Forecasting Basis Using Kriging Extrapolation and Markov Chains: Which is More Accurate? Ward E. Nefstead Associate Professor

1
Forecasting Basis Using Kriging Extrapolation and
Markov Chains Which is More Accurate?Ward E.
NefsteadAssociate ProfessorExt. EconomistU. of
Minnesota
2
Role of Basis in Grain Marketing Decisions

• Knowledge of Basis allows decisionmaker to
  estimate local prices in the future
• Forward contract prices can be evaluated relative
  to normal basis
• Basis levels can signal changes in future
  prices-narrow vs. wide

3
Components of Basis

• Transportation cost
• Interest rates
• Storage and related costs
• Competition among elevators

4
Law of One Price Revisited

• Prices should vary based on transportation costs
  only given assumptions of perfect competition and
  a uniform product.
• Narrow basis nearest resale(terminal) points and
  wider basis further away from resale
• Kevin McNew estimates spatial basis points with
  an equation-based on flow pattern to market

5
Structural Change changes Basis Patterns

• Growth of ethanol, other processing plants
  creates a narrow basis
• Basis in narrowest along NW and SW Minnesota vs.
  traditions pattern-along Mississippi River- SE
  Minnesota

6
Minnesota Corn Basis 2004
7
Small Changes in Basis not easily visible in
Macro-spatial basis patterns

• AgManager, CARD products use data smoothing
  techniques to average basis
• Precision Ag Software- Vesper- allow small area
  basis changes
• Variograms show small area changes

8
(No Transcript)
9
(No Transcript)
10
Computer Software for Basis Comparison

• ESRI- Spatial Analyst,Business Map Pro
• Spreadsheet- Basis Tool
• Other

11
Methods of Forecasting Basis

• Traditional- use 3 or 5 year average at location.
  Assumption-past pattern will feature regression
  toward mean. Changes in structure, costs will
  dramatically alter future basis patterns.

12
Other Basis Forecasting Methods

• Markov Chains- allows an evolving structure to be
  included in forecast. Several states are possible
• Large supply, weak demand(state 1)
• Large supply, normal or strong
  demand(state 2)
• Small supply, weak demand(state 3)
• Small supply, normal or strong
  demand(state 4)
• Normal supply ,normal demand(state 5)

13
State Diagram
14
Markov Chains identify which state will
predominate

• Probabilities of movement from state to state can
  be estimated
• Narrow or wider basis will be associated with
  each state-example at MN location(Hutchinson)-corn
  average
• state 1-..70
• state 2- .50
• state 3- .45
• state 4- .30
• state 5- .35

15
State Progression 1-5
16
Markov Chain software

• Measures the transition probabilities
• New state probabilities can be estimated
• Spreadsheet based software

17
(No Transcript)
18
(No Transcript)
19
(No Transcript)
20
(No Transcript)
21
(No Transcript)
22
Another Method of Basis Forecasting- Kriging
Extrapolation

• Spatial software uses Kriging methods
• Kriging and related measures- estimate surfaces
• Use of ESRI Spatial Analyst and other programs
  allows surface estimation
• Changes in surfaces can be projected forward on a
  weighted basis

23
Vesper-Used for Kriging
24
Kriging

• Uses data points for project a surface
• Variations are used- CoKriging ,etc
• Variograms show the change in the surface with
  distance

25
(No Transcript)
26
(No Transcript)
27
Kriging price surface

• Minnesota price surface
• Iowa price surface
• Changes in basis-3 years(Ag Manager, other)
• Surface can be extrapolated(projected ahead)

28
(No Transcript)
29
Kriging Surface Extrapolation

• Changes must be weighted
• Example weight changes for the past five years
• Point forecasts are smoothed to create new
  projected surface

30
corn5596355350560555626556378
31
How has Basis Changed

• Increased competition for supplies- feed vs other
  uses
• Changes in rail and other surface transportation-
  DME railroad
• Growth of ethanol and biodiesel plants
• Higher energy costs(widen basis)

32
Quicktime video of spatial changes
33
(No Transcript)
34
(No Transcript)
35
(No Transcript)
36
(No Transcript)
37
(No Transcript)
38
(No Transcript)
39
(No Transcript)
40
(No Transcript)
41
Now, time for comparison- Which is more accurate
in forecasting 2006 MN Basis- One
locationCompare Traditional, Markov, and
Kriging Methods
42
1 stepTransition Matrix
43
(No Transcript)
44
Hutchinson Mn Spot Corn Basis

• Traditional
• Markov estimation
• Kriging Point/Surface
• Which is Best? You Decide!

45
Following estimates of 2007 Basis at Hutchinson
,MN
46
proj.basis-5 yr ave0.55650.231840.602040.58128
0.64108
47
Kriging Extrapolation0.5567470.2379520.589125
0.5691090.631987
48
(No Transcript)
49
Markov estimate.52500.55250.5150.4750.45
50
Farmer Consultations/Meetings

• Review traditional basis for area
• Present alternate methods of basis forecasting
• Discuss impact of alternative forecasts on grain
  marketing decisions

51
Summary
52
So, How do the Methods Compare?
53
Summary

• More methods needed to forecast basis
• Changing structure and economic variables
• Basis may become more variable than price levels
  due to Just-In-time acquisitions

54
References

• Taylor, Duyvetter and Karstens Incorporating
  Current Information in Historical-Average Based
  Forecasts to Improve Crop Basis Forecasts-
  NCR-134, April 2004
• Manfredo M, and Saunders D. Is Basis Really
  Local- NCCC-134, April 2006
• McNew K. Spatial Market Integration
  Definition, Theory and Evidence, AgResource
  Econ. Review,

55
Questions?