• No results found

Does temperature analysis generate a positive alpha in a corn trading strategy?

N/A
N/A
Protected

Academic year: 2021

Share "Does temperature analysis generate a positive alpha in a corn trading strategy?"

Copied!
28
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

Does temperature analysis generate a positive

alpha in a corn trading strategy?

Jitse Schol

Student number: 10781463

Bachelor’s Thesis Finance and Organization

Faculty: Economics and Business

Supervisor: dr. P.J.P.M. (Philippe) Versijp

Date: 31 January 2018

Abstract

Using temperature analysis in a corn trading strategy did not provide a positive alpha. The strategy did generate positive returns in some timeframes, however returns were not large enough to outperform a normal market portfolio. The Fama and French three-factor model and the Fama and French five-factor model were used to analyze the returns of the trading

strategy. Using a timeframe of 22 years (1995-2017), results showed the alpha was not significantly different from zero.

(2)

Statement of Originality

This document is written by Student Jitse Schol, who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text

and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

Table of Contents

1: INTRODUCTION 4

2: LITERATURE 6

3: METHOD AND DATA 9

3.1:TRADING STRATEGY 9

3.2:TEMPERATURE DATA 9

3.3:FINANCIAL DATA 11

3.4:TESTING DETAILS 12

4: HYPOTHESIS AND MODEL 16

5: RESULTS 18

6: CONCLUSION 24

APPENDIX 26

(4)

1: Introduction

Corn is a commodity that has an influence on many aspects in our lives. Besides from being used as food for livestock, it is used for a long list of things, such as food, ethanol, beer, wine, ink, glues, leather seats, and much more (The Balance, 2017). Because it is such an important commodity in our lives, its price is closely watched on the financial markets. Furthermore, Hamilton and Wu (2015) write “The last decade has seen a phenomenal increased

participation by financial investors in commodity futures markets”. This means the popularity of commodity investing has increased.

In November 2017, the World Agricultural Production report by the United States Department of Agriculture stated the following: The United States is the largest corn producer of the world, having produced 384.78 million metric tons in 2016/2017. The corn world production in 2016/2017 was 1,074.76 million metric tons. China as number two in the list, produced 219.55 million metric tons of corn in 2016/2017. As a result, the United States accounts for roughly 36% of the world’s corn production.


 Because the United States is such a dominant player in the corn market, it is very interesting to analyze what temperature levels, specifically in the United States corn belt, do to corn prices. The reason temperature is chosen as a possible forecasting variable is that hot weather (usually accompanied by drought) negatively affects corn production. By using daily temperature data, it might be possible to forecast the direction of movement of corn prices, which in turn could lead to a trading strategy that can provide a positive alpha. 


Aside from the United States being such a dominant player in the corn market, there are four other reasons why this is a topic worth investigating. Firstly, climate change is something that has been in the news a lot the last few years. In an online article by NASA, it says that “Scientists have high confidence that global temperatures will continue to rise for decades to come, largely due to greenhouse gases produced by human activities.”. It is therefore important to analyze whether temperature can be used to forecast movement direction of corn prices, since this is such a relevant commodity. Secondly, Chen, Rogoff & Rossi (2010) have written a paper on exchange rates as forecasting variables for commodity prices. In the conclusion of their paper they suggested that it would be interesting if further research could be done on forecasting commodity prices by using other financial and macroeconomic variables. Temperature is not necessarily a financial or macroeconomic variable, but it is widely known that it influences financial instruments. Thirdly, Cargill & Rousser (1975) have done research on temporal price behavior in commodity futures markets.

(5)

They concluded that the random walk theory does not apply to commodity futures markets. This suggests that there might be a way to predict commodity prices movement, since they are not random. Corn is also traded on the commodity futures markets. Therefore, it is interesting and of scientific value to find out whether temperature is a variable that can be used in a corn trading strategy, hopefully generating a positive alpha. Lastly, this research can be useful for investors and traders. Traders and investors are always looking for strategies that can generate a positive alpha. Corn futures are traded on the financial markets. The corn futures prices are closely linked to the commodity corn. As mentioned, the random walk theory does not apply to commodity futures markets. Therefore, the trading strategy might be able outperform a normal market portfolio. If that is the case, traders and investors can decide to use the trading strategy to make larger profits.

(6)

2: Literature

Commodities as a financial investment have gained popularity. As mentioned in the introduction, Hamilton and Wu (2015) indicate that in the last decade, participation by financial investors in the commodity futures market has increased phenomenally.

Furthermore, Büyükşahin and Robe (2014) state “The last decade also witnessed growing commodity-market activity by hedge funds, commodity index traders, and other financial traders”.

