Norway

UNITED NATIONS ECONOMIC COMMISSION FOR EUROPE

CONFERENCE OF EUROPEAN STATISTICIANS

Expert meeting on Statistical Data Confidentiality 26–28 September 2023, Wiesbaden

A Disclosure-Based Framework for Comparing Frequency Table Protection Daniel P. Lupp and Øyvind Langsrud (Statistics Norway)

{dlu, oyl}@ssb.no

Abstract For the protection of the dissemination tables from the 2021 population census, Eurostat recommended a combined use of the cell-key method and targeted record swapping. As part of a grant awarded to Statistics Norway on multi-grid geographical data, we compared this recommendation to alternative methods (in particular small count rounding) on dissemination of frequency data over multiple grid systems. This was done using Norwegian census data as a use case. In this work, we present the findings of this project, as well as discuss the comparison framework used. This framework is based on a suite of disclosure scenarios that can occur in frequency tables. Using established notions from information retrieval, disclosures are counted and evaluated for each scenario, providing measures of risk. Given an acceptable threshold for risk, methods deemed satisfactory are compared using common utility measures. Of the remaining methods, only those preserving enough utility are considered as viable protection methods.

1 Introduction

Population statistics is disseminated using multiple overlapping grid systems at various resolutions: Statistics Norway uses a national grid system, whereas another grid system is used for European census delivery. This provides multiple challenges with regards to disclosure control. The official Eurostat recommendation for the 2021 publication is a combination of targeted record swapping (TRS) along with the cell-key method (CKM). The former identifies records deemed to have a high risk of disclosure and swaps them with similar records from nearby regions, whereas the latter is a post-tabular perturbation method designed to handle such differencing attacks. In the 2011 population census publications, Statistics Norway employed a rounding procedure described by Heldal (2017). Since then, the algorithm has been improved upon by Langsrud and Heldal (2018), and was subsequently named small count rounding. This method bears a resemblance to methods used by others in previous publications (for example, the UK in 2001 as described in a report by Spicer (2021)), but promises to address many of the complaints users had: specifically, small count rounding maintains additivity in a way that attempts to minimize information loss. As part of a grant on multi-grid geographical data, Statistics Norway wished to compare the Eurostat recom- mendation of the combination of CKM and TRS with the small count rounding method. In order to make a rigorous comparison, we employed a comparison framework designed to compare how each method performs with respect to different kinds of disclosure. This paper presents that framework, as well as a brief summary of the results of the evaluation. Finally, we discuss ways in which the framework could be improved upon.

2 Comparison Framework

The comparison framework we present is based on common practice in statistical disclosure control: maximize utility given an acceptable level of risk. It consists of two parts: a suite of disclosure scenarios used to quantify risk, and measures for measuring loss of utility. In general, the disclosure scenarios one chooses for the comparison may vary depending on the needs of the entity publishing the statistics. Within this framework, the only requirement posed to the disclosure scenarios are that they must be “countable": for a given disclosure scenario and a data set, one must be able to count the number of disclosures. In the following section, we present four disclosure scenarios that capture different flavors of classical disclosures. In particular, the presented scenarios were used to evaluate different protection methods for the dissemination of the 2021 population census.

2.1 Disclosure Scenarios

We adopt a disclosure-centric approach to measuring risk. In particular, we consider the following four types of disclosure:

Ordinary attribute disclosure: When all records in a table marginal share the same attribute for a given variable, group attribute disclosure occurs. This is the case when there is only one category with a non-zero frequency within a marginal, and thus the exact category membership can be revealed. Then the cell with the non-zero frequency is considered disclosive.

Attribute disclosure when the original total is 1: Similar to ordinary attribute disclosure, but limited to marginal cells where the population total is 1. This is of particular interest in sparsely populated countries such as Norway, because one can often assume it to be known that the population total in a grid cell is 1.

Negative attribute disclosure: When no record contributing to a marginal cell has a certain attribute, negative attribute disclosure occurs. Any frequency that is zero is disclosive in the sense that no record can have that category. When the frequencies in all categories are zero, the zeros are no longer

2

Table 1. Example frequency table illustrating different disclosure scenarios. The first row exhibits ordinary attribute disclosure: all units in M1 are unemployed. The second row demonstrates negative attribute disclosure; no units in M2 are self-employed (indeed, all zero cells represent negative disclosures). Additionally, all non-zero cells are existence disclosures.

Municipality Unemployed Employed Self-employed Total M1 12 0 0 12 M2 5 6 0 11

considered disclosive. Note that for two-level categorical variables the measurement of ordinary and negative attribute disclosure will in practice be the same.

Disclosure of existence: Any non-zero frequency discloses that at least one record has a certain attribute. That is, all cells with non-zero frequencies are considered disclosive.

In related work, Geyer et al. (2022) compare multiple methods based on the preservation of singletons, i.e., frequencies of 1, due in part to the increased risk of identification, but also the perceived feeling of vulnerability a unit might have even in cases where the attribute disclosure is incorrect. This approach is not a suitable measure for our study, given the methods and parameters: none of the methods allow for publication of frequencies less than 3, and hence no singleton cells are preserved. Rather, our study focuses on the possibility of actual disclosures (an attacker disclosing information about a different unit) as opposed to perceived disclosures (a unit being able to identify themselves in a data set), and the extent to which the different methods preserve real disclosures for each of the above disclosure types. The above disclosure scenarios are intentionally broad: indeed, it is, realistically, far too restrictive to require no possibility of disclosure of any kind. The intention is not to ensure prevention of each of the disclosure scenarios, but rather the ability to measure the performance of protection methods in different situations. This provides a solid basis for deciding which method is best suited to the given publication. For example, though we considered and measured all of the above scenarios for the evaluation of the 2021 population census, the first two scenarios (attribute disclosure, and attribute disclosure where the original total is 1) were deemed far more important to protect against than the other two.

2.2 Measuring risk

A common framework can be used to assess all these types of disclosure risk, based on measures used in information retrieval and machine learning: precision and recall. For each disclosure type as discussed in the previous section, all cells to be published can either be marked as disclosive or non-disclosive. Then we count the number of disclosive cells in the original and the perturbed data.

𝑎 = #disclosive cells in original data 𝑏 = #disclosive cells in perturbed data 𝑐 = #common disclosive cells

That is, 𝑐 is the number of disclosive cells in the original data that are still disclosive after perturbation. With these counts we can calculate precision and recall for each method as follows:

precision = 𝑐/𝑏 recall = 𝑐/𝑎

These measures provide two different views on the protection provided by the perturbation methods considered. Intuitively, precision provides a measure for how many of the disclosures in the perturbed data set are actual disclosures, whereas recall provides a measure for how many of the disclosures in the original data set are

3

preserved. Thus we use these measures as the primary means of measuring the risk for each method and disclosure type. Occasionally, we limit the calculation to selected cells of interest. For instance, we may look at certain categories that are considered more sensitive than others. Likewise, we can limit the calculations based on cell frequencies in the original or the perturbed data. However, in this case one must then keep in mind some of the measures become degenerate.

2.3 Measuring Utility

Given an acceptable level of risk, we wish to choose the method that provides the highest utility. High utility in this context means the same as low information loss. A measure of utility is therefore also a measure of information loss. In this paper we consider three measures of utility loss:

Maximum absolute deviation = 𝑛max 𝑖=1

|𝑦∗𝑖 − 𝑦𝑖 |

Average absolute deviation = 1 𝑛

𝑛∑︁ 𝑖=1

|𝑦∗𝑖 − 𝑦𝑖 |

Hellinger distance =

√√ 1 2

𝑛∑︁ 𝑖=1

(√︃ 𝑦∗ 𝑖 − √

𝑦𝑖

)2

Here, 𝑛 is the total number of cells to be published, and the 𝑦𝑖’s and 𝑦∗ 𝑖 ’s are the original and perturbed

frequencies, respectively. The Hellinger distance is a common measure that provides a good overall assessment. The average absolute deviation is a number that is very easy to understand and interpret. By looking at that number, you get quick information about the degree of perturbation. In addition to overall measures, we will also make sure that there are no single deviations that are too large. Some large deviations may render the data essentially useless to some users. Therefore we also consider the maximum absolute deviation. In practice, we may also look more closely at several of the biggest deviations.

3 The Framework Applied: Evaluating 2021 Grid Data Protection Methods

The framework presented above was the basis for comparing multiple perturbative methods for the the dissemi- nation of multi-grid data in the 2021 population census. In this section, we present a shortened summary of the findings. All the details and results will be presented in a future publication. We begin by giving an overview over the methods and parameters used, before moving on to the evaluation.

3.1 The Perturbation Methods Considered

3.1.1 Cell Key Method. The cell-key method (CKM) introduced in Thompson et al. (2013) is a perturbative method which produces a cell’s noise based on its contributing units: each record is stochastically assigned a record key, which are in turn used to determine a cell key. This cell key is used together with a reference noise table to determine a cell’s noise. In this manner, two cells with exactly the same contributors are guaranteed to to be perturbed with the same noise. This is particularly beneficial for dynamic table generators such as the Australian Table Builder and the functionality in the microdata.no platform, where this feature ensures consistency of noise across multiple tables sharing cells. CKM, in combination with targeted record swapping which we briefly discuss in the following section, are the current Eurostat recommendation for statistical disclosure control in the 2021 population census.

4

This method does not preserve table additivity, however it does provide a fixed bound on how much noise is added on the cell level. Extensions to the approach that preserve or fix table additivity do exist, though come with their own caveats: for example, increasing time complexity of the algorithm, or yielding non-integer frequency counts. Furthermore, the base version of the cell-key method does not perturb cells with no contributors (as the noise is dependent on its contributors). Again, extensions exist which take this into consideration, where cell keys also rely on categorical information relating to the cell (in addition to its contributing records). However, for the sake of the evaluation of the 2021 population census, we refer to the original method where cell keys are determined solely based on contributing record keys. Zero frequency cells are therefore not perturbed by CKM, and this must rather be handled by other means, such as targeted record swapping. The evaluation is performed with noise ranging from −5 to 5, with a minimum allowed non-zero frequency of 3. The noise table is generated using the ptable R package (Enderle and Giessing, 2022).

3.1.2 Targeted Record Swapping. Targeted record swapping is a SDC method for protecting sets of microdata. This method aims at protecting locally unique records from disclosure by identifying unusual/unique records in a region and swapping them with similar records from neighboring regions. Common implementations of the technique rely on 𝑘-anonymity for determining similarity between records. In our evaluation, targeted records are swapped with nearby regions according to the Norwegian national grid squares. To achieve this, we rely on the implementation in the sdcMicro R package (Templ et al., 2022). Our evaluation included both random and targeted record swapping. The former selects a percentage of records at random and swaps them with similar records in nearby regions (achieved by setting the k_anonymity parameter to 0). The latter identifies unique records and prioritizes them for swapping with similar records in nearby regions (achieved by setting the k_anonymity parameter to 2). In this paper, we only present the results for targeted record swapping, as it had overall better performance than random record swapping and is the official recommendation from Eurostat. Furthermore, the evaluation was run for both 1% and 10% swap rates as input parameters. However, the actual swap rates after running the method differed greatly from the input parameters: 22.8% and 25.3% respectively. Therefore for the sake of brevity, we present only the results for targeted record swapping with 1% swap rate in this paper.

3.1.3 Small Count Rounding. The small count rounding method (SCR) described in Langsrud and Heldal (2018) is a perturbation method aimed to produce consistent and additive frequency tables without small counts. The method is about changing frequencies of the inner cells, which are the microdata aggregated into frequencies. Identical rows in the microdata are replaced with a single row and a frequency value. A heuristic algorithm ensures good and fast solutions. The method is implemented in the R package SmallCountRounding (Langsrud and Heldal, 2022), which has been continuously updated with new functionality. In this paper we consider three variants of the method. SCRsimple: This is the basic method with three as rounding base. Ones and twos in the published tables are avoided by changing a limited number of ones and twos in the inner cells to zeros and threes. SCRzeros: This is similar to the method above (SCRsimple) an in addition inner cells with zero frequencies are treated as candidates to be rounded up. This way some zeros will be perturbed. The method requires that the inner cell data includes zeros. However, including all possible zeros is not feasible for this type of data. Instead, a limited number of random inner cells with zero frequency were added using the Extend0 function provided by the SSBtools package (Langsrud and Lupp, 2022). Care was taken to avoid introducing structural zeros, such as impossible combinations of geographical areas. SCRforceInner: This is similar to the method above (SCRzeros), but with the change that all inner cells are rounded. Thus, the feature that limits the number of inner cells to be rounded has been disabled. Additionally, it is worth noting that in this study, we utilized the weight parameter of the algorithm. Small original frequencies were downweighted. Given the large size of the dataset, a special looping feature, known as the PLSroundingLoop function, was also employed.

