The quality bias of inflation measurement in the Netherlands during 2005-2017 : acase study of the cellphone industry

(1)

1

Bachelor Thesis

The quality bias of inflation measurement in the

Netherlands during 2005-2017: A case study of the

cellphone industry

By: Jesper Zuurbier, 10888012

Supervisor: C.W. Haasnoot

(2)

2

Statement of Originality

This document is written by Student Jesper Zuurbier who declares to take full responsibility

for the contents of this document.

I declare that the text and the work presented in this document are original and that no

sources other than those mentioned in the text and its references have been used in creating

it.

The Faculty of Economics and Business is responsible solely for the supervision of completion

of the work, not for the contents.

(3)

3

different worth now than in ten years. This affects financial decision making by people as the value of money can deteriorate. The complexity of measuring inflation has caused that there are multiple ways to measure inflation and these different methods cause different inflation numbers (Triplett, 2004). If these differences are large, the consequences are large as well.

For example, if inflation is largely understated, consumers can buy less products than numbers would imply. This in turn can result in the fact that companies to produce too many products relative to the demand. Companies cannot sell all the products and end up with lower profit. Companies will then try to cut costs and because of rigid wages companies will fire employees. Jobless have less to spend and this results in even less products sold by companies and as a result the economy can end up in a recession. This is called a deflation cycle (Williams, 2009). This potential consequence makes that policymakers need to make sure that consumers have enough purchasing power to keep up their consumption. This is only possible if government institutions responsible for the data supply reliable inflation numbers. The primary objective of the European Central Bank (ECB) is price stability, meaning that the inflation within the eurozone is around but maximum 2% over the medium term (European Central Bank [ECB], n.d.). This can only be obtained if the inflation numbers are accurately measuring inflation.

Earlier research identified the two most prominent biases that can arise when measuring inflation: These are: The substitution bias and the quality bias. The substitution bias is a bias that is the result of new products being introduced into the market and old products disappearing. The quality bias is a bias that is the result of products being improved constantly, which also has an effect on the value of money (Schmitt-Grohe & Uribe, 2009). According to Cecchetti (1993) the discussion on the substitution bias is well documented. In practice, the Centraal Bureau voor de Statistiek (CBS) actively tries to correct for the substitution bias in the Netherlands. Contrary to the substitution bias the CBS stated that the quality bias is corrected for by multiple methods and those methods are not without criticism. That is why the focus of this research will be on the quality bias in the Netherlands.

The most dominant view in literature is that inflation is overstated because of the quality bias (Boskin et al., 1998). Boskin et al. reasoned that the quality increases faster than measured by the Consumer Price Index (CPI). On the other hand, Gordon & van Goethem (2005) used a time series

(5)

5

analyses to look at rental prices during 1975-2003 and concluded that the quality correction used in the CPI was too large. Because of this, the CPI understated inflation. The main focus of this research will be the quality bias in the Netherlands during 2005-2017, focused on the consumer electronics and cell phone segment. CPI and pricing data in the Netherlands shows a steep declining CPI during a period of price increase in an example of the cell phone market.

The research is structured as follows. First, inflation will be defined. Then I will describe why quality improvements occur and use the theories of the quality ladders and the product life cycle to do so. I then explain how inflation can be measured, what difficulties and trade-offs there are, and where the quality bias and substitution bias come from. After that I describe all possible quality correction methods and the hedonic function1 _{specifically. With a clear picture of what the options to}

correct for quality improvements are, cell phone and CPI data will be used of the Netherlands during the period 2005-2017 to analyse how the CBS corrects for quality improvements. This research will end with the conclusion as to what the data implies and what possible consequences are for consumers and policy makers in the Netherlands.

1_{A hedonic function is a relation between the prices of different varieties of a product and the quantities of}

characteristics in them (Triplett, 2004). Many authors describe the hedonic function to be the most objective and most accurate tool to measure inflation.

(6)

6

Chapter 2: Literature review

The Centraal Bureau voor de Statistiek (CBS) in the Netherlands provides a different inflation number for the Netherlands, when commissioned to do so by the Dutch government, than for the European Union (Rabobank, 2014). From this we can derive that there are different methods to measure inflation and these different measurement methods can result in different numbers (Triplett, 2004).

The term inflation means an increase in the general level of prices. Deflation is a decrease in the general level of prices and can be regarded as negative inflation. Henceforth, the term inflation will be used for both scenarios.

