Rijksuniversiteit Groningen
Faculty of Economics
The Induced Innovation Hypothesis:
Evidence of Induced
Innovation in the Automotive
Industry
Svend van der Vlugt (S1199277)
Doctoral Thesis International Economics & Business
August 2007Under supervision of Dhr. Bart Los
This thesis focuses on the Induced Innovation Hypothesis in the Automotive Industry. Assuming that the costs perceived by the consumer act as indirect costs for manufacturers, manufacturer data of car and engine characteristics are used to show
that the basic principles of induced innovation hold. I establish a significant relation between the oil price and innovation done towards more fuel efficient automobiles. This
confirms the notion as set out by Griliches (1957) stating that demand for a certain innovation actually spurs innovation in that area. I also show that innovation towards
fuel efficiency increases vis a vis another form of automotive innovation in times of increasing oil prices, confirming the notion as set out by Ahmad (1966) that innovation
will take place in the area where the relative costs are highest. Further research when taking into account the relative wealth of customers, changing the relative importance of
Table of Contents
1. Preface...3
1.1 Introduction... 3
1.2 The Automotive landscape ... 4
2. Induced innovation, theoretical background ...6
2.1 General theory... 6
2.2 Review of existing comparable studies... 10
2.3 Applicability of current research on the automotive market... 13
3. Key Questions & Hypotheses ...16
4. Data Collection ...19
4.1 Automobile Database... 19
4.2 Selecting cars for the data set... 19
4.3 Oil price data... 22
5. Descriptive analysis ...23
5.1 Segmenting the cars ... 23
5.2 Transforming technical data into useful technical indicators... 24
5.3 Converting technical indicators into growth rates... 26
6. Statistical analysis...31
6.1 Tests being used... 31
6.2 Results testing for hypothesis one, correlation analysis... 32
6.3 Results testing for hypothesis one, linear regression analysis... 34
6.4 Conclusion on hypothesis one ... 36
6.5 Results testing for hypothesis two, correlation analysis. ... 36
6.6. Results testing for hypothesis two, regression analysis... 38
6.7. Conclusion on hypothesis two ... 39
6.8 Testing the third hypothesis ... 40
6.9 Results testing for hypothesis three, correlation analysis... 41
7. Final conclusion...42
7.1. Interpreting combined results... 42
7.2. Recommendations on current research ... 44
7.3. Recommendations for future research... 45
1. Preface
1.1 Introduction
Economics is all about trying to explain why certain economic phenomena happen. What is the cause of a change in the exchange rate, what variables determine market prices, what causes employers to hire more or less people? Scientific economic research revolves around those issues, trying to find an explanation if possible. An interesting subject in the world of economics is innovation. What induces innovation, what determines the type of innovation which is being undertaken by companies? Can we identify the main determinants which spur innovation and possibly use these to make reliable predictions about the future? In this thesis I will try if I can establish relationships between an economic parameter which is always in the center of attention due to it’s importance in the world’s economic and political spectrum, and the innovation of a product which is always at the centre of debate when it comes to
environmental issues: the price of crude oil versus the innovation of automobiles.
The push for development of leaner, cleaner automobiles is continuously gaining momentum as oil reserves are not infinite and environmental pollution becomes a more important issue. This results in increased pressure on manufacturers to produce more fuel efficient vehicles. A part of this pressure is translated through legislation concerning maximum emission levels. In practice, most brands sold in Europe have no issues complying with these regulations, so any extra innovation must be spurred by other factors. This notion is known as the induced
innovation hypothesis.
In this thesis I will investigate if one of those possible factors, the price of crude oil,
influences innovative behavior. Two types of innovation of automobiles will be focused on: fuel efficiency of the car as a whole and the relative power output of the engine, so I can test whether the price of crude oil affects the type of innovation. I will also perform research on whether the type of innovation done for manufacturers aiming at wealthier or less wealthier target groups varies at one specific moment.
Preceding the actual research I will first present a brief overview of the landscape of
innovation will be dealt with in section 2. Section 3 will deal with identifying the key questions and hypotheses of this thesis. Data collection will discussed in section 4, while subsequently section 5 will deal with a descriptive analysis of the data. Statistical analysis and the accompanying conclusions will be presented in sections 6 and 7.
1.2 The Automotive landscape
In the European market a couple dozen of smaller and larger manufacturers are operating. Smaller manufacturers often specialize their car catalogue on a niche in the market. Turnover is relatively low, and because of their size they outsource a lot of the designing and
manufacturing technical components. The highly capital intensive nature of the automotive industry as well as the relatively high market concentration and unionization of their
workforce implies that larger firms have an innovative advantage, and most innovation will be done by the larger manufacturers1. Therefore I need to focus on larger brands who perform
the majority of their innovation in-house, and at the same time have an incentive to be innovative because of tense competition. The market of larger manufacturers can be
characterized by competition in every customer segment. Most manufacturers offer products ranging from small family cars towards larger executive cars2.
It must be noted however that the automotive market has undergone some significant changes in recent years, especially when it comes to market concentration. In the 1990’s we witnessed a large wave of mergers as well as strategic partnerships. Various independent manufacturers sought cooperation although individual brands would still remain3. In some cases this undoubtedly resulted in the pooling of innovative activities. I have to take this into account when interpreting research results later in this thesis. However, the vast majority of brands still operates on a fully independent basis.
In this paper I will also focus on the differences between Asian and European manufacturers with regard to innovation. Although there is not a very broad array of scientific work
available aimed specifically at the automotive industry, some interesting papers do arise
1 Acs & Audretsch 19871 as well as Koeller 1995
2 As we will see later in this thesis, the data set contains 17 options for an entry level family car from different
manufacturers, while some manufacturers were still left out of consideration.
3 Renault & Nissan in 1999, Volvo & Ford in 1999, Ford & Jaguar in 1989 or the Volkswagen – Audi Group
which might serve to explain certain phenomena which I will encounter in the actual research. A few characteristics attributed to manufacturers from certain regions persistently arise in papers dealing with the automobile industry. They mainly have to do with efficiency. It is the Asian automotive industry which is continuously described as more efficient4. This
efficiency advantage results in the Asian manufacturers being more healthy financially. Where European automobile factories are facing tough times, often being supported financially by local governments, Asian manufacturers have improved their processes time over time, resulting in sound financial results. This was mainly done by streamlining
production processes and integration of the vertical supply chain (introduction of the Just-In-Time strategy). This decreased inventories and allows Asian companies to be more flexible, resulting in lower costs. European manufacturers often are a lot more sluggish. Because European industries tend to be less integrated vertically as opposed to the large Japanese conglomerates, they face a bigger challenge when trying to adopt those practices when they strive for a higher efficiency5.
