Edinburgh Research Explorer

Correlating the chemical engineering plant cost index with

macro-economic indicators

Citation for published version:

Mignard, D 2014, 'Correlating the chemical engineering plant cost index with macro-economic indicators',
Chemical Engineering Research and Design, vol. 92, no. 2, pp. 285-294.
https://doi.org/10.1016/j.cherd.2013.07.022

Digital Object Identifier (DOI):

10.1016/j.cherd.2013.07.022

Link:
Link to publication record in Edinburgh Research Explorer

Document Version:
Peer reviewed version

Published In:
Chemical Engineering Research and Design

General rights
Copyright for the publications made accessible via the Edinburgh Research Explorer is retained by the author(s)
and / or other copyright owners and it is a condition of accessing these publications that users recognise and
abide by the legal requirements associated with these rights.

Take down policy

The University of Edinburgh has made every reasonable effort to ensure that Edinburgh Research Explorer
content complies with UK legislation. If you believe that the public display of this file breaches copyright please
contact openaccess@ed.ac.uk providing details, and we will remove access to the work immediately and
investigate your claim.

Download date: 26. Sep. 2019

 1 Correlating the Chemical Engineering Plant Cost Index with
2 macro-economic indicators
3

4 Dimitri Mignard
5 School of Engineering, University of Edinburgh,
6 Edinburgh EH9 3JK, Scotland, UK
7 D.Mignard@ed.ac.uk, Tel. +44 131 651 9024
8
9 Abstract
10
11 The Chemical Engineering Plant Cost Index (CEPCI) is widely used for
12 updating the capital costs of process engineering projects. Typically,
13 forecasting it requires twenty or so parameters. As an alternative, we suggest
14 a correlation for predicting the index as a function of readily available and
15 forecast macro-economic indicators:
 n 
16 CEPCI (n)  0.135  CEPCI (k o )  exp  A   ik   B  Poil  C ,
 k  ko 
17 with ko the first year of the period under consideration, ik the interest rate on
18 US bank prime loans in year k, and Poil the US domestic oil price in year n.
19 Best fit was obtained when choosing distinct sets of values of the constants
20 A, B and C for each of the three periods 1958 to 1980; 1981 to 1999; and
21 2000 to 2011. These changes could have resulted from the impact of the oil
22 shocks in the 1970’s and very high interest rates in the 1980’s, which
23 perhaps heralded changes to the index formula in 1982 and 2002. The error
24 was within 3% in any year from 1958 to 2011, and within 1% from 2004 to
25 2011 after readjusting the weighting of the price of oil. The correlation was
26 applied to forecast the CEPCI under different scenarios modeled by the
27 Energy Information Administration or predicted from oil futures contracts.
28
29 Keywords: Chemical Engineering Plant Cost Index, capital cost
30 estimates, price of oil, interest rates, inflation
31

32 1. Introduction
33
34 1.1 The Chemical Engineering Plant Cost Index
35 Process engineers often require to forecast or update the capital cost of new
36 plants as a function of historical data on plants that were previously built.

1/24
 1 Cost indices are available for estimating the escalation of costs over the
2 years, from a year m where the known or estimated cost is Cm and the index
3 takes the value Im, to a year n where it is Cn and the index takes value In. the
4 projected cost in year n is then
5
6 Cn = ( In / Im ) x Cm Eqn (1)
7
8 Several indices are available to the process engineer; for example the
9 Nelson-Farrar Refinery Cost Index published in the Oil&Gas Journal is widely
10 used in the oil and gas industry; the Marshall and Swift equipment cost index,
11 which was published monthly in Chemical Engineering until April 2012 and is
12 now made available online (Marshall & Swift/Boeckh, LLC, 2013) is intended
13 for the wider process and allied industries (chemicals, minerals, glass, power,
14 refrigeration etc.); and the Process Engineering Plant Cost Index published
15 by the UK monthly Process Engineering provides data not just for the UK but
16 also for 16 other OECD countries.
17
18 However, it seems that the best known process plant cost index worldwide is
19 the Chemical Engineering Plant Cost Index (CEPCI), which has appeared
20 every month in the publication Chemical Engineering since 1963. Although it
21 is primarily based on US cost data, the relative lack of local and specialised
22 cost indices for the process industries amongst the countries in the world
23 (according to The Institution of Chemical Engineers, 2000) might explain its
24 widespread adoption. The dominance of the US dollar as an international
25 currency has also favoured the use of an index based in the US. Often, the
26 CEPCI is used alongside a location factor to transpose the estimate from one
27 country to another.
28
29 The CEPCI is a composite index, made up from the weighted average of four
30 sub-indices, and currently calculated from the following equation:
31 CEPCI = 0.50675 E + 0.04575 B + 0.1575 ES + 0.290 CL Eqn. (2)
32 where E is the Equipment index, B is the Buildings index, ES is the
33 Engineering and Supervision index, and CL is the Construction Labour index
34 (Vatavuk, 2002).
35 The Equipment index E itself is in fact a weighted average of seven
36 components, including: Heat exchangers and tanks; process machinery;
37 pipes, valves and fittings; process instruments; pumps and compressors;
38 electrical equipment; structural supports and miscellaneous.
39 In turn, each sub-index is the weighted average of sub-indices, derived from
40 monthly Producer Price Indices (PPIs, that are compiled by the US
41 Department of Labor’s Bureau of Labor’s Statistics (BLS) from about 100,000
42 price quotations issued by about a quarter as many domestically producing
43 companies. Sub-indices or components for which labour costs have a