2.1 Quality increase

Our standard of living has increased dramatically since men first set foot on earth. And ever since upgrading of our standard of living plays a central role in our society. Houses were built,

transportation for mankind improved and we can fly from Amsterdam to Sydney in 21 hours. Besides the scientific breakthroughs, existing products are also improved. A car bought today has more value of use than a car bought ten years ago. The cars nowadays are safer, faster, cleaner and more fuel-efficient. All these improvements happened over time. The view that innovation of a product takes place gradually is called a quality ladder (Boldrin & Levine, 2009). According to Grossman & Helpman (1989) almost every product exists on a quality ladder.

On a quality ladder, companies are incentivised to improve their products constantly because of competition in the market. If the competitor can produce a better product, consumers have an incentive to switch to the competitor and the company will lose market share. This ensures that companies strive to be a competitor in the market. Companies try to achieve this by improving the quality of their products constantly. The process of improving products is both cyclical and continual (Grossman & Helpman, 1989). According to Grossman and Helpman (1989), every step up the quality ladder requires research and development. This means that the quality improvement of products takes time and money. Therefore, products have to stay long enough on the market in order to generate enough revenue to finance the research and development cost for the next improved product. This creates a trade-off between improving products and having products long enough on the market to offset the research and development cost. Because of the continual improvement, products have a finite lifespan. The development and decline of products is described by the product life cycle theory.

The product life cycle is a description of the evolution of a product measured by sales over time (J.W. Cox, 1967).An improved product, with other words a product that has made a step up the quality ladder can be viewed as a new product for the product life cycle. The product life cycle

(7)

7

distinguishes four stages: introduction, growth, maturity, decline.

In the introduction stage the product is new. Sales are low, the costs per consumer are high and there are few competitors. In the growth phase sales are increasing, costs per consumer are falling and more competitors are entering the market. In the maturity phase sales are peaking, costs are lowest per consumer and there is a mass market. In the declining phase sales are declining, costs per consumer are still low and the number of competitors fall again. It is common for companies to introduce new or improved products when old products are in the declining phase. Profit and revenue of the old product is shrinking so the company is forced to adapt and come up with new products in order to survive in the market (Klepper, 1996). The quality ladder theory and product life cycle theory explain why products change, new products are introduced, and old products disappear from the market. One of the difficulties arising from these three developments is the fact that old products need to be compared with new products in inflation measurement.

2.2 Matching

To measure a difference in price level between a group of products2_{, products need to be linked over}

time. This means that if inflation of a product is studied over two periods, the product must stay the same over the two periods. This is called matching. The term matching implies that the products compared in the two periods are similar. Since not every aspect of the production process and the selling process is observed, it is assumed that they are similar. For example, it could be the case that the customer service of a certain product improved and that may be a reason why the product is sold more often.

2.3 Price indexes

Matching plays a vital part in calculating inflation. In practice, inflation is calculated by indexes. These indexes try to match one period to another period. The index representing inflation is a weighted index composed of (almost) all the consumption of consumers. Each product gets a weighing in the index according to how much percent of total consumption can be attributed to that product.

Indexes that are not weighted are outside of the scope of this research. Unweighted indexes are not practical as they give distorted inflation numbers for countries in which many different products are consumed (Silver & Heravi, 2002).

The two most common indexes are the Laspeyres index and the Paasche index. For both indexes data has to be collected on the consumption. A Laspeyres index calculates the price difference between a base period and the purchasing of the same group of goods in the present. A

(8)

8

Paasche index calculates the price difference between a group of goods in the present compared to the same group of goods in the past (Diewert, 1998). The advantage of the Laspeyres index is that inflation estimates can be computed in a timely fashion. The prices and weights for the base year are already determined, which is not the case in the Paasche index. The weights of a Laspeyres index are fixed in January and stay the same until January next year, while during the year consumption can shift to cheaper or better products. This causes the Laspeyres index to overstate inflation (Boskin et al., 1998). In the case of the Paasche index the weights are fixed at a base year and a recent year will be compared. This also ignores the consumption shift during a year. Consumers could have shifted consumption to better alternatives during the year, causing the Paasche index to understate inflation.