Illustrating the presence of Asian manufacturers in the European market is the increasing penetration rate of Japanese cars on European soil. In 1992, 11% of all cars on the European market were of Japanese descent (the Japanese manufacturers are by far the most dominant force in the Asian segment of manufacturers), and this figure is improving6.
2. Induced innovation, theoretical background
2.1 General theory
My aim is to research the effect of the price of crude oil on automobile innovation as well as the effects of customer preferences on the type of innovation. Through the years a great deal of attention has already been given to induced innovation in scientific literature.
There are several perspectives which try to explain why innovation is taking place, of which three have a prominent place in scientific research7. The first perspective is the demand pull
perspective, and mainly covers the rate of innovation. It basically argues that when there is a high demand for a certain good, innovation for that good will increase8. The supply of knowledge (the pushing force of exogenous scientific research) is considered as being less important for the rate of innovation9.
The second and third perspective deal with factor endowments and the direction in which innovation is heading. The starting point for both is the publication Theory of Wages by Hicks10. It argued that if the cost of capital increased in comparison to the cost of labor, innovation would be focused on reducing the capital share of a production process. This notion became known as the Induced Innovation Hypothesis.
This version of the Induced Innovation Hypothesis was criticized11, stating that an
entrepreneur is not interested in innovating per se, but just wants to decrease costs. This cost decrease might also be achieved by substituting between the use of labor and capital, and that any changes in factor prices wouldn’t necessarily induce innovation. In defense of the
Induced Innovation Hypothesis, further elaboration on Hicks’ theory was done by Kennedy12.
Kennedy argued that Salter’s criticisms could not hold, pointing out that in case of
substitution there would be no constancy in distributive shares, which is the case in advanced real world economies. A theory of induced bias towards different types of innovation along an Innovation Possibility Function should be able to explain this constancy. The Innovation Possibility Function is illustrated graphically in graph 1.
Graph 1. The Innovation Possibility Function
Graph taken from Kennedy, 1964. pp. 545.
The proportion of labor reduction is indicated by p, the proportion of capital reduction is indicated by q. If p or q is negative this implies an increase in the proportion of labor or capital used. It is possible that a change which requires an increase in p and a decrease in q is still an overall improvement which leads to a total cost decrease.
Various arguments against the views of Kennedy were brought forward. It was argued that when interpreting Kennedy’s views on induced innovation, innovation must be seen as exogenous, while in reality firms always face an endogenous tradeoff between innovation and
production13. If this tradeoff is not present and innovation is considered as exogenous,
Kennedy’s notions do not explain what actually induces innovation. Other authors also further elaborated on the lack of relevance between Kennedy’s IPF and the real world14. Examining Kennedy’s model, Binswanger shows that it assumes a constraint on research budgets, fixed research costs and revenues. Binswanger argues that no firm will operate exactly on a frontier which specifies that innovation will take place as long as there are productivity gains. In reality decision makers will face a tradeoff between marginal revenues and marginal costs, and base innovation decisions on maximizing firm profits. Indicating that marginal revenues and marginal costs are the main determinant of innovation also creates a situation which makes Binswanger’s comments suitable for research; these indicators can be measured. This is also relevant for the case of the automobile industry. If marginal revenues and costs are relevant for the characteristics of innovation, it implies that according to Biswanger indirectly consumer behavior influences policy on innovation: consumers influence prices and thus marginal revenues through the mechanism of demand versus supply, consumers behavior has an effect on the profitability of a firm.
The other perspective of Hicks’ notions is the Micro-economic model. The micro-economic model argues that changes in innovative behavior are a result of changes in factor prices. When the cost of a factor of production increases, innovation will take place to counter that movement in order to make that factor of production more economical again. While the ideas of Kennedy with regard to innovation focus on the ratio between capital and labor, in the Micro-Economic model the varying costs of capital and labor are taken into account, implying the inclusion of the interest- and wage-rates15. What this means for innovation is illustrated graphically in graph 2.
Graph 2. Ahmad’s Micro-economic model.
Graph taken from Ahmad, 1966. pp. 349.
At time (n-1) a technical process I(n-1) is in use, optimizing the situation P(n-1). This technical progress was ‘found’ on C(n-1) line, also called the Innovation Possibility Curve. Once I(n-1) is developed, it will shift inward towards the I(n) line. However, if in a
subsequent period the relationship between capital and labor costs changes from P’(n-1) towards P’(n), a new technical process on the Innovation Possibility Curve C(n) needs to be found. This will result in a shift on line C(n) towards process I’(n), indicating a change in innovative preferences
2.2 Review of existing comparable studies
In the past decades a vast amount of research has already been done with regard to induced innovation. I present an overview of empirical research relevant for my thesis in this section. Especially a lot of papers can be found concerning fossil fuel prices, taxes and innovation. I found two key papers which deal with products which are sold in large quantities to a large quantity of businesses or consumers. That implies that I neglected papers which deal with induced innovation in a more general way (e.g. at the factory level; innovation in steel plants16)
The first study we turn to has been conducted by David Popp. Popp17 aims to estimate the effects of energy prices on energy-efficient innovations. Demand side effects are taken into account as well as supply-side factors such as scientific advancements. In order to estimate those effects Popp performs a patent analysis, using U.S. patent data from the period 1970 – 1994. Eleven technology groups related to energy supply and energy demand were identified, and patenting activity was measured for each of those years. Prices of several energy sources related to the several technology groups were also gathered. Also, some other factors
specifically related to some of the eleven technology groups were gathered. With this data, a regression analysis is performed. Energy prices, the current knowledge stack as well as the ‘other factors’ are the independent variables. The ratio between new patents in one of the eleven technology fields versus all new patents is the dependent variable.
Popp comes up with two conclusions: a significant relation between energy prices and the amount of innovation for certain industries somewhat related to the automotive industry can be established. Innovation speeds up when energy prices increase for products such as solar cells, fuel cells and certain types of engines. He also finds that the quality of existing knowledge has a positive effect on innovative activity. What is also important to note is that he mainly establishes a short term effect, citing diminishing returns when the amount of R&D is increased on the long term.
A second paper which revolves around innovation, energy prices and mandates with regard to energy consumption has been written by Stavins, Jaffe & Newell18. An analysis is done on the development of gas water heaters as well as on room air conditioners and central air
conditioners. More specifically, conclusions are drawn about the rate of overall innovation and the direction of innovation. Also the effect of the energy price on the characteristics of the models offered for sale are discussed, as well as the effect of government regulations
(compulsory product labeling indicating energy use). Changes in the characteristics of the model range, when no innovation takes place, is seen as substitution.
Stavins, Jaffe & Newell use the notion of Transformation Surfaces to indicate the specific types of innovation. See graph 3. The line 0 indicates a Transformation Surface at t = 0, while the line 1 indicates a Transformation Surface at t = 1. Movements along a line imply that no innovation takes place, but certain characteristics of a product are substituted for other characteristics (here energy flow versus price). The shift from 0 to 1 indicates innovative activity (the rate of innovation), while the different slopes of 0 and 1 indicate a change in the direction of innovation.
