

72

De La Salle University – Dasmariñas

2015

2016

October November December January February March April May June July August September October November December January February March April May June July August September October November

2.60 2.55 2.53 2.49 2.48 2.51 2.51 2.52 2.52 2.53 2.54 2.53 2.50 2.47 2.45 2.44 2.47 2.50 2.50 2.53 2.55 2.56 2.55 2.51 2.44 2.41

-

Month

Actual Price (in PhP/kg.)

Forecasted Price (in PhP/kg.)

December January February March

2.40 2.40 2.40 2.40

-

Continued … Year

2017

73

De La Salle University – Dasmariñas

2018

April May June July August September October November December January February March April May June July August September October November December

2.41 2.42 2.41 2.41 2.44 2.50 2.50 2.52 2.52 -

2.54 2.55 2.55 2.55 2.53 2.52 2.51 2.51 2.51 2.53 2.55 2.56

74

De La Salle University – Dasmariñas 2.7 2.6 2.5 2.4 2.3 2.2 2.1 07

08

09

10

11

12

LFPRICESA_AR

13

14

15

16

17

18

LFPRICESA

Figure 8. Actual and forecasted prices using AR

The graph above showed the actual and forecasted prices using the AR model. Contrary to what is shown by the results of using ARIMA model, the AR model predicted that prices will continuously increase throughout the entire 2018. This only means that if the corn farmers were to start selling corn at present, they will earn profit by the end of 2018. This graphical result, as far as earning profit is concerned, agrees

75

De La Salle University – Dasmariñas that ARIMA (3,1,3) model is a better model in forecasting the price of corn in the Philippines.

Table 11 Actual and Forecasted Farmgate Prices of Corn Using AR Model Year

Month

2007

January February March

Actual Price (in PhP/kg.) 2.21 2.22 2.27

Forecasted Price (in PhP/kg.) -

76

De La Salle University – Dasmariñas

2008

2009

April May June July August September October November December January February March April May June July August September October November December January February March April May

2.31 2.34 2.34 2.32 2.33 2.34 2.34 2.32 2.31 2.31 2.32 2.35 2.38 2.35 2.38 2.37 2.37 2.41 2.43 2.45 2.47 2.54 2.57 2.48 2.47 2.42

-

Month

Actual Price (in PhP/kg.) 2.32 2.30 2.22 2.20

Forecasted Price (in PhP/kg.) -

Continued … Year

June July August September

77

De La Salle University – Dasmariñas

2010

2011

October November December January February March April May June July August September October November December January February March April May June July August September October November

2.21 2.28 2.28 2.30 2.34 2.40 2.40 2.38 2.42 2.42 2.42 2.40 2.40 2.44 2.49 2.51 2.51 2.50 2.44 2.43 2.43 2.42 2.41 2.43 2.50 2.57

-

Month

Actual Price (in PhP/kg.)

Forecasted Price (in PhP/kg.)

December January February March

2.57 2.59 2.57 2.50

-

Continued … Year

2012

78

De La Salle University – Dasmariñas

2013

2014

April May June July August September October November December January February March April May June July August September October November December January February March April May

2.50 2.52 2.51 2.51 2.52 2.55 2.56 2.56 2.54 2.50 2.44 2.44 2.45 2.46 2.47 2.46 2.47 2.52 2.52 2.49 2.48 2.46 2.44 2.47 2.50 2.54

-

Month

Actual Price (in PhP/kg.)

Forecasted Price (in PhP/kg.)

June July August September

2.58 2.58 2.66 2.62

-

Continued … Year

79

De La Salle University – Dasmariñas

2015

2016

October November December January February March April May June July August September October November December January February March April May June July August September October November

2.60 2.55 2.53 2.49 2.48 2.51 2.51 2.52 2.52 2.53 2.54 2.53 2.50 2.47 2.45 2.44 2.47 2.50 2.50 2.53 2.55 2.56 2.55 2.51 2.44 2.41

-

Month

Actual Price (in PhP/kg.)

Forecasted Price (in PhP/kg.)