A third option, the Fisher index combines both the Laspeyres and Paasche indices and takes the geometric mean of the two. Resulting in a theoretically better measurement of inflation

(Aizcorbe & Jackman, 1993). The Fisher index needs the data from the Paasche index and the data from the Laspeyres index. This makes the Fisher index less cost efficient than the Paasche and Laspeyres index and therefore the Fisher index is not often used in practice (Balk, 1995).

2.4 Trade-off and substitution bias

CBS, which is responsible for the inflation numbers in the Netherlands uses the Consumer Price Index (CPI). The CPI is a Laspeyres index3_{, which compares the price of a virtual basket of products in}

a base period to the price of the virtual basket of products in another period. The virtual basket represents the average of products bought by an average household in the Netherlands. The goal is always to measure the price level in a country as accurate as possible, but the research should not get too expensive or time consuming. If research gets too time consuming, the products an average person is buying now differs from the products that a person was buying 10 years ago. New products are introduced and old products disappear from the marketplace. If this is not corrected for it can create a bias in the CPI: The substitution bias. In this example, if the measurement takes 10 years to complete the result would be dubious at best. The replacement of the products and services for new products and services is called the substitution effect. It can happen that products become obsolete and are replaced by new products. This causes the value of money to increase in most cases (Bryan & Cecchetti, 1993).

(9)

9

2.5 Quality bias

Furthermore, it can happen that the quality of existing products increases. In a hypothetical case that someone is willing to pay €699 for an iPhone 6, which is also the price in this case. It can happen that Apple makes an iPhone 6s which has a higher quality than the regular iPhone. Because of the higher quality, the consumer now wants to pay €1398 for the iPhone 6s. Ergo the price he wants to pay has doubled, since he gets more value than before. If the iPhone 6s also costs €699, he will get more value for his money. In other words, there is a case of deflation, since his money is worth more to him at this point. If we do not correct for this, there will be an upward bias in the CPI. This is called the quality bias. The quality bias arises only if products and services are not homogenous (Triplett, 2004). When there is the case of homogeneity a simple method which compares price to quality can be used to correct for any change.

The quality bias can happen in two ways. Failing to identify a quality increase causes a measurement to overstate the price a person must pay for the value he is getting. Failing to adjust the measurement appropriately has the same effect. Improving quality in homogenous products is often not hard to measure. The quality improvement of more complex products can be often more difficult to measure precisely. Phenomena such as fashion trends can have the same effect. They are in theory a quality improvement since they provide more value at a certain time, yet there is no objective quality improvement. This quality improvement cannot be captured by the CPI at this point, after all no physical characteristic has changed from the product.

2.6 Quality corrections

In practice there are implicit and explicit methods to correct for quality. Implicit methods estimate the correct prices without using the real data. On basis of the estimation one can look at the real data and estimate the quality difference. The explicit method uses the real data immediately and the quality difference is estimated on basis of these data (CBS, 2008).

2.6.1 Explicit methods

1. The direct comparison: When there is no real difference between the new and old product, only the price change will be noted. No direct quality correction takes place.Some regard this as a quality correction option, while most authors omit this method (Triplett, 2004).

2. Quantity adjustment: If products are of the same quality but are now sold in different quantities. This correction is also straightforward. Often the price quality ratio is linearly used to compare the old and new products and quantities.

3. Option pricing: This method can be applied when a product and the same product with extra options is available at the same time. If only the product with extra options becomes available, no comparison can be made with the old one. This is because consumers are now forced to buy the

(10)

10

product with options and therefore it becomes impossible to determine how much quality is added by the extra options. In practice this is a problem for the option pricing model. If the extra options are always build into the product, the costs of making that option will decrease. This will happen because of economies of scale. Levy et al. (1999) talk about an anti-theft alarm for cars that was made standard on all cars and was disabled for consumers who did not want the option. This was cheaper for the factory than producing cars with and without the alarm. Because it is hard to estimate what the quality increase of the option is, the option model attributes 50% of the price difference to a quality increase.

4. Production costs: The increase in quality will be measured in an increase in production costs. When trying to correct for an increase in quality it can sometimes be the best method to look at the production costs. In markets where producers come up with new models and/or improvements on a predictable basis the production cost increase can be checked with competitors. A difficulty can lie in obtaining the right data. ‘’For the production costs method to be accurate scale of production, labour and input costs and production technology should all be held constant’’ (Triplett, 2004). In practice these variables are almost never constant. In theory a good market for this would be the smartphone market, since competitors come up with similar changes and improvements to products. This provides a good basis to compare the production cost changes among companies. (Greenlees, 2000). However, the Bureau of Labour Statistics (BLS) has tried to compare computers using the production costs method, but could not make a comparison due to the fact that

production costs were declining while the prices were increasing. This would produce a negative correlation between production costs and prices.