Commodity trading strategies, in which financial/economic variables are used, have been discussed in multiple financial research papers. Two famous papers that were inspiring for this research paper were written by Chen, Rogoff & Rossi (2010) and Driesprong,

Jacobsen & Maat (2008). As described in the introduction, Chen, Rogoff & Rossi (2010) have written a paper on exchange rates as forecasting variables for commodity prices. The

currencies that were used in the research were the Australian, New Zealand, Canadian dollars, the South African Rand and the Chilean peso. These currencies were chosen, because the related countries had a sufficiently long history of market-based floating exchange rates and therefore the dynamic relation between exchange rates and commodity prices could be explored. In their empirical research, they found that exchange rates have robust power in predicting global commodity prices. The reason for this is that the exchange rate is forward-looking and contains information about commodity price movements in the future. When market participants foresee future commodity price shocks, this expectation will be reflected in the exchange rate, through anticipated impact on future export income and exchange rate values. In contrast, commodity prices tend to be sensitive to current global market conditions, since both demand and supply are quite inelastic. As a result, commodity prices tend to be a less accurate reflection of future conditions than exchange rates.

Furthermore, Driesprong, Jacobsen & Maat (2008) did empirical research on the link between the commodity oil and stock market returns. Using stock market data of 48 countries and a price series of several types of oil, it was found that changes in oil prices can predict worldwide stock market returns, both in developed and emerging markets. It was concluded that a rise in oil prices lowers stock returns.

Schenkler and Roberts (2006) have written an article on the relationship between weather and corn yields. In their empirical research, most US counties from 1950-2005 were investigated. Results show that temperatures above 30 degrees Celsius are harmful for corn yields. Using their results, it is fair to say that when temperatures are above 30 degrees

(7)

Celsius, corn supply will drop. This should cause the price of corn to move upwards, since there will be less corn on the markets. The results of this research paper made temperature an interesting variable to investigate.

Lastly, it was mentioned in the introduction, that Cargill & Rousser (1975) have done research on temporal price behavior in commodity futures markets. They focused on whether the random walk theory applies to the commodity futures markets. The random walk theory means that financial securities move independently and historical price trends cannot be used to predict the future. In other words, prices of financial securities cannot be predicted and are random. Cargill & Rousser found that the random walk theory was not applicable to the commodity futures markets. This indicates that prediction of commodity futures prices is possible. As a commodity, corn is also traded on the financial markets in the form of futures. Prices of corn futures are highly correlated to the price of corn. Together with the results found by Schlenker and Roberts (2006) and the rejection of the random walk theory by Cargill & Rousser, there is reason to investigate whether a trading strategy can be build, in which temperature is used to predict the price movement of corn.

The details of the trading strategy will be explained in the next section. However, every trading strategy must be tested against a normal market portfolio, to see whether the returns are exceptional. An insight in how this will be tested can be given at this point

already. The literature that is needed to test this was found in the articles written by Fama and French. Fama & French have done research on a three-factor model (1993) and a five-factor model (2015). Fama & French have investigated common risk factors in the returns on stocks and bonds. The common risk factors that are accounted for in the three-factor model are Rmkt

RFt, SMB and HML. In the five-factor model, the two factors that are added are RMW and

CMA. The purpose of these factors is to explain the returns of an asset or a portfolio. This can be done by doing a regression analysis on the factors. The regression analysis is performed with a constant, which is referred to as the alpha. The three-factor model and the five-factor model can be seen below.

Three-factor model:

Rit - RFt = 𝛼i + bi(Rmkt - RFt)+ siSMBt + hiHMLt + eit

Five-factor model:

(8)

On the website of K. French, the data of the factors can be found. By using these numbers, the performance of a trading strategy/asset can be evaluated. When the regression is performed, the excess return of the asset can be calculated. If there is an excess return of the asset (given a certain risk) compared to the normal market portfolio, the alpha of the regression is positive. To define the alpha, when it is significantly different from zero, it indicates a superior return of an asset/portfolio, compared to the normal market portfolio.

The usage of the three-factor model and the five-factor model has been explained, however it is important to describe why the certain factors are in the model to explain the return of an asset. All the factors have something in common, namely that according to Fama and French the factors explain the returns of an asset. The three-factor model was built in 1993, on the knowledge of the already existing CAPM model, which tried to explain the return of an asset by comparing the asset’s risk with the market risk. The higher the risk, the higher the Beta of an asset would be. Rmkt - RFt is therefore the first factor of the model. Fama

and French believed that there should be two more factors explaining the return of an asset, which were the return of a diversified portfolio of small stocks minus the return on a diversified portfolio of big stocks (SMB) and the difference between the returns on

diversified portfolios of high and low B/M (book-to-market-ratio) stocks (HML). The reason these factors were added, was because Fama and French found a relationship between size and return of an asset and price ratios (such as B/M) and return of an asset (Fama & French, 2015).

