Econometric Analysis of Seasonality June 8, 2010
Table of Content 1 Introduction: Futures and Futures Spreads 2 Description of Data 3 Application of Seasonal Models Deterministic Model Linear Stationary Seasonal Model Seasonal Unit-Root Non-Stationary Model 4 Results and Discussion
Introduction Aim of this empirical analysis Estimation and forecast of corn futures spreads based on the fact of seasonal time series Application of different seasonal models to the time series
Introduction Futures Contracts A futures contract is an agreement between two parties to buy or sell an asset at a certain time in the future for a certain price. Suppose that, on June 8, the July futures price of corn at the Chicago Merantile Exchange (CME) is quoted as 350 $Cents per bushel (contract size is 5,000 bushels 127 t.). This is the price, exclusive of commissions, at which traders can agree to buy or to sell Corn for December delivery.
Introduction July and December Corn Futures Prices 2006 Future Prices 220 230 240 250 260 270 280 July Corn Contract December Corn Contract Sep Nov Jan Mär Mai Jul 2006
Introduction Corn Futures Prices Corn Futures Prices 200 300 400 500 July Corn Contract December Corn Contract 1990 1995 2000 2005 1989 2006
Introduction Futures Spread A futures spread is a technique in which a trader buys one contract and sells another contract of the same commodity with another delivery date. Suppose that, on June 8, the July futures price of corn at the Chicago Merantile Exchange (CME) is quoted as 350 $Cents per bushel and the December futures price of corn is quoted as 360 $Cents per bushel. A trader sells July Corn and buys December Corn. The spread has as value 360 350 = 10 $Cents per bushel. The goal for the trader is that the July contract declines and that the December contract increases in order to increase the spread value.
Introduction Corn Futures Spreads Dec Jul (every Spread beginning with 0) Spread (beginning with 0) Dec Jul in points 150 100 50 0 50 1990 1995 2000 2005 1989 2006 Last 250 days of both legs
Introduction Economic Reasons for Seasonality in Agricultural Markets There is obviously seasonality in the underlying agricultural business. For example the bulk of the US corn crop is planted April/May and harvested October/November.The seasonal pattern for the corn market should assume therefore a specific path. Price and perceptions of supply tend to be inversely related, with price often lowest when supply is greatest, at harvest and with price often highest in May when the market is anxious about the potential for new production.
Description of Data Data are the July and December Corn Futures Contracts At the Chicago Merantile Exchange (CME) During the years 1989-2006 Unit of measurement: $Cents per bushel (contract size is 5,000 bushels 127 t.) Relevant time series are the last 250 days of the Dec - July Spread
Description of Data Corn Futures Spreads Dec Jul (in absolute numbers) Spreads Dec Jul in points 150 100 50 0 50 0 50 100 150 200 250 Last 250 Days 1989 2006
Description of Data Corn Futures Spreads Dec Jul (in absolute numbers) Spreads Dec Jul in points 150 100 50 0 50 mean of spreads median of spreads 0 50 100 150 200 250 Last 250 Days 1989 2006
Description of Data Mean/Median of Corn Futures Spreads Dec Jul (in absolute numbers) Mean/Median of Spreads Dec Jul in points 0 5 10 mean of spreads median of spreads 0 50 100 150 200 250 Last 250 Days 1989 2006
Description of Data Average spread in time classes: mean(spread[x[i 1]=>Days<x[i]]) 100 50 0 Average Spreads over Time Classes Days until maturity > 250 > 225 > 200 > 175 > 150 > 125 > 100 > 75 > 50 > 25
Description of Data Mean of average Spread over Time Classes Mean of average spread over time classes 2 4 6 8 > 250 > 225 > 200 > 175 > 150 > 125 > 100 > 75 > 50 > 25 Days until maturity
Description of Data Average Spreads over Time Classes Average spreads over time classes 100 50 0 > 250 > 225 > 200 > 175 > 150 > 125 > 100 > 75 > 50 > 25 1989 1991 1993 1995 1997 1999 2001 2003 2005
Application of Seasonal Models Discussion on useful Seasonal Model Classes Data follow a deterministic seasonality ( summer remains summer ) and are non-stationary. Deterministic model class: non-stationary maybe a good model for data Linear stationary model class: stationary not such a good model for data Unit-root model class: non-stationary but summer may become winter maybe only a good model with deterministic parts