5. Expert guess: An expert will be called in to estimate the quality difference between the new and old product. This may be someone who works in the branch or works directly with the product. It is argued that these experts have access to information about the product that is not available to the public. Though it can be an inconsistent, since this method is not founded in science but depended on the skill, experience and state of a human being. Greenlees (2000) among other researchers does not recognize the expert guess method as a viable method for estimating the quality of a product. The CBS stated that the agency does use the expert guess method (CBS, 2008).

6. The Hedonic regression method (hedonic function): A regression will be made of the price-quantity relation. Most often this will be a logarithmic function. The goal of the regression is to remove all subjective (hedonic) product features. The hedonic regression splits the product into all the observable characteristics.

There are 2 major types of hedonic regressions: The characteristics price method as depicted in equation (1) and the dummy variable method in equation (2).

(11)

11

An example of the characteristics price method is:

(1) Pt SMARTPHONE = β0+ β1RESOLUTION + β2RAM + β3CAMERA + β4DESIGN + ε

In which the price of smartphone at time t is determined by a constant, the screen quality, the memory quality, the quality of the camera, the design and a random error term.

This formulation leads to a definition of the product quality in terms of the amount of each characteristic that each variety has (Hulten, 2003). The price of a variety at a time t is a function of its characteristics plus an error term. If a new smartphone model enters the market, for example the iPhone X, equation (1) will give an estimated price. This estimated price represents the mean price of a product with those specific characteristics and can also be viewed as the equilibrium between supply and demand (Rosen, 1974). This also means that external shocks can have an effect on the characteristics.For example, a positive income shock can cause the coefficients of the characteristics to change. This can cause all coefficients to increase, but can also mean a shift in coefficients, meaning that in this example β4 can increase and β3 can decrease.

The hypothesis of the hedonic regression is that all the price difference that is not

explained by the regression is regarded as inflation. If the estimated price of the iPhone X is at €999 and the iPhone X is sold for €1099 the hedonic regression will consider the difference: €100 as inflation. The difference in the estimated price of an older iPhone in comparison to the iPhone X is regarded as quality increase by the characteristics price method and is therefore not regarded as inflation.

The second method is the dummy variable approach. In this approach 1 or more binary variables are added which represents the difference in pricing between periods, while holding the characteristics of the product constant (Piccolo & D’Elia, 2008). An example of a dummy variable regression is:

(2) Pt SMARTPHONE = αO + α1 RESOLUTION + α2 CAMERA + γ SUMMER + ε

In which the Price of a smartphone at time t is determined by a constant, the screen quality, the quality of the camera, the dummy variable summer which is either 1 when the it is summer or 0 when it is not and a random error term. A dummy variable regression can contain 1 to ∞ number of dummies. For example, Combris, Lecoq & Visser (1997) use over 25 dummy variables for their hedonic function regarding the price of Bordeaux wine. Time dummies can be added in case comparisons are made over time. This implies that the coefficients α1 and α2 are held constant over

time, which is the difference between the dummy variable method and the characteristics price method. Triplett (2004) argues that this restriction is controversial as coefficients change over time due to changing preferences and circumstances. After all the hedonic regression is an equilibrium between supply and demand. An advantage of the dummy variable method is that the effect of

(12)

12

changes in time like a change in season can be taken into account.

While the hedonic regression is regarded ‘’as the most promising technique for explicitly adjusting observed prices to account for changing product quality ‘’ (CNSTAT Panel Report, pg. 122, 2002), there is one persistent critique to the method: Hedonic indexes fall too fast (Hulten, 2003). While this is true for the first hedonic regression studies that are done (the studies that are often referred to), no evidence is showing that there is such a trend in hedonic regressions across all sectors. This critique is regarded by Triplett (2004) as out of date and ill informed.

According to the CBS these are specialized procedures which require much data and are therefore not often used in practice. A lot of data is needed because there is the danger of omitted variable bias. The more data there is, the higher the chance is that no variables are left out.

Furthermore, the regression becomes more accurate with more data. As the sample grows, the representability towards the population grows as well.