5

Table 2. Attribute disclosure risk (ordinary and where original population is 1) measured as precision and recall.

Precision Recall Method Ordinary Total is 1 Ordinary Total is 1 CK 61.09 100.00 67.50 26.27 CKswk2r01 42.88 58.36 42.38 58.04 SCRsimple 59.84 100.00 63.14 24.43 SCRzeros 55.95 85.05 57.54 23.97 SCRforceInner 52.92 79.48 53.15 22.48

3.2 Results

For each method and variant, we generate a perturbed data set containing all cells across all the different grid systems. For each of these data sets, we measure precision and recall according to the defined disclosure scenarios, as well as utility loss according to the measures defined in the previous section. In order to determine the risk one considers acceptable, one must consider how to prioritize the scenarios and precision/recall measures. For this evaluation, we consider precision a more important metric than recall: intuitively, precision represents how certain an attacker can be about a disclosure. Clearly, the lower the risk measure, the better. Therefore, we automatically disqualify all methods that have 100% precision or recall: with 100% precision an attacker can know with certainty that a disclosure is real. With 100% recall an attacker would have access to all real disclosures. Though 100% recall can, in theory, be somewhat mitigated by low precision, for the sake of this evaluation we consider this problematic. In the full evaluation, all disclosure scenarios were considered. Indeed, one can incorporate more granularity be considering certain variables or categories as sensitive. Then one can measure the risk associated to the disclosure scenarios for these. However, in the context of this article we present only the values that had the greatest impact, and illustrate the largest differences between the considered methods. Tables 2, 3, and 4 show the results of the evaluation.

Ordinary attribute disclosure: Precision performance with respect to ordinary attribute disclosure dif- fered only slightly, as seen in Table 2. Considering this risk measure in isolation, the cell-key method combined with targeted record swapping (CKswk2r01) performed best, with a precision of 42.88%, whereas the cell-key method alone (CK) had the worst performance, with a precision score of 61.09%. All of these values can be considered acceptable, as an attacker can at most be approximately 61% sure that a disclosure is real.

Attribute disclosure where total is 1: Norway is a sparsely populated country. Therefore, many grid cells contains few people, and it is reasonable to assume that an attacker can know that a grid cell contains only one person. All methods that leave zeros unperturbed will result in 100% precision, which we deem unacceptable. This can be seen in Table 2, where the standard cell-key method (CK) as well as the simple small count rounding method (SCRsimple) have 100% precision. Of the remaining methods, a combination of the cell-key method with targeted record swapping performs best with precision at 58.36%, whereas the remaining variants of small count rounding (SCRzeros and SCRforceInner) have similar values at 85.05% and 79.48% respectively.

Disclosure of existence: Disclosure of existence is not about specific individuals, and thus in this context is only problematic when cell totals are low. This is a common occurrence is countries such as Norway. In Table 3, we see that if a grid cell’s published frequency is 4 or 5 persons, the SCRzeros has 100% precision for disclosure of existence. As we deem this unacceptable for low frequency cells, we must exclude SCRzeros.

Two methods remain which have an acceptable level of risk: CKswk2r01 and SCRforceInner. We wish to determine which of the methods maintains the greatest utility. Table 4 summarizes the results of utility

6

Table 3. Risk of disclosure of existence measured as precision and limited to cases where the perturbed frequency has a specific value (1-6). Values less than 3 are not published, hence the first two columns are empty.

Method 1 2 3 4 5 6 CK 100.00 100.00 100.00 100.00 CKswk2r01 80.99 84.88 88.89 92.89 SCRsimple 100.00 100.00 100.00 100.00 SCRzeros 92.77 100.00 100.00 99.71 SCRforceInner 89.65 99.16

Table 4. Utility measures for each perturbative method.

Method

Maximum absolute deviation

Average absolute deviation

Hellinger distance

CK 5 0.989 886.6 CKswk2r01 13401 2.098 1245.8 SCRsimple 17 0.944 891.5 SCRzeros 15 1.008 968.6 SCRforceInner 19 1.304 1052.3

measurements. Here, CKswk2r01 performs considerably worse across the board as compared to the flavors of small count rounding, with a maximum absolute deviation of 13401, compared to 19 for SCRforceInner. Analyzing the underlying data more closely, and considering the deviation relative to the original frequency, the large deviation in CKswk2r01 is approximately 35% of the original value, which we deem unacceptable. Thus we are left with SCRforceInner as the preferred method.

4 Conclusion

In this paper, we present a general framework for comparing SDC methods for frequency table protection. It was applied on multi-grid publication of 2021 population census data in order to compare different flavors of the cell-key method, small count rounding, and targeted record swapping. The results indicate that, in this particular case in Norway, small count rounding where inner cells are forced to be rounded (SCRforceInner) is the preferred method. The presented framework has focused on perturbative methods. However, in general the framework could be applied to comparing both non-perturbative and perturbative methods. The only condition posed to the disclosure scenarios used to measure risk are that one can count how many disclosures there are, something that is also possible for non-perturbative methods such as cell suppression. The main challenge when adapting the framework to include non-perturbative methods is in finding utility measures that are suitable for both non-perturbative and perturbative methods. Furthermore, both the choice of utility measure and the choice of risk measure (precision and recall) can likely be fine-tuned or adapted. One could extend the utility measurements by including measures of information loss, for example Kullback-Leibler divergence or by measuring the variation of information. Regarding risk, there are multiple ways of combining precision and recall into a single measure, allowing for instance a prioritization of one over the other. This would have the benefit of a single value for comparison. However, we consciously decided not to do so, as both precision and recall provide different, orthogonal insights, and summarizing this into a single value appeared to obfuscate some of the nuance. Despite this decision, it is likely a fruitful direction of future research and experimentation.

7

Finally, the results of the evaluation should not be interpreted as a conclusive answer as to which method is best. The comparison framework in this paper is intended to illustrate the different effect various protection methods have given different situations. Including other methods and variants in the comparison, such as cell-key methods with perturbation of zeros or additivity modules, is an obvious candidate for future research.

References

Enderle, T. and S. Giessing (2022). ptable: Generation of perturbation tables. R package on github.com/tenderle/ptable Version 0.3.3.

Geyer, F., R. Tent, M. Reiffert, and S. Giessing (2022). Perspectives for Tabular Data Protection: How About Synthetic Data? In J. Domingo-Ferrer and M. Laurent (Eds.), Privacy in Statistical Databases, Volume 13463, pp. 77–91. Cham: Springer International Publishing. Series Title: Lecture Notes in Computer Science.

Heldal, J. (2017). The European Census Hub 2011 Hypercubes - Norwegian SDC Experiences. In Work Session on Statistical Data Confidentiality. Skopje, The former Yugoslav Republic of Macedonia, September 20-22 , 2017.

Langsrud, Ø. and J. Heldal (2018, 09). An algorithm for small count rounding of tabular data. Privacy in statistical databases, Valencia, Spain.

Langsrud, Ø. and J. Heldal (2022). SmallCountRounding: Small Count Rounding of Tabular Data. R package version 1.0.2.

Langsrud, Ø. and D. Lupp (2022). SSBtools: Statistics Norway’s Miscellaneous Tools. R package version 1.3.4. Spicer, K. (2021). Statistical Disclosure Control (SDC) for 2021 UK Census. In

https://uksa.statisticsauthority.gov.uk/wp-content/uploads/2020/07/EAP125-Statistical-Disclosure-Control- SDC-for-2021-UK-Census.docx.

Templ, M., B. Meindl, A. Kowarik, and J. Gussenbauer (2022). sdcMicro: Statistical Disclosure Control Methods for Anonymization of Data and Risk Estimation. R package version 5.7.4.

Thompson, G., S. Broadfoot, and D. Elazar (2013). Methodology for the automatic confidentialisation of statistical outputs from remote servers at the Australian Bureau of Statistics. Joint UNECE/Eurostat Work Session on Statistical Data.

8

A Disclosure-Based Framework for Comparing Frequency Table Protection

DANIEL P. LUPP AND ØYVIND LANGSGRUD

Background

• Work as part of a grant for publishing multigrid geographical

data on the 2021 population census

• Goal: compare perturbative methods to determine best solution

for Norway

This talk:

• Present comparison framework

• Discuss results for 2021 population census

Comparison framework

• Measure risk

◦ Describe disclosure scenarios

◦ Count and compare disclosures in original and perturbed data

◦ Exclude methods with unacceptable risk

• Measure utility

◦ Of remaining measures, keep methods with highest utility

Disclosure scenario Attribute disclosure

• all records in a table marginal share the same attribute for a

given variable

Municipality Unemployed Employeed Self- employed

Total

M1 12 0 0 12

M2 5 6 0 11

Disclosure scenario Attribute disclosure when total is 1*

• Similar to ordinary attribute disclosure, but limited to marginal

cells where the population total is (known to be) 1

*relevant in, e.g., sparsely populated countries, where even large geographical areas can contain

very few inhabitants

Municipality Unemployed Employeed Self- employed

Total

M1 3 0 0 3 (1)

M2 5 6 0 11

Value is known to be 1

Disclosure scenario Negative attribute disclosure

• When no record contributing to a marginal cell has a certain

attribute

Municipality Unemployed Employeed Self- employed

Total

M1 12 0 0 12

M2 5 6 0 11

Disclosure scenario Disclosure of existence

• Any non-zero frequency discloses that at least one record has a

certain attribute

Municipality Unemployed Employeed Self- employed

Total

M1 12 0 0 12

M2 5 6 0 11

Measuring risk

• Use measures from information retrieval

• Precision: 𝑐

𝑏 Approx. probability that a disclosure is real

• Recall: 𝑐

𝑎 Proportion of real disclosures in «visible» data

Disclosures in original data

Disclosures in protected data

a bc

Measuring utility

• Maximum absolute deviation

max 𝑖

𝑦𝑖 ∗ − 𝑦𝑖

• Average absolute deviation

1

𝑛 σ𝑖=1 𝑛 𝑦𝑖

∗ − 𝑦𝑖

• Hellinger distance

1

2 σ𝑖=1 𝑛 𝑦𝑖

∗ − 𝑦𝑖 2

Applied to 2021 population census grids

• Cell key method

◦ Idea: noise based on which records contribute to a cell

◦ No additivity in tables

• Targeted record swapping

◦ Idea: swap units with risk of disclosure with units from neighboring areas

• Small count rounding

◦ idea: round the inner cells (microdata aggregated to frequencies)

◦ Maintains additivity within and consistency across tables

Different flavors considered

Comparison was done with many (combinations of) methods. For

illustration, we only show:

• Cell key method with/without targetted record swapping:

◦ Labels CK and CKswk2r01 respectively

• Small count rounding:

◦ SCRsimple: simple method, inner cells that are 1 or 2 are rounded to 0 or 3

◦ SCRzeros: same as simple, but zeros rounded as well

◦ SCRforceInner: all inner cells are rounded to multiple of 3

Risk threshold

Need to define what is «acceptable risk»

• This was difficult, so we rather defined «unacceptable risk» as

precision or recall at 100%

• This was actually sufficient to reach a conclusion

Results: Risk

Results: Utility

Approx. 35% of original cell value

Concluding remarks

• Comparison done with Norwegian use case in mind

• Possible refinements: consider loss of information as utility

measure

◦ E.g., Kullbach-Leibler divergence, variation of information

• Risk measures can work on non-perturbative measures, but

work is needed to compare utility loss between non-perturbative

and perturbative measures.

Takk!

Abstract

Sharing economy or just utilization of new business models?

Authors: Camilla Rochlenge, Randi Johannessen

The year 2019 was when the sharing economy and its collaborative consumption was starting to make

a bigger impact on Norwegian society and way of life. With international hospitality and mobility

services leading the way, also several digital platforms developed domestically saw noticeable growth

in its users and revenue. New legislation was put in place to support an orderly transition to an

economy that makes better use of idle resources. However, the COVID-19 pandemic caused a major

temporary setback to this development.