However, in 2015, Fama and French wrote in their research paper: “The evidence of Novy-Marx (2013), Titman, Wei, and Xie (2004), and others says that the three-factor model is an incomplete model for expected returns because its three factors miss much of the variation in average returns related to profitability and investment.”. Motivated by the evidence, Fama and French decided to include two new factors in their model, leading to the five-factor model. The two new factors that were added were the difference between the returns on diversified portfolios of stocks with robust and weak profitability (RMW) and the difference between the returns on diversified portfolios of the stocks of low and high

investment firms (CMA). Adding these factors to the model results in accounting for the relationship between asset returns and the profitability of firms and the level of investments of firms and asset returns.

Now that the Fama and French model has been fully explained, it is time to describe the trading strategy and the dataset that will be used for this research.

(9)

3: Method and data

This section consists of three parts. Firstly, the trading strategy will be introduced. This part clarifies how the trading strategy works and what is needed to succeed in testing the strategy. Secondly, the datasets that are used will be described, along with the calculations that are necessary to prepare the dataset for testing. Lastly, after having described the datasets and calculations, the final details of the testing phase will be explained.

3.1: Trading Strategy

As described in the literature section, when temperatures are too high, corn production will be negatively influenced (Schlenker and Roberts, 2006). In other words, when temperature levels are too high, corn supply will drop. This should cause the price of corn to move upwards, since there will be less corn on the markets. The geographical focus area of the research is the US corn belt area, which produces 76.267% of total US production (table 1). Since the

expectation is that prices will move upwards when temperatures are too high, it is interesting to see whether a trading strategy based on temperatures will provide a positive alpha. In the trading strategy, corn will be bought when a (high) temperature threshold is reached and corn will be sold when the temperature drops below the threshold.

3.2: Temperature Data

To be able to build the trading strategy there are two variables necessary, daily temperature levels and corn prices. To begin with, daily average temperature levels (in degrees

Fahrenheit) of all corn belt states were obtained from the University of Dayton, who in turn received the data from The National Climatic Data Centre. It would have been beneficial for the research if there were also maximum and minimum temperature levels of each given day, however this was not available. As mentioned above, the geographical focus area is the corn belt area. The corn belt area consists of the following states: Illinois, Indiana, Iowa,

Minnesota, Nebraska, Michigan, Ohio, Pennsylvania, South Dakota and Wisconsin. The reason for using the corn belt area is that it produces the largest amount of the corn in the entire US. In 2016, the corn belt produced 76.267% of total US corn production (table 1). It would make no sense to include states that hardly produce any corn. The reason for this is that traders and investors that move the market will most likely not look at temperatures in states that produces almost no corn, such as Alaska. Therefore, the trading strategy will only focus on corn belt states.

(10)

The timeframe of the dataset is from 1995 until 2017 and all daily average temperature levels are notated in degrees Fahrenheit. Weekend days were removed from the dataset, since there was no corn price data for weekends. In the temperature dataset, there were 11

measurements of –99 degrees Fahrenheit, which is obviously a measurement error in the data. This error was solved by not acting (no transactions) on these 11 days. In total, there are 5892 measurement days that are used, so excluding 11 days in the trading strategy should not have a large effect.

Table 1:

The table shows the corn production in (1,000) bushels for every corn belt state, the total corn belt and the United States. Also, per corn belt state, the percentage of total US production and

the percentage of total corn belt production is shown.

Source: Crop production 2016 summary (January 2017), USDA, National Agricultural Statistics Service: Corn area planted for all purposes and harvested for grain, yield and production -- States and United States, page 11

After sorting the daily temperature data and excluding weekends, the production weighted temperature levels are calculated. A state that produces large amounts of corn should have a larger price effect when temperature changes, than a state that produces smaller amounts of corn. To make this more clear, Iowa and Pennsylvania are used as an example. Of all the corn belt states, Iowa produces the largest amount of corn and Pennsylvania the

smallest amount (table 1). If temperatures in both states are too high, Iowa will lose more corn yields than Pennsylvania, because of its higher volumes of corn production. A weather effect should therefore have a stronger relationship to the market supply of corn when a state

(11)

produces large volumes of corn.

To include this assumption in the trading strategy, production weighted temperature levels are used. As is shown above in table 1, for all states the production level relative to total corn belt production level is calculated in percentages. This gives every corn belt state a weight. The weight of each corn belt state is then multiplied by the daily average temperature of the given state. By taking the sum of each state’s weight multiplied by its daily average temperature, the production weighted daily temperature average of the corn belt is calculated. This action is performed for all days between 1995 and 2017 (excluding weekends).

Mathematically, the calculation for production weighted temperature at day x is as follows:

𝑃𝑟𝑜𝑑&∗ 𝑇& *

&+,

Where:

Prodi = production level state i relative to total corn belt production level in %

Ti = average temperature level of state i at day x

n = number of states

The result of this formula is the production weighted average daily temperature that will be used in the trading strategy. By using this calculation method, the productivity of each state is considered when using its temperature level. This solves the productivity level issue of the example described before, about Iowa and Pennsylvania.