2.6.2 Implicit Methods

1. Bridged overlap of class mean imputation: In this case the quality change is estimated by using the price development of comparable products. By calculating the average price of comparable products a price estimate will be made for a product. This is only relevant for goods in which quality change is a small component. If the quality increases are large, products will lose comparability and this method cannot be used anymore.

2. Overlap method: In this method it is assumed that there is a period in which the old product and the new product are available at the same time. It is also assumed that the quality difference is equal to the price difference in that period. This method can only be applied if the two products that are compared are in the same product life cycle phase. Products that are not in the same product life cycle are not comparable on the ground that they attract a different type of consumer. In the

introduction phase consumers might not know the product yet or are hesitant to buy the product until they have seen other consumers buy the product. This means that less consumers buy the product compared to other product life cycle phases, while it does not mean that the consumers get less value from the new product. Hence the products can only be compared when they are in the same product life cycle phase.

2.7 Differences between price indexes and hedonic functions

We now have seen two general categories to measure inflation: The more traditional indexes and the hedonic regressions. Do empirical differences between the two categories exist, what are the differences and where do they come from? From the research of Okamoto and Sato (2001) comparing hedonic regressions with traditional matched model indexes it showed that there was

(13)

13

virtually no difference between both indexes in the case for PCs. In the case of cameras, the hedonic regression index declined more slowly (10.4% per year vs 22% per year). The study by Berndt et al. (1995) shows that for the PC market in ’89-’92 in the United States the hedonic regression index drops with an annually 30% while a matched model index declined 19% on average each year. These differences can have huge implications in the long run for consumers and policy makers.

According to Triplett (2004) it is very hard for outside researchers to analyse a statistical agency’s price index sample. Statistical agencies lack transparency, which hinders the research possible. Triplett does advocate researchers trying to analyse statistical agency’s price index sample to build a sample as to why and how hedonic regressions differ from regular price indexes.

(14)

14

Chapter 3: A case study of the cell phone industry

In this section I will look at the quality correction of cell phones in the Netherlands in the period of 2005-2017. All information presented in the section below comes from the CBS, besides the data from the iPhone, either from the CBS website or from personal correspondence. According to Cecere, Corrocher & Battaglia (2015) the competition in the smartphone industry makes it a very innovative sector. The more innovations there are, the more quality adjustments need to be made. Therefore, smartphone industry an interesting industry to study for quality biases.

The CBS gets inflation numbers by sending questionnaires to many companies and based on this sample they try to get representable numbers for the population. The questionnaires are checked on quality by an intern questionnaire, which dictates the guidelines for the questionnaire sent to the companies. Based on the responses of these companies the CBS will try to compose the CPI, which as stated before should represent the inflation in the Netherlands. The CBS will check the given data for errors and plausibility. The CBS creates the virtual basket of goods in January, this basket stays constant throughout the year. That means that the virtual basket is different every year. The CBS claims that the virtual baskets are never the same, but always comparable. The CPI in the Netherlands is a Laspeyres index.

Since 2005 the CBS is using scanner data from supermarkets. Scanner data gives accurate and detailed information for the CPI. Each item scanned at the registry is immediately sent to the CBS. This way the CBS has very fast and very complete information on prices and quantities sold. Scanner data has shown that the markets for commodities are very dynamic. Normally the CBS changes the virtual basket each January, while with scanner data the CBS can update the virtual basket almost instantly. This result in a virtual basket that more accurately represents the purchases of the consumers and therefore lowers the substitution bias in the CPI. Scanner data provides a complete and timely source of data. This makes it perfectly fit for the hedonic regression. The CBS does not use it for this purpose yet and expects only to use implicit correction methods often in the future (Grient & Haan, 2010).

Below are 3 figures. All data below is either retrieved as and index or transformed to index numbers to facilitate comparing the data. In Figure 1 we will look at the inflation data from the CBS for the cell phone market and the inflation data for consumption in general. To analyse the quality

(15)

15

Figure 1: The Dutch CPI and the partial Dutch CPI for Cell phones. Data is retrieved from CBS Statline.

In this figure we see that the partial CPI for phones has declined by 63,09% during 2005-2017, while the CPI increased by 19,18% during this 13-year period.