Graph 3. Stavins, Jaffe & Newell’s Transformation Surface.
Graph taken from Jaffe, Newell & Stavins, 1999. pp. 944.
The following data was gathered about gas water heaters and the two types of air
conditioners: product price, energy flow, cooling/heating capacity, the amount of fan speed settings (air conditioners only) and storage capability (gas water heaters only). As a dependent variable the up front price of a unit, corrected for inflation, is used. With this data a regression analysis was performed, the results were used to ‘interpret’ various Transformation Surfaces over time.
No attention is given towards short or long term effects in this paper, but more interestingly a division is made between innovation induced by energy-prices, innovation induced by mandated energy standards, and autonomous innovation. They conclude that a only minority part of the innovations can be related to energy prices or mandated energy standards. A majority is considered as autonomous, which in this case implies that there is no apparent relation between innovation and energy prices or government legislation.
It is interesting to see how Stavins, Jaffe & Newell measure which part of autonomous development goes towards energy saving innovation, and which part doesn’t: “In the early
part of the period, autonomous improvement in these products appears to have been biased away from energy-efficiency. That is, the up-front costs of the products was decreasing faster than their operating costs. But the significant increase in energy prices that occurred in the 1970s and 1980s had noticeable effects, slowing or reversing this process.“19
This notion provides an interesting idea in how to analyze innovation citing decreasing operating costs as a measure of energy price related innovation.
Overlooking the papers which deal with the theory of induced innovation, as well as having looked into papers which actually deal with empirical research into induced innovation, it seems that there is definitely room for further specific empirical research. In the introduction I already mentioned the importance of the automotive industry in the current debate concerning a cleaner environment. Energy prices are frequently used as a factor when researching
induced innovation. The automotive industry, despite it’s importance in the world economy, has however not yet been the subject of research with regard to the notions about induced innovation in combination with energy prices. As I showed in section 1.2 there is literature
available describing especially the differences in manufacturing process innovations.
Scientific work on innovation of product characteristics seems to be largely absent. This gap provides plenty of opportunity to check whether evidence of induced innovation as a result of changing energy prices can be found in the automotive industry.
2.3 Applicability of current research on the automotive market
Both papers mentioned in section 2.2 analyze the effect of rising energy prices on innovation, albeit both use a different approach. David Popp use a patent analysis, while Stavins, Jaffe and Newell focus on the actual product characteristics to gather data about innovative progress. Both papers use a regression analysis in order to measure the influence of the various parameters. If I want to research innovation in the automotive industry I have to determine what data will give me the most reliable research results. Let’s turn to patents first. The main advantage of using patents, at least at first glance, is that patents are classified according to strict rules. This, at least in theory, allows you to make a clear division between patents intended to increase the fuel efficiency of an automobile as opposed to patents which for example increase safety or decrease production costs. In practice, even when being able to search on specific classifications, it turned out that it is very difficult to discern between patents aimed at increasing the efficiency of fuel usage and patents aimed at other
characteristics of a car. I ran a series of analysis at the Esp@cenet20 patent database which holds over 50 million patents. It turned out that there are thousands of patents available for certain aspects of combustion engines alone. Unfortunately using the various classifications it is not possible to isolate innovations which are made in response to for example rising energy prices. Take the category ‘Controlling the engine output power by varying inlet or exhaust
valve operating characteristic’. Over the complete time period available in the database this
category renders over 6000 patents. Most of them will to some extent increase the efficiency of the engine, however also a lot of patents in the same category can be found which clearly don’t. For example, patents which describe the way the engine is controlled by the accelerator pedal fall under the same category.
20 http://ep.espacenet.com. This is the patent database of the European Patent Office, which is an organ of the
What I also should take into account is that the nature of patenting seems to change. Over time, an equal amount of R&D expenditure apparently leads to less patenting. However, the patents are of higher quality, meaning that more actual inventive output is being associated to a patent. This might distort any data derived from a patent analysis 21. Also the few mergers which can be witnessed in the automotive industry will change the absolute amount of new patents. It is unlikely that two merging companies will just add up their R&D budgets and spend as much on research together as they used to when they were separate companies. Even if they do, (dis)economies of scale will increase or decrease the absolute amount of patents under the assumption of added up budgets, although it is hard to test in which direction this effect works22. Therefore applying patent analysis on specific innovations in the automotive
industry is not a feasible approach.
I therefore opt to create a data set using the method used by Stavins, Jaffe & Newell, which implies focusing on practical outcome of those patents; the actual efficiency performance of automobiles. It should be fairly easy to construct a data-set which holds the actual
performance of combustion engines using the amount of fuel used per kilometer as well as indicators such as the power output of an engine.
I then however still need a variable for which I can test a possible innovative trend against. As mentioned earlier, Stavins, Jaffe & Newell use the operating costs (which can be seen here as the amount of fuel used per kilometer combined with the price of fuel) as an indication of innovation on the environmentally innovative front, and they use the up-front costs of buying a certain unit as a measure of other innovation (such as more efficient ways of producing). This method will probably work for the automobile industry as well. I run into another problem here. In the case of air conditioners and gas water-heaters the characteristics of the product remain largely the same over a longer period of time. Cars however evolve a lot over time. Not only luxury characteristics such as a more spacious interior, sunroofs and air-conditioning have become the norm, but also other technical innovations such as electronic engine management, ABS and Traction Control are nowadays more or less standard. These features come at a price, and correcting for them would be worth a thesis on it’s own, as has been done in the past23.
21 See Griliches, 1990. 22 See Fisher & Termin, 1973
I therefore need to look into another indicator to which I can compare the innovation in terms of operating costs. Apart from fuel efficiency, an important feature of a car/engine
combination is the power output24. The tradeoff between more output power and higher fuel efficiency is an interesting one. In times of rising energy prices, the induced innovation hypothesis should result in more innovative development for the fuel efficiency of engines, and less on the power output of engines. The reasoning behind this is the following. If the price of a factor becomes too high, the price for the consumer becomes too high, eroding any competitive advantage which a company might have. It is in the end the consumer’s demand which steers what innovation takes place. It the perceived cost of a durable good, such as a car, is too high (this includes the up-front sale price, but also includes the operating costs, being the price of oil), a customer will look for a cheaper alternative25. Therefore the oil price must be seen as an indirect cost for a manufacturer. When the indirect cost (the oil price) increases, a larger innovation effort will go towards fuel efficiency, and a smaller innovation effort should go towards power output efficiency.
What I also have to take into account is that innovation does not take place instantaneously. New models and types are introduced every few years. New developments are not created at the flick of a finger. When doing statistical analysis I must take into account that it takes some time before an innovation initiated a few years ago finally hits the market. I expect to see some lag here. Also, since the data gathered per brand might not necessarily change each year, some data smoothing might be needed in order to find a general trend.
