Model Results

The three-stage approach of the previous sections showed that cracker maintenance activities have an influence on prices, specifically on price spreads. The effect is positive, indicating that in times of high maintenance activity, the spread between naphtha and polyethylene widens. Most of the petrochemical players are exposed to precisely these prices as they purchase naphtha and sell, next to other derivatives, polyethylene. Hence, the difference between both prices determines an upper bound for the profit margin. Supposing further that variable and fix costs

remain stable, maintenance activities have a direct impact on a producer’s profit even if they are not undergoing maintenance themselves.

A forecast regression model taking into account planned maintenance activity adds substantial value compared to a model solely based on (historic) crude oil price and GDP growth. The models show that around 25% of the variance in the relevant price spreads can be explained by maintenance activities. Assuming safely that naphtha prices are exogenous for the petrochemical industry, the variance in question is in fact those of ethylene respectively polyethylene prices.

This underlines the importance of cracker maintenance in the upstream petrochemical industry.

Not only the process itself requires a great and costly deal of planning and coordination, but also the profit margins for those who maintain operations are affected.

4.3.1 INFLUENCE OF MARKET CAPACITY

In addition, the findings show that an argumentation solely based on market utilisation levels does not hold. The average effective utilisation of European cracker production units rarely surpassed 90% since the Lehman Shock in late 2008. Compared to previous average rates of 95% and more, one could argue that sufficient capacity is available to buffer an outage. However, the findings show that such a high-level argument does not hold and indicate that a cracker outage goes beyond issues of market capacity. Challenges which increase uncertainty and thus effort are higher need for coordination or longer reaction times to only name a few.

Further, market utilisation can describe local capacity balanced only insufficiently. Because ethylene is relatively difficult to transport, it cannot easily be substituted by distant production.

The capacity utilisation of regions likely plays an important role as well. As discussed in 1.3.2 and the European regions are fairly imbalanced. However, utilisation data on unit or regional level is not available and thus subject to further research.

4.3.2 FORECAST ABILITY

By using historical oil price values as independent variable, the model can be extended to a price-spread forecast model. GDP growth and the actual maintenance extent do not require historical values as they are either easy to forecast (GDP) or known well in advance (maintenance). In the longest lag investigated, the oil price of six months prior to the maintenance activity, the predictive power of the model is still fairly high accounting for around 50% of the variance in the product price spreads. One has to argue, however, that the forecast is more sensitive to the oil price than the maintenance activity. For the -1 lag polyethylene model the 𝐵-values are .823 for planned and 2.954 for Crude Oil Brent (see Table C-6). This indicates that an increase of 1 unit (1 kt of ethylene capacity) of maintenance activity increased the price spread by 0.82€ compared to 2.95€ for a 1€ increase of the oil barrel price. A full turnaround for an average European cracker takes around 45 kt of ethylene capacity out of the market.

Following the previous reasoning, this would result in a price spread increase of 36.9€. The same effect on the price spread has an oil price increase of 12.49€ per barrel¹⁴.

14 The highest observed increase in the period in scope was an increase of 9.47€ from July to August 2012.

31 Basic Supply Chain model

5 B

S

C

The model used for the supply chain is a linked series of Single Echelon Models. The name stems from the fact that an echelon represents a tier of competing firms manufacturing the same product rather than a single firm. Consequently, in the context of this study the aggregated European industry is modelled. The model is an extension with backlogs of Sterman’s Manufacturing Supply Chain Model (2000, pp. 709-755) and has been used for similar approaches (see Chapter 2). In the basic version the only external model input is end market demand. A calibration algorithm against industry reports is applied and cross-validated (see 5.1.3). The model describes all relevant physical (production, inventory and shipping volumes) and information (order and forecast) flows immanent in the supply chain. Prices are not part of the basic version but are implemented in a merged approach and discussed in Chapter 6.

Each echelon is connected to the direct pre- and successor in the supply chain, receiving orders and shipping products to the downstream party and executing orders and receiving material from the upstream party. The imitated supply chain is based on the generic four echelons and one end-market as discussed in 1.5. The echelons are numbered in descending order starting from 4 at the most upstream echelon. Figure A-4 illustrates the detailed model including the split up into three different products after Echelon 3. Echelon 2 ships to generic OEMs dedicated to an end market, diverging the chain further. When the chain is diverging, an additional integer after a full stop is added to the variables. A superscript 𝑡 depicts the time period. The Vensim model is clocked in weeks. The superscript is omitted if possible.