In figure 2 data from the cost side of the production will be displayed. This data is needed in case a production cost method correction is to be used. Since the PPI for the smartphones is not directly available from the CBS, while the consumer electronics data is. Therefore, consumer

electronics data will be used in this case. The PPI stands for Producer Price Index and can be regarded as prices producers pay to produce or import a product. The data is broader than the smartphone industry, but the consumer electronics is also an industry where there are constant innovations (The Economist, 2018). These similarities combined with the fact that smartphones are part of the consumer electronics data, makes this data a good substitute.

Most consumer electronics are not made in the Netherlands and consumer electronic

manufactures do not publish the manufacturing costs, import prices for intermediaries are regarded as the PPI in this case. The import prices for intermediaries will probably fluctuate with the production costs. Wages in figure 2 are the Dutch wages during 2005-2016. These are also included since they are costs from the seller’s perspective.

0 50 100 150 200 250 CPI (Phones) CPI

(16)

16

Figure 2: The partial PPI for intermediaries and Wages in the Netherlands 2005-2016. Data is retrieved from CBS Statline.

The PPI decreased by 29,2% over the period 2005-2016, therefore the production cost method cannot be used. This data implies that the import of consumer electronics became cheaper for the importers in the Netherlands.

Figure 3 shows the introduction price of new models of the iPhone over time. So far we have seen the CPI of smart phones in the Netherlands and the iPhone is and was a dominant phone in the Dutch market since its release with market share between 13-23% for the last 8 years (Statista, 2018). The data from Apple (retrieved via AAPLinvestors.net) is in $ originally, but because of the law of one price we can assume that the correlation is near perfect for the $/€ exchange ratio.

0 20 40 60 80 100 120 140 PPI (Consumer electronics) Wages

(17)

17

Figure 3: new model iPhone prices from iPhone to iPhone X. Data retrieved from Apple

The data shows that the although the prices of iPhones went up by 150% in 11 years’ time. The CPI of smartphones in the Netherlands did not go up by the same percentage, but even went down by 48,11%. This difference is caused by quality corrections. As far as I know there is no alternative explanation for a difference of this magnitude. In correspondence with the CBS they stated that they used different methods for quality correction ‘Whichever is the best for a particular scenario’.

The first thing to note from that statement is that it can be quite dubious to use multiple correction methods on the same data at different times. Each method has its pros and cons as stated in the section above. The con of each method makes that the method cannot be used each time or that it will not result in a good correction for the data. Mixing methods can result in decrease in transparency of the CPI. If there can be biases from multiple methods and the results cannot be checked by third parties, the numbers will lose credibility and representability. The second thing to note is that we can try to deduce what methods the CBS does use for quality correction in cell phones.

The direct comparison method is not used, because there is a case of quality correction. The quantity adjustment method is not used, because there is not a difference in quantity but only in price and quality. The option method could have been used, since new models and old models are

sometimes in the market at the same time but note that both products are often not in the same phase of the product life cycle. While the old model is often in the mature phase, the new model is often in the introduction or growth phase. The option model usually estimates the quality difference as 50% of the price increase. In our case the quality difference is almost 250% while the price increase is 150%. The CBS could also have used the production costs method. Although the cost of production, here measured by the import prices of consumer electronics, has gone down, the price of cell phones has

0 50 100 150 200 250 300 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017

iPhone index

(18)

18

gone up. This fact makes it hard to estimate a relation between production costs and pricing. The CBS could have used this method in some years when prices went down. The expert guess method could have been used. This method is not transparent and not consistent, so it is impossible to rule this method out. In case the expert’s guess is published, others can confirm or criticize the estimate. This is not the case with the CBS, so we cannot analyse this method further.

The CBS stated that the hedonic model is almost never used by the CBS. The CBS recognises the potential of the hedonic regression but argues that there is often not enough data available to make one. This fact makes it unlikely that either of the hedonic regression methods has been used. The bridged overlap of class mean method is not likely to have been used, since the quality correction in the data is so large. The overlap method is not likely to have been used, because the quality correction is between products which succeed one another. This makes that the overlap method cannot be used in this case. From the above we can conclude that three methods could have been used: The overlap method, the production costs method and the expert guess method. The overlap method has a correction between 0-100% and the production cost method has a correction based on growing production costs in case of increasing prices. If decreasing costs are regressed on increasing prices the correlation would be negative and that would imply that lowering costs would increase prices.

Production costs in consumer electronics has been declining. Combining these facts, we can conclude that the expert opinion method has been used more than the other two methods.