The sharing economy offers a quick and cheap way of matching supply with demand for goods and

services. The main innovation in the business model of the sharing economy lies in the technological

platforms such as smartphone apps which bring demand and supply together. There are two main

types of sharing platforms: peer-to-peer (P2P) and business-to-consumer (B2C). In P2P demand and

supply are matched via a digital platform developed and operated by a third entity who usually charges

a fee of a fixed percentage of each transactions’ payment. Typical examples are platforms such as

Airbnb and Uber, two major players in the sharing economy. Due to the growing popularity of the P2P

business models, more traditional commercial firms are also adapting their economic model to

incorporate this concept of “sharing” into their companys portyfolio. This type of business (B2C)

implies direct contact between the commercial provider and their customers via sharing platform apps

or by adapiting the providers own app or platform.

The aim of the paper is to define and delineate sharing economy within the P2P and B2C plattforms.

We find that although the underlying business model of the sharing economy keeps growing, the

consumption within the P2P segment in Norway is still limited, while there is an increase in the B2C

segment. Further, based on data from the Norwegian Tax Authority, the paper will demonstrate the

limitations and the challenges of estimating a proper price index for accommodation within the sharing

economy.

1 Introduction The sharing economy as a sizable phenomenon is relatively new and due to Norway being a small

country with a small market, it may be subject to international companies operating with platforms

based on mature technology after testing their set up in other countries first. New legislation was put

in place in 2018 to support an orderly transition to an economy that makes better use of idle resources.

And 2019 was the year when the sharing economy and collaborative consumption was starting to make

a bigger impact on Norwegian society and way of life. With international hospitality and mobility

services leading the way, also several digital platforms developed domestically saw noticeable growth

in the numbers of users and income. However, the occurance of the COVID-19 pandemic in 2020 dealt

a major temporary setback to the development.

The sharing economys business models utilization of technological platforms such as smartphone apps

provides an enviroment where demand and supply can meet at “an instance” independent of time

zones and geography. The business model is found in a wide range of sectors, although currently most

noteworthy within tourist accommodation and personal transport, such as taxi services and sharing

of vehicles. Since the term “sharing economy” appeared around 2008 the phenomenon has grown

alongside the rise of peoples’ omnipresent connection to the web through smartphones, all while the

activity within the sharing economy has evolved during the same period of time. In this paper we will

describe multiple definitions existing in Norway of what is considered as sharing economy. We also

aim to identify which economic activity is covered within the sharing economy platforms in Norway,

and that the sharing economy business model is widespread both in the B2C and P2P segments, also

showing that the P2P segment for the time being is rather small. We will furthermore demonstrate the

limitations and challenges of estimating a proper price index for accommodation within the sharing

economy based on data from the Norwegian Tax Authority. As of now it is still preferable to obtain

data from Airbnb or other platforms directly. But in the future with sufficient adjustments the

Norwegian Tax Authority data may prove useful as a source for price information.

2 Definition of sharing economy A random search online for the definition of the sharing economy results in “noun: an economic

system in which assets or services are shared between private individuals, either free or for a fee,

typically by means of the internet”. However in Norway other, more specific definitions exist, among

them the definition by a Norwegian Official Report (Government.no, 2017:3) which states that the

sharing economy is “economic activity enabled or facilitated via digital platforms that coordinate the

provision of a service or the exchange of services, skills, assets, property, resources, or capital without

transferring ownership and primarily between private individuals.”

Next, the sharing economy is defined by The Norwegian Tax Administration as “a business model

where private individuals sell services or rent out assets directly or through intermediary companies”

(The Norwegian Tax Authorities, 2022). Payment may be returned as services in kind, instead of money.

As a clear distinction between a hobby and a commercial activity is not defined, to identify what

category the activity falls within the Tax Administration suggest the following assessment to be carried

out in effort to identify whether the activity:

- is carried on at the business’s own expense and risk

- has a certain scope

- is likely to generate a surplus over time

- is aimed at having a certain duration

Another definition that is based upon three key features that characterize the sharing economy is

provided by Fafo (an independent social science research foundation associated with the largest

Norwegian labour union) (Jesnes et al., 2016:7):

- An intermediary in the form of a digital platform.

- Which helps to connect complementary players, which can be considered as providers and

customers.

- Who exchange a set of benefits from the provider to the customer. There can be a wide variety

of benefits, from services and asset/property sharing to capital, expertise, and labour.

In all the above definition the peer-to-peer (P2P) transaction is a defining characteristic of the sharing

economy, however in the last definition by Fafo it is the contact facilitation of the P2P transaction

which defines the sharing economy, not the sharing element itself.

3 B2C and P2P There are two main types of sharing platforms: peer to peer (P2P) and business to consumer (B2C). In

P2P demand and supply are matched via a digital platform developed and operated by a third entity

who usually charges a fee of a fixed percentage of each transactions payment. Typical examples are

platforms such as Airbnb and Uber, two of the best-known examples of sharing economy models0F

1.

Following the strong growth of P2P business models, two trends have occured. Some suppliers expand

operations to investing in more rental units, thereby transcending from a P2P supplier to becoming an

owner of several units and operating as a B2C supplier within the same platform. Simultaneously, the

traditional commercial firms adapt their economic model to incorporate similar concepts of “sharing”.

This type of B2C business implies direct contact between the commercial provider and their customers

either via the providers own platform or through an established sharing economy platform. According

to the definition from FAFO, these activities are not included in the definition of the sharing economy

as these business models are relatively similar to those of traditional traders. Contact through well-

established web sites such as booking.com between hotels and guests are easily defined as

B2C,however when booking.com also include listings of lodging by private owners the distinction

between the two segments becomes less clear as the web sites trancends into also providing stays

P2P.

Given the connection between the National Accounts (NA) the Consumer Price Index (CPI)1F

2 a

collaboration between the price and the NA communities is preferable to ensure progress and

consistency in both statistics. Digitalisation leads to a shift in the production boundary with more

activities taking place within the household. The traditional assumption in NA is that firms create value

added as producers, while households/individuals are consumers only. Due to the limited role

of households as producers, their value added is recorded in the informal economy (IMF Committee

on Balance of Payment, 2020). We now face an increasing number of individuals who participate

directly as “producers” in activities related to the sharing economy. For instance, we see a growing

trend of trading second-hand goods like clothing, furniture, electronics, books, etc. This trend is

facilitated by the simplicity brought to the second-hand market by P2P sharing platforms. The practice

of P2P in general not being measured in NA applies for every area of the economy with one exception;

1 Consumers are also using digital networks to lend office space, parking spots, boats, bicycles, cameras and more. 2 Throughout the paper CPI also refers to the Harmonised Index of Consumer Prices HICP)

for accommodation services where a correction is performed to the housing service by owner

occupiers of houses which otherwise is registered as production in the NA.

4 Accommodation Arguably among the most well-known sharing economy models are Airbnb, which has been said to

have disrupted the industry of accommodation when entering the market as a competitor operating

under rules differing from the ones existing in the established market. Airbnbs P2P offered

accommodation service may feel different from stays provided through traditional accommodation.

Differences are present through the accomodations physical attributes and its less visible ones such as

different requirement, such as for instance building risk assessment and other similar national

regulations mandatory to the existing accomodation service while not required in the regular housing

market. Hence, the two service options should be seen as different products in price statistics. As the

market share of Airbnb and the likes differs between countries, the inclusion of these services in the

CPI sample must be determined individually by each country. In Norway short-term rental services

like Airbnb are still rather small. Based on figures from 2017, NA currently estimates the household

expenditure share to be below 0.1 per cent of total household consumption, but there are indications

that the share is steadily growing and is expected to grow in the years to come. The question about

whether rentals through Airbnb are to be considered B2C or P2P remains to be answered.

Traditional accommodation services such as hotels, motels, inns, and their likes operate within a legal

context supporting the supplier and consumers in the existing markets. However, the existing

legislation did not fully cover the activity made possible by sharing platforms which enabled peers to

easily offer lodging under the safeguarding of the platforms terms and condition all while connecting

the host to the “whole world” in an instance. As platforms such as Airbnb offer user profiles at no fee,

the barrier was lowered substantially for peers to put an offer out for lodging while at the same time

increasing the awareness of these possibilities for potential hosts.

The economic efficiency from the sharing economy model, which make it easier to rent out underused

assets, is in general welcomed by the Norwegian government who appointed a Sharing Economy

Committee in March 2016. The committee was asked to evaluate opportunities and challenges

presented by the emerging market phenomenon (Government.no, 2017). Among other things, the

Committee was tasked with identifying and assessing regulatory provisions challenged by the sharing

economy, identifying the consequences of the sharing economy on the labour market and finally, the

Committee was requested to consider consumer protection rules and the objective of consumer

safety.

In the wake of the committee’s findings the government took legislative action. In effect from 2 April

2019 a new short-term rental law was effectuated allowing apartments in housing cooperatives to be

rented out for a total of 90 days per year, while previously these types of short rentals were not

allowed at all. Furthermore, the law made it illegal to own more than one unit in each housing

cooperative. The intention of the new law is to balance the interests of those who wish offer lodging

in their home and their neighbouring residents. The new rules state that for rentals where the length

of stay is less than 30 days for each individual letting, the revenue is taxable under the standard method

i.e.: that revenue from rental up to NOK 10,000 (around 850 € April 2023) is tax-free, while 85 percent

of the remaining surplus revenue is considered taxable income. Rental revenue equals the total fee

paid by the renter to the host including all additional cost related to the individual letting (The

Norwegian Tax authorities, 2021).

5 Transport services Among the most well-known and highest profile companies within the sharing economy are the

ridesharing companies Uber and Lyft. However, Norway has chosen a legislation which acts counter-

current to many other countries in the transportation field within the sharing economy.

5.1 Taxi services As of November 2020 a new taxi market reform took effect in Norway, postponed from July 2020 due

to the pandemic (www.government.no, 2021). The main elements in this reform was linked to shifting

rights and responcebilities from the taxi license holders to the taxi drivers, in addition to deregulationg

the numerical restriction on number of licences.

The deregulation of the market due to the taxi reform has led to a huge increase in the number of taxi

licenses by 45% on a national basis and as much as 69% only in Oslo. New companies, like Yango, Bolt

and Uber, providing taxi services have entered the market since the reform took place. These new

companies in the market are all foreign, which limits the Norwegian Statistics Act legal force to to

oblige data delivery to Statistics Norway. Due to the taxi reform the taxi companies are not required

to be connected to a dispatcher. This makes it difficult to obtain data for taxi rides, probably even for

the regular taxi rides in the future, as comprehensive data ideally should be obtained from each taxi

driver.

Uber provided their services prior to the reform, but had to abandon their operation in Norway after

a damaging court case in 2017. There are reasons to believe that the increase in the number of taxi

licenses is connected to the establishment of these new platform companies. However, since the

taximeter requirement is not yet removed, none of the new companies can operate entirely within the

P2P business model, as one is required to holding a professional taxi licence2F

3 as well as registering the

vehicle as a taxi. The latter requirement includes a yearly EU periodic roadworthiness check, as

opposed to a biannual check of roadworthiness required for a car purposed for private usage. Regular

taxi drivers have also started driving for the new companies in addition to dispatching central. Yango

and Bolt have registered as transport companies and not as taxi operation as given in NACE, probably

to avoid some of the rules that a taxi driver/taxi company is subject to, among others the requirement

of a taximeter.

According to the Tax Authorities, if you decide to make driving your main source of income, you must follow the general tax and reporting rules that apply to businesses. The general rule will then be that the income from the driving is taxable from the first NOK, and the expenses associated with the driving are deductible.

As the current regulation does not allow a taxi service purely through the P2P segment, the platforms

are not able to fully make use of the P2P business models. In the future, if the taximeter requirement

are replaced by digital platforms, P2P offered transportation services may reach significant market

shares.

The question of whether or not taxi fares from the regular taxi companies work as a proxy for prices in

the P2P segment still remains unanswered. Most likely, to gain market share in the Norwegian market

the price level for taxi rides by the new companies will not surpass the fares in the existing taxi market.

However, we do not have any information whether the price development differ from the regular taxi

rides.