3.3: Financial Data

Above is shown how temperature data is transformed to be used for the trading strategy. Using the productivity weighted temperature should give more accurate results than using temperature as an unadjusted variable on its own. Following the temperature data, corn price data is needed for the trading strategy. The three most common ways of trading corn on the markets are futures, options and ETFs. Futures and options are defined in an excellent way by Berk & DeMarzo (2014): “A futures contract is an agreement to trade an asset on some future date, at a price that is locked in today (p995). A financial option contract gives its owner the right (but not the obligation) to purchase or sell an asset at a fixed price at some future date (p707).” The main difference between an option and a future is that futures imply an

(12)

obligation to buy or sell, whereas an option gives the right (not the obligation) to buy or sell. As mentioned above, ETFs are also a possibility when trading commodities. Again, Berk & DeMarzo (2014) provided a clear definition: “An Exchange Traded Fund (ETF) is a security that trades directly on an exchange, like a stock, but represents ownership in a portfolio of stocks (p404)”. ETF’s are an excellent way to invest in indexes of commodities.

To analyze the profitability of the trading strategy, it will be tested on historical data, also a process known as back testing. Back testing on futures and options is incredibly

complex, because of contract end terms and rollovers. Therefore, the initial idea was to use an ETF of corn, which would track the index of corn. The problems with the ETFs that were found, is that their data go back only a few years, because they are fairly recent. This timeframe would be insufficient for the analysis and therefore it is decided to use the corn index itself (Corn #.2 Yellow CBOT USA 1st Futures). The financial data was obtained from Datastream. The advantage of using this index is that the data timeframe is from January 1995 until August 2017. The disadvantage is that an investor cannot trade this index directly and would need to use an ETF for this. However, a corn ETF represents the corn index and is therefore highly correlated with its price movements. Therefore, if the tests show that the trading strategy outperforms a normal market portfolio, it will also be tested by including high transaction costs in the trading strategy. When high transaction costs are included, the

assumption is that it will present a reliable view of how the trading strategy would perform in the financial markets. Yet, if the trading strategy does not outperform the normal market portfolio, which means generating an alpha that is not significantly different from zero, transaction costs will not be calculated. The reason for this is that if the trading strategy does not work, it makes no sense to include transaction costs in the analysis, since this will only further decrease the profits of the trading strategy. If the strategy does not work, it will be of no scientific value to test the strategy with transaction costs. Only if the strategy outperforms a normal market portfolio, it is of scientific interest to analyze how it would perform when transaction costs are included.

3.4: Testing details

Up until now, the trading strategy has been introduced and it has been explained how the temperature data was collected and transformed to productivity weighted data.

Furthermore, it also has been made clear why the corn index is used as financial data. The ingredients for the trading strategy are present, now it will be described how the strategy is

(13)

tested. As mentioned in the beginning of this section, daily average temperatures are used and above 30 degrees Celsius, which is equal to 86 Fahrenheit degrees, corn yields will be

harmed. When yields are harmed, supply is harmed, meaning the corn price should go up. Therefore, this would be the optimal threshold for buying. However, since there are no daily maximum and minimum temperatures in the dataset, there are two problems that arise. First, 86 degrees Fahrenheit in the dataset will mean that that was the average temperature of the day. This means that during the hottest moments of the day, when the sun was up,

temperatures were probably much higher than that. To compensate for this, a lower temperature is used as buying threshold. In table 2, the percentiles of the daily average

temperatures of the entire dataset (1995-2017) can be seen. Three temperature thresholds will be analyzed, which are 76, 80 and 82 degrees Fahrenheit. The temperature thresholds are in the 90th, 95th and 99th percentiles, respectively (table 2). Furthermore, another argument for

using multiple thresholds is that it can serve as a robustness test.

The second problem is also easily solved, but important to mention nonetheless. It is also related to the problem of daily average temperature data. Because daily averages can only be measured at the end of the day, reacting to the daily average temperature can only be done the next day. Therefore, buying corn based on a given threshold will be done the day after the average daily temperature was above the threshold.

To test the trading strategy, the most structured way is to buy or sell 1 contract in each transaction. The result of the trading strategy will therefore explain what would have

happened if the investor would have bought and held 1 contract each time the temperature was above the threshold and would have sold 1 contract when the temperature was below or equal to the threshold.

(14)

Table 2: Weighted Temperature (F)

The table shows the descriptive statistics of the productivity weighted temperature (Fahrenheit). Percentiles Smallest 1% 5.939 -99 5% 17.033 -99 10% 23.631 -99 Obs 5,892 25% 34.427 -99 50% 52.339 Largest Mean Std. Dev. 50.315 20.807 75% 68.626 87.108 90% 74.979 87.295 Variance 432.945 95% 77.663 87.702 Skewness -0.934 99% 81.687 87.717 Kurtosis 6.764

To summarize the trading strategy, it is easiest understandable in so called pseudo-code. Pseudo-code is used by computer programmers to write down a computer program in normal English, in a very structured way. This gives a solid and, most importantly, an organized way to present a back-testing strategy:

If Temperature > threshold AND not in possession of corn Buy 1 corn contract next day

Else if temperature ≤ threshold AND in possession of corn Sell 1 corn contract next day

Else

No Transaction (wait)

Now that the trading strategy has been formalized, it is ready to be tested. To give a proper overview of whether the trading strategy works well, it will be tested in multiple periods. First, the entire timeframe will be tested (1995-2017). After the entire timeframe has

(15)

been tested, it will also be investigated how the strategy behaves before the 2008 financial crisis, during the 2008 financial crisis and after recovery of the 2008 financial crisis. A lot of research has been done on financial crises, and especially on the start and end dates. Since, it is very difficult to determine whether a financial crisis has ended, there is much disagreement between researchers about the end date of the financial crisis of 2008. In this paper, to

determine when the economy has recovered from a crisis, the S&P 500 will be used. The S&P 500 is used, because it gives an accurate representation of the value of the 500 largest

companies in the US. Since these are the 500 largest companies in the US, it should give an accurate representation of the economy. Pre-crisis period will be from the 1st of January 2004 until the start of the financial crisis of 2008. 2004 is chosen, because the economy surely had recovered from the dotcom crisis by then, which can be seen from the graph in the appendix. As the date for the start of the 2008 crisis, Lehman Brothers’ bankruptcy on 15th of September

2008 will be used, because this had such a massive impact on the economy. The crisis period will be from Lehman Brothers’ fall until 31st of December 2009. The short duration for the crisis is used, because it is very difficult to determine when a crisis has ended. If the short period of 2008 until end 2009 is used, it is certainly within the crisis period and it is

interesting to see how the trading strategy behaves in a financial crisis. Lastly, the post crisis period will start the 1st of January 2014 until the end of the dataset timeframe. The reason 2014 is chosen, is that the economy surely had recovered by then, so it will give a reliable view on how the trading strategy performs in a post crisis scenario, but most importantly, in a strong economy. By using these timeframes, the analysis will give an accurate overview of how the trading strategy behaves in the long run (1995-2017), in a pre-crisis period (2004-2008), in a crisis period (2008-2009) and in a post crisis period when the economy had recovered (2014-2017). Last, but not least, another period will be tested, which is the period between 1st of January 2010 until 31st of December 2013. The reason this period is also tested,

is that it is not used in the before crisis, during crisis and post crisis timeframe. It would be wasteful to neglect this period. In the next section, the expected outcome of the trading strategy will be discussed, along with the hypothesis.

(16)

4: Hypothesis and Model

Now that the basis for the trading strategy has been formalized, a hypothesis can be formed based on the expected outcome of the trading strategy. As explained previously, above 30 degrees Celsius, corn production is expected to be negatively affected. Because of this, corn supply will drop, resulting in a price increase. Therefore, the expectation is that when

temperatures exceed a certain threshold, buying corn will result in positive returns. However, even if the strategy generates positive returns, the question is whether it provides a return that is beyond a fair compensation for the risks that are taken. To test this, the Fama and French three-factor model will be used (Fama & French, 2015). The Fama and French model is used because the commodity is looked at from an investment viewpoint, as an asset rather than a product. The Fama and French three-factor model is used to analyze whether a portfolio outperforms a normal market portfolio.The Fama and French three-factor model is as follows:

Rit - RFt = 𝛼i + bi(Rmkt - RFt)+ siSMBt + hiHMLt + eit

Where: 𝛼i = alpha

Rit - RFt = Excess return corn strategy

Rmkt - RFt = Excess return market portfolio

SMBt = Return diversified portfolio of small stocks minus the return on a diversified

portfolio of big stocks

HMLt = Difference between the returns on diversified portfolios of high and low B/M

(book-to-market-ratio) stocks eit = Zero-mean residual

The most important coefficient for testing is the alpha. The alpha is the excess return relative to a normal market portfolio. If the strategy works, the alpha will be significantly different from zero. This results in the following hypothesis:

H0: 𝛼i = 0 v H1: 𝛼i ≠ 0

The hypothesis will be tested for all temperature thresholds. The alpha will indicate whether the strategy outperforms the normal market portfolio.

(17)

K. French has done research on normal market portfolios, providing the data on all the coefficients needed for the Fama and French three-factor model. Daily data of the factors was retrieved from the website of K. French. However, the daily data could not be used with the corn-temperature dataset. This was due to a misfit in the daily data. The reason for the misfit in the data was that the corn index data included holidays (for instance 25th of December). French’s data did not include holidays, because the markets are closed then. However, this was simply solved by using monthly data on the three factors instead, which was also retrieved from the dataset provided by K. French. Every year, only April until December is tested, because these months make up the planting-harvesting period of corn(United States Department of Agriculture, 2010). Temperature is only expected to have an effect in the planting-harvesting months, therefore it would be unfair towards the trading strategy results to include months that are not in the planting-harvesting period, since the trading strategy will not be used in these months.