3. Key Questions & Hypotheses
The theoretical framework in which I’m going to conduct my research has now been established. I have explained three different mechanisms through which induced innovation takes place. I have explained how I expect the notion of induced innovation to have an effect on automotive innovations. I have identified the relevant parameters, being the oil price, engine power output and automobile fuel efficiency. Within this framework I can ask myself the following questions:
1) Do changes in the oil price affect the development of the fuel efficiency of an automobile?
I explained above how changes in the oil price should affect innovation into more fuel efficiency. If the operating costs are too high, then the part of the production process which is the engine places a too high burden on the perceived retail price for a customer. Following the theory of induced innovation an increased effort to make the engine more fuel efficient will then be made. This leads to the first hypothesis:
“An increase in the price of oil will have a positive effect on the growth of fuel efficiency of automobiles”.
In order to be able to make sensible conclusions when testing for the first hypothesis I need to test for another parameter which gives an indication of innovation. Even if the first hypothesis turns out to hold, it is impossible to say whether only this specific type of innovation changed as a result in changes of the oil price, or if other types of innovation increase at the same time. The induced innovation hypothesis argues that an increase in the oil price will shift focus towards fuel efficiency development and away from other forms of development. Therefore I need to compare any changes in the fuel efficiency against a second indicator. The question I’m asking myself is;
As discussed earlier, the two main characteristics of an engine are it’s output power and it’s fuel efficiency. The easiest way to create a more fuel efficient engine is to sacrifice on it’s output power. Injecting less fuel in an otherwise equal engine implies a lower power output. Less innovation on power output will be done in times of increasing oil prices. Therefore we expect a negative relation between the oil price and the relative power output of an
automobile’s engine. This leads to the second hypothesis:
“An increase in the price of oil will have a negative effect on the growth of relative power output of automobile’s engines.”
In case the second hypothesis does not hold, what does this mean with regard to the
development of the output power of automobile’s engines? When I find no significant relation I can conclude that the development of an automobile’s relative output power is independent of oil price growth rates. In case I discover a positive relation between the oil price and output power there must be other forces influencing development, or my assumption that
manufacturers differ between specific types of engine development is wrong. While there is no direct incentive to increase output power in times of increasing oil prices (a customer will prefer less fuel costs, the induced innovation hypothesis therefore indicates a strong
preference towards fuel efficiency and not higher relative power output), some other mechanisms might indeed cause an increase in the relative power output of an automobile’s engine. Perhaps decision makers at automobile manufacturers shift attention towards engine development in general (versus for example innovation on safer or more comfortable cars) in times of rising oil prices. They might not differ between fuel efficiency and power output efficiency but just increase innovation budgets for the engine department as a whole.
The induced innovation hypothesis claims that a change in the relative price of the factors of production pushes innovation into a certain direction. When constructing the first and second hypothesis I assumed that the manufacturers were all approached by one and the same virtual customer and that his preferences and response to oil price changes was equal for all
segments with the lower average retail prices (Asian and cheap European) will attract less affluent customers.
For more affluent customers fuel costs, assuming they spend as much on fuel per year as less affluent customers, will comprise a smaller part of their shopping basket. An increase in the oil price will therefore result in a relatively smaller burden on their total budget. For less affluent customers, an oil price increase will result in a relatively larger burden on their budget. According to the notions of Ahmad I discussed earlier, when the relative cost of a factor of production increases, innovation for that factor will increase. However, for a more affluent customer the ‘factor’ oil price is relatively less important when compared to the total costs of buying and operating a car. Therefore, in theory, innovation of expensive cars should be less inclined to innovate specifically to increase fuel efficiency. Instead of increasing the fuel efficiency (km/L) and paying less attention towards output efficiency (kW/cm3),
automotive development in general will be done, not favouring increasing km/L or kW/cm3 at any time, or at least not changing their favouritism too much. Therefore I expect
manufacturers of cheaper mobiles to differ between innovation in fuel efficiency and output efficiency in times of rising oil prices, whereas manufacturers of expensive automobiles focus more on engine development in general. There is simply less pressure from their customers to focus on fuel efficiency only. The third key question is therefore:
3) Is there a difference in the innovative behaviour when comparing cheaper brands to more expensive brands?
This leads to the final hypothesis:
“For more expensive automobiles there will be a stronger relation between the growth of fuel efficiency and the growth of output efficiency when compared to cheaper automobiles.”
4. Data Collection
4.1 Automobile Database
In order to test for my proposed hypotheses I needed to gather a uniform dataset covering most automotive brands containing information such as fuel efficiency, power output, engine capacity, price etcetera. I was able to find a database which fulfilled those requirements; the Carbase-database from the Dutch Autoweek Magazine26. This database contains information on most brands sold on the Dutch market. Every car which was brought to the market by those brands in the period 1982 – 2006 is listed. Even more or less equal cars but with slightly differing specifications are listed separately, think for example of cars with a slightly different engine capacity or cars with the same capacity but with a different package of extra’s. The technical data is provided by manufacturers themselves. This does cause some issues with the reliability of the data. There is however no other database readily available which is just as extensive as Autoweek Carbase that contains data from an independent entity.
4.2 Selecting cars for the data set
In order to keep my dataset as uniform as possible I had to choose a certain type of car which was sold by most if not all manufacturers. When comparing development behavior of
different manufacturers it doesn’t make a whole lot of sense to compare for example a Peugeot 106 to a BMW 7-series. Both cars are aimed towards a different target audience in terms of use. The BMW 7-series is clearly an executive car, while the Peugeot 106 is aimed at small commuting households traveling short distances. Since the influence of changes in the oil price on demand for cars must be equal for all manufacturers it is paramount to pick a car which can be bought by one uniform customer searching for an equal set of features which a car offers. The cars in my data-set must be designed for more or less the same use. Expensive brands often don’t sell the smallest type of car (the earlier mentioned Peugeot 106), small manufacturers often don’t sell large executive cars (such as the earlier mentioned BMW 7-series). It turned out that the 17 manufacturers which I selected in my sample all sold an entry-level or small-sized family car in every year for the period 1982-2006 27. This car is characterized by being either a 4-door sedan of a 5-door hatchback and it is, apart from an
exception here and there where a slightly larger model was chosen, the smallest 4-door sedan or 5-door hatchback available.
After having established the brands which would be included in my database as well as the type of car I needed to determine what exact model would be included in my data-set. I took the following steps in order to select a certain car for a certain year:
- First, in order to pinpoint the latest and most recent developments done by a manufacturer, in a certain year I would only ‘shop’ for the newest model available. Example: In 1998 Ford introduced the Focus-range. The Escort-range was slowly being phased out, but was sold for another few years after the introduction of the Focus. I can expect the latest development work to be present in the Focus-range, since the Escort-range is in effect becoming obsolete, so I pick a certain type of car from the Focus range. See table 1 for an overview of brands and models selected.