3 The fee for getting a taxi license issued is at present NOK 3400 (around 300 € April 2023)

Following the changes in the market brought to us by the new reform, some political parties are now advocating a reversion of the reform due to complaints about too many licenses in the market, leaving drivers without enough clients to reach decent wages. According to economic theory prices should drop in the face of supply overbidding demand, and the digitalisation within the taxi market has opened for more differeansiated prices (Aftenposten, 2023).

5.2 Vehicle sharing Several companies in Norway are offering vehicle sharing within the B2C segment. The companies are

a mix of Norwegian and foreign.

As an alternative to private car ownership, organised carsharing is a system that offers people to rent

cars locally available at any time and for any duration. Carsharing has existed in Norway for over two

decades, however the number of users is still limited. The first carsharing providers in Norway were

member-owned cooperatives in Norways three largest cities: Oslo, Bergen, and Trondheim. The

carsharing stations were almost always located in central areas with a high enough residential or

business density to sustain a viable customer base. Currently, seven carsharing providers within the

B2C segment operate in Norway. It is estimated that around half of the members are passive members.

One platform offers carsharing within the P2P segment, with about 10 000 cars in their registry (figures

from 2021).

Carsharing users generally tend to be more urban, wealthy, educated and younger than the general

population (link.springer.com, 2023). A typical user is between 30-40 years old, has higher education

and fewer cars in the household. The biggest motivations for memberships are related to convenience,

the financial aspect and the environment. Carsharing is primarily used for holiday and leisure trips as

well as for shopping heavy goods, and rarely used for everyday travel such as commute. As more of

the following generations grow up in families who do not own a car the phenomenon of carsharing

may increase.

Allthough the station based cooperative model is the most established model, newer types of

platforms, both in terms of organizational model and operational model, have entered the market

since 2015. What remains to be seen is who will be the dominant players and what the dominant

platforms will be in the future.

According to the Tax Authorities you do not have to pay tax on renting out your car if your rental

income is up to NOK 10,000 (around 850 € April 2023) per year. It makes no difference whether you

rent out the vehicle yourself privately or through an agent.

Smaller vehicles such as e-scooters was legalized in Norway in 2018, and since then several e-scooter sharing companies have established themselves in Norwegian cities offering around 20 000 vehicles. In Oslo, and elsewhere, unregulated e-scooter markets create challenges with respect to traffic safety and littering of excess vehicles resulting in an introduction of a new regulation in 2021 which limmit the number of companies in Oslo to three and the number of vehicles reduced to 8000, down from previously 23 000. All companies that operate in Norway are within the B2C segment, as the vehicles are owned by commercial companies. The same is the case for bicycles, both regular and electric,which also operate commercially.

Table 1. Types of sharing economy and share of total private consumption

Type of service Expenditure share, CPI (%)

P2P’s share of expenditure

Comment

Accommodation 0.8 Not significant* Both B2P and P2P at play, legislation adapted to both business models

Taxi services 0.3 Not significant* P2P business models not fully utilized yet as the requirement are like regular taxies

Carsharing (rental car) 0.1** Not significant* One companies offering P2P services. Several B2C companies are established in the market

*Less than 0.1 % of total private consumption according to NA **The expenditure share is for rental cars

6 The Covid-19 pandemic and the sharing economy The rapid development of the sharing economy in Norway was dealt a major setback due to the COVID-

19 pandemic of 2020 (Halvorsen, 2021). In 2023 it is still not fully clear what of the pandemics impacts

remain permanent, and whether the rapid changes experienced pre-pandemic may soon return when

the COVID-19 pandemic converges towards an endemic stage.

Figures from Statistics Norway’s accommodation statistics show a sharp decline in guest nights at

commercial accommodation establishments in 2020. Norwegian guest nights declined by 17 per cent,

while foreign guest nights declined by 69 per cent. Increased guest nights by Norwegians in the

summer, especially at camping sites and holiday dwellings and youth hostels, did not compensate for

the absence of foreigners. As restrictions were loosened and the willingness to travel domestically

rose, the number of guest nights increased by 14 per cent from 2020 to 2021. In 2022, when most

restrictions were liftet worldwide, the total number of guest nights rose to almost 3 per cent above

the pre pandemic level in the year of 2019.

Similar data for Airbnb lodging in Norway is not public, but figures for Airbnb nights & bookings

worldwide (FourWeekMBA, 2023) describe a substantial rise from 2017 to 2019, while figures dropped

to reach the 2017 level in 2020, before once again climbing steeply through 2021 to reach an all-time

number of bookings in 2022. Although the rules and regulations differed between countries

throughout the pandemic, some similarities were present; rules and regulations which serve to limit

the contact between people and reduce the likelihood of spreading the virus. It is likely that the decline

observed in commercial accommodation in Norway corresponds to a decrease in the activity facilitated

by Airbnb worldwide.

Also the transportation services were hit hard by the Covid-19 pandemic. The level of restrictions induced a reduction in demand, with activity increasing during the summer months of 2020, although variations between different segments were observed; the street segment was hit hardest, while the contract market segment3F

4 seems to have performed better.

4 The taxi service industry can be divided into two segments, the single trip segment, and the contract segment, e.g. contract driving for public authorities or companies who negotiate fares for multiple trips. In the single trip segment customers either order a taxi through a dispatching service companies or hail a taxi from a taxi rank or from the street.

For taxi owners and employed drivers, the reduced demand in the early phase of the pandemic led to many temporary lay-offs and parked cars. Many taxi owners applied for compensation, with those who own multiple cars having a much higher chance of getting their claim for compensation accepted. Some of the temporarily laid-off drivers likely received an equal or larger sum in unemployment benefits than they would have been paid in wages if they were to continue to work in a market with severely reduced demand. Combined with the deregulation of the taxi market in November 2020, the pandemic made many taxi owners and employed drivers leave the industry.

While sales of new cars in Norway faced new records in 2021 and the government instructed people

to avoid the use of crowded public transport as an attempt to stop the spread of the Covid-19

culminating in historic low passenger demand for public transport use especially in the capital of Oslo,

there are reasons to believe that private driving increased during the pandemic, and therefore also the

use of carsharing in 2020 and 2021.

No data are found for city-bicycles during the pandemic period in Norway, however dealerships of new

electrical bicycles reportet new sales records during this period.

7 Taxable income data, - a possible data source? Legislative action was introduced in 2018 to address short-term rentals (defined as rental periods of

30 days or less) resulting in a softening of the regulation of the housing market. The deregulation

opened up for subletting apartments in housing cooperatives for 90 days per year, as opposed to

earlier restriction which forbid renting out these types of self-owned apartments.

Income from the sharing economy is liable to taxes, and as of February 2021 all platforms providing

connection and facilitating payment between parties involved in renting lodging services in Norway,

both Norwegian and foreign, are obliged to report information about each unique rental, regardless

of the duration of the stay. Statistics Norway was granted full access to these data from the year 2020

through an agreement with the Norwegian Tax Authorities.

In its most severe form travel bans due to the Covid19 pandemic, effectuated in the spring of 2020,

restricted inhabitants to stay within their registered municipality. Hence the figures derived from the

Norwegian Tax Authority data for 2020 must be viewed as highly affected by the rules and regulations

imposed on international travel and national movement in the period the data covers. In comparison

several hotels shut down during the early stages of the pandemic. Most likely the following year is also

affected by the pandemic as restrictions were imposed with variable strength and strictness in Norway

and the rest of world throughout 2021 . This hypothesis is supported by figures for number of nights

conveyed through Airbnb throughout the last six years, where the number of nights in 2021

accumulated to less than the most recent pre-pandemic year of 2019. Analysis on the year of 2021 will

be released by Statistics Norway later in 2023, henceforth this paper will only describe figures from

the first pandemic year of 2020.

In total slightly more than 400 000 unique rentals4F

5 were registered in the Tax Authorities data for the

year 2020 with about 90 per cent of them related to short-term rentals for up to 90 consecutive days.

The numbers do not include information about rentals where the platform only arrange for the

connection between the provider and the buyer of lodging services as verifiable transaction prices

related to them registered in their system, as these platforms are relieved from the duty to report.

5 Unique rentals are equivalent to each transaction between the one that rents out and the ones renting. Every transaction is considered unique. The data does not identify the renter leaving no option for Statistics Norway to identify renters that repeats their rental on several occasions.

Since then COVID-19 hit, Airbnb launched its “Live Anywhere theme” in 2020, and said: As a result of

the pandemic, millions of people can now live anywhere. They’re using Airbnb to travel to thousands

of towns and cities, staying for weeks, months, or even entire seasons at a time. We want to design

for this new world by making it even easier for guests to live on Airbnb. We believe that living

somewhere enables deeper connections to local communities and the people who live there. In Q4

2022 stays of at least 28 nights accounted for 22% of gross nights booked, while 47 per cent of gross

nights booked were from stays of at leas 7 nights (rentalscaleup, 2022).

8 Analysing the data from the Norwegian Tax Autorities

8.1 Number of stays and revenue In economic figures revenues from lodging reported to the Norwegian Tax Authorities was 1.7 billion

NOK for 2020, while the corresponding total revenue for hotel accommodation from official statistics

was 9.4 billion NOK. On average the price for lodging was 1100 NOK per night stay, while the average

price per night in a hotel room in 2020 was 979 NOK (Statistics Norway, 2021). Be aware that these

figures do not say anything about the size and location of the rental, not the number of people staying

in the unique rental object; all factors that may influence the observed prices.

Further drilling in to the length of stay dimension in the Tax Authority data show that most rentals are

substantially shorter than 90 days5F

6. This corresponds to the general right of vacation days granted

within the EU being 4 weeks, some member states and EFTA members, like Norway, operate with more

vacation days, while the US and Canada have considerable weaker standard rights to paid vacation

days (EurDev, 2021). The observed data showed a boost in the length of stay at exactly one-week

rentals, while the numbers consistently decreased for each more added night of stay. By selecting only

stays consisting of one-week rentals or less we are left with about 84 per cent of the original data

material.

Table 2. Share of stays by lenght

All platforms providing rental agreement in Norway are represented in the Tax Authority data. In this

analysis we aim to identify the ones represented by platforms as defined by the sharing economy

phenomenon. As defined above an important aspect of these exchanges is the distinction of

transaction made “primarily between private individuals”.

6 The Norwegian Tax Authority’s definition of short-term rental (less than 90 days per stay) is adapted to the purpose of tax liability.

Night(s) stay Percent Cumulative Percent

1 27,3 27,4

2 23,3 50,7

3 14,3 64,9

4 7,6 72,6

5 3,9 76,5

6 1,9 78,4

7 5,8 84,3

8 0,7 85,0

9 0,6 85,6

The better part of the data were rentals arranged via Airbnb. Booking.com were well represented in

the data too. However, booking.com operate also in the segment were rentals made through them are

targeted at established brands and entrepreneurs of all sizes (Booking.com, 2022).

8.2 Rental object number The data included information about rental object number. To secure the aspect of transactions

“primarily between private individuals” we assume that private individuals most likely do not operate

with several rental objects and decided to include only rental object numbers of one or two. The higher

rental object numbers from 3 and up covered about 10 per cent of the original data material, leaving

almost 90 per cent for further analysis.

By selecting only rentals with a length of stay of a week or less, only Airbnb and securing the number

of objects rented out by each host to be no more than 2 we believe we are left with a subset of data

that is well within the definition of sharing economy where transactions are made “primarily between

private individuals” as well as it represents lodging acquired by private households in Norway through

this channel. The subset of data after this selection is done accounts for more than 180 000 unique

rentals, equivalent to about 45 per cent of the original dataset. The number of nights of stay in the

subset of data were about 500 000, about 1/3 of the original data material, totalling up to 500 million,

about 30 per cent of the total revenue in the full data set.

In compiling a price index the first step was to derive a unit price per night per unique rental. As the

same rental object may have been rented out several times within one period (month) the unit price

was aggregated to a monthly unit price before a timeline per unique rental object was constructed.

Rental numbers vary throughout 2020. About 60 000 unique rental subjects have a monthly unit price

registered which are unequally distributed throughout the months of 2020.

Table 3. Monthly overview of rentals, per cent.