The three-factor dataset timeframe is from April 1995 until July 2017. After having retrieved the data needed for the Fama and French three- factor model, the daily returns of the corn strategy were transformed into monthly returns. The daily returns of the corn strategy within a given month were compounded, resulting in compounded monthly returns. The reason for compounding is that when using the strategy, it is possible that an investor does multiple transactions in a month. To be able to compare this with a normal market portfolio, the transactions should be compounded to get a monthly return. Compounded monthly returns thus means that the daily returns within a month are compounded, but the monthly returns are not compounded with other monthly returns.

After the compounded monthly returns are calculated, a regression analysis can be used to calculate the alpha. The regression is done using STATA, a statistical analysis program. The regression analysis will provide an answer to the question whether the alpha is significantly different from zero for the entire timeframe of the dataset. If the alpha is

significantly different from zero, this would imply that the trading strategy outperforms the normal market portfolio.

As mentioned in the Method and Data section, transaction costs will be only calculated and included in the trading strategy if the alpha is significantly different from zero. If the trading strategy does not provide an alpha that is significantly different from zero, it makes no sense to include transaction costs, since this will only decrease the performance of the

(18)

5: Results

As explained in the previous sections, first the trading strategy’s returns will be calculated. This is followed by analyzing the returns in a Fama and French three-factor model, to find out whether the strategy outperforms the normal market portfolio for the entire timeframe. There is a reason why the Fama and French regression is only performed for the entire timeframe instead of for all time periods. It is difficult for an investor to determine when a crisis will start or when it ends. Since market timing is so difficult, it would be more realistic to use the trading strategy in the long run than to market time pre-crisis, crisis and post-crisis periods. This is the reason why it is more interesting to see whether the trading strategy outperforms a normal market portfolio in the long run. Therefore, only the entire timeframe is tested with the Fama and French model. For the smaller time periods, only the compounded returns are calculated.

To begin with, it was very interesting to find out what the returns of the strategy would be. As mentioned before, the strategy was divided into five timeframes. Daily data was used to calculate the compounded return of each timeframe, which can be seen in table 3 below. The trading strategy which used 76 degrees Fahrenheit as threshold performed the worst at first glance, however it did manage to get a return of 47.3% in the period between the 1st of

January 2010 and the 31st of December 2013. The 80 degrees Fahrenheit threshold performed

best when looking at the entire timeframe. In the period between the 1st of January 2010 and the 31st of December 2013 it also had a high return of 36.5%. The 82 degrees Fahrenheit threshold strategy performed adequate, but the returns were quite small. Even though some of the returns are positive, and for the 80 and 82 thresholds there were positive returns for the entire timeframe, it is interesting to see whether the strategy outperformed the market.

(19)

Table 3:

The table shows the compounded returns for each threshold in different time periods. The compounded returns are in decimals.

As indicated earlier, the Fama and French three-factor will be used to analyze whether the trading strategy outperformed the normal market portfolio. The Fama and French three-factor data was only useable in monthly form for this trading strategy. Therefore, the daily returns of the trading strategy for each threshold were transformed into monthly data as well. Furthermore, the trading strategy was only analyzed in the months from April until

December. From January until March, an investor could choose from a variety of options, such as investing in risk free treasury bonds, market indices, stocks, etc. This choice is left to the investor, since the only goal of this paper is to show how the corn trading strategy

performs and the risk preference of the investor is unknown. During the three months that the corn trading strategy is not used, it would be better if the investor used its knowledge of its risk preference to decide how to invest. To test the hypothesis, whether alpha is significantly different from zero, a regression analysis was used. In the regression analysis, the excess return of the trading strategy was the dependent variable. The three factors RMkt – RF, SMB

and HML were the independent variables and the constant is the alpha. Furthermore, robust standard errors were used for the regression, to avoid heteroscedasticity issues. Below, table 4 shows the results of the regression analysis.

(20)

Table 4:

The table shows the regression analysis of the Fama and French three-factor model.

(1) (2) (3)

Excess Return 76 Excess Return 80 Excess Return 82

MktRF 0.00668 (0.12) 0.0274 (0.42) -0.113 (-0.86) SMB -0.152* (-2.01) -0.103 (-0.43) -0.195 (-0.99) HML -0.0274 (-0.29) 0.0777 (0.48) -0.221 (-1.06) Constant 0.00225 (0.40) 0.0143 (1.57) 0.0150 (1.49) R2 0.004 0.001 0.006 N 202 202 202 t statistics in parentheses * p < 0.05, ** p < 0.01, *** p < 0.001

The results in table 4 show that for each threshold, the trading strategy does not provide an alpha that is significantly different from zero. The null hypothesis (𝛼i = 0) will

therefore not be rejected at a 5% significance level for each temperature threshold. This implies that for each threshold, the trading strategy did not outperform the normal market portfolio. Even though the trading strategy did provide positive returns in some periods, it is not enough to be used for an actual investment strategy. However, in their paper, Fama and French (2015) write “The evidence of Novy-Marx (2013), Titman, Wei, and Xie (2004), and others says that the three-factor model is an incomplete model for expected returns because its three factors miss much of the variation in average returns related to profitability and