2/24
 1 significant influence are discounted by multiplying their labour cost
2 component by a productivity factor (calculated from an average yearly
3 increase of 2.2% in productivity since 2002). Baselines are taken as values
4 of 100 in 1957-1959 for the composite CEPCI and all four sub-indices
5 (Vatavuk, 2002). Finally, although the CEPCI underwent overhauls in 1982
6 and 2002 which affected the selection of PPIs, the productivity factor and the
7 weighting coefficients in equation (2), it remained unchanged in its basic form
8 and adjustments were made to provide revised indices in years prior to the
9 changes (Vatavuk, 2002).
10
11 1.2 Forecasting the Chemical Engineering Plant Cost Index
12
13 1.2.1. Micro-economic approach
14
15 The composite make-up of the CEPCI suggests that forecasting it requires a
16 piecemeal approach to each of its four components as per Eqn. (2), given
17 that each component is likely to respond differently to factors such as
18 inflation on raw materials, productivity gains, labour costs, etc. In turn, each
19 component could be disaggregated into the relevant sub-indices from which
20 it is made. However, when taken too far, this disaggregation can become
21 difficult. All 53 PPI inputs would require tracking and forecasting, not to
22 mention the added inconvenience that at times some of the PPI components
23 can be modified or even discontinued by the BLS.
24
25 These difficulties would suggest using a reduced number of sub-indices as
26 proxies for the whole set. This ‘micro-economic’ approach was first
27 advocated by Caldwell and Ortego (1975), who proposed a surrogate index
28 that could track the CEPCI by using only five BLS indices: four wholesale
29 price indices (metal tanks; general purpose machinery and equipment;
30 electrical machinery and equipment; and processing materials and
31 components for construction), and one chemical engineering labour index.
32 Earl (1977) found that Caldwell and Ortego’s index failed to keep up well with
33 historical data after 1974, and advocated a more disaggregated approach.
34 He kept the main sub-indices and their respective weightings in the CEPCI
35 but substituted 24 variables for the 70 or so that the CEPCI was then using.
36 Importantly, he selected the 24 proxy variables from those amongst the
37 BLS’s PPIs for which both historical records and forecasts were available.
38 This basic approach appears to have been retained in modern practice: for
39 example Hollmann and Dysert (2007) quoted that in their experience, no
40 more than 20 or so relevant proxies are applicable to estimating cost
41 escalation of a process plant.
42
43 1.2.2. Macro-economic approach
44

3/24
 1 As an alternative to the disaggregation method, straightforward prediction of
2 the CEPCI from more general economic indicators on the cost of materials
3 and labour could also be attempted. Cran (1976) suggested two component
4 indices as effective proxies for major construction engineering indices,
5 including the CEPCI. The two indices that he proposed tracked the costs
6 associated with steel and labour respectively, with the proxy index a weighted
7 average of the two. He found that the resulting index was following the
8 CEPCI pretty closely. However, these correlations may then become too
9 simplistic to withstand major changes in technology, productivity, market or
10 other macroeconomic factors. In the same year as Cran’s paper, Styhr
11 Petersen and Bundgaard-Nielsen (1976) observed that his two-component
12 index could not account for productivity gains in assembling plant
13 components, leading to an overestimate for the capital cost of plants in
14 Western Germany between 1973 and 1975. In spite of its flaws, Cran’s
15 approach was followed by the PEI index, which was published by the journal
16 Process Economics International for 36 countries, and formerly called the
17 Engineering and Process Economics (EPE) index. Styhr Petersen and
18 Bundgaard-Nielsen also suggested that to a lesser extent other multi-
19 component indices would be affected in a similar manner, including the
20 CEPCI.
21
22 Nevertheless, the idea that wider macro-economic data can be the sole input
23 parameters is attractive because of the wide availability of data and forecasts
24 for these. In fact, the wider economic activity is not just indicated by the cost
25 of materials and labour as in Cran’s model, but can be linked with more
26 general indicators. This type of approach seems to have been initially
27 advocated by Caldwell and Ortego (1975), as an alternative to their own
28 micro-economic approach. They found that simple linear correlations held
29 between the CEPCI and any of the following: the Gross National Product
30 deflator; the Consumer Price Index; the Wholesale Price Index; and other
31 price indices. In all cases the slope of the correlation was close to 1.
32 However, they observed that the actual values of the CEPCI significantly
33 swung cyclically above and below the values predicted by those simple linear
34 correlations. Since then, literature on the topic of correlating the CEPCI with
35 macro-economic indicators appears extremely scarce. A more recent
36 example that we found regarded the Nelson-Farrar refinery cost index rather
37 than the CEPCI, but it evidenced again the type of difficulty Caldwell and
38 Ortego faced when trying this type of approach: Parker (2008) presented a
39 graph where he plotted the fuel cost index against the construction cost index
40 of the Nelson-Farrar refinery cost index from 1930 to 2007. While on a
41 logarithmic scale the construction cost index seemed to be a broadly linear
42 function of the fuel cost index with a slope of 1.00, there were wide swings
43 away from this parity ratio, with vertical and horizontal segments indicating
44 periods of rapid surges and drops of one factor apparently independently
45 from the other. The two indices were correlated to some extent, but they
46 were visibly subject to different influences too. As we shall see later, this may
47 be explained by the fact that correlations are not immediately apparent
48 unless at least two parameters are considered, and the right selection of

4/24
 1 these parameters is made, including careful appraisal of their degree of
2 mutual correlation.
3
4 In fact, econometric methods have been developed since the 1970’s outside
5 the field of engineering that more generally model economic variables. A
6 good introduction to these methods for the non-specialist can be found in
7 (Koop, 2000). Of critical importance to these methods is a rigorous handling
8 of time series, in particular with respect to the autocorrelation of the
9 variables, which is the influence that the past values of a variable have on its
10 current value. Another critical aspect of these methods lies in the avoidance
11 of spurious correlations of trended variables (i.e. variables that tend to either
12 increase or decrease monotonically with time), which will inexorably occur as
13 the sample size of the series increases (if only as the ratio of the average
14 rates of change with time of the series). In fact, Caldwell and Ortego’s as well
15 as Parker’s seemingly good correlations (op. cit.) may have been affected by
16 this flaw. Spurious correlation can often be resolved by differencing the
17 variables, however testing for its presence (denoted by the existence of a
18 ‘unit root’, i.e. the observed variable being correlated with its lagged value
19 with a slope of 1) requires appropriate statistical testing which is not always
20 conclusive if root values are close to 1. In the end, models may be obtained
21 that predict the observed variable as a function of its past values as well as
22 current and past values of the explaining variables, each variable and its
23 lagged values being tested for statistical significance and retained if
24 appropriate. From a practical viewpoint, building and testing these models
25 require specialised software (e.g. Microfit® or Stata®). They may also require
26 as many adjustable parameters as there are variables, including lagged
27 values, thus potentially being as cumbersome as the models derived from
28 the micro-economic approach.
29
30 Therefore, it is the aim of this paper is to present a simpler approach that can
31 be readily used by engineers without the requirement for specialist tools;
32 takes into account the influence of past values but with a very small set of
33 adjustable parameters; and still allows effective modelling and prediction of
34 the CEPCI.