Action taken by the government during the spring of 2020 to restrict the spread of COVID-19 is visible

through the low activity seen in the spring months of March, April and May. Followed by a summer

where the mobility within Norway was unrestricted and numbers again rose, the lodging numbers

declined as COVID-19 numbers rose through the fall and the government once again enforced strict

regulationsto reduce the spread of the virus to a maintainable scale as the year moved towards

Christmas celebration. Most likely also non-pandemic figures would vary throughout the year, with

high numbers associated / coinciding with national holidays and summer vacation in Norway mid-June

to mid-August, and the following summer holiday season for southern Europe lasting through August.

When measuring hotel prices, the services followed are consistent over time with regards to location,

interior and amnesties included. It is to be assumed that the same rooms are either rented out or

offered for rent. This is opposite to the sharing economy lodging which offers non-commercial

accommodation by private households at the time when the rental object is available for the hosts to

offer the public. Whereas it is possible to measure the same service over time offered by traditional

accommodation services at hotels or other established facilities, the very idea of sharing economy

imposes challenges through its diversity in object offered or actually rented out which may differ

greatly between periods.

8.3 Matching unique rental objects Aggregating the about 60 000 unit price observation per night per unique rental to an annual time

series leaves us with shortly less than 20 000 unique hosted lodgings during the year. Among these

slightly 30 per cent of the unique lodgings were present during only one of the months in the year of

2020. To measure prices over time the lodging object must at least be present in two consecutive

periods (months) or more. The data shows that only very few object (less than 1 per cent) were rented

out throughout every month of 2020, with an increasing percentage of lodging objects appearing when

moving from occurring in twelve months during the year towards only twice.

Table 4. Unique rentals per month, per cent.

As a large proportion of the data are only present in one month of 2020 only a small share are available

for a match with a previous period. The figures illustrate the increase in matches when shifting from

matches towards a fixed base period to matches between two consecutive months.

Numbers of matches during a full year can at maximum reach 11 for one unique lodging object. More

than 70 percent of the lodging object does not match with a fixed base period of January. While the

range differs between a good 6 per cent for a match between January and two other months during

the year of 2020 and below 1 per cent for a match between a unique lodging object in January and the

following eleven months.

Table 5. Overview of unique rental objects rented per month and in January

The numbers of matches increase for unique lodging object when matches are made for two

consecutive months. Almost 36 per cent are unique lodging object rented out only during one of the

months in 2020, while about 23 per cent are found to have been rented out for two consecutive

months and almost 17 per cent were rented out for two consecutive months twice6F

7. The numbers

evenly decrease for each added possible match with only 0,2 per cent of the unique lodging object

being rented out for 11 of the all years twelve months, and a slight rise to 0,7 per cent of all unique

lodgings accounted for were rented out at least once in every month of 2020.

Table 6. Overview of unique rental objects rented in a particular month and the month before, per cent.

Having price observations for the same service is one step along the way to compiling a price index.

We also need to make sure the services we measure prices for in the whole universe of services offered

and consumed are representative services consumed by private households, both with regards to

location, length of stay, size of lodging and the standard of the service provided. When we have access

to data which accounts for all activity which fall under the Norwegian Tax Authority terms of sharing

economy for lodging, covering the geographical boundaries of Norway and channelled both through

nationally and abroad owned platforms, the challenge moves from traditional sample issues towards

more limitations in the data source. Other P2P rentals are probably prevalent throughout the year, but

7 This can either be 3 consecutive months or two consecutive months twice.

rentals which are channelled via a sharing app or website are most likely registered in these data unless

the platform operates under illicit terms. Hence, the Tax Authority data provides a complete overview

over the accommodation activity in Norway under the terms of sharing economy apps and websites.

Lacking in the data is information on the purpose of the stay, if it is to be considered business or

recreational purpose. The lack of such information is well known when measuring prices with the

intention to include in a price index. As long as there is no discrimination between the two consumer

purposes, leaving one of the two with a different price development then the prices measured are

accurate enough. However, if the rental objects are strongly related to the purpose of business, which

does not fall the scope of the CPI, then a bias towards including non-relevant rental objects may be

introduced when the corresponding prices are not properly identified and excluded from the price

material that enters the index. For instance, when booking at Airbnb.com there is an option to mark

an Airbnb reservation as a business trip resulting in Airbnb providing an receipt for expenses while also

providing Airbnb with valuable information about the purpose of the stay. This information is not

conveyed to the Norwegian Tax Authority. In the fall of 2018 Airbnb launched a work program aiming

for an increase from 15 per cent of total bookings stemming from business travel to a 30 per cent share

in 2020 (curbed.com, 2018). For the period of the Tax Authority data which we analyse in this paper,

it is safe to assume that if the business segment it present, then it is limited as the rules and restriction

for work forced typical travel activity to become digital. However, in a situation with no pandemic

restriction these are issues that should be addressed.

The most basic form of an unweighted monthly chained price index for the year 2020 shows a more

volatility and a higher index level throughout the year of 2020 compared with the published series for

11201 Hotels, motels, inns and similar accommodation services.

Chart 1. Price index lodging by Airbnb 2020. January 2020=100.

The experimental index is a monthly chained index for the year 2020. The published index is a weighted

Laspeyre, with December the previous year as the price reference month. For comparability the

published index is re-referenced to January 2020=100 from the official 2015=100.

As described above the experimental index is tainted by several challenges, while the published index

series for some of the months are affected by the pandemic directly through how we treated

consumption which fell close to zero in the period.

First and foremost, the number of prices entering the experimental index varies strongly between the

periods. In general, missing observations in the published index series are imputed in line with rules

according to the principle of nearest neighbour imputation; starting with the most detailed level within

the region the missing price is observed, then drilling upward in the hierarchy . In the experimental

index no such imputation is performed, and prices enter where they exist. Additionally, the published

index is affected by imputation of the overall index of the CPI consisting of the remaining consumption

based on real price observations for the (pandemic) period (Statistics Norway, 2020). With regards to

homogeneity, in the published series homogeneity are ensured as the respondents are asked to price

a representative service of a specified standard, equally stated to all respondents who provide these

services. In the Airbnb data the aspects of the rental object beyond regionality is not registered. The

variation of unique rental object whose prices enter the index may vary substantially both within a

month and over time time. Not performing imputation of the missing basic data in the experimental

index forfeit the possibility to follow the unique rental object over time as missing price observation in

one period introduces a breach to the timeline.

9 Further work The sharing economy within the P2P platforms is for the time being rather small in Norway. Most of

what is described as sharing economy is within the B2C plattforms, just indicating that traditional

business are utilizing the new business models. Worldwide, we find accommodation and transport

services as major services within the P2P plattforms. Through NA we are able to identify an

expenditure share for Airbnb, which is still less than 0.1 per cent of total private consumption. No

data is available to identify a significant expenditure share for taxi services from the platforms

operating within the P2P segment. However, restrictions in the taxi market, making it not so easy to

use your own car, indicates an almost non-existing market share of the total taxi market. The

experimental work on the Norwegian Tax Authority data shows the new possibilities that occur when

access to a new data source appears. Although several challenges remain unanswered, the data

available from the Norwegian Tax Authorities are detailed enough to compile a simple version of a

price index retrospectively.

Data for the following year are yet to be analysed, however we are already aware of more granulated

details introduced in the data for 2021. Utilizing the added level of detail in the data source are

expected to enable further improvements in the processing and delineation of the data, maybe even

increasing the subset of data potentially entering a price index as the added granularity of detail may

prove usefull to subtract the P2P arranged stays from the B2P segment for platforms such as

booking.com which currently are catgorized as fully operating withing the B2P segment in lack of

information to categorize a stay differently.

The primary challenge with the Norwegian Tax Authority data stems from timeliness, as these data

are a one-time extraction for the whole year of 2020, this does not satisfy the timeliness needed in a

price index which should register the prices in the period the service commences.

If or when this data source may be of the right timeliness and quality to be used as a source of price

information to produce the CPI is too early to conclude on. However, these data will be a much-

needed new source of information for NA in their calculation of the production level for Airbnb-

related activities in Norway. And the data in its current set up does shed light on aspects regarding

traditional sampling issues such as specifying the population and selecting a sample with regards to

regionality.

Even though the aggregated expenditure shares for the variety of services provided though the P2P

measured by the NA are yet less than of 0.1 per cent of total consumption, we anticipate a future

need to measure these prices as the sharing economy activity in Norway most likely will become

more prevalent.

References (

Aasestad K, K. J. (2021, November 30). Accommodation offered via online collaborative economy

platforms. Norway 2020. Retrieved from https://www.ssb.no/en/transport-og-

reiseliv/reiseliv/artikler/accommodation-offered-via-online-collaborative-economy-

platforms.norway-2020

Aftenposten. (2023, 05 12). Retrieved from

https://www.aftenposten.no/meninger/kommentar/i/9zMBaq/smarte-drosjekunder-har-

faatt-det-bedre

Andreotti, A. A. (2017). bo.edu. Retrieved from European Perspectives on Participation in the Sharing

Economy: https://www.bi.edu/globalassets/forskning/h2020/participation-working-paper-

final-version-for-web.pdf

Booking.com. (2022). Booking.com. Retrieved from About Booking.com:

https://www.booking.com/content/about.html?aid=318615&label=Norwegian-NO-

131246328204-

NGQNXKWfg44q8FRM%2A4MHhwS562363086939%3Apl%3Ata%3Ap1%3Ap2%3Aac%3Aap%

3Aneg%3Afi2657853280%3Atidsa-

1227182654382%3Alp1010826%3Ali%3Adec%3Adm&sid=3ef3ce6a69ac1cce78b18a7b5f

curbed.com. (2018, OCT 4). Retrieved from Airbnb expands services to corner profitable business

travel market: https://archive.curbed.com/2018/10/4/17938076/hotel-airbnb-meeting-

business-travel

EurDev. (2021, January 22). EurDev. Retrieved from Paid Vacation Days Europe 2021:

https://blog.eurodev.com/paid-vacation-days-europe-2021

Eurostat. (2018, November). ec.europa.eu. Retrieved from Harmonised Index of Consumer Prices

(HICP) Methodological Manual:

https://ec.europa.eu/eurostat/documents/3859598/9479325/KS-GQ-17-015-EN-

N.pdf/d5e63427-c588-479f-9b19-f4b4d698f2a2

Eurostat. (2018). Harmonised Index of Consumer Prices (HICP). Methodological manual. Retrieved

from https://ec.europa.eu/eurostat/documents/3859598/9479325/KS-GQ-17-015-EN-

N.pdf/d5e63427-c588-479f-9b19-f4b4d698f2a2

FourWeekMBA. (2023, February 19). Retrieved from https://fourweekmba.com/airbnb-bookings/

Government.no. (2017). Retrieved from NOU 2017: 4 Sharing Economy - Opportunities and

challenges: https://www.regjeringen.no/en/dokumenter/nou-2017-4/id2537495/

Government.no. (2017:3). Retrieved from NOU Norges offentlige Utredninger: Delingsøkonomien -

muligheter og utfordringer :

https://www.regjeringen.no/contentassets/1b21cafea73c4b45b63850bd83ba4fb4/no/pdfs/

nou201720170004000dddpdfs.pdf

Government.no. (2021, 10 14). Retrieved from Spørsmål og svar om nytt drosjeregelverk:

https://www.regjeringen.no/no/tema/transport-og-kommunikasjon/ytransport/sporsmal-

og-svar-om-nytt-drosjeregelverk/id2641640/

Government.no. (2021). Retrieved from The coronavirus situation:

https://www.regjeringen.no/en/topics/koronavirus-covid-19/id2692388/

Halvorsen, T. C. (2021, OCTOBER). sharingandcaring.eu. Retrieved from The Sharing Economy in

Norway: Emerging Trends and:

https://sharingandcaring.eu/sites/default/files/files/ebook/Chapter_18_The_Sharing_Econo

my_in_Norway_Emerging_Trends_and_Debates.pdf

IMF Committee on Balance of Payment. (2020). Statistical Framework for the Informal Economy.