Table 1.
Brands, and models per brand selected.
Brand Type / number of doors / year of introduction into dataset Toyota Corolla / 4D / 1982
Honda Accord / 4D / 1982
Mazda 626 / 4D / 1982 6 / 4D / 2002
Mitshubishi Lancer / 4D / 1982
Suzuki 28 Alto / 4D / 1982 Swift / 4D / 1984
Nissan Cherry / 5D / 1982 Sunny / 4D / 1986 Almera / 4D / 1995 Opel Kadett / 5D / 1982 Astra / 5D / 1991
Ford Escort / 5D / 1982 Focus / 5D / 1998
Volkswagen Golf / 5D / 1982
Peugeot 305 / 4D / 1982 309 / 4D / 1987 306 / 5D / 1993 307 / 5D / 2001
Renault 9 / 4D /1982 11 / 5D / 1983 19 / 5D / 1988 Megane / 5D / 1996
Alfa Romeo Alfasud / 5D / 1982 33 / 5D / 1983 146 / 5D / 1995 147 / 5D / 2001 Mercedes29 200 / 4D / 1982 190 / 4D / 1983 C-class / 4D / 1993 BMW 3-series / 4D / 1982 Audi 80 / 4D / 1982 A4 / 4D / 1995 Volvo 345 / 4D / 1982 360 / 4D / 1983 S40 / 4D / 1996 Jaguar XJ / 4D / 1982
28 For Suzuki I switched to the Swift-model from 1984 onwards. I considered the Alto as too small to be suitable
for sensible comparison towards other types of cars.
29 For 1981-1982 I picked the larger 200 model. From 1983 until the introduction of the C-class the smaller 190
- Secondly, in order to pinpoint the latest and most recent developments done by a manufacturer, always the latest series of a certain model was chosen. Example: Volkswagen started off with the Golf I, then came the Golf II, then the Golf III and so on. Since I can expect the latest development work to be present in the latest
evolution, a car from the latest series was always picked despite older versions sometimes still being sold on the market.
- Thirdly, effects of oil prices on development work can most visible be seen in the most fuel efficient model on sale. Therefore, after having taken step 1 and 2, the type with the best combined fuel efficiency (L/100km) was chosen30.
- Fourth of all, when there were similar types with similar fuel efficiency available, the cheapest type (judging only on the upfront sale price) was chosen. The sale price does not have a lot to do with fuel efficiency, but it is of use when putting the cars into segments.
This work yielded a data set containing data from 17 manufacturers representing the latest, most fuel efficient, entry level family car (size-wise) containing data such as the year when the car was released to the market, engine capacity (cm3), combined fuel efficiency
(L/100km), maximum engine power (kW) and price over the period 1982-2006. Data for 17 manufacturers for a period 25 years should imply that I had 425 observations. It turned out that for 5 of those 425 observations I had insufficient data available caused by a gap in the availability of a certain car. If in a certain year no suitable car was available on the market for a certain brand which met the requirements discussed above (for example, no 4-door sedan of 5-door hatchback was sold by a specific brand in a specific year), data from the last year in which a suitable car was available for that brand was used for that certain year31.
30 The Autoweek Carbase database defines the combined fuel efficiency as 37,5% of the city fuel efficiency and
62,5% of the highway fuel efficiency.
4.3 Oil price data
Data on historical oil prices was also gathered32. The yearly averages I gathered are based on
monthly averages. The monthly averages are based on the average price per barrel of crude oil sold in a certain month. The nominal yearly average oil price indicated in dollars per barrel was corrected for inflation using the Consumer Price Index for all households in the Netherlands33. Although it would have been more logical to use EU data (since car
manufacturers won’t base their innovation policies on a single country), it turned out to be more useful to use Dutch CPI data because of it’s consistency. The EU data is based on continuously changing circumstances (such as the ever increasing amount of countries being part of the EU). Also, the data for the Dutch CPI was gathered conform the same definition for the whole period which was covered in my sample for oil prices. Data for the EU Consumer Price Index was not presented according to a uniform definition over time. The monthly data indicating the one-year increase or decrease in prices was transformed to a yearly average inflation rate. They were in turn used to correct the oil price towards real values. If I wouldn’t have applied this correction, oil price growth rates might have been exaggerated for some years while oil actually became relatively cheaper when compared to other products. It is of course the price of oil versus other consumer products sold at that time which triggers certain consumer behavior (and therefore, as I argued, manufacturer behavior).
32 Source US DOE/ www.economagic.com and www.imperialoil.com
5. Descriptive analysis
5.1 Segmenting the cars
From the third hypothesis it shows that I want to investigate differences in response between manufacturers of cheaper and more expensive automobiles. In order do determine which manufacturer could be classed as ‘cheap’ and which one could be indicated as being ‘expensive’ I calculated the mean price per brand over a period of 25 years of the selected entry-level or small-sized family cars. This leads to the segments as described in table 2. The Asian segment holds six cars, the European segment holds eleven cars. To ensure that the amount of data from my dataset would be divided between different segments as equal as possible in order to attain an as equal as possible situation when conducting statistical analysis, I maximized the amount of cars in each segment. When deciding which one of the two European segments would hold six instead of five cars I opted to choose the cheap segment for that. In doing so it offered the best fit price-wise with the already 6-car Asian segment. The difference between the mean value of those segments is only 552,43 Euro, when comparing either segment to the European expensive segment the latter is far more expensive.
Table 2.
Brand segmentation. Prices in Euro’s.
Brand Mean Price in
Euro’s Segment
Jaguar 59512,56 European Exp.
Mercedes 27287,72 European Exp.
BMW 22612,92 European Exp.
Audi 20653,96 European Exp.
Honda 18545,56 Asian
Volvo 17420,04 European Exp.
Mazda 16475,40 Asian
Volkswagen 14377,20 European Cheap
Alfa Romeo 13709,04 European Cheap
Toyota 13635 Asian
Nissan 13385,56 Asian
Renault 13189,52 European Cheap
Ford 13236,12 European Cheap
Opel 13220,8 European Cheap
Peugeot 13008,04 European Cheap
Mitshubishi 12754,39 Asian
Suzuki 9259,4348 Asian
The unweighted mean price of the European expensive segment is 29497,44 Euro, for the European cheap segment is is 14009,22 Euro and for the Asian segment it is 13456,79 Euro. Apart from the European cheap segment I can also characterize the Asian segment as being cheap, since it has an even lower mean price. The only outlier here is Honda, which is more expensive on average than Volvo, which resides in the European expensive segment.