35 2. Methodology
36
37 Values for the CEPCI from 1958 to 2010 were taken from Vatuvuk (2002)
38 and Chemical Engineering, (2009) and (2012).
39
40 From consideration of the process of constructing a plant, we first determine
41 the likely macro-economic factors that seem to impact directly on the capital
42 cost of chemical plants: firstly, finance costs when paying for the project;
43 secondly, market forces such as the balance of supply and demand of
44 materials, equipment, and even labour during design and procurement,

5/24
 1 contracting and construction; and thirdly, labour productivity and costs during
2 design, construction and commissioning. (We chose to neglect other factors
3 such as taxation and subsidies as they are more site-specific, particular to a
4 given state or region of the world.) The interdependence of these factors
5 means that care is required in selecting macroeconomic indicators that will
6 act as independent parameters in a model for the CEPCI.
7
8 2.1 Financial costs
9 Finance costs play a critical role in the construction of process plants. Prior to
10 the decision on whether or not to invest in the plant and build it, they are
11 typically factored in as “cost of capital” for the purpose of calculating a Net
12 Present Value (NPV). In order for a project to be viable, the NPV must be as
13 high as possible, highlighting the prime importance of financing costs to the
14 industry.
15 The considerable extent to which financial costs have an impact on cost
16 escalation has been known for a while. Often ‘real’ interest rates in which
17 inflation has been discounted are used in NPV calculations when inflation is
18 not explicitly applied to the data. However in this paper we are seeking to
19 correlate an inflation indicator (the CEPCI) with interest rates, and therefore
20 we wish to exploit this relationship, rather than nullify it through the use of a
21 ‘real’ interest rate. For this reason, we only consider uncorrected interest
22 rates.
23 The question is then, what is the observed relationship between inflation and
24 interest rates in historical data? Back in 1981, Remer and Gastineau
25 remarked that interest rates (taken as US AAA corporate bonds) and inflation
26 (taken as the rate of increase of the EPE index) tended to cancel each other
27 out for the purpose of calculating NPV on engineering projects, due to a
28 certain degree of correlation between the two. We found that this still applied
29 to some extent throughout the period from 1958 to 2011, for example we
30 found a linear regression coefficient R2 = 0.19 between the CEPCI inflation
31 rate and the yearly averaged rate on prime loans, as shown in Figure 1.
32 This result is not surprising. It can be expected that any rise in the cost of
33 financing will affect the CEPCI at several levels, from the costs to the
34 company commissioning the plant to the cost of contractors and equipment,
35 with everyone passing on their financing costs to their customers unless
36 competition is significant and margins are wide enough to cushion any rise in
37 interest rates. Conversely, market forces will also influence the cost of
38 financing: depending on inflation figures, Central Banks like the Federal
39 Reserve in the US will sell or buy back securities on the open market and in
40 competition with private investors. While their mandated aim in doing so is to
41 achieve an interest rate that they have set, ultimately the intended
42 consequence is to keep the economy within a safe and fairly narrow window
43 of inflation by controlling the availability of money.
44 When looking for a suitable indicator for finance costs, we considered both
45 the rates on US AAA corporate bonds (long term) and the rates on US prime
46 loans (short term), the data being collated by the Federal Reserve Bank of

6/24
 1 St. Louis and found on their website (Federal Reserve Bank of St. Louis.
2 2012a) and b)). Both rates are expected to be representative of the range
3 that would be available to industry. In this work, we tested both as
4 parameters, and settled for the one that gave the best fit correlations.
5
6 2.2 Market forces and the price of oil
7 Alongside financing costs, market forces like the balance of supply and
8 demand for raw materials, for plant components and for labour are expected
9 to affect prices significantly. While we have just seen that finance costs are
10 connected to some extent to market forces, we found that the yearly change
11 in the price of oil seems to bear no apparent correlation with interest rates
12 (for example, R2 = 0.0004 with rates on prime loans from 1958 to 2011, as
2
13 shown in Figure 2; and still R = 0.014 when replacing the yearly change in
14 the price of oil by the yearly % change in the price of oil). Therefore, we
15 chose the price of the barrel of oil as our second indicator, as a major driving
16 force for inflation that will gauge the state of the market fairly independently
17 from interest rates. The historical data for US domestic crude oil prices was
18 taken from (Illinois Oil and Gas Association, 2012), however other
19 benchmarks could also be used (e.g. Brent or WTI). For the same type of
20 reason that we chose to use raw interest rates rather than real interest rates,
21 the yearly averages for the price of oil were taken without discounting
22 inflation, i.e. we believe that the CEPCI being an escalation index it may as
23 well be accounted for by inflation in the price of its contributing factors.
24
25 2.3 Productivity and the cost of labour
26 Finally, we also attempted to consider productivity and the cost of labour as a
27 factor influencing the CEPCI. The relevant index that combine these two
28 elements is the unit labour cost (in US $ labour cost per US $ output)
29 published by BLS. However, we found some degree of linear correlation
30 between the % change over a year of the unit labour cost and the interest
31 rates (R2 = 0.45 over the period 1958 to 2011 when considering prime loan
32 rates, as shown in Figure 3). It might be that low interest rates encourage
33 investments that increase productivity, and therefore push down the unit
34 labour cost; conversely, high interest rates might discourage investments in
35 productivity. Therefore, including interest rates might indirectly account for
36 some at least of the effects of changes in productivity.
37
38 2.4 Parameters and method of the model
39 Starting with these datasets, we attempted to correlate the CEPCI linearly
40 with either the yearly average of the price of US crude oil (in $US/bbl), Poil(n),
41 or the yearly average of interest rates on prime loans (%), in. In order to
42 distinguish between temporary effects and long term effects of the changes
43 in the price of oil and in the interest rates, we also introduced integrated
44 indices of these parameters. This approach was inspired by the following
45 consideration: all along the supply chain that leads to the construction of a

7/24
 1 chemical plant, one would expect successive suppliers to pass on their
2 operating and financing costs to their customers, who themselves are
3 suppliers to other customers; so it is not unreasonable to expect that hikes in
4 costs are more likely to be fully passed on than savings.
5
6 With oil, we simply assumed a certain percentage of the costs could translate
7 into long term inflation. The resulting cumulative oil index Ioil(n) in year n is
8 the integral of the price of oil, Poil from 1958 to year n, with a basis value of
9 100 ($US/bbl)∙yr in 1958,
k n
10 I oil (n)  100  P
k 1958
oil (k ) (Eqn. 3),