Retrieved from

https://www.unescwa.org/sites/default/files/event/materials/Informal%20Economy%20Tas

k%20Team-concept-note.pdf

Jesnes et al. (2016:7). Retrieved from Aktører og arbeid i delingsøkonomien:

https://www.fafo.no/images/pub/2016/10247.pdf

link.springer.com. (2023, 03 25). Retrieved from https://link.springer.com/article/10.1007/s11116-

023-10386-0

Newlands, G. L. (2019). The conditioning function of rating mechanisms fro consumers in the sharing

economy. Retrieved from biopen.bi.no: https://biopen.bi.no/bi-

xmlui/handle/11250/2602833

Ranzini, G. E. (2017 - II). bi.edu. Retrieved from Privacy in the Sharing Economy: European

perspective: https://www.bi.edu/globalassets/forskning/h2020/privacy-survey-working-

paper-for-web.pdf

Ranzini, G. N. (2017 - I). bi.edu. Retrieved from Millennials and the sharing economy: European

perspectives.: https://www.bi.edu/globalassets/forskning/h2020/focus-group-working-

paper.pdf

rentalscaleup. (2022, 02 17). Retrieved from https://www.rentalscaleup.com/2022-airbnb-strategy/

Statistics Norway. (2020). Retrieved from Corona consequences for CPI:

https://www.ssb.no/en/priser-og-prisindekser/artikler-og-publikasjoner/corona-

consequences-for-cpi

Statistics Norway. (2021). ssb.no. Retrieved from 12897: Revenue and utilisation of rooms at hotels,

by region, contents and month:

https://www.ssb.no/en/statbank/table/12897/tableViewLayout1/

Statistics Norway. (2021). Travel Survey . Retrieved from https://www.ssb.no/en/transport-og-

reiseliv/reiseliv/statistikk/reiseundersokelsen

Thackway, W. T. (2021). Airbnb during COVID-19 and what this tells us about Airbnb’s Impact on

Rental Prices. Retrieved from Findings: https://findingspress.org/article/23720-airbnb-

during-covid-19-and-what-this-tells-us-about-airbnb-s-impact-on-rental-prices

The Norwegian Tax authorities. (2021). Tax rules for short-term letting of homes and holiday homes.

Retrieved from https://www.skatteetaten.no/en/person/taxes/get-the-taxes-right/property-

and-belongings/houses-property-and-plots-of-land/letting-of-houses-and-property/short-

term-letting-of-dwellings-and-holiday-homes/tax-rules-for-short-term-letting-of-homes-and-

holida

The Norwegian Tax Authorities. (2022). Sharing economy. Retrieved from www.skatteetaten.no:

https://www.skatteetaten.no/en/person/taxes/get-the-taxes-right/employment-benefits-

and-pensions/hobby-odd-jobs-and-extra-income/sharing-economy/

Utleiemegleren.no. (2022). Retrieved from Om Utleiemegleren: https://www.utleiemegleren.no/om-

oss

www.government.no. (2021, 10 14). Retrieved from https://www.regjeringen.no/no/tema/transport-

og-kommunikasjon/ytransport/sporsmal-og-svar-om-nytt-drosjeregelverk/id2641640/

www.ssb.no. (2021). Retrieved from https://www.ssb.no/en/omssb/lover-og-

prinsipper/statistikkloven

Ydersbond. (2023). Erfaringer med lov om utleie av små elektriske kjøretøy på offentlig grunn.

Retrieved from https://www.toi.no/publikasjoner/erfaringer-med-lov-om-utleie-av-sma-

elektriske-kjoretoy-pa-offentlig-grunn-article38033-8.html

https://one.oecd.org/document/STD/CSSP/WPNA(2017)9/En/pdf

Appendix Examples of sharing economy in Norway

Following is a list of some of the sharing apps in the Norwegian market which ranges from singular

focused sharing apps to the all-consumer area apps:

Lodging services: www.airbnb.com

Child care services: www.sitly.no

Transportation by car: www.uber.com

Cleaning services: www.weclean.no

FINN online market (almost everything): https://finn.no

Book market (used and new): https://bookis.com (skal brukt være med?)

Carsharing: https://nabobil.no/en

Services provided by neighbours: www.obos.no/Nabohjelp

Clothes, decoration and furniture (used and redesign): https://tiseit.com

Sharing goods: https://www.hygglo.no/

Sharing economy or just utilization of new business

models? MEETING OF THE GROUP OF EXPERTS ON CONSUMER PRICE INDICES

07- 09 JUNE 2023

CAMILLA ROCHLENGE & RANDI JOHANNESSEN

Sharing economy and digitalization «an economic system in which assets or services are shared between private

individuals, either free or for a fee, typically by means of the internet.»

Digital platforms:

• Key enabler for the emergence of sharing economy

• Neutralises the importance of geography and time for a connection

• Safeguard the quality and the payment of the trade

Several definitions In Norway we operate with (at least) two distinct:

• The Norwegian Tax Administration:

◦ Sale or rental is made by private individual, either directly or through intermediary companies

• Alternative definition (by a large research institute):

◦ The intermediary must be a digital platform which aim to connect providers and customers

with the intent to exchange benefits from one to the other

P2P - a founding pillar Peer-to-peer transactions are defining

characteristics, but definitions vary:

• Sharing element

• Contact element

Repackaging consumerist impulses in a more

appealing message:

• P2P  B2P

Sharing assets

Consumer - cheaper deal

than traditional provider

Provider – seeking surplus of underused

assets

«Sharing is caring» Key examples

◦ Networks providing transportation services (Uber)

◦ Short-term rentals (Airbnb)

Norway today:

◦ Deregulation of taxi market (Nov 2020) - true P2P not fully possible

◦ A new short-term rental law (Apr 2019) welcome more P2P accommodation

- Balance interest of those who wish to offer lodging in their home and their neighbouring residents

- Regulate taxation of “the emerging” activities

Taxable income data - a possible date source? No data from Airbnb (not covered by the Norwegian Statistics act)

Data customized tax purpose a potential data source?

◦ 400 000 unique rental in 2020

◦ 1 week or less = 85 per cent of total

◦ Airbnb, max 2 rental object per host & 1 week or less:

- 1/3 all guest nights

- 30 per cent of total revenue

Experimental index Tax report data

◦ Variability of number of prices

◦ No imputation of missing prices

◦ Homogeneity of rental object

◦ Covid-19 influence

◦ Timeliness not ideal

Published index

◦ Imputation during pandemic with

overall index of CPI excl. zero

consumption

What now? • Tax data year 2 not yet fully

processed

◦ Improved level of detail, however no

change to timeliness

◦ In the future though….

• Web scrape Airbnb listings

◦ Not actual transactions

◦ Maintenance

• Data from Airbnb preferable

• National accounts

◦ Monitor development in expenditure

shares

◦ Conceptual clarifications needed

households = consumers

businesses = producers

◦ Ongoing revision of SNA

Thank you

Sharing economy or just utilization of new business

models? MEETING OF THE GROUP OF EXPERTS ON CONSUMER PRICE INDICES

07- 09 JUNE 2023

CAMILLA ROCHLENGE & RANDI JOHANNESSEN

Sharing economy and digitalization «an economic system in which assets or services are shared between private

individuals, either free or for a fee, typically by means of the internet.»

Digital platforms:

• Key enabler for the emergence of sharing economy

• Neutralises the importance of geography and time for a connection

• Safeguard the quality and the payment of the trade

Several definitions

In Norway we operate with (at least) two distinct:

• The Norwegian Tax Administration:

◦ Sale or rental is made by private individual, either directly or through intermediary companies

• Alternative definition (by a large research institute):

◦ The intermediary must be a digital platform which aim to connect providers and customers

with the intent to exchange benefits from one to the other

P2P - a founding pillar

Peer-to-peer transactions are defining

characteristics, but definitions vary:

• Sharing element

• Contact element

Repackaging consumerist impulses in a more

appealing message:

• P2P → B2P

Sharing assets

Consumer - cheaper deal

than traditional provider

Provider – seeking surplus of underused

assets

«Sharing is caring»

Key examples

◦ Networks providing transportation services (Uber)

◦ Short-term rentals (Airbnb)

Norway today:

◦ Deregulation of taxi market (Nov 2022) - true P2P not yet legal

◦ A new short-term rental law (Apr 2019) welcome more P2P accommodation

- Balance interest of those who wish to offer lodging in their home and their neighbouring residents

- Regulate taxation of “the emerging” activities

Taxable income data - a possible date source?

No data from Airbnb (not covered by the Norwegian Statistics act)

Data customized tax purpose a potential data source?

◦ 400 000 unique rental in 2020

◦ 1 week or less = 85 per cent of total

◦ Airbnb, max 2 rental object per host & 1 week or less:

- 1/3 all guest nights

- 30 per cent of total revenue

Experimental index

Tax report data

◦ Variability of number of prices

◦ No imputation of missing prices

◦ Homogeneity of rental object

◦ Covid-19 influence

◦ Timeliness not ideal

Published index

◦ Imputation during pandemic with

overall index of CPI excl. zero

consumption

What now?

• Tax data year 2 not yet fully

processed

◦ Improved level of detail, however no

change to timeliness

◦ In the future though….

• Web scrape Airbnb listings

◦ Not actual transactions

◦ Maintenance

• Data from Airbnb preferable

• National accounts

◦ Monitor development in expenditure

shares

◦ Conceptual clarifications needed

households = consumers

businesses = producers

◦ Ongoing revision of SNA

Thank you

Sharing economy or just utilization of new business

models? MEETING OF THE GROUP OF EXPERTS ON CONSUMER PRICE INDICES

07- 09 JUNE 2023

Sharing economy and digitalization «an economic system in which assets or services are shared between private

individuals, either free or for a fee, typically by means of the internet.»

Digital platforms:

• Key enabler for the emergence of sharing economy

• Neutralises the importance of geography and time for a connection

• Safeguard the quality and the payment of the trade

Several definitions In Norway we operate with (at least) two distinct:

• The Norwegian Tax Administration:

◦ Sale or rental is made by private individual, either directly or through intermediary companies

• Alternative definition (by a large research institute):

◦ The intermediary must be a digital platform which aim to connect providers and customers

with the intent to exchange benefits from one to the other

P2P - a founding pillar Peer-to-peer transactions are defining

characteristics, but definitions vary:

• Sharing element

• Contact element

Repackaging consumerist impulses in a more

appealing message:

• P2P  B2P

Sharing assets

Consumer - cheaper deal

than traditional provider

Provider – seeking surplus of underused

assets

«Sharing is caring» Key examples

◦ Networks providing transportation services (Uber)

◦ Short-term rentals (Airbnb)

Norway today:

◦ Deregulation of taxi market (Nov 2022) - true P2P not yet legal

◦ A new short-term rental law (Apr 2019) welcome more P2P accommodation

- Balance interest of those who wish to offer lodging in their home and their neighbouring residents

- Regulate taxation of “the emerging” activities

Taxable income data, - a possible date source?

No data from Airbnb (does not fall within the Norwegian Statistics act)

Tax report data utilized

as a data source?

◦ Data customized tax

purposes

◦ Timeliness of data

◦ Quality change

◦ Influenced by Covid-19

What now? • Tax data year 2 not yet fully

processed

◦ Improved level of detail, however no

change to timeliness

◦ In the future though….

• Web scrape Airbnb listings

◦ Not actual transactions

◦ Maintenance

• Data from Airbnb preferable

• National accounts

◦ Monitor development in expenditure

shares

◦ Conceptual clarifications needed

households = consumers

businesses = producers

Thank you

CPI weights in light of the COVID-19 pandemic

Author: Kjersti Nyborg Hov1, Statistics Norway

Meeting of the Group of Experts on Consumer Price Indices UNECE 7 – 9 June 2023, Geneva, Switzerland

1. Introduction

As in many other countries, the COVID-19 pandemic hit Norway full force mid-March

2020. The Norwegian government put in place comprehensive national restrictions in an

effort to prevent the coronavirus to spread: social distancing, working from home,

shutting down non-essential in-person services and more. This resulted in non-

availability of several services and an abrupt shift in consumer spending. This sudden

shift in consumption starting March 2020 had implications for the compilation and

calculation of the CPI in 2020, but also in the years to come2.

The main objectives in this study has been analysis of the sudden change in consumer

spending during 2020 and how it affected the Norwegian CPI. This paper documents the

actions taken during 2020 and an analysis of the aftermath. The paper is structured as

follows: Chapter 2 describes the challenges that occurred when the pandemic hit, with

emphasis on the treatment of missing price observations for wider product3 groups

related to the COVID-19 pandemic. In chapter 3 the shift in consumer spending will be

1 Statistics Norway, email: [email protected] The views expressed in this paper are those of the author and do not necessarily reflect the views of Statistics Norway. The author would like to thank Ragnhild Nygaard and Espen Kristiansen, Statistics Norway, for valuable input during the writing process. 2 The Norwegian CPI is closely linked to the European Harmonized Index of Consumer Prices (HICP). In the following, the challenges related to the CPI also applies to the HICP. 3 In the following the term product will be used for goods and services

analysed, and a recalculation of weights for 2020 will be presented using final annual

National Accounts (NA) data for private consumption in 2020. In chapter 4 an

experimental recalculated CPI will be presented using the recalculated weights, and

results will be analysed. In the end some concluding remarks.