(21)

five-factor model by Fama and French. It works the same way as the three-factor model, however, two new factors are added. The two new factors are RMW, the difference between the returns on diversified portfolios of stocks with robust and weak profitability, and CMAt, which is the difference between the returns on diversified portfolios of the stocks of low and high investment firms (Fama & French, 2015). Adding these two factors leads to the

following five-factor model:

Rit - RFt = 𝛼i + bi (RMT - RFt) + siSMBt + hiHMLt + riRMWt + ciCMAt + eit

Where:

𝛼i = alpha

Rit - RFt = Excess return corn strategy

Rmkt - RFt = Excess return market portfolio

SMBt = Return diversified portfolio of small stocks minus the return on a diversified

portfolio of big stocks

HMLt = Difference between the returns on diversified portfolios of high and low B/M

(book-to-market-ratio) stocks

RMWt = The difference between the returns on diversified portfolios of stocks with

robust and weak profitability

CMAt = The difference between the returns on diversified portfolios of the stocks of

low and high investment firms eit = Zero-mean residual

Again, data of the five factors was retrieved from the website of K. French.

Furthermore, the regression analysis works the same as with the three-factor model. Robust standard errors were used to avoid heteroscedasticity issues. The five-factor dataset timeframe is from April 1995 until July 2017, also in monthly form. Again, only the months April until December are used. The output of the regression can be seen below, in table 5.

(22)

Table 5

The table shows the regression analysis of the Fama and French five-factor model.

(1) (2) (3)

Excess Return 76 Excess Return 80 Excess Return 82

MktRF -0.139 (-0.69) -0.183 (-0.92) -0.317 (-1.27) SMB -0.241 (-1.36) -0.338 (-1.25) -0.328 (-1.22) HML 0.215 (0.82) 0.257 (1.02) 0.147 (0.60) RMW -0.423 (-0.81) -0.812 (-1.35) -0.573 (-1.10) CMA -0.266 (-1.28) 0.0813 (0.25) -0.456 (-1.25) Constant 0.00539 (0.61) 0.0190 (1.61) 0.0195 (1.55) R2 0.016 0.015 0.015 N 202 202 202 t statistics in parentheses * p < 0.05, ** p < 0.01, *** p < 0.001

(23)

The results of the five-factor regression show that, for each temperature threshold, the alpha (constant) is not significantly different from zero at a 5% significance level. Therefore, again, the null hypothesis (𝛼i = 0) will not be rejected. Due to low returns, the trading strategy

does not perform better than the Fama and French normal market portfolio.

As mentioned in the Method and Data section, the corn prices are based on the corn index (Corn #.2 Yellow CBOT USA 1st Futures). Normally, an investor or trader would need to buy an ETF to track the corn index, which would lead to transaction costs. However, it was also explained that transaction costs will only be calculated if the trading strategy generates an alpha that is significantly different from zero. If it generates a positive alpha, it is very

interesting to see whether it will still outperform the normal market portfolio when transaction costs are added. However, the results show that the alpha is not significantly different from zero. Including transaction costs in the analysis would be of no scientific value, since the trading strategy would only perform less when doing so. Since the results have made it clear that the trading strategy does not outperform the normal market portfolio, and therefore should not be used by investors or traders, transaction costs will not be calculated.

(24)

6: Conclusion

The corn trading strategy provides positives returns for the entire timeframe 1995-2017, using the 80 and 82 degrees Fahrenheit thresholds. For 2010-2013, all the thresholds provide

positive returns. However, by using a Fama and French three-factor model, the regression analysis shows that the trading strategy does not outperform a normal market portfolio, in the period 1995-2017. Furthermore, a five-factor Fama and French model was also used to test the trading strategy. However, again the null hypothesis was not rejected, meaning the trading strategy still did not outperform the normal market portfolio, in the period 1995-2017.

The temperature levels that were used for the data analysis were productivity weighted temperature levels. Using productivity weighted levels seems logical, since a state that

produces large volumes of corn should cause a larger price effect when temperature changes, than a state that produces smaller amounts of corn. However, it is of course not clear how the typical trader/investor analyzes temperature levels. Traders and investors doing transactions on the markets will be the cause of price changes in corn. Therefore, it would be interesting for further research to investigate how the typical trader/investor uses temperature in their trading strategy. This can be tested by investigating whether the corn trading strategy performs better using the average temperature level of the US or the average temperature of corn belt states. It could be the case that this leads to better returns for the corn trading strategy, than when productivity weighted temperatures are used.