11
12 For the cumulative interest rate index Iint(n) in year n, we first considered
13 compounding the yearly interest rates on prime loans from 1958 to year n,
14 with a basis value of 100 %∙yr in 1958. However, we found it difficult with this
15 approach to take into account the proportion of escalation in costs that was
16 represented by financial costs, and also maintain the proportion of financial
17 costs to the CEPCI into a reasonably narrow range. We resolved these
18 difficulties after we stepped back to a differential formulation of the problem,
CEPCI (n)
19  A  in (Eqn. 4)
CEPCI (n)
20 where ∂CEPCI(n) is the variation of CEPCI attributable to interest rate in on
21 year n when all other variables are held constant, and Ais a proportionality
22 constant between the relative increase in the CEPCI and the interest rates.
23 On integrating,
 n

24 CEPCI (n)    CEPCI (1958)  exp  A   ik   F ( X , n) (Eqn. 5),
 k 1959 
25 in which  is a constant representing the proportion of CEPCI in 1958 that
26 could have been attributed to current year and prior financing costs, and
27 F(X,n) is a function of all the variables other than the ik’s. The resulting
28 cumulative index then takes the values
29
I int (1958)    CEPCI (1958)
30  n
 (Eqn. 6)
I int (n) n1959  I int (1958)  exp  A   ik 
 k 1959 
31
32 Having now selected our parameters Poil(n), Ioil(n), in, and Iint(n), we first
33 attempted correlating the CEPCI with each of these separately, and if the fit
34 was promising we tried to correlate the difference between the CEPCI and
35 the fit given by the first parameter, by a second parameter.

8/24
 1

2 3. Results
3
4 In the following we considered the rates on US prime loan rather than US
5 AAA corporate bonds. Both gave similar results.
6
7 In table 1, we reported the extent of the linear correlations between the
8 CEPCI and each of the four parameters that were introduced in the previous
9 section. It can be seen that the best fits were achieved for the cumulative
10 index for interest rates and the cumulative price of oil, with (R2) values in the
11 range 0.93-0.96. The price of oil was also quite a strong factor. Interest rates
12 in themselves did not seem a determinant at all at R 2 = 0.0025, but the
13 cumulative index for interest rates achieved the highest fit at R2 = 0.958 (with
14 A and  arbitrarily set at 0.2 and 1, respectively – these values constituted an
15 initial guess, and it so happened that they produced a good enough
16 correlation to warrant retaining the cumulated interest rate as a parameter
17 before further optimising them).
18 However, the very high coefficient of correlation for the CEPCI with each of
19 the two cumulated indices must be taken with caution: all three of these time
20 series were in fact trended, in the sense that they all mostly increased with
21 time, and hence they would necessarily present an apparent degree
22 correlation with each other of a spurious nature (if only by the ratio of their
23 respective average rates of change). A more meaningful indication of
24 correlation between any two of them could be obtained by correlating their
25 yearly rates of change, which tends to remove the effect of yearly trends
26 (Koop, 2000). Thus, when correlating the % rate of change of the CEPCI with
27 the rate of change of the cumulated interest rate index, we obtained the
28 previously reported Figure 1, i.e. R2 = 0.19. On the other hand, correlating
29 the % rate of change of the CEPCI with the price of oil (i.e. the rate of
30 change of the cumulated price of oil) gave a value R2 = 0.05 (Figure 4), i.e.
31 the cumulated price of oil was in fact barely worth considering on its own.
32
33 By contrast, the price of oil on the current year of the CEPCI seemed to have
34 a much larger impact than its cumulated impact over the previous years. A
35 plot of the CEPCI against the prices of oil (Figure 5) showed three distinct
36 periods: From 1958 to 1980, the CEPCI seemed to increase with the price of
37 oil in an approximately linear fashion; from 1981 to 2003, the CEPCI seemed
38 to vary independently from the price of oil; and from 2004 to 2011, the
39 CEPCI again seemed to increase with the price of oil in an approximately
40 linear fashion but at less than half the rate that was observed during the first
41 period.
42
43 Therefore, tentatively we retained the price of oil and the integrated interest
44 rate index as parameters. We then attempted to optimize the linear fit for the

9/24
 1 function
2 f(A) = CEPCI(n) – Iint(n) (Eqn. 7)
3 to match the price of oil, separately for each of the three periods that were
4 identified in Figure 5 (allowing, if necessary, minor changes of boundaries).
5 The result is shown in Figure 6. We found that the following correlation
6 applied:
 n

7 CEPCI (n)  0.135  CEPCI (1958)  exp  A   ik   B  Poil  C
 k 1958 
8 (Eqn. 8)
9 with the parameters A, B and C taking the following values over the following
10 periods:
11  From 1958 to 1978, A = 1.7; B = 1.616 (bbl/US $); C = 79.5,
12 with A dropping to 0.34 in 1979-1980.
13  From 1981 to 1999, A = 0.34; B = 0.322 (bbl/US $); C = 160.
14  From 2000 to 2011, A = 0.54; B = 1.806 (bbl/US $); C = 95.7.
15
16 The match between this model and the CEPCI can be seen in Figure 7. The
17 deviation of the model with respect to CEPCI is shown in Figure 8. The error
18 was within ±3% over the period 1958-2011 (or slightly over at 3.2% in 1965
19 and 1989), and it could be reduced to within ±1% over the period 2004-2011
20 after readjusting the weighting of the price of oil – this is discussed in the
21 next section.
22
23 The same type of approach failed when trying to correlate the function
24 CEPCI(n) – ∙Ioil(n) (with  an adjustable parameter) with any of the indices
25 related to the interest rates.
26

27 4. Discussion
28
29 4.1 Respective influence of oil prices and interest rates
30 On Figures 5 and 6, the lack of influence of the price of oil on the CEPCI in
31 the 1980’s and 1990’s may simply reflect the decrease in energy intensity
32 within the process industries after the two oil shocks of the 1970s (the two
33 successive rises of the oil prices during this period are very visible on Figure
34 5). By the early 1980s, the process industries had adapted and built
35 resilience to high oil prices, but the subsequent drop in prices did not result in
36 a drop in CEPCI. Instead, the CEPCI remained relatively flat for a few years
37 even though the price of oil dropped significantly during the 1980s (Figure 1).
38 When the prices began rising again, the CEPCI was seemingly unaffected up