December January February March

2.40 2.40 2.40 2.40

-

Continued … Year

2017

80

De La Salle University – Dasmariñas

2018

April May June July August September October November December January February March April May June July August September October November December

2.41 2.42 2.41 2.41 2.44 2.50 2.50 2.52 2.52 -

2.52 2.53 2.53 2.53 2.53 2.53 2.54 2.54 2.54 2.54 2.54 2.55

Differences Between the Actual Data and Predictive Data This last part of the chapter discussed about the gap between the actual data and the predictive data using AR and ARIMA (3,1,3) models. The previous discussions showed how to estimate the forecasting models ARIMA (3,1,3) and AR, and how accurate these two models are in forecasting future prices. Provided in this section is a discussion on which model has the closer predictive data to the actual data.

81

De La Salle University – Dasmariñas The results indicated that there is a huge gap between the predictive data using ARIMA (3,1,3) model and the actual data. While the de-seasonalized logarithmic data had a lowest value of 2.20 and a highest value of 2.66, the predictive value revolved around 0.0. This is not a good indication for the farmers to plant corn if they are to adapt this model in their production decisions. On the other hand, there is a considerable gap between the predictive data using AR model and the actual data. The predictive value had a starting value of 2.29 in 2007, and 2.52 in 2017. The farmers will most likely earn profit from producing corn at present, ceteris paribus.

Table 10 ARIMA Model (3,1,3) OLS Estimation

Variable

Coefficient

Std. Error

t-Statistic

Prob.

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

0.002216 1.648502 -0.983811 -1.396783 0.658443

0.003366 0.023556 0.023713 0.096188 0.137901

0.658180 69.98140 -41.48812 -14.52142 4.774740

0.5116 0.0000 0.0000 0.0000 0.0000

82

De La Salle University – Dasmariñas MA(3) SIGMASQ

0.186109 0.000688

0.090752 8.31E-05

2.050744 8.285646

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

0.300043 0.266174 0.026967 0.090175 289.7120 8.858944 0.000000

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

Inverted AR Roots Inverted MA Roots

.82-.55i .79-.58i

.82+.55i .79+.58i

0.0424 0.0000 0.002399 0.031480 -4.316214 -4.162578 -4.253785 2.003855

-.19

Dependent Variable: D(LFPRICESA) Method: ARMA Maximum Likelihood (OPG - BHHH) Date: 05/10/18 Time: 13:08 Sample: 2007M02 2017M12 Included observations: 131 Convergence achieved after 44 iterations Coefficient covariance computed using outer product of gradients Table 10 showed the OLS estimation of ARIMA model with AR(3), I(1) and MA(3) terms. This is the exact ARIMA model used to check the gap between the actual and predictive prices. The R-squared is the measurement of how successful the regression model is. In this model, it means the fraction of the variance of D(LFPRICESA) that can be explained by the independent variables. If it shows a value of 100%, it means that the regression fits perfectly, which is not practical in real life situations. In this model the R-squared value is 30.0043%, and this means that the independent variables in this model explain 30.0043% of the changes in the dependent variable, D(LFPRICESA).

83

De La Salle University – Dasmariñas

3.0 2.5 2.0 1.5 1.0 0.5 0.0 -0.5 07

08

09

10

11

LFPRICESA

12

13

14

15

LFPRICESA_ARIMA

Figure 9. Forecast graph

16

17

84

De La Salle University – Dasmariñas

It can be seen from the graph the comparison between the actual and forecasted values of LFPRICESA. There exists a considerable gap between the actual value and the forecasted value. The actual values exhibited a nearly horizontal trend, with few price fluctuations. On the contrary, the forecasted values exhibited a somewhat volatile, yet neither increasing nor decreasing, trend. Overall, the assumption is that in the longrun, due to the forecasted prices being low, it is not advisable for the corn farmers to produce corn at present as they will only be suffering huge losses in doing so.

85

De La Salle University – Dasmariñas

2.7 2.6 2.5 2.4 2.3 2.2 2.1 07

08

09

10

11

LFPRICESA

12

13

14

15

LFPRICESA_AR

Figure 10. Forecast graph

16

17

86

De La Salle University – Dasmariñas

The graph above showed that both the actual and forecasted values exhibit upward trends. The obvious difference is that the actual values have a lot of price fluctuations, while the forecasted values showed a steady increasing trend at an increasing rate. This only suggests that in the long-run, with the right timing, the farmers somehow might be able to enjoy earning a profit from producing corn at present. However, it can be observed that the profit the farmers will gain in the future will not be enough to cover the losses that they will most likely incur.