5.2 Transforming technical data into useful technical indicators.
With a sizable dataset available now, divided into three different segments, I had to transform the data so I could put it to good use doing statistical analysis. From the hypotheses it turns out that I need to develop two indicators: one for fuel efficiency (how far can I travel on a given amount of input) and one for output efficiency (how much power does an engine generate on a given amount of input). Combined fuel efficiency data (L/100km) was easily transformed to km/L (kilometers driven per liter of fuel). In this way an increase in efficiency would imply a positive change in the fuel efficiency indicator. A positive change indicates a benefit for the customer, since he is able to travel more kilometers on the same amount of fuel.
For the second indicator (how much power does an engine generate on a given amount of input) it took some more contemplating when trying to come up with a sensible solution. Engine output in kilowatts (kW) is highly sensible to for example the engine capacity (in cm3), but also to the electronic management of the engine and the construction of the engine. Improvements to either the construction of an engine or the management of electronics are clearly innovations. There is however not much innovation going on when one simply
The most suitable one turned out to be power density. Power density is defined as a measure
of the power-carrying capability of a mechanical component. It is the ratio of power capability to volume and is expressed in units of power per volume34. As a result, when
innovations increase the power output of an engine, while keeping the engine capacity equal, our indicator will show an increase. Again, an increase in this ratio is beneficial for the customer looking for more power from an equal capacity engine.
It has to be noted that no indicator to express power output for a given amount of input is perfect. Power density might show a positive growth rate even when there is no innovation done. In the case of an engine which has a very lean fuel usage, it’s power output might be increased by just allowing more fuel into the engine. If this is the case, the ratio for power density will show a positive growth rate, while the growth rate for fuel efficiency will be negative35.
Also, engine development inhibits increasing returns to scale with regard to engine capacity. A 100kW 1000cm3 engine might actually be more developed than a 200kW 2000cm3 engine, yet they show the same power density. What follows is that when the power density is
increased by a factor 1,1 (resulting in 110kW and 220kW engines), this does not per se imply that it took the same development effort.
There are more accurate indicators available which describe fuel efficiency and output efficiency at the same time, and therefore eliminate some of the issues mentioned above. An example is Specific Fuel Consumption36. This indicator involves the creation of fuel maps: for each amount of power output the fuel efficiency usage should be known. No sufficient data is readily available to calculate this indicator for the amount of cars I intend to use in my data set. Also, the creation of such a data set goes well beyond the scope of this thesis since it involves a vastly higher amount of modelling technical models and processing technical data. For the economic scope of this thesis the km/L and kW/cm3 ratios will be sufficient.
34 In numbers: This meant dividing the maximal power value (in kW) by the engine capacity (cm3) times 1000
(to get the amount of power per 1 litre or 1000cm3 of engine cylinder capacity). In this thesis I will refer to Power Density as kW/cm3.
35 As we will see in my research results this will not be the case. Significant responses for power density and fuel
efficiency do not show opposite signs.
36 Specific Fuel Consumption is defined as the amount of fuel needed to produce a certain output over a fixed
5.3 Converting technical indicators into growth rates.
The indicators km/L and kW/cm3 were grouped per year according to the segmentation described above. Per segment and for the sample of all cars as a whole the mean value of an indicator for a certain year was calculated. From the mean values I distilled the growth rates from year to year. This resulted in some pretty volatile results when graphing the growth rate of the indicators per segment over a period of time. Taking into account that every segment contains five or six cars, and that the development cycle of a new car or engine is a few years, I also decided to calculate a three year moving average in order to smooth the data. In other words, the three year moving average segment growth rate for 1990 consisted out of the yearly segment growth rate from 1989, 1990 and 1991, all of equal weight. The growth rates using three year moving averages were indeed less volatile.
Graph 4. Growth rate of the km/L ratio over time. -0,03 -0,02 -0,01 0 0,01 0,02 0,03 0,04 0,05 0,06 0,07 19 83 19 84 19 85 19 86 19 87 19 88 19 89 19 90 19 91 19 92 19 93 19 94 19 95 19 96 19 97 19 98 19 99 20 00 20 01 20 02 20 03 20 04 20 05 20 06
X-axis: year. Y-axis: km/L growth rate. One year moving averages (dark line) and three year moving averages (lighter line).
Graph 5. Growth rate of the kW/cm3 ratio over time. -0,01 -0,005 0 0,005 0,01 0,015 0,02 0,025 0,03 0,035 0,04 19 83 19 84 19 85 19 86 19 87 19 88 19 89 19 90 19 91 19 92 19 93 19 94 19 95 19 96 19 97 19 98 19 99 20 00 20 01 20 02 20 03 20 04 20 05 20 06
X-axis: year. Y-axis: kW/cm3 growth rate. One year moving averages (dark line) and three year moving averages (lighter line).
Graph 6. Growth rate of the oil price over time. -0,6 -0,4 -0,2 0 0,2 0,4 0,6 0,8 19 78 19 80 19 82 19 84 19 86 19 88 19 90 19 92 19 94 19 96 19 98 20 00 20 02 20 04 20 06
X-axis: year. Y-axis: Oil price growth rate. One year moving averages (dark line) and three year moving averages (lighter line).
It is evident from graph 6 that some periods were characterised by a sustained growth of the oil price, while other periods were marked by continuous decline of the oil price. When looking at the three year moving averages, four periods in which the oil prise showed an increase in three subsequent years 37 can be identified. The first period, from 1979 till 1981, showed an increase due to instability in the Middle East and especially Iran. The Iranian Revolution resulted in less oil being exported, decreasing the overall amount of oil supplied to the market. Although the OPEC tried to offset the loss of Iranian production, it took some time before oil prices returned to their pre-1979 levels.
The second period of sustained oil price growth can be identified in the period 1988-1990. This was a direct result of the first Gulf War in which Iraq invaded Kuwait and the tensions preceding it. A lot of damage was done to the Kuwait oil industry, again diminishing supply to the market, leading to increasing prices.
The third period of sustained growth is from 1999 till 2001. It has to be noted that one year showing a severe decrease in the price of oil (1998) prevented a longer period of sustained growth. The dip in 1998 coincides with the Asian financial crisis, which resulted in a short term drop in demand for oil. For the other years of the period 1995 till 2001 we witnessed a steady growth of the oil price, probably as a result of a booming economy in most parts of the western world.
6. Statistical analysis
6.1 Tests being used
In order to test if the first two hypotheses hold I decided to do two statistical tests. In order to test for any significant correlations between the growth rate of the oil price (independent variable) and the growth rate of the km/L-ratio and the kW/cm3 ratio (dependent variables) I calculated the Pearson Correlation Coefficient between these independent and dependent variables for each of the 3 automobile segments. The tests were accompanied by a 2 tailed significance test. I tested for both the data-set using yearly average growth rates, as well as the data-set using 3 year moving average growth rates. Since I also want to measure the effects over time I also tested for correlations between the oil price growth rate in a certain year and the ‘development’ growth rates 1 year later, 2 years later, 3 years later, 4 years later and 5 years later.