10/24
 1 until 2004 when it reached the value of its previous peak again (yearly
2 average US$ 37.7/bbl in 2004) and stayed above this value in the following 7
3 years. During this later period, the CEPCI increased almost linearly with the
4 price of oil. From 2004 to 2011, the CEPCI varied linearly with the price of oil
5 within 5%.
6
7 Perhaps surprisingly at first, the weighing of the price of oil in the correlation
8 for the CEPCI from 2000 to 2011 is close to what it was in the period 1958-
9 1980. At this point it is good to remember that Poil was chosen as an indicator
10 of the general health of the market regarding demand and supply, rather than
11 just reflecting energy costs.
12
13 A comparison of Figure 2 with Figure 1 shows that the intermediate region
14 between 1980 and 1999 has been flattened in Figure 2 once the high interest
15 rates of that period have been discounted.
16
17 Changes in the value of the parameter A between the three periods, 1958-
18 1978; 1979-1999; and 2000-2011 will be related to levels of investment in the
19 manufacturing industry. The drop in value of A from 1.7 to 0.34 after 1978
20 might be explained in part as a consequence of exceptionally high interest
21 rates in that period (mostly above 8% from 1979 till the late 1990s). These
22 high rates would have discouraged new investments by companies
23 concerned about reining in their financing costs. Consistent with this view,
24 the average yearly rise in productivity was only 1.6% during the period 1979-
25 1999, as compared with 2.5%.over the previous period (1958-1978) and
26 2.36% over the next period (2000-2011) (these figures were computed from
27 BLS data taken from series PRS85006091 for the non-farm business sector,
28 which is also the one used for evaluating labour costs in the CEPCI).
29
30 4.2 Further improvements to CEPCI and its estimates
31
32 We found that the accuracy of the correlation over a limited number of years
33 could be excellent when using US prime loans. The following formula was
34 found to predict the CEPCI within less than 1% from 2004 to 2011:
 n

35 CEPCI (n)  0.135  CEPCI (1958)  exp  A   ik   1.64  Poil  107
 k 1959 
36 2004  n  2011 (Eqn. 9).
37 where the values for A were specified when introducing Eqn. (8).
38
39 This is much better that the 5% or so accuracy that can be found from
40 linear interpolation using the price of oil over the same period.

11/24
 1
2 4.3 Accuracy of Eqn. (9) when applied to 2012 data
3
4 2012 data became available after this paper was submitted for
5 publication. Encouragingly, Eqn. (9) still held when applying the 2012
6 values for US prime loan rates and US domestic oil prices to it (3.25% and
7 US$ 86.46, respectively): The computed value was 589.6, compared with
8 the actual value of 584.6, i.e. less than 1% error.
9
10 4.4 Forecasting the CEPCI in the next few years
11
12 For ease of use, Eqn. (9) is rewritten by substituting the value of
 2012

13 0.135  CEPCI (1958)  exp  A   i k   340.7 into Eqn. (9), thus giving
 k 1959 
 n

14 CEPCI (n)  340.7  exp 0.54   i k   1.64  Poil  107
 k  2013 

15 2013  n (Eqn. 10).

16
17 Forecasting the CEPCI using Eqn. (10) requires a forecast of future
18 interest rates and oil prices.
19
20 4.4.1 Forecasting interest rates
21
22 Forecast data on interest rates is available from some institution who
23 developed suitable macro-economic models, e.g. the Financial Forecast
24 Centre (at the time of writing, three year extended forecast can be bought
25 from them for a fairly modest fee of about US$ 30; clients seem to include
26 well known worldwide companies) (Financial Forecast Centre, 2013).
27 However, one can attempt a guess from statements issued by the Federal
28 Reserve.
29
30 US interest rates are strongly influenced by the target federal funds rate
31 set by the Federal Open Market Committee (FOMC), which comprises the
32 seven members of the Board of Governors of the Federal Reserve
33 System and the 12 presidents of the Federal Reserve Banks. In its recent
34 report to Congress, (Federal Open Market Committee, 2013), the FOMC
35 stated that the majority of its members felt that the current economic
36 climate required maintaining the base rate at 0.25% till 2015. In 2015, all
37 but one member suggested a rise in interest rates would occur by the end
38 of the year, the majority of them suggesting that the rates should be set

12/24
 1 between 0.5 and 1.25% by the end of 2015, the median being at 1%. The
2 ‘long term’ value was expected to be set between 3.5% and 4.5% by all
3 but one of the members, with a median value of 4%.
4
5 In practice, the effective rate may not quite match the target rate but
6 should be quite close. In fact, the data for target rates can be
7 painstakingly collated from press releases by the FOMC, and that for
8 effective rates is available in tabulated form from e.g. (Federal Reserve
9 Bank of St. Louis, 2012 c)). We performed a spot check on the period
10 2003-2008 and found that the yearly averages for the two agreed within
11 3% relative difference between 2003 and 2007, although 2008 produced
12 an error of 9% as the financial crisis unfolded (Figure 9). In turn, the prime
13 bank loan rate correlates very well with the effective federal funds rate
14 with a slope of practically 1 and an offset of 3% point, as shown in Figure
15 10 for the period 2000 to 2012 (R2 = 0.9993). In conclusion, it seems
16 reasonable to forecast the prime loan rate by simply adding 3% point to
17 the forecast target federal funds rate.
18
19 4.4.2 Forecasting oil prices
20
21 Oil companies will have their own forecast of oil process which will be
22 necessary for planning their operations, but these are not publicly
23 available. Forecast is also made by governmental and intergovernmental
24 institutions using macroeconomic models, e.g. by the US Energy
25 Information Administration under certain assumptions that include US and
26 global GDP growth as well as monetary policies and a host of other
27 factors. It has also been proposed to use Oil Futures Contracts as
28 forecast for the price of oil (Chinn and Coibion, 2013), since their values
29 represent a compromise between the estimates of those who buy them
30 having factored in the risk that the prices will be higher, and those who sell
31 them having factored in the exact opposite risk.
32
33 4.4.3 Forecasting the CEPCI
34
35 For the purpose of illustrating the application of Eqn. (9) to forecast and
36 sensitivity analysis, we used the 2013 forecast produced by the US Energy
37 Information Administration (EIA) for interest rates and oil prices, under a
38 number of different scenarios:
39 - “Reference” (US real GDP growth is 2.5% p.a. and Brent spot prices
40 rise from US$ 96.81 in 2013 to US$ 163 in 2040 (in 2011 US$))
41 - “High growth” (associated with lower interest rates; as in reference
42 case, but US real GDP growth is 2.9% p.a.)
43 - “Low growth” (associated with higher interest rates; as in reference