87

De La Salle University – Dasmariñas

3.0 2.5 2.0 1.5 1.0 0.5 0.0 -0.5 07

08

09

10

11

12

13

14

LFPRICESA LFPRICESA_AR LFPRICESA_ARIMA

Figure 11. Forecast comparison graph

15

16

17

88

De La Salle University – Dasmariñas Figure 11 presented the graphical comparison summary of the performance models of ARIMA and AR models. In the graphs presented above, it was proven that the AR model would most likely be more helpful to the corn farmers as opposed the ARIMA model. However, the AR model only suggests that the farmers will be able to earn much profit if they were to rely on this model in their production decisions. If there are any other models that will give the farmers the opportunity to earn more and lose less, then it will be more strongly preferred.

CHAPTER VI

89

De La Salle University – Dasmariñas SUMMARY, CONCLUSIONS, AND RECOMMENDATIONS

Summary The research was primarily concerned in determining the model that would best forecast the farmgate prices of corn in the Philippines. The researcher gathered the monthly farmgate corn prices directly from the PSA. 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 2007 to 2017. Tables 1 showed the trend of the farmgate prices of corn in the Philippines from 2007 to 2017. It was found out that regardless of the varying price fluctuations, the farmgate 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. In the case of ARIMA model, there were six steps followed in order to provide a good ARIMA

90

De La Salle University – Dasmariñas model. Overall, it was needed to convert the prices into logarithmic form, and so far there was no need to difference the dependent variable. After the ARIMA model, the AR model was estimated first, 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, LFPRICESA(-1) and LFPRICESA(-2) are significant enough individually to explain the changes in the dependent variable, LFPRICE, and it turned out that both LFPRICESA(-1) and LFPRICESA(-2) are statistically significant enough to explain the changes in LFPRICE given the probability of t-statistics values. From there, a forecasting model based on AR can be estimated. Conclusions The univariate study was conducted in order to compare the performances of ARIMA and AR models, and to 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, the AR performed well and that both the independent variables are significant. The AR proved to be non-spurious because its R-squared value exceeds its Durbin-Watson statistic value. The ARIMA (3,1,3) model, however, had an R-squared value of 30.2949%, and this is most likely due to the lack of independent variables accounting for many different factors that can possibly affect D(LFPRICESA). The deciding factor of these two models is when their MAPE, 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

91

De La Salle University – Dasmariñas 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 1. Lastly, the lower the value of MAPE is, the lesser the forecast error becomes. The results showed that the RMSE value of ARIMA Model is lower than that of AR Model’s, the former model’s Theil’s Inequality Coefficient value is closer to 1 than the latter model, and the latter model’s MAPE value is way lower than that of the former model’s. In this study, satisfying two among the three criteria, the ARIMA model proved to be the best model to use in forecasting the farmgate prices of corn in the Philippines. Recommendations The research was able to show clearly that the ARIMA Model is better than the AR Model. However, some things have to be taken into consideration. First, the ARIMA model was only able to outperform the AR model in two out of three criteria. This only suggests that not all cases will the ARIMA model perform better against the AR model. This is a good study which is not thoroughly looked up to in the Philippines. Given that situation, it is advised for the future researchers to work on other models, aside from ARIMA and AR, which can be used to compare performances with the two models in this study.

92

De La Salle University – Dasmariñas Next, 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. This also became a problem because the ARIMA (3,1,3) forecasting model had a low R-squared value. So adding more independent variables will do much help. 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. This study is only limited to the corn prices in the Philippines. The future researchers, and the government as well are recommended to gather the data of corn prices, assuming they are available and accessible, 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.

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-factorsthat-move-the-price-of-corn/.

93

De La Salle University – Dasmariñas Ahmad, F. (2013). The Effect of Oil Prices on Unemployment: Evidence from Pakistan. Retrieved from http://www.berjournal.com/wp-content/plugins/downloadsmanager/upload/BERJ%204(1)13%20Article%203%20pp.43-57.pdf. 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. Agricultural Management Assistance. (2008). http://eol.org/pages/1115259/details.

Zea

mays.