If a correlation is established, I also want to know how the variables are related. In other words, I want to know the size of the effect the independent variable has on the dependent variable. For this goal I used a simple linear regression model. Significance was tested by means of a T-test. I tested for both the data-set using yearly average growth rates, as well as the data-set using 3 year moving average growth rates. It has to be noted that in most cases, if a significant correlation is established, a regression analysis will also often show highly significant results.
In order to be able to interpret the results from the linear regression analysis I’ll show the model first;
1)
G
x= C + B : (oil pricegrowth rate)
x+ f
xG( ) is the growth rate of either the km/L ratio or the kW/cm3 ratio in the period ( ). It indicates the following relation (km/L can be substituted for kW/cm3):
2)
(km/L)
x
C is the constant. The constant is the change of G irrespective of changes in the oil price growth rate. If the oil price growth rate is zero, there still is innovation, indicated by the value of the constant. B indicates the strength of the effect of the oil price growth rate on G, and ( ) is a random error term. Furthermore, we assume the strength of the relation to be constant over time.
For the sake of testing for lagged development, I didn’t only test the segment-year
combinations which resulted in significant correlations, but I tested for all the combinations of the oil price growth rate in a certain year and the ‘development’ growth rates 1 year later, 2 years later, 3 years later, 4 years later and 5 years later.
6.2 Results testing for hypothesis one, correlation analysis.
The first hypothesis states that I expect ‘an increase in the price of oil will have a positive
effect on the fuel efficiency of automobiles’. This implies that I expect to see a positive and
significant correlation between the growth rate of the oil price and the growth rate of km/L-ratio. Taking into account that the development cycle of a new car or engine lasts at least a few years I noted earlier that I expect to see some sort of lag before the growth rates start to correlate. First I present the results of the Pearson Correlation Coefficient test using two tables, with the results discussed after each table. After that I’ll move to the regression analysis.
Table 3.
Km/L & Oil price growth rate. Pearson Correlation Coefficient – 1 year average growth rates. N = 24
Lag all cars asian eurcheap eurexpensive
No lag -,238 -,184 -,377 ,087 1 year lag -,128 -,149 -,095 -,044 2 year lag ,122 -,298 ,153 ,358 3 year lag ,604** ,466* ,390 ,472* 4 year lag ,564** ,636** ,422* ,181 5 year lag ,019 ,215 ,037 -,191
** Correlation is significant at the 0.01 level (2-tailed). * Correlation is significant at the 0.05 level (2-tailed). All significant values are highlighted in bold.
Asian manufacturers as well as the more expensive manufacturers in the European markets show a medium correlation38(0,466 and 0,472 respectively) which is significant at the 0,05 level. When taking into account a lag of 4 years, Asian manufacturers show a strong
correlation (0,636) which is significant at the 0,01 level. The cheaper European manufacturers show a medium correlation of 0,422 significant at the 0,05 level. The sample containing all cars shows a strong correlation of 0,564.
Table 4.
Km/L & Oil price growth rate. Pearson Correlation Coefficient – 3 year average growth rates. N = 22
Lag all cars asian eurcheap eurexpensive
No lag -,132 -,488* -,126 ,399 1 year lag -,070 -,374 -,133 ,434* 2 year lag ,279 -,110 ,162 ,614** 3 year lag ,779** ,492* ,595** ,586** 4 year lag ,778** ,772** ,540* ,320 5 year lag ,454* ,715** ,265 -,060
** Correlation is significant at the 0.01 level (2-tailed). * Correlation is significant at the 0.05 level (2-tailed). All significant values are highlighted in bold.
I argued earlier that growth rates on a yearly basis looked pretty volatile because the five or six manufacturers in a certain segment only release new cars or engines every few years. Therefore I smoothed out the data using a three year moving averages. While I discovered some significant results when analyzing with yearly averages, even more significant results arise when using the three year average growth rates. There seems to be one outlier though. When testing with a three year moving average without any lag I found a negative correlation of -0,518 at the 0,05 significance level for Asian. When testing with one year of lag I obtain our first positive significant result. The expensive European manufacturers show a significant medium correlation coefficient of 0,434 and it is significant at the 0,05 level. In the second lagged year there is a strong correlation of 0,715 which is significant at the 0,01 level. Just as with testing with the data using yearly averages, I get the most significant results with the three year moving averages when taking into account a lag of three years. All segments including the complete sample show a positive correlation coefficient between 0,492 and 0,779, and they’re all significant at at least the 0,01 level. When adding another year of lag
38 I judge a correlation between 0,1 and 0,29 to be weak, a correlation between 0,3 and 0,49 and medium, and
(four years in total), the Asian segment, the cheaper European segment and the sample as a whole still show a strong correlation.
With a lag of five years only the Asian manufacturers still show a significant strong correlation of 0,688 (significant at the 0,01 level). Also the sample as whole still shows a significant medium correlation of 0,454.
6.3 Results testing for hypothesis one, linear regression analysis.
Now I turn to the linear regression analysis for hypothesis one. I again highlighted the significant values in bold.
Table 5.
Km/L & Oil price growth rate. Linear Regression – 1 year average growth rates. N = 24
All cars Asian Eurcheap Eurexpensive
B Std. Error B Std. Error B Std. Error B Std. Error
No lag (Constant) 0,06 ,003 ,005 ,004 ,007 ,005 ,006 ,006 oilgrowthrate -,014 ,013 -,012 ,014 -,033 ,017 ,009 ,022 1y Lag (Constant) ,005 ,005 ,004 ,019 ,006 ,005 ,006 ,006 oilgrowthrate -,008 -,010 ,014 ,018 -,008 ,019 -,005 ,023 2y Lag (Constant) ,005 ,005 ,003 ,018 ,005 ,005 ,005 ,006 oilgrowthrate ,007 -,019 ,013 ,016 ,013 ,018 ,037 ,020 3y Lag (Constant) ,004 ,003 ,003 ,017 ,004 ,005 ,004 ,005 oilgrowthrate ,034** ,028 ,011* ,016 ,032 ,016 ,046* ,018 4y Lag (Constant) ,004 ,003 ,003 ,003 ,004 ,005 ,005 ,006 oilgrowthrate ,031** ,028 ,037** ,010 ,034* ,016 ,017 ,020 5y Lag (Constant) ,005 ,003 ,004 ,004 ,005 ,005 ,007 ,006 oilgrowthrate ,001 ,012 ,013 ,012 ,003 ,017 -,018 ,020
** Correlation is significant at the 0.01 level (2-tailed). * Correlation is significant at the 0.05 level (2-tailed). All significant values are highlighted in bold.
When using the one year average growth rates the only significant B for the expensive
Table 6.