13/24
 1 case, but US real GDP growth is 1.9% p.a.)
2 - “High oil price” (as in reference case, but Brent spot prices rise to
3 US$ 237 in 2040 (in 2011 US$), pushed by global growth and tight
4 supply of oil)
5 - “Low oil price” (as in reference case, but Brent spot prices drop to
6 US$ 75 in 2040 (in 2011 US$), depressed by slow global growth and
7 oversupply of oil)
8 (US Energy Information Administration, 2013)
9 Noting that the EIA used the West Texas Intermediate (WTI) benchmark
10 as its reference for crude oil price, we first correlated to it the Illinois Oil
11 and Gas Association (IOGA) which we had used for Eqn. (9). Historical
12 data for the WTI was taken from (BP, 2013), and it was found that the
13 IOGA was obtained by multiplying the WTI by 0.9114 to a very good
14 approximation (R2 = 0.994 between 1976 and 2012). In addition, the oil
15 prices were brought back to current year values using the price index from
16 the same set of EIA data. We also noted that the EIA forecast for effective
17 federal funds rates in the reference scenario was consistent with the
18 FOMC’s recent report to congress (FOMC, 2013) except for 2015 where it
19 was higher. For the sake of using a consistent set of data for oil and rates,
20 we chose to retain the EIA forecast rates, to which we added 3% point to
21 estimate the prime loan rates as described in section 4.4.1. it is worth also
22 mentioning that on inspecting the EIA forecast data, switching between
23 the high and low oil price scenarios resulted in fairly minor changes to the
24 interest rates up to 2020 when compared with the effect of GDP growth,
25 consistent with our finding in section 2.2 that the two parameters were
26 fairly independent from each other.
27
28 In addition to the EIA scenarios, we added a sixth scenario, “Futures”, in
29 which we combined the interest rates from the EIA reference case with the
30 forecast given by the oil future markets (CME group, 2013). A plot of how
31 the oil prices compare for the different scenarios is given in figure 11.
32 Interestingly, the markets seem to “think” that the EIA’s reference forecast
33 is an overestimate, and is in closer agreement with the low oil price
34 scenario.
35
36 Table 2 and Table 3 present the values of the prime loan rates and oil
37 prices, respectively, under the different scenarios. Table 4 and Figure 12
38 present the forecast values of the CEPCI under these different scenarios
39 as predicted by eqn. (10).
40
41 The results clearly indicate that the CEPCI forecast will be strongly
42 influenced by the forecast for the price of oil, with values ranging from 605
43 in the low oil price scenario to 754 in the high oil price scenario, i.e. an
44 increase of 25%. This amount of variation could impact on the economic

14/24
 1 viability of a project. However one could perhaps follow the trend indicated
2 by the oil futures contracts as an indication of the most likely outcome
3 according to the markets (indicated by the dashed line on Figure 12).
4
5 Figure 12 would seem to suggest that interest rates hardly matter when in
6 fact they do: it is just that there is too little variance in interest rates
7 between the scenarios that have been considered here to show any
8 impact. If interest rates were kept at the very low levels where they
9 currently are (3.25% for prime loan), this would remove 40 points from the
10 CEPCI compared with the reference scenario in 2020, as can be found
11 from Eqn. (10).
12
13 4.5 Limitations, enhancements and extensions to this approach
14
15 In spite of the apparent success of this approach in describing the
16 changes in the CEPCI with a very small number of variables, there will
17 remain omitted variables. Although the effect of some of these omitted
18 variables on the index will be partly or wholly transmitted through co-
19 linearity with the used variables, some influence may still remain
20 unaccounted for. Another difficulty of the approach lies in accounting for
21 ‘structural breaks’, i.e. abrupt changes in economic fundamentals such as
22 oil shocks or deep recessions that may affect the values of the constants
23 in the correlation.
24
25 It must also be noted that the actual rates available to the industry may be
26 still different than those indicated here. Markets for the chemical industry
27 are known to be cyclical and volatile, with alternating periods of tight
28 supply and overcapacity – the resulting rates may reflect the associated
29 risk by being even higher than the ones we used.
30
31 More generally, it is important to reflect on the inherent limitations in accuracy
32 of an index like the CEPCI. In recent years, the CEPCI and other indices
33 seem to have struggled to follow very volatile prices. The Association of Cost
34 Engineers (ACostE) announced in 2008 that it would stop publishing updates
35 to its Cost Engineers’ Index pending further review and consultation with
36 members (ACostE, 2008). It stated that ”significant price increases (were)
37 being reported anecdotally from various sources” of a magnitude such that
38 the index “may no longer provide an appropriate guide to changes in the
39 erected costs of process plants in the UK”. In the same year, Hollman and
40 Dysert (2008) also emphasized that real costs were the result of “competitive
41 bidding (usually with few bidders in a tight market)”, and will depart from
42 estimates that are based on government-input measures like the BLS’s
43 PPIs. Thus, they concluded, indices such as the CEPCI are missing out on
44 market intelligence that becomes particularly critical to the pricing of large

15/24
 1 projects which only a limited number of players can deliver, and these indices
2 cannot account for a situation where bidders are few and demand is
3 rocketing up (as it already was in 2008 on the part of the Asian economies).
4 These authors demonstrated the use of a ‘capex market adjustment factor’
5 to correct the CEPCI as a function of the state of the market, an approach
6 that could easily be applied to the CEPCI estimates as obtained by the
7 methods that have been proposed in this paper.
8
9 Ultimately, it is hoped that the method presented in this work should allow
10 fairly straightforward and accurate forecasting of capital costs for industrial
11 process plants. If the relationship that is presented here between the CEPCI,
12 interest rates and the price of oil is upheld in the future, this model could help
13 remove or manage uncertainty on the forecast of capital cost for new
14 projects, since it clearly pegs the CEPCI to the wider economic outlook. One
15 can model how the CEPCI varies depending on forecast oil prices and
16 interest rates, thus informing investment decisions on building or not building
17 a plant.
18
19 The implications concerning chemical plants for the manufacturing of
20 alternative fuels are worthy of interest: while the price of oil sets a benchmark
21 against which the production cost of an alternative fuel must compare
22 favourably, it also impacts on the capital expenditure that is required for
23 building the plant. The evaluation of this type of impact for the techno-
24 economic assessment of such projects was in fact the starting point for this
25 work.
26
27 Finally, it is interesting to note that the method proposed here could be
28 extended to other cost indices.