Application of Seasonal Models: Deterministic Model Simple Model for Deterministic Seasonality y t = S δ st m s + ɛ t s=1 δ st = 1 if t falls to season s, and δ = 0 otherwise; m s is the mean for season s; S is the numer of seasons and in the example 250; ɛ t is zero-mean stationary; t = 1,..., T = 4500 Estimation: SSE 1.508 10 6
Application of Seasonal Models: Deterministic Model Corn Futures Spreads Dec Jul Spread Dec Jul in points 150 100 50 0 50 Seasonality Estimation 1990 1995 2000 2005 1989 2006 ( 2007 estimate) Last 250 days of both legs
Application of Seasonal Models: Linear Stationary Model Simple Stationary Seasonal Model y t = φ S y t S + ɛ t, φ S < 1 S is the numer of seasons and in the example 250; ɛ t is zero-mean stationary; t = 1,..., T = 4500 Estimation: Estimate t value p-value ˆφ S 0.10872 6.846 < 10 3 AIC 37, 197 SSE 1.642 10 6
Application of Seasonal Models: Linear Stationary Model Corn Futures Spreads Dec Jul Spread Dec Jul in points 150 100 50 0 50 Seasonality Estimation 1990 1995 2000 2005 1989 2006 ( 2007 estimate) Last 250 days of both legs
Application of Seasonal Models: Linear Stationary Model Another Stationary Seasonal Model y t = 5 φ is y t is + ɛ t i=1 S is the numer of seasons and in the example 250; ɛ t is zero-mean stationary; t = 1,..., T = 4500; φ is < 1 for each i
Application of Seasonal Models: Linear Stationary Model Another Stationary Seasonal Model Estimation: Estimate t value p-value ˆφ 1S 0.22493 11.808 < 10 3 ˆφ 2S 0.11821 6.039 < 10 3 ˆφ 3S 0.09586 5.007 < 10 3 ˆφ 4S 0.01669 0.865 0.387 ˆφ 5S 0.01464 0.790 0.429 AIC 28, 957 SSE 1.695 10 6
Application of Seasonal Models: Linear Stationary Model Corn Futures Spreads Dec Jul Spread Dec Jul in points 150 100 50 0 50 Seasonality Estimation 1995 2000 2005 1989 2006 ( 2007 estimate) Last 250 days of both legs
Application of Seasonal Models: Unit-Root Model Seasonal Unit-Root Model (SWR) y t = y t S + ɛ t S is the numer of seasons and in the example 250; ɛ t is zero-mean stationary; t = 1,..., T = 4500 Estimation: SSE 3.376 10 6
Application of Seasonal Models: Unit-Root Model Corn Futures Spreads Dec Jul Spread Dec Jul in points 150 100 50 0 50 Seasonality Estimation 1990 1995 2000 2005 1989 2006 ( 2007 estimate) Last 250 days of both legs
Application of Seasonal Models: Unit-Root Model Another more general seasonal unit-root model y t = φy t S + S δ st m s + ɛ t, s=1 assuming φ to be well behaved (all roots outside unit circle). S is the numer of seasons and in the example 250; m s is the mean for season s; ɛ t is zero-mean stationary; t = 1,..., T = 4500 Estimation: Estimate t value p-value ˆφ 0.19682 13.11 < 10 3 AIC 36, 721 SSE 1.632 10 6
Application of Seasonal Models: Unit-Root Model Corn Futures Spreads Dec Jul Spread Dec Jul in points 150 100 50 0 50 Seasonality Estimation 1990 1995 2000 2005 1989 2006 ( 2007 estimate) Last 250 days of both legs
Application of Seasonal Models: Unit-Root Model A third general seasonal unit-root model y t = 5 S φ is y t is + δ st m s + ɛ t, i=1 s=1 assuming φ is to be well behaved (all roots outside unit circle). S is the numer of seasons and in the example 250; m s is the mean for season s; ɛ t is zero-mean stationary; t = 1,..., T = 4500
Application of Seasonal Models: Unit-Root Model A third general seasonal unit-root model Estimation: Estimate t value p-value ˆφ 1S 0.30125 16.678 < 10 3 ˆφ 2S 0.20608 11.104 < 10 3 ˆφ 3S 0.19126 10.535 < 10 3 ˆφ 4S 0.12277 6.708 < 10 3 ˆφ 5S 0.07595 4.325 < 10 3 AIC 28, 612 SSE 1.920 10 3
Application of Seasonal Models: Unit-Root Model Corn Futures Spreads Dec Jul Spread Dec Jul in points 150 100 50 0 50 Seasonality Estimation 1995 2000 2005 1989 2006 ( 2007 estimate) Last 250 days of both legs
Results and Discussion Model comparison Model AIC SSE Simple Deterministic Model y t = S s=1 δ stm s + ɛ t - 1.508 10 6 Simple Stationary Seasonality Model y t = φ S y t S + ɛ t 37, 197 1.642 10 6 Another Stationary Seasonality Model y t = 5 i=1 φ isy t is + ɛ t 28, 957 1.695 10 6 Seasonal Unit-Root Model(SRW) y t = y t S + ɛ t - 3.376 10 6 Another Seasonal Unit-Root Model y t = φy t S + S s=1 δ stm s + ɛ t 36, 721 1.632 10 6 A third Seasonal Unit-Root Model y t = 5 i=1 φ isy t is + S s=1 δ stm s + ɛ t 28, 612 1.920 10 6
End Thank you for the attention!