Km/L & Oil price growth rate. Linear Regression – 3 year average growth rates. N = 22
All cars Asian Eurcheap Eurexpensive
B Std. Error B Std. Error B Std. Error B Std. Error
No lag (Constant) ,004 ,002 ,005 ,002 ,004 ,003 ,002 ,003 oilgrowthrate -,009 ,014 -,040* ,016 -,012 ,020 ,041 ,021 1y Lag (Constant) ,004 ,002 ,004 ,003 ,004 ,003 ,003 ,003 oilgrowthrate -,005 ,015 -,032 ,018 -,013 ,021 ,047* ,022 2y Lag (Constant) ,003 ,002 ,004 ,003 ,003 ,003 ,003 ,003 oilgrowthrate ,020 ,016 -,010 ,021 ,017 ,072 ,021** ,020 3y Lag (Constant) ,003 ,001 ,003 ,002 ,003 ,002 ,003 ,003 oilgrowthrate ,055** ,010 ,044* ,017 ,059** ,018 ,066** ,020 4y Lag (Constant) ,002 ,001 ,001 ,002 ,002 ,003 ,003 ,003 oilgrowthrate ,046** ,008 ,057** ,011 ,045** ,016 ,030 ,020 5y Lag (Constant) ,002 ,002 ,001 ,002 ,003 ,003 ,004 ,003 oilgrowthrate ,025* ,011 ,050** ,011 ,021 ,017 -,005 ,020
** Correlation is significant at the 0.01 level (2-tailed). * Correlation is significant at the 0.05 level (2-tailed). All significant values are highlighted in bold.
Table 6 shows the same analysis, now using three year moving averages. Using the three year moving averages the strength of the value of B is less varied. If I take the mean values of the significant B’s for every segment and the sample as a whole I end up with a mean B for the whole segment of 0,042. The mean beta for the Asian segment is 0,050. For the cheaper European segment the mean value of B is 0,052. For the expensive European segment it is 0,044. Here the European expensive segment has the lowest mean value of B for the years in which it shows a significant correlation.
6.4 Conclusion on hypothesis one
For hypothesis one I obtain at least significant medium correlations for all three segments and significant strong results for some segments, including the sample as a whole. When the data is smoothed out, correcting for the spikeyness of the raw data, I obtain significant strong results for all three segments and the sample as a whole. The expensive European segment is quickest when responding towards changes in the oil price growth rate, launching adapted models within one year. The cheaper European and Asian segment follow suit two year later. The cheaper European segment brings their ‘response’ within three to four years. After that, no significant correlations can be found. The Asians bring their response starting three years after a growth rate change and continue until at least the fifth year after a change in the oil price growth rate with adapting their products. For the sample as a whole significant response is found in the third, fourth and fifth year.
I can therefore conclude that when using a data set with three year moving averages, although response is lagged due to the length of a development cycle, all manufacturer segments respond to variations in the oil price. The first hypothesis holds. All three segments show a significant response at at least the 0,05 level for at least two years in a row after a change in the oil price when using the three year averages. The analysis shows that the Asians are late with finishing their response. The negative correlation at t = 0 is difficult to explain. Perhaps because of the cyclical nature of the oil price growth rates (see graph 2) and the large lag in the response of the Asian manufacturers, the Asians are still finalizing their response towards negative oil price growth rates. Anyhow, the large amount of positive significant results indicate a certain robustness, allowing us to discard the importance of this single negative correlation.
6.5 Results testing for hypothesis two, correlation analysis.
The second hypothesis states that ‘an increase in the price of oil will have a negative effect on
the growth of relative power output of automobile’s engines’. This implies that I expect to see
for correlations using data with one year growth rate averages, followed by data with three year moving averages. The results are below in table 7.
Table 7.
kW/cm3 & Oil price growth rate. Pearson Correlation Coefficient – 3 year average growth rates. N = 22
Lag all cars asian eurcheap eurexpensive
No lag -,100 ,123 -,088 -,140 1 year lag -,011 -,090 ,199 -,098 2 year lag ,247 -,200 ,011 ,257 3 year lag ,473* ,415* ,184 ,376 4 year lag -,146 ,151 ,026 -,295 5 year lag -,242 ,-,211 ,103 -186
** Correlation is significant at the 0.01 level (2-tailed). * Correlation is significant at the 0.05 level (2-tailed). All significant values are highlighted in bold.
No significant negative correlations can be found when analyzing the 1 year growth rates. There is a significant and positive medium correlation for the sample as a whole and for the Asian segment in the third year. Please note that the correlation is less strong than most of the correlations found for the km/L growth rates.
Table 8.
kW/cm3 & Oil price growth rate. Pearson Correlation Coefficient – 3 year average growth rates. N = 22
Lag all cars asian eurcheap eurexpensive
No lag ,080 ,296 ,149 -,116 1 year lag ,299 ,350 ,211 -,154 2 year lag ,487* ,499* ,119 ,414 3 year lag ,411 ,489* ,083 ,376 4 year lag -,018 ,228 -,084 -,107 5 year lag -,257 -,134 ,118 -,258
** Correlation is significant at the 0.01 level (2-tailed). * Correlation is significant at the 0.05 level (2-tailed). All significant values are highlighted in bold.
6.6. Results testing for hypothesis two, regression analysis.
Using the linear regression model I can again say something about the strength of the effects for the moments when there was a significant correlation. The results are in table 9.
Table 9.
kW/cm3 & Oil price growth rate. Linear Regression – 1 year average growth rates. N = 24
All cars Asian Eurcheap Eurexpensive
B Std. Error B Std. Error B Std. Error B Std. Error
No lag (Constant) ,010 ,002 ,008 ,002 ,010 ,003 ,012 ,005 oilgrowthrate -,004 ,008 ,004 ,007 -,004 ,010 -,012 ,018 1y Lag (Constant) ,010 ,002 ,009 ,002 ,010 ,003 ,012 ,005 oilgrowthrate ,000 ,008 -,003 ,007 ,009 ,010 -,008 ,018 2y Lag (Constant) ,010 ,002 ,008 ,002 ,010 ,003 ,011 ,005 oilgrowthrate ,009 ,007 ,007 ,007 ,000 ,010 ,021 ,017 3y Lag (Constant) ,004 ,003 ,008 ,002 ,010 ,003 ,010 ,004 oilgrowthrate ,034* ,028 ,013* ,006 ,008 ,009 ,030 ,016 4y Lag (Constant) ,010 ,002 ,008 ,002 ,010 ,003 ,013 ,005 oilgrowthrate -,005 ,007 ,005 ,006 ,001 ,009 -,023 ,016 5y Lag (Constant) ,010 ,002 ,009 ,002 ,010 ,003 ,012 ,005 oilgrowthrate -,008 ,007 -,006 ,006 -,004 ,009 -,015 ,016
** Correlation is significant at the 0.01 level (2-tailed). * Correlation is significant at the 0.05 level (2-tailed). All significant values are highlighted in bold.