Retrieved

from

Bäckman, S. and Sumelius, J. (2009). Identifying the Driving Forces Behind Price Fluctuations and Potential Food Crisis. Retrieved from http://www.helsinki.fi/taloustiede/Abs/DP35.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. Corpuz, P. G. (2010). Philippine Grain and Situation Outlook. Retrieved from https://gain.fas.usda.gov/Recent%20GAIN%20Publications/Grain%20and %20Feed%20Annual_Manila_Philippines_2-5-2010.pdf. Dalinina, R. (2017). Introduction to Forecasting with ARIMA in R. Retrieved from https://www.datascience.com/blog/introduction-to-forecasting-with-arima-in-rlearn-data-science-tutorials. 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.

94

De La Salle University – Dasmariñas Food and Agriculture Organization Corporate Statistical Database. (2012). Zea mays. Retrieved from http://eol.org/pages/1115259/details. Frankel, J. A. and Rose, A. K. (2009). Determinants of Agricultural and Mineral Commodity Prices. Retrieved from https://www.rba.gov.au/publications/confs/2009/frankel-rose.html. 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-dealscentral-phl-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. Gwirtz, J. A. and Casal, M. G. (2014). Processing Maize Flour and Corn Meal Food Products. Retrieved from https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4260129/pdf/nyas13120066.pdf. 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. Hyndman, R. J. (2014). Measuring Forecast http://citeseerx.ist.psu.edu/viewdoc/download? doi=10.1.1.434.2646&rep=rep1&type=pdf.

Accuracy.

Retrieved

from

Hyndman, R. J. and Koehler, A. B. (2005). Another Look at Measures of Forecast Accuracy. Retrieved from http://citeseerx.ist.psu.edu/viewdoc/download? doi=10.1.1.154.9771&rep=rep1&type=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.

95

De La Salle University – Dasmariñas 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.pd f. Jorgensen, N. (2014). A Comparison of Corn Yield Forecasting Models. Retrieved from http://openprairie.sdstate.edu/cgi/viewcontent.cgi?article=1339&context=etd. Juliano, B. O. and Gonzales, L. A. (n.d.). Rice and Corn Technology and Policy in the Philippines. Retrieved from http://www.nast.ph/images/pdf %20files/Publications/NAST%20Transactions/NAST%201986%20Transactions %20Volume%208/Plenary%20Paper%201%20Rice%20and%20Corn %20Technology%20and%20Policy%20in%20the%20Philippines %20%20%20Bienvenido%20O.%20Juliano%20and%20Leonardo%20A. %20Gonzales.pdf. Kulkarni, S. and Haidar, I. (2009). Forecasting Model for Crude Oil Price Using Artificial Neural Networks and Commodity Futures Prices. Retrieved from https://arxiv.org/ftp/arxiv/papers/0906/0906.4838.pdf. Latham Hi-Tech Seeds. (2012). The Importance of Corn. Retrieved from http://www.thefieldposition.com/2012/06/the-importance-of-corn/. 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. 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.

96

De La Salle University – Dasmariñas Off The Grid News. (2015). The Importance of Corn in The American Economy. Retrieved from http://www.offthegridnews.com/off-grid-foods/the-importanceof-corn-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-andtheir-economic-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-lowercommodity-prices/. Sims,

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

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_ Detection_from_a_Moving_Video_Camera_Using_Neural_Network_and_Kalm an_Filter. Teucrium Trading, LLC. (2015). Seasonal Price Patterns in Corn: December Might be the Best Month to Consider Investing in Corn. Retrieved from

97

De La Salle University – Dasmariñas https://seekingalpha.com/article/3727656-seasonal-price-patterns-corndecember-might-best-month-consider-investing-corn. Tiffany, D. G. (2009). Economic and Environmental Impacts of U.S. Corn Ethanol Production and Use. Retrieved from https://files.stlouisfed.org/files/htdocs/publications/red/2009/01/Tiffany.pdf. 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/266a0bf3-51424545-b806-ef9fd78b9c2f/jec-agriculture-report.pdf. Wisner, R. (2014). Ethanol, Gasoline, Crude Oil and Corn Prices: Are the Relationships Changing?. Retrieved from https://www.agmrc.org/renewableenergy/ethanol/ethanol-gasoline-crude-oil-and-corn-prices-are-the-relationshipschanging/.