Plastics Models

Vasant,

I wanted to keep you up-to-date on the redevelopment of forward curves for the suite of 7 plastics (ethylene contract, ethylene spot, propylene contract, propylene spot, polypropylene, HDPE, LLDPE). As you may recall, we talked about fitting all of the components into a VAR model and solving the system of simultaneous equations. Unfortunately, with a limited dataset (say 5 years or so), our degrees of freedom in such a model are dramatically reduced. Alternatively, Mauricio and I have decided to construct individual models for each of the plastics commodities while utilizing results from the initial VAR analysis as a guide.

Based on the petrochemicals flowsheet, both propylene and ethylene are fashioned directly from raw materials such as crude oil, coal, and natural gas. In turn, polypropylene may be made from propylene while HDPE and LLDPE use ethylene as a raw material. Given this logical flow, we restricted our model specifications in the following manner (all price variables tested for stationarity--first difference used in all cases):

1) Construct models for ethylene contract and ethylene spot based on AR terms, WTI, and natural gas. Determine the "best" model between ethylene contract and ethylene spot as a function of these variables (based on R2).

2) Supplement the model above with the smaller R2 value with the contemporaneous and lagged values of the dependent variable from the other model (e.g., construct ethylene contract as a function and WTI and natural gas--construct ethylene spot as a basis to ethylene contract).

3) Repeat the process for propylene refinery grade and propylene chemical grade.

4) For polypropylene, construct a model based on AR terms, contemporaneous and lagged values of propyleneRG, propyleneCG, WTI, and natural gas.

5) For HDPE, construct a model based on AR terms, contemporaneous and lagged values of ethylene contract, ethylene spot, WTI, and natural gas.

6) For LLDPE, construct a model based on AR terms, contemporaneous and lagged values of ethylene contract, ethylene spot, HDPE, WTI, and natural gas.

Note that we are aware of the "cascading error" problem as we move from model to model. Alternatively, we can construct each model based only on WTI and natural gas (see first-cut of polypropylene model). We do have monthly fundamental supply and demand information for 3 of the plastics (polypropylene, HDPE, and LLDPE) and will incorporate this data as well. Alan Engberg has forwarded the latest 12 month forward curves from the current model that is being used. He has also offered his own projections for these commodities over the next 12 months. We'll compare the revised model projections with both the old model projections and Alan's numbers and then make a decision on next steps.

Nelson