29 5. Conclusions
30
31 From 1958 to 2011, an effective correlation was
 n 
32 CEPCI (n)  0.135  CEPCI (k o )  exp  A   ik   B  Poil  C (Eqn. 8)
 k  ko 
33 with ko the first year of the period under consideration, ik the interest rate on
34 US bank prime loans in year k, and Poil the price of oil in year n. Best fit was
35 obtained when choosing a distinct set of values of the constants A, B and C
36 for each of the three periods 1958 to 1980; 1981 to 1999; and 2000 to 2011.
37 The error was within ±3% over the whole period (1958-2011). The error was
38 reduced to within ±1% over the period 2004-2011 after readjusting the
39 weighting of the price of oil, compared with ±5% from linear fitting as a sole
40 function of oil prices. The same model also correctly predicted the 2012

16/24
 1 value with the same accuracy. Forecasts were also presented for a range of
2 scenarios, using available forecasts for interest rates and oil prices.
3
4 Future research could focus on developing this approach using actual
5 interest rates as experienced by industry. It will also be interesting to track
6 the robustness of the correlation when the circumstances change, for
7 example rising interest rates, greater competitiveness of bidders from rising
8 world economies, or further improvements in energy efficiency in the process
9 industries as a whole.
10
11 References:
12
13 ACostE, 2008. Cost Engineers’ Index, notice created on 15/05/2008 and
14 posted on http://www.acoste.org.uk/uploads/CEindexNotice.pdf
15 (last visited 05/09/2012)
16
17 Board of Governors of the Federal Reserve System, 2013a). Monetary
18 Policy Report to the Congress, February 26, 2013.
19 http://www.federalreserve.gov/monetarypolicy/mpr_20130226_part3.htm
20 Last accessed 16/07/2013
21
22 Board of Governors of the Federal Reserve System, 2013 b). Monetary
23 Policy Releases,
24 http://www.federalreserve.gov/newsevents/press/monetary/2003monetary.
25 htm
26 Last accessed 16/07/2013
27
28 BP, 2013. Statistical Review of World Energy 2013. Data is available for
29 download on http://www.bp.com/en/global/corporate/about-bp/statistical-
30 review-of-world-energy-2013/review-by-energy-type/oil/oil-prices.html
31 Last accessed, 16/07/2013.
32
33 Bureau of Labor Statistics, 2012. United States Department of Labor,
34 Consumer Price Index History Table:
35 ftp://ftp.bls.gov/pub/special.requests/cpi/cpiai.txt, last visited on 17/08/2012
36
37 Caldwell D.W. and Ortego J.H., 1975. A Method for Forecasting the CE Plant
38 Cost Index, Chemical Engineering, 7: 83-85.
39

17/24
 1 Chemical Engineering, Aug. 2009. Economic Indicators, Chemical
2 Engineering, 116(8): 56-56.
3
4 Chemical Engineering, Jan. 2012. Economic Indicators, Chemical
5 Engineering, 119(1): 56-56.
6
7 Chinn M.D. and Coibion O., 2013. The Predictive content of Commodity
8 Futures, The Journal of Future Markets, in press (doi: 10.1002/fut.21615)
9
10 CME group, 2013. Crude oil financial futures,
11 http://www.cmegroup.com/trading/energy/crude-oil/light-sweet-crude-cash-
12 settled.html
13 Last visited 19/07/2013
14
15 Cran, J., 1976. Cost indices, Engineering and Process Economics, 1: 13-23.
16
17 Earl, P.H., 1977. Specifying Escalation Indexes to Aid in Process Plant Cost
18 Forecasting, Engineering and Process Economics, 2: 281-294.
19
20 Federal Reserve Bank of St. Louis. 2012a)
21 Moody's Seasoned AAA Corporate Bond Yield
22 http://research.stlouisfed.org/fred2/data/AAA.txt
23 last visited on 26/09/2012
24
25 Federal Reserve Bank of St. Louis. 2012 b)
26 Bank Prime Loan Rate
27 http://research.stlouisfed.org/fred2/data/DPRIME.txt
28 last visited on 26/09/2012
29
30 Federal Reserve Bank of St. Louis. 2012 c)
31 Effective Federal Funds Rate
32 http://research.stlouisfed.org/fred2/data/FEDFUNDS.txt
33 last visited on 16/07/2013
34
35 Hollmann J.K., and Dysert L.R., 2007. Escalation Estimation: Working with
36 Economics Consultants, AACE International Transactions, Proceedings of
37 the 51st Annual Meeting in Nashville, Tennessee USA; Association for the
38 Advancement of Cost Engineering International (Publ., Morgantown, WV,
39 USA), EST.01.1-6
40

18/24
 1 Hollmann J.K., and Dysert L.R., 2008. Escalation Estimating: Lessons
2 Learned in Addressing Market Demand, AACE International Transactions;
3 Proceedings of the 52nd Annual Meeting of AACE International and the 6th
4 ICEC World Congress on Cost Engineering, Project Management, and
5 Quantity Surveying; Association for the Advancement of Cost Engineering
6 International (Publ., Morgantown, WV, USA), EST.08.1-7
7
8 Illinois Oil and Gas Association, 2012. Oil prices, 1946-present, reported
9 by InflationData.com,
10 http://inflationdata.com/inflation/inflation_rate/historical_oil_prices_table.as
11 p, last visited on 17/08/2012
12
13 Koop G., 2000. Analysis of Economic Data, John Wiley & Sons Publ.
14
15 Parker H.W., 2008. Cost Shift Signals Changes in Energy Investment, Use,
16 Oil&Gas journal, 106(29): 20-23.
17
18 Remer, D.S., and Gastineau, C., 1981. The role of Interest and Inflation
19 Rates in Present Worth Analysis for Eleven Industrialized Countries of the
20 Free World, Engineering Costs and Production Economics, 5: 255-263
21
22 Styhr Petersen H.J., and Bundgaard-Nielsen M., 1976. A Note on the Use of
23 Cost indices in the Chemical Industry, Engineering and Process Economics,
24 1: 313-315.
25
26 The Institution of Chemical Engineers, 2000. Guide to Capital Cost
27 Estimating, A.M. Gerrard editor, Institution of Chemical Engineers (Publ.,
28 Rugby, U.K.)
29
30 U.S. Energy Information Administration, 2013. Annual Energy Outlook 2013,
31 Data Tables for reference case and for side cases,
32 http://www.eia.gov/forecasts/aeo/tables_ref.cfm
33 http://www.eia.gov/forecasts/aeo/data_side_cases.cfm
34 Last visited 18/07/2013
35
36 Vatavuk W.M., 2002. Updating the CE Plant Cost Index, Chemical
37 Engineering, 109(1): 62-70.
38
39