Furthermore, in this research, daily average temperature levels were used. Since information was not that easily accessible in the early years of the timeframe that was used (1995-2017), it is reasonable that daily average temperature levels were used. A daily average temperature level can only be calculated at the end of the day, meaning the corn transaction reacting to the daily average is bound to be the next day. This replicates the early years of the dataset, when information also had a time lag. However, as data becomes more accessible for traders and investors, it would be fair to say that temperature levels in the US are certainly available at multiple times of the day and in some places even every minute. Because of the increase in data availability, it would also be interesting for further research to see how the trading strategy performs when it reacts to temperature levels on the same day, instead of doing a transaction the next day.

The results of this research have shown that this corn trading strategy did not

outperform a normal market portfolio. However, as shown above, a few adjustments could be made to the strategy, which is something that can be used in further research. It would be

(25)

fascinating to see whether the trading strategy would outperform the market if the average temperature level of the US or the average temperature of corn belt states were used instead of the productivity weighted temperature levels. Furthermore, reacting to temperature on the same day is a change that can be made to the trading strategy. Another reason, for doing further research on this topic is that it was noticed that there has not been a lot of research on corn trading strategies. The amount of information on this topic was relatively small

compared to more popular commodities, such as oil. Corn is a very important commodity in our lives, so it was quite surprising that not much research has been done on this topic. Therefore, it would be very meaningful if further academic research would be done in this area.

(26)

Appendix

(27)

References

Berk, J., & DeMarzo, P. (2014). Corporate Finance (3rd edition). Harlow: Pearson Education Limited.

Büyükşahin, B., & Robe, M.A. (2014). Speculators, Commodities and Cross-Market Linkages. Journal of International Money and Finance, 42, 38-70.

Cargill, T.F., & Rausser, G.C. (1975). Temporal Price Behavior in Commodity Futures Markets. The Journal of Finance, 30(4), 1043-1053.

Chen, Y.C., Rogoff, K.S., & Rossi, B. (2010). Can Exchange Rates Forecast Commodity Prices? The Quarterly Journal of Economics, 125(3), 1145-1194.

Daily Corn price (Corn #.2 Yellow CBOT USA 1st Futures) data was obtained from Datastream.

Driesprong, G., Jacobsen, B., & Maat, B. (2008). Striking Oil: Another Puzzle? Journal of Financial Economics, 89(2), 307-327.

Fama, E.F., & French, K.R. (1993). Common risk factors in the returns on stocks and bonds. Journal of Financial Economics. 33(1), 3-56.

Fama, E.F., & French, K.R. (2015). A five-factor asset pricing model. Journal of Financial Economics, 116(1), 1-22.

US Research returns data by Kenneth French. CRSP Monthly Data Fama-French three factors. Retrieved from

http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html.

US Research returns data by Kenneth French. CRSP Monthly Data Fama-French five factors. Retrieved from http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html.

(28)

Hamilton, J.D., & Wu, J.C. (2015). Effects of Index-Fund Investing on Commodity Futures Prices. International Economic Review, 56(1), 187-205.

Hecht, A. (2017, 15th of February). Grain Opportunities for the Future — Corn. The

Balance. Retrieved from https://www.thebalance.com/grain-opportunities-for-the-future-corn-808903

Roberts, M.J., & Schlenker, W. (2006). Nonlinear Effects of Weather on Corn Yields. Review of Agricultural Economics, 28(3), 391-398.

National Aeronautics and Space Administration (NASA). The Consequences of Climate Change. Retrieved from 
https://climate.nasa.gov/effects/.

Novy-Marx, R. (2013). The other side of value: The gross profitability premium. Journal of Financial Economics, 108(1), 1-28.

Titman, S., Wei, K., & Xie, F. (2004). Capital investments and stock returns. Journal of Financial and Quantitative Analysis, 39(4), 677-700.

United States Department of Agriculture. (2010). Field Crops: Usual Planting and Harvesting Dates.

Referenties

GERELATEERDE DOCUMENTEN

Studying implementation fidelity of OHL-interventions, their moderators, including barriers and facilitators affecting implementation, and long-term outcomes, are

Een lange adem en veel overleg bleek nodig voor de natuurontwikkeling van het Hunzedal, maar de resultaten van de afgelopen 20 jaar en plan- nen voor de komende 15 jaar

In this study, the chemical composition of ambient particles and the particles in fresh biomass burning plumes were studied at a savannah environment in Botsalano, South Africa..

These three factors are the Market factor; measured as the return of the market portfolio over the risk-free rate, the Size factor; measured as the difference between the

The dependent variable is the value weighted average stock return of the portfolio sorted by size and book-to-market ratio minus the riskfree interest rate in the period.. Size,

By allocating stocks into portfolios based on the cash flows between investors and companies, the FF5 model can explain the cross- sectional variation in returns and identify

The goal of this research is to test whether the FFM, CAPM and four-factor model are able to account for different risk factors that influence stock returns and on how

The final results of this research could not confirm that when an organization uses CRM and celebrity endorsement to communicate to their consumer, it will