2. CPI compilation during troubled times

2.1 The Norwegian CPI

The Norwegian CPI is defined as a measure of the change in the cost of purchasing a given

set (a “basket”) of consumption goods and services offered to Norwegian residents. The

associated expenditure weight shares of the goods and services should reflect the relative

importance of the good or service in the CPI basket. Both the composition of goods and

services and their associated weight shares are updated annually in order to stay relevant

to private consumption.

The Norwegian CPI is a so-called Lapseyres type index where weights are based on

household final consumption expenditures from National Accounts (NA)4. More precisely,

the Norwegian CPI is calculated by a Young formula where expenditure shares of period

y-1 is held constant, and the price reference month is December of the previous year. As

of January 2011, National Accounts (NA) replaced the Household Budget Survey (HBS) as

the primary data source for weight information at sub-class and higher levels. At lower

levels weight shares are derived from HBS, scanner data, other statistics, industry reports

and other.

As the Lapseyres type/Young formula used in the Norwegian CPI indicates, CPI weights

should reflect the household consumption expenditure pattern of the previous year5. The

most recent NA data at the level of detail necessary for the compilation of CPI weights

refer to the year two years prior to the year for which the weights will be used in the CPI.

In normal pre-pandemic years, changes in expenditure have proved relatively small from

4 A true Laspeyres price index implies using quantity data which relate to exactly the same period as the price reference period. The NO CPI uses December of previous year as the price reference month while the weights are based on a 12-month period, prior to December y-1. 5 Annually chained Laspeyres type/Young formula

one year to another. In ordinary years it is therefore reasonable to assume that the

consumption pattern in year y-2 is a good approximation of the consumption in year y-1.

Thus, it has been regarded as unproblematic to use lagged NA data to update the weights.

From 2011 to 2020, annual NA from year y-3 in combination with the growth rate in

quarterly NA y-3 to NA y-2 was used to derive CPI weights. During 2020 however, Norway

as many other countries, experienced an abrupt shift in the consumption pattern due to

national and regional lockdowns and other restrictions related to the outbreak of the

COVID-19 pandemic. The assumption of relatively small changes in consumption pattern

between years no longer held. This had implications for the compilation of the CPI in year

2020, and also the years to come.

2.2 Year 2020 – the pandemic hit

The COVID-19 pandemic and its implications on consumption hit Norway full force mid-

March 2020. The Norwegian government held a press conference 12th March 2020

concerning comprehensive national restrictions, affected immediately, in an effort to

prevent the coronavirus to spread. In the initial phase of the outbreak the restrictions

were quite limiting for all residents:

• Kindergartens and schools were forced to shut down and reallocate to remote

online teaching

• Non-essential workers were forced to work from home where possible. For

workers not able to perform work from home strict limitations on the number of

persons were put in place

• Recreational, cultural and sports arrangements and services such as gyms, theatre,

cinema and more were forced to closed.

• Non-essential in-person services such as hair dresser, personal trainers, nail

saloons were forced to close

• The national border was closed for private international travel, and domestic

travel was strongly advised against, thus air traffic, accommodation and restaurant

services were heavily reduced

In effect these government restrictions resulted in an abrupt non-availability of several

services and private consumption of these services fell sharply. The direct effects were a

sudden fall in spending on non-essential in-person services, recreational and cultural

services and also services related to travel. In addition, there were some indirect effects

on the consumption pattern, shifting the consumption from out-of-house to in-house:

• Increased expenditure on groceries and takeout food at the expense of canteens,

cafes, restaurants and bars

• Increased expenditure on the State Wine Monopoly for wine and liquor at the

expense of restaurants and bars. The increase was also a result of closed national

borders and the sudden halt of cross-border shopping in the neighbouring country

Sweden in particular6.

• Increased expenditure on consumer electronics and furniture, both likely a result

of remote work and school

• Increased expenditure on recreational and sports activity goods, likely a result of

closed gyms and the need to use the outdoors more

The direct and indirect effects of the government restrictions on consumption led to a

general shift from expenditure on services to increased expenditure on goods.

2.3 CPI compilation 2020

The consequences of the restrictions put in place concerning private consumption

naturally led to challenges for the CPI compilation in 2020. According to the Young

formula, expenditure weights are fixed for the time period in question, in the Norwegian

CPI this equals to one calendar year. This meant that even though consumption shifted

abruptly due to the COVID-19 restrictions, the weights underlying the CPI were kept fixed

during 2020. This was in line with international recommendations concerning the COVID-

19 pandemic7. However, as certain goods and services in the CPI basket experienced close

to zero consumption, alterations to the CPI compilation were needed. It is reasonable to

6 Norway and Sweden share an extensive country border, and cross-border shopping in Sweden of especially groceries, alcohol and tobacco is common for Norwegian residents. The different tax schemas in combination with generally lower price levels in Sweden makes it beneficial for Norwegian residents 7 See e.g. Eurostat (2020) and IWGPS (2020)

expect that goods and service experiencing close to zero consumption should not impact

the measure of the CPI, i.e. the effects should be neutralized.

Neutralization – treatment of non-available, non-seasonal products

Three alternatives were discussed as to how to best neutralize the effects of goods and

services that were no longer available for consumption, but still remained in the CPI

basket.

1. Carry forward the last observed price observations for the elementary

aggregate(s)

2. Omit the goods and services from the basket and recalculate the weights mid-year

3. Impute the missing price observations

Alternative 1, carry forward the last observed price method was dismissed as this is

generally not a desired solution for missing price observation. It could be reasonable to

believe that the prices would not change during the lockdown and therefore be justified

as a method. However, carry forward the last observed price on a good or service would

mean that we put emphasis, in this case a zero-percentage change, on a product that is not

available. Including a zero-percentage change for the not available product would entail a

bias in the month-to-month (period-to-period) index; if prices in general were increasing,

carry forward the last observed price for the non-available products would cause a

downward bias in the index. And likewise, carry forward would entail an upward bias if

the prices in general were falling. In general, carry forward should only be used in if the

prices are regulated or otherwise known not to change, see the Consumer Price Index

Manual: Concepts and Methods (IMF et al., 2020), hereafter named the CPI manual.

The second alternative, omitting the goods and services from the basket and recalculate

the weights, could be a viable option, however this is not in line with international

recommendations nor the legal and conceptual framework of the HICP which the

Norwegian CPI is closely related to8.

8 See Commission Implementing Regulation (EU) 2020/1148

In addition, changing weights in the midst of a crisis such as the COVID-19 pandemic is

not necessarily straight forward. The shifts in consumption is challenging to monitor real-

time, especially when the changes are sudden, sharp and unknow for the near future. In

the initial phase of the pandemic restrictions were severe and applied to the entire nation.

Later, the restriction could vary, both in scope, but also across regions. The capital city

Oslo and other larger cities generally experienced tighter regulations than less populated

areas in Norway, and restrictions would vary in intensity and time. This would make it

challenging and time-consuming to recalculate weights, re-introduce product groups

once restrictions lifted - possible multiple times - during the year. Also, changing weights

entails chaining, and there is a higher risk of chain drift in the index if the weight shifts

are substantial and price movements at the point of chaining fluctuates, which is plausible

to be the case during a crisis, see Reinsdorf (2020). Given the above, the quality of the CPI

could be compromised, and the option of recalculating weights mid-year was therefore

dismissed.

The third option was to impute the missing price observations, a well-known method for

treatment of temporarily missing products, see the CPI manual. A general method for

imputing missing price observations is to calculate the average price change for the prices

available in the elementary aggregate, or by calculating the average price change of

targeted comparative varieties. The implicit assumption of imputing prices by similar

products is that when a product is no longer available, consumers will substitute by

similar products. However, in this particular case we were experiencing a non-availability

of not only single price observations, but for wider product groups and also higher

aggregate product groups. The non-availability of several products could not (easily) be

substituted by similar products as they were also not available, therefore it seemed

inadequate to impute by nearest higher aggregates. The decision was to impute by the

highest aggregate, i.e. the all-items CPI, containing all reliable price indices.

The decision of using the all-items CPI as the imputation factor for the missing price

observations was based on the assumption that the substitution for the non-available

products were evenly distributed, in relative terms, on all the other consumer products

available. In effect, this would give the same output as omitting missing products and

recalculating the weights, however without causing bias to the indices. The solution was

relatively easy to implement and also easy to monitor, and it made it possible to do

changes month-to-month according to the shift in restrictions and availability of products.

However, one challenge remained: the treatment of missing price observations for

seasonal products.

Neutralization – treatment on non-available, seasonal products

The treatment of missing price observations for products that did not show a pronounced

seasonal pattern over time, were to impute by the all-items CPI. For products showing

pronounced seasonality however other considerations were needed. The price

movements for seasonal products are volatile, and by definition the variation repeats

itself and occurs during the same time period every year. Hence, the absence of a seasonal

price variation will affect the annual rate of change.

The challenge can be described by an example: Airfares for international travel inherit

pronounced seasonality related to the summer vacation, winter - and fall break and the

holiday seasons, and the price movements are generally substantial. The prices naturally

increase with the increase in demand, and the increase normally occurs during the same

time periods every year. If the price increase in percentage terms, in e.g. July year y was

the same as July year y-1, the effects on the all-items CPI annual rate of change for July

would be neutral. However, if the expected price increase were not to occur in July year y

then the contribution to the annual rate of change for the all-items CPI would be negative.

I.e. imputing the airfares for international travel by the all-items CPI during the summer

months would contribute to pulling the inflation rate down, measured by the annual rate

of change. This is in line with the findings in IWGPS (2020) and Lamboray et al. (2020).

Considering the annual rate of change, the imputation of seasonal products by the all-

items CPI could severely affect the annual rate of change, which would make it difficult to

interpret the results. The possible decline in inflation would not be a reflection of less

price pressure in the economy, or that residents experience less inflation, but would

merely be a result of technicalities. According to international recommendations the

imputation of missing products should therefore not break the seasonal pattern of the

product9. Eurostat (2020) recommended the following two options for treatment of

missing price observations for seasonal products:

1. Impute with the annual rate of change of all reliable price observations

2. Carry forward with a seasonal correction factor

In order to properly capture the seasonality of the indices in question, seasonal correction

factors were obtained by estimating econometric models using X-13ARIMA-SEATS (U.S.

Census Bureau), based on a minimum of 5-10 years of time series data. The computed

seasonal component was used to estimate the monthly seasonal correction factor.

Seasonal correction factors were calculated for the following product groups:

• Passenger transport by air, international flights

• Package holidays

• Accommodation services, hotels

It could be argued that imputing by a seasonal factor for products (here services) not

available incorrectly affects the monthly rate of change. Imputing by a seasonal factor

favours the annual rate of change at the expense of the monthly rate of change, therefore

it was important to keep our user well informed of the chosen method and the

implications it had on the indices for the affected months. The affected time periods and

the chosen imputation methods for the affected indices were marked in the Statistics

Norway Statbank data tables, and also noted in the monthly dissemination reports for the

affected months.

3. Recalculating 2020 – Expenditure weight shares

3.1 Expenditure weight shares

It has been around three years since the first wave of the COVID-19 hit, and restrictions

have been lifted in most countries. Comprehensive data on actual consumption during

9 See for example Eurostat (2020), IWGPS (2020), UNECE (2021)

2020 by final NA is now available, thus making it possible to analyse the abrupt shift in

consumption pattern during this period in relation to the CPI.

To analyse the shift in consumer expenditure in Norway in 2020 an alternative set of

weight shares for the CPI 2020 were calculated using final NA 2020 figures for household

consumption expenditure. The experimental recalculated weight shares will in the

following be named recalculated CPI weights, while the actual weight shares used for

computing the CPI in 2020 will be named published CPI weights.

