Research Methodology

This work is based on two conceptually independent models each addressing a different angle of the research questions as stated in 3.2: A Maintenance-Price Regression Model and a Basic Supply Chain Model based on System Dynamics. Structure and findings of both models are then combined in an Advanced Supply Chain Model. Figure 15 illustrates the framework.

Maintenance-Price Regression Model (Chapter 4)

This statistical model investigates the relationship between cracker maintenance activities and commodity price spreads. In a multiple linear regression model the extent of maintenance activities, the oil price and quarterly GDP growth are used to explain the price spreads of ethylene/naphtha as well as polyethylene/naphtha. A multiple regression model is used because it matches the objective of predicting “changes in the dependent variable in response to changes in the independent variables.” (Hair, et al., 2009, p. 17). Further, only one dependent variable (price spread) is considered, it is a metric variable and the size is of interest. All of which are requirements for a multiple regression analysis.

19 Research Contribution

FIGURE 15: RESEARCH MODEL STRUCTURE

In an extension the forecast ability is enhanced by using historical dependent variables instead of data from the same time period. Monthly benchmark reports of the industry are used to derive the maintenance data. Prices are extracted from information provider databases (IHS/CMAI, 2005-2013).

Basic Supply Chain Model (Chapter 5)

The basic supply chain model is similar to the one discussed by Udenio et al. (2012) who based their model on Sterman’s One-Echelon Model (2000, pp. 709-755). It is extended by an order release function and a capacity limitation as an effect of cracker maintenance. System

Dynamics has been used frequently to model supply chain dynamics such as Bullwhip and synchronised destocking effect. A forecast feature is implemented and the extension of incorporating capacity limitation and product imports evaluated. In addition to forecast applications the model allows for precise policy or scenario analysis, the very core of System Dynamics modelling (Forrester, 1961).

The model uses end market demand data (Eurostat, 2013) and is calibrated against observed production output of crackers, the highest echelon.

Advanced Supply Chain Model (Chapter 6)

The advanced supply chain model is in its core the basic model extended by a stock effect.

The stock effect adjusts an echelon’s order volume according to observed price movements based on previous studies (Corbijn, 2013). This price response is introduced at the converter Echelon 2 for all three different product streams (HDPE, LDPE, and LLDPE). In addition, the response is also introduced for feedstock purchases of the cracker Echelon 4.

21 Maintenance-Price Model

4 M

-P

M

Cracker turnarounds affect operations and planning in a reoccurring pattern which led to the establishing of best practice processes across the industry. The effect on feedstock and product prices is, however, less well understood. Sales management and business intelligence agree that there “is an effect on prices” but this effect has not been quantified yet. Turnarounds are reported as part of a monthly industry overview report listing “loss of capacity in kilo tons ethylene”. They are thus not listed as binary but as a planned nominal capacity loss. Next to the reported planned figures, also unplanned maintenance is reported to the industry. However, due to the fact that market product prices are settled on a monthly basis (see 1.4) short-term capacity losses are expected to have little influence on monthly prices. As further discussed in 1.4, the monthly settled contract price is only a guideline for producers and firm-specific discounts can be and are applied. This is where a short-term maintenance activity could be reflected, but remains unnoticed by industry reports.

In order to isolate this abovementioned effect of (planned) maintenance activities on prices one first has to understand the influence of it on the planning process of different products. As shown in Figure 5 and discussed in 1.6.1, ethylene is the most important cracker product in terms of volume. Planning decisions in production are mainly driven by ethylene and its derivatives. The production volumes for other by-products are determined by ethylene. Hence, as ethylene is the most relevant product for planning and, due to its far-reaching impact (see 1.3.3), scheduled maintenance is taken into account in planning decisions, maintenance activity is expected to have an effect mainly and strongest on ethylene and derivative prices. Because capacity is taken out of the market, supply availability decreases. Assuming an independent demand, this creates a short market situation. Because naphtha price can be seen exogenous (see 1.4), the ethylene as well as polyethylene prices are expected to rise which eventually leads to a widening of the relevant price spreads and the following hypothesis:

HYPOTHESIS 1A: MAINTENANCE ACTIVITIES OF CRACKER UNITS HAVE A POSITIVE EFFECT ON PRICE SPREADS OF COMMODITY PRODUCTS.

Figure 16 summarises the hypothesis and the corresponding model steps which are used to confirm the premise. In 4.1 the hypothesis is further detailed and the statistical model explained.

4.2 then investigates the practical implications, namely the influence on customer order behaviour.

FIGURE 16: HYPOTHESIS ARGUMENT AND CORRESPONDING MAINTENANCE-PRICE MODEL

Hypothesis

Due to high volume share, ethylene price spread is affected the most by cracker maintenance.

Because of its strong impact on capacity, maintenance can explain a substantial part of the price variance.

Due to long-term planning of cracker turnarounds, the model can be extended to forecast price spreads.

Model

(adaption for historic oil prices)