The expert opinion as stated in the section above is a disputed method for measuring quality. The experts can be subjective or have a different opinion from day to day. Wright & Bower (1992) quote multiple researches which conclude that ‘’Emotions can have a substantial impact on the retrieval of knowledge from memory and on the processing of information. This makes the expert opinion method volatile. Furthermore, it can be an expensive method to use, as experts need to be compensated for their time.

Based on the data above it is not possible to measure any bias in the CPI. There simply is not enough information on the measurement methods and the case for cell phones and consumer electronics is so specific that most methods cannot be used accurately to correct for quality. This also makes that it is hard to say if the CPI is a good measurement of the inflation of consumer electronics.

(19)

19

Chapter 4: Conclusion

Inflation in the Netherlands is measured by the CPI. The CPI is a Laspeyres index. Because of the rapid innovations within industries the quality of products is increasing fast. The quality ladder theory explains this as a result of the competitiveness of the market. This is the case in phone industry. Consumers get more value from these quality enhanced products. This creates the need for quality adjustments in the CPI to accurately reflect the value a consumer can buy.

To correct for quality the CBS stated that it uses multiple methods. This is probably a mix of the option method, production costs method and the expert opinion method, in which the expert opinion method would be most used. Using multiple methods creates a non-desirable situation as it decreases transparency and reliability of the CPI. Also, the expert opinion is a volatile and non-transparent method to use. Furthermore, the CBS does not use another method, for instance the hedonic regression, to verify the CPI. For transparency and consistency reasons it might be a good idea to use one quality correction method and afterwards verify the CPI number with a hedonic regression. In case the hedonic regression is too expensive, the CBS can also use the hedonic

regression to identify biases in a cheaper method. The CBS then can keep using the cheaper method and correct it for the found bias. The hedonic regression is viewed as the most accurate

measurement tool for the CPI. On the other hand, much data is required for a hedonic regression to be accurate.The CBS can use scanner data to enlarge the dataset on consumer electronics. With the scanner data, which is complete and immediate, the regressions would be more accurate.

While the price of the iPhone went up by 150% in a little over ten years, the CPI price of phones went down with about 50%. This implies that the iPhone costs 2.5 times as much as it did before and the utility the average consumer gets out of the iPhone is 5 times as high as before. In other words, people in 2005 would have paid roughly $5000 for the latest iPhone.

Since the CBS will not give openness about the methods used. We cannot check if the numbers stated above are correct. The methods probably used are widely disputed and sometimes not even regarded as valid methods for quality adjustment. This in turn would mean that consumers and policy makers cannot know that these numbers are correct. This also means that if policy makers want wages and inflation to stay at the same level, that consumers would lose purchasing power. If the trend of price increases and partial CPI decreases continues, fewer and fewer consumers are able to buy for example an iPhone. This decrease in consumption can slow down company growth and start a deflation cycle.

(20)

20

Chapter 5: References

Aizcorbe, A. M., Corrado, C. A., & Doms, M. E. (2003). When do Matched-model and Hedonic Techniques Yield Similar Measures? SSRN Electronic Journal.

Aizcrobe, A. M., & Jackman, P. C. (1993). The Commodity Substitution Effect in CPI Data, 1982-91. Monthy Labor Review, 116(12), 25-33.

Balk, B. M. (1995). Axiomatic Price Index Theory: A Survey. International Statistical Review / Revue Internationale de Statistique, 63(1), 69.

Berndt, E. R., Griliches, Z., & Rappaport, N. J. (1995). Econometric estimates of price indexes for personal computers in the 1990s. Journal of Econometrics, 68(1), 243-268.

Boldrin, M., & Levine, D. K. (2009). Quality Ladders, Competition and Endogenous Growth (Rep.). Department of Economics, Washington University in St. Louis.

Bradford, W. D. (1974). Price-Level Restated Accounting and the Measurement of Inflation Gains and Losses. The Accounting Review, 49(2), 296-305.

Boskin, M. J., Dulberger, E. R., Gordon, R. J., Griliches, Z., & Jorgenson, D. W. (1998). Consumer Prices, the Consumer Price Index, and the Cost of Living. Journal of Economic Perspectives, 12(1), 3-26.

Bryan, M., & Cecchetti, S. (1993). The Consumer Price Index as a Measure of Inflation. The National Bureau Of Econonomic Research.

CBS. (2008). Indexcijfers [PDF]. Den Haag: Centraal Bureau voor de Statistiek.