A similar model has been successful in describing volume flows in the process industry. Udenio et al. (2012) applied the model to a specialised chemical company echelon operating further downstream. Corbijn (2013) effectively used the model to describe a polymer supply chain in commodity context. Both approaches were able to accurately describe the collective destocking behaviour and resulting Lehman Wave. Extending the supply chain scope by moving one echelon further upstream than Corbijn is likely to yield similar results due to the high level of integration of ethylene and polymer producers (see 1.5.1). Hence, the following hypothesis can be phrased:

HYPOTHESIS 2A: A LINKED SINGLE ECHELON MODEL TRIGGERED BY END-MARKET DEMAND CAN ACCURATELY DESCRIBE OBSERVED DE-STOCKING EFFECTS IN THE PLASTICS SUPPLY CHAIN.

In the aftermath of the Lehman Wave the environment in the upstream process industry changed and the post-Lehman period shows significantly difference compared to the situation before. Diminished demand in combination with unaltered high supply capacity (see 1.1.1) tightened margins and enlarged pressure on financial performance. Prices and market sentiment are increasingly nervous leading to high volatility (see 1.4.2). Furthermore, prices render supply chain dynamics such as large-scale maintenance only in the post-Lehman period (see 4.1). This leads to the following hypothesis:

HYPOTHESIS 2B: A MODEL NOT TAKING INTO ACCOUNT PRICE PERFORMS WORSE IN THE POST-LEHMAN PERIOD THAN BEFORE.

Large-scale maintenance is reported in “capacity loss” and thus can be interpreted as capacity limitation. The strong significant effect of such capacity limitation on prices discussed in 4.1 suggests an effect can also be observed on production output. The next hypothesis is stated:

HYPOTHESIS 2C: LARGE-SCALE MAINTENANCE ACTIVITIES ARE A DYNAMIC REASON FOR UPSTREAM SUPPLY AND DEMAND FLUCTUATION.

Imports of polyethylene have increased over time (see 1.5.3). Managers report that in particular LLDPE grades arouse the European market and put additional pressure on local producers. This statement is tested by a fourth hypothesis:

HYPOTHESIS 2D: IMPORTS OF POLYETHYLENE TO EUROPE INFLUENCE LOCAL PRODUCTION OUTPUT.

The four hypotheses will be tested in consecutive order by first developing the basic version of the model and then extending it sequentially. Section 5.1 explains the basic model in detail, section 5.2 and 5.3 analyse the discussed extensions. The baseline version is named Model 1A, the extensions for capacity shortage and imports are based on this model. The forecast model is named Model 1B. The model which still gives volume as output but takes into account price as input variable is named Model 2. Figure 20 shows the different calibration and forecast periods of the different models. Model 1A is used to mimic observed behaviour as closely as possible and hence the entire period is used to calibrate the model. The model, used calibration technique and verification as well as validation are discussed in section 5.1.1 to 5.1.4. Model 1B includes a forecast feature; hence the calibration horizon is shortened to give room for a 6-month forecast.

The model is discussed in 5.1.5. Due to the increased influence of price in the post-Lehman period (see 4.1.1 and 5.1.4), Model2, which partly relies on prices as input, omits the years 2007 and 2008. It is discussed in a different chapter (Chapter 6) as it applies findings of the Maintenance-Price Model from Chapter 4.

FIGURE 20: DIFFERENT CALIBRATION AND FORECAST HORIZONS OF THE SUPPLY CHAIN MODELS 0

50 100

50.00 100.00

Jan 2007 Jan 2008 Jan 2009 Jan 2010 Jan 2011 Jan 2012 Jan 2013

Model 1A

Jan 2007 Jan 2008 Jan 2009 Jan 2010 Jan 2011 Jan 2012 Jan 2013

Model 1B

Jan 2007 Jan 2008 Jan 2009 Jan 2010 Jan 2011 Jan 2012 Jan 2013

Model 2 (to test Hypothesis 3A)

33 Basic Supply Chain model