19/24
 1
2 TABLE 1
3
4 Table 1: Initial attempt at correlating the CEPCI with the selected macro-
5 economic indicators
First
parameter Interest rates Cumulative Price of oil Cumulative
interest rate price of oil
Linear
regression 0.0025 0.958 0.719 0.934
coefficient
2
(R )

7
8 TABLE 2
9
10 Table 2: Prime loan rates (%) in the 2013 Energy Information Administration
11 scenarios (“reference”, “high growth”, “low growth”).

Year Reference High growth Low growth

2013 3.11 3.3 3.06
2014 3.17 3.73 3.17
2015 4.81 6.48 4.83
2016 6.56 6.52 6.83
2017 6.89 6.29 7.44
2018 6.92 6.31 7.76
2019 6.96 6.37 8.2
2020 7.04 6.5 8.52
12

13

14

15

16

17

18

20/24
 1
2 TABLE 3
3
4 Table 3: IOGA, non-deflated oil prices in the 2013 Energy Information
5 Administration scenarios (“reference”, “high price”, “low price”) and in the
6 “futures” scenario. All values are in US $ / bbl.

Year Reference High price Low price Futures

2013 82.78 82.78 82.78 91.15
2014 84.45 109.03 72.69 87.63
2015 85.64 122.3 68.93 80.96
2016 90.06 135.1 67.06 77.01
2017 96.21 142.9 66.09 74.83
2018 100.4 148.9 67.46 73.45
2019 104.7 155.0 68.84 72.86
2020 108.8 161.0 70.27 72.10
7
8 TABLE 4
9
10 Table 4: CEPCI forecast for the different scenarios (“reference”; “high
11 growth”; “low growth”; “high oil price”, “low oil price”; and “Futures”).

Year Reference High Low High Low Futures

growth growth oil price oil price
2013 583.7 584.0 583.6 583.7 583.7 597.4
2014 586.77 588.1 586.6 627.0 567.4 591.9
2015 592.0 597.0 592.0 652.2 564.6 584.3
2016 606.0 610.5 606.4 679.8 568.2 584.6
2017 623.5 627.0 625.0 700.1 574.1 588.4
2018 638.1 640.4 641.3 717.6 584.0 593.8
2019 652.9 654.1 658.8 735.4 594.2 600.8
2020 668.0 668.1 677.1 753.6 604.8 607.8
12

13

14

15 CAPTIONS FOR FIGURES

21/24
 1
2 Figure 1: Yearly % change in CEPCI plotted against yearly averaged
3 prime loan interest rates, from 1958 to 2011.
4
5 Figure 2: Yearly change in price of oil plotted against yearly averaged
6 prime loan interest rates, from 1958 to 2011.
7
8 Figure 3: Yearly % change in unit labour cost plotted against yearly
9 averaged prime loan interest rates, from 1958 to 2011.
10
11 Figure 4: Yearly % change in CEPCI plotted against yearly averaged price
12 of oil, from 1958 to 2011.
13
14 Figure 5: CEPCI plotted against the price of oil. Key: Δ 1958-1980; X
15 1981-1999; ♦ 2000-2011.
16
17 Figure 6: Optimized correlation between the CEPCI minus the cumulative
18 interest rate, and the price of oil from 1958 to 2011 (with parameter value
19  = 0.27 optimized over the whole period, and A optimized for each of the
20 three periods 1958-1978; 1979-1999; 2000-2011). Key: Δ 1958-1980; X
21 1981-1999; ♦ 2000-2011.
22
23 Figure 7: Comparison between the CEPCI (plain, thick line), and the
24 CEPCI reconstructed from the optimized model (symbols). Key: Δ 1958-
25 1980; + 1981-1999; ♦ 2000-2011.

26 Figure 8: % error between the result of applying the model and the CEPCI
27 between 1958 and 2011.

28 Figure 9: correlation between effective and target federal funds interest

29 rates between 2003 and 2008

30 Figure 10: correlation between prime loan interest rates and effective
31 federal funds rate between 2000 and 2012.

32 Figure 11: Forecast of oil prices in the different scenarios. Key: - - -

33 Futures; O Reference; ♦ Low oil prices; ◊ High oil prices.

34 Figure 12: Forecast of CEPCI in the different scenarios. Key: - - -

35 Futures; O Reference; + Low growth; x High growth; ♦ Low oil prices;
36 ◊ High oil prices.
37

38
39

22/24
1

2 FIGURES

20
Yearly % change in CEPCI

15

10

0
0 5 10 15 20
-5
Prime loan interest rate
-10 (yearly average), %
-15 y = 0.5772x - 0.9358
R² = 0.1912

4 Figure 1

40
Yearly change in price of oil,

30
20
10
US $ / bbl

0
-10 0 5 10 15 20
-20 Prime loan interest rate
-30 (yearly average), %
-40
-50 y = 0.0529x + 1.1554
R² = 0.0004

7 Figure 2

23/24
 14 y = 0.6338x - 1.7566
R² = 0.4494
in unit labour cost 12
10
Yearly % change

8
6
4
2
0
-2 0 5 10 15 20
-4 Prime loan interest rate
(yearly average), %

2 Figure 3

0.2
Yearly % change in CEPCI

0.15

0.1

0.05

0
0 20 40 60 80 100
-0.05
Price of oil, US $ /barrel
-0.1

-0.15
y = 0.0004x + 0.0242
R² = 0.0508

5 Figure 4

24/24
1

2 Figure 5

4 Figure 6

25/24
1

2 Figure 7

4 Figure 8

26/24
1

2 Figure 9

4 Figure 10

27/24
1

2 Figure 11

4 Figure 12

28/24