Cecere, G., Corrocher, N., & Battaglia, R. D. (2015). Innovation and competition in the smartphone industry: Is there a dominant design? Telecommunications Policy, 39(3-4), 162-175. Cecchetti, S. (1996). Measuring Short-Run Inflation for Central Bankers. The National Bureau Of

Econonomic Research.

Combris, P., Lecocq, S., & Visser, M. (1997). Estimation of a Hedonic Price Equation for Bordeaux Wine: Does Quality Matter? The Economic Journal, 107(441), 390-402.

Cox, J. W. (1967). Product Life Cycles as Marketing Models. The Journal of Business, 40(4), 375. Diewert, W. E. (1998). Index Number Issues in the Consumer Price Index. Journal of Economic

Perspectives, 12(1), 47-58. doi:10.1257/jep.12.1.47

Document CBS: Consumentenprijsindex (CPI), 27-10-2006. http://www.cbs.nl/nl-

NL/menu/themas/prijzen/methoden/dataverzameling/korte-onderzoeksbeschrijvingen/2006-cpi-art.htm.

European Central Bank, (n.d.). Definition of price stability. Retrieved February 10, 2018, from https://www.ecb.europa.eu/mopo/strategy/pricestab/html/index.en.html

(21)

21

Het verschil in Nederlandse inflatiecijfers verklaard. (2014, June 18). Retrieved January 07, 2018, from https://economie.rabobank.com/publicaties/2014/juni/het-verschil-in-nederlandse-inflatiecijfers-verklaard/

Gordon, R., & Vangoethem, T. (2005). A Century of Housing Shelter Prices: Is There a Downward Bias in the CPI? National Bureau of Economic Research

Greenlees, J. (2000). Consumer Price Indexes: Methods For Quality and Variety Change. Statistical Journal of the United Nations Economic Commission for Europe, 17(1), 59-74.

Grossman, G. M., & Helpman, E. (1989). Quality Ladders in the Theory of Growth (Working paper No. 3099). Cambridge: National Bureau of Economic Research.

Hulten, C.R. (2003). Price Hedonics: A Critical Review. Economic Policy Review, 9(3)

Klepper, S. (1996). Entry, Exit, Growth, and Innovation over the Product Life Cycle . The American Economic Review, 86(3), 562-583.

Levy, F., A. Beamish., R.J. Murnane and D. Autor (1999), “Computerization and Skills: Examples from a Car Dealership”, presented at the Brookings Institution Program on Output and

Productivity Measurement in the Service Sector

National Reasearch Council. 2002. At What Price?: Conceptualizing and Measuring Cost-of-Living and Price Indexes. Washington, DC: The National Academies Press.

Piccolo, D., & D’Elia, A. (2008). A new approach for modelling consumers’ preferences. Food Quality and Preference, 19(3), 247-259.

Rosen, S. (1974). Hedonic Prices and Implicit Markets: Product Differentiation in Pure Competition. Journal of Political Economy, 82(1), 34-55.

Schmitt-Grohe, S., & Uribe, M. (2009). On quality bias and inflation targets. Cambridge, MA: National Bureau of Economic Research.

Silver, M. & Heravi, S. (2002). A failure in the measurement of inflation: results from a hedonic and matched experiment using scanner data, ECB Working Paper, No. 144

Silver, M., & Heravi, S. (2003). The Measurement of Quality-Adjusted Price Changes (Rep.). University of Chicago Press.

Solutions, E. D. (n.d.). Consumer goods. Retrieved January 28, 2018, from

http://www.eiu.com/industry/consumer-goods/theme/20/innovation-in-consumer-electronics

Triplett, J. E. (2001). Should the Cost-of-living Index Provide the Conceptual Framework for a Consumer Price Index? The Economic Journal, 111(472), 311-334.

(22)

22

Science, Technology and Industry Working Papers. doi:10.1787/643587187107

Van der Grient, H., & De Haan, J. (2010). The Use of Supermarkt Scanner Data in the Dutch CPI (Rep.). Statistics Netherlands.

Williams, J. C. (2009). The Risk of Deflation (Rep.). FRBSF Economic Letter.

Wright, W. F., & Bower, G. H. (1992). Mood effects on subjective probability assessment. Organizational Behavior and Human Decision Processes, 52(2), 276-291.

The quality bias of inflation measurement in the Netherlands during 2005-2017 : acase study of the cellphone industry

Bachelor Thesis