Final NA 2020 figures were used to redistribute the CPI weights at COICOP level 1-4,

creating the recalculated 2020 weights. The relative weight distribution at lower levels

were kept fixed using the fixed relative distribution according to the published CPI

weights in 2020. Thus, the recalculated weights on COICOP level 1-4 are based on actual

2020 consumption while the weight distribution on lower level aggregates are based on

pre-pandemic information up until 2019. The latter part is a weakness in the analysis and

could be explored further in subsequent analyses.

Changes in weight shares - what we didn’t do

Comparing the published and recalculated weights for 2020 we clearly see the effects of

the sudden shift in consumption that took place during 2020. As expected, we found large

declines in expenditure for travel and leisure related activities in the recalculated weights.

At 3-digit COICOP level we found the largest decreases in both percentage and absolute

terms for group 09.6 Package holidays and 07.3 Transport services for the recalculated

weights compared to the published weights. Both groups were severely affected by the

pandemic and the government restrictions on travel both domestically and

internationally. For transport services we found the largest drop in consumption for air

fares, but also passenger transport by railway, road and sea experienced large decreases.

In addition, the government restrictions included keeping a one-meter distance between

people and closing down in-person service of alcoholic beverages, in effect shutting down

many restaurants and bars. Non-essential medical help and other in-person services were

also forced to close. Also, recreational, sport events and cultural services with audience

were banned. As expected, comparing recalculated and published weights we found the

larger deviations between published and recalculated weights for group 11.1 Restaurant

services and 09.4 Recreational and cultural services.

Figure 1: Recalculated and published weights 2020, in percentage points. Selected series based on largest

differences.

Source: Statistics Norway

The electricity market in 2020

One of the largest deviations between recalculated and published weights are found for

group 04.5 Electricity, gas and other fuels. Electricity is the main energy source for

Norwegian households, thus carry a large weight share in the CPI, and is also the main

driver behind the difference in recalculated and published weights for 2020. The

deviation is however not related to the pandemic. The deviation is rather a result of using

lagged information about a current weather situation: Electricity is a volatile component

in the CPI, highly affected by weather conditions. From the beginning of 2020 Norway

experienced a record amount of precipitation, resulting in full water reservoir coverage

and abundant amount of snow in the mountains that later would melt and re-fill the water

reservoirs (NVE 2020). The weather condition lasted well into the fall keeping the prices

0 1 2 3 4 5 6

11.1 Restaurant services

09.6 Package hollidays

09.4 Recreational and cultural services

07.3 Transport services

04.5 Electricity, gas and other fuels

Published weights 2020 Recalculated weights 2020

on electricity low during 2020. In addition, warmer climate than most years resulted in

less use of electricity for heating during the fall and winter months.

Full water reservoir coverage and less need for electricity resulted in less consumer

expenditure on electricity than what was estimated at the time of compiling the weights

for CPI 2020. It should be noted that it is not uncommon to deviate from expected to actual

consumer expenditure on electricity, however for 2020 the deviation was more severe

than most years.

Changes in weight shares – what we did do

The deviation between recalculated and published weights clearly show the effects of the

restrictions and what we were no longer allowed to do, but they also show clearly what

we actually did do during these pandemic months. In short, we stayed at home and did

activities related to the residence. Most meals were made and consumed at home,

including beverages, at the expense of restaurants, cafes and bars both domestically and

international. This included also the cross-border grocery shopping in Sweden in

particular, that experienced an abrupt halt. Another effect related to more cooking at

home was a considerable increase in expenditure on kitchen appliances.

Apart from cooking, many homes also became offices, kindergartens, schools and the like,

thus increased expenditure on IT equipment, but also on furniture such as chairs and

desks. In addition, staying more at home also seemed to have increased the expenditure

on home décor, household textiles and other refurbishing related activities.

Figure 2: COICOP divisions, recalculated and published weights 2020, in per cent.

Source: Statistics Norway

Food and non-alcoholic beverages

Comparing recalculated and published weights for 2020 we found one of the largest

increases in weights for COICOP division 01 Food and non-alcoholic beverages, for

reasons explained above. The index compilation of food and non-alcoholic beverages is

entirely based on scanner data. According to regular weight update procedure weights at

4-digit COICOP are fixed according to the NA figures. Weights on lower level aggregates

are distributed according to scanner data turnover information (prices and quantities) for

a whole year10. For 2020 that meant using scanner data turnover information from 2019

to distribute weight shares at 5-digit COICOP and the lower level aggregates. According to

the published weights, food and non-alcoholic beverages received a weight share of 11.9

per cent in 2020, while for the recalculated weights the weight share accounted to 13.8

per cent, a deviation of 1.9 percentage points.

Comparing the published and recalculated weights at 4-digit COICOP level for food and

non-alcoholic beverages we found that the deviation for all sub-groups vary between 10-

10 Scanner data is considered a comprehensive data source of information, but it should be noted that scanner data doesn’t differentiate between private and public consumption, which could alter the results somewhat.

0 5 10 15 20 25

12 Miscellaneous goods and services

11 Restaurants and hotels

10 Education

09 Recreation and culture

08 Communications

07 Transport

06 Health

05 Furnishings, household equipment and routine…

04 Housing, water, electricity, gas and other fuels

03 Clothing and footwear

02 Alcoholic beverages and tobacco

01 Food and non-alcoholic beverages

Published weights 2020 Recalculated weights 2020

30 per cent, se figure 3. The largest deviation, in percent, is found for sub-group 01.2.2

Mineral waters, soft drinks, fruit and vegetable juices. The largest deviation in percentage

points are found in 01.1.2 Meat and 01.2.2 Mineral waters, soft drinks, fruit and vegetable

juices. For 01.2.2 we also find the largest deviation between published and recalculated

weights in relative terms.

Figure 3: Sub-groups within COICOP 01 Food and non-alcoholic beverages, recalculated and published

weights 2020, in per cent.

Source: Statistics Norway

As explained above, all sub-groups within COICOP division 01 showed increase when

using recalculated weights, maybe not so surprisingly given the large increase in weight

in general for COICOP division 01. Another interesting point is to look at the relative

difference in the increases, i.e. to analyze what products showed the highest (lowest)

increase in relative importance. To do so the weight shares at 4-digit COICOP level were

kept fixed according to published weights 2020, and then the weight shares at lower level

aggregates were analyzed with respect to the weight information present when compiling

the published weights (scanner data turnover information for 2019), i.e. the published

weights for 2020, and compare it to a new set of weights containing weight information

0 5 10 15 20 25

Mineral waters, soft drinks, fruit and vegetable juices

Coffee, tea and cocoa

Food products n.e.c.

Sugar, jam, honey, chocolate and confectionery

Vegetables

Fruit

Oils and fats

Milk, cheese and eggs

Fish and seafood

Meat

Bread and cereals

Recalculated weight 2020 Published weight 2020

now available for 2020 (scanner data turnover information for 2020). The latter weight

series will in the following be named redistributed weights 2020. Comparing redistributed

weights 2020 and published weights 2020 show the difference between using 2020 vs.

2019 scanner data turnover information (respectively) for the distribution of weights at

lower level aggregates. This gives insight into the relative change in consumption.

One of the sub-groups showing the largest increase in weights was COICOP 01.1.2 Meat.

For meat we found that especially beef and veal, pork and poultry received the largest

increase in weights when using recalculated weights. Looking at the redistributed weights

for 2020 we find that, even though the weights increased, the relative importance of lamb

and goat, and dried, salted and smoked meat fell.

Figure 4: 5-digit COICOP in sub-group 01.1.2 Meat, weight shares, in per mille.

Source: Statistics Norway

Another group showing large increase in weights was COICOP 01.2 Non-alcoholic

beverages. For non-alcoholic beverages we found that COICOP 01.2.2.2 Soft drinks

received the greatest increase in weights in 2020 compared to the published weights.

0 1 2 3 4 5 6 7 8 9

Other meat preparations

Dried, salted or smoked meat

Edible offal

Other meats

Poultry

Lamb and goat

Pork

Beef and veal

Redistributed weight 2020 Recalculated weight 2020 Published weight 2020

Looking at the redistributed weights compared to the published weights 2020 however

we find that the weight increase in COICOP 01.2 Non-alcoholic beverages is fairly evenly

distributed across all 5-digit COICOP groups. Thus, the large weight increase for soft

drinks is more related to its initial size of weight share than an exceptionally increase of

consumption of soft drinks compared to the other non-alcoholic beverages.

Figure 5: 5-digit COICOP within group 01.2 Non-alcoholic beverages, weight shares, in per mille.

Source: Statistics Norway

4. Recalculating 2020 - Index calculation

4.1 Experimental CPI 2020

As shown above, and as was expected, we found large differences between published CPI

weights in 2020 compared to actual consumption in 2020, measured by final NA 2020. In

the following a recalculated experimental CPI index series for 2020 using the recalculated

weight shares has been computed. It should be noted that the recalculated index series is

an experimental index series and not an official index, thus should therefore not be viewed

0 1 2 3 4 5 6 7 8 9 10

Fruit and vegetable juices

Soft drinks

Mineral or spring waters

Cocoa and powdered chocolate

Tea

Coffee

Redistributed weight 2020 Recalculated weight 2020 Published weight 2020

as a “true” CPI index series for 2020. The experimental index series also inherit

shortcomings on the compilation of weights, for example no adjustments to the relative

distribution of weights at lower level aggregates (below the 4-digit COICOP level). In

addition, the abrupt changes in consumption patterns varied largely during 2020, the

weights reflecting consumption for a whole year will therefore not adequately reflect

actual consumption for the individual months. This is also true for CPI compilation and

weight calculations in general, but for 2020 the discrepancy and variance are larger than

usual.

Comparing the published CPI index series and the recalculated experimental CPI index

series we found that the published CPI lies below the recalculated experimental index

series throughout the year. This indicates that the published CPI in 2020, using weight

shares based on lagged consumption data, somewhat underestimated inflation in 2020.

This is in line with other studies such as Reinsdorf (2020) and Lamboray et al. (2020).

We see however an upward level shift for the recalculated experimental series in the

beginning of 2020, a period not related to the COVID-19 pandemic. It is reasonable to

believe that the recalculated weights for the period January to mid-March 2020 might be

less representative than the published weights in the same period as consumer

expenditure was not yet affected by the restrictions during the pandemic, thus these

results must be handled with care.

Figure 6: Published and recalculated experimental CPI index (DEC2019=100). January - December 2020.

Source: Statistics Norway

The first wave of the COVID-19 pandemic hit Norway mid-March 2020, thus both

consumption and expenditure were less affected by the pandemic in the first quarter of

2020. Therefore, the figure above might be overestimating the effects of using the

published weights compared to the recalculated weights. March 2020 was the last semi-

normal pre-pandemic month, thus starting the recalculated index series in March 2020

will give a better assessment of the difference between published and recalculated

experimental CPI index series.

Comparing the published and recalculated experimental CPI index series starting March

2020 we found that the deviation between the two indices are less than when comparing

the year as a whole. This indicates that the underestimation of the CPI we found using the

whole year might be misleading to a certain degree; the size of the divergences was

reduced when isolating the months mostly affected by the pandemic. It should be noted

that a weakness in the analysis is that the recalculated weights are based on the whole

year of 2020, including both pre-pandemic months and months affected by the pandemic.

It is reasonable to believe that excluding the pre-pandemic months from the data could

alter the results further, leading to larger deviations, however this has not been tested in

the analysis.

98.5

99.0

99.5

100.0

100.5

101.0

101.5

102.0

102.5

202001 202002 202003 202004 202005 202006 202007 202008 202009 202010 202011 202012

Published CPI Experimental CPI

Figure 7: Published and recalculated experimental CPI index (MAR2020=100). March – December 2020.

Source: Statistics Norway

As shown above, the published and recalculated experimental CPI index series show close

to identical development from March to July 2020. From August we found a level-shift

were the recalculated experimental index series show a smaller decrease than the

published index. A similar larger drop in the published CPI is found in November, before

the two indices diverge back to each other in December 2020.

4.2 Contributing factors

To examine the driving factors behind the differences in the two indices we calculated and

compared the contribution factors for each month in 2020. The contribution factor is

defined as the products contribution to the rate of change in the all-items CPI, either the

monthly or the annual rate of change, and is related to both the price development of the

product and the products weight share. In order to examine which COICOP divisions that

contributed to pulling the recalculated experimental CPI index up (down) compared to

the published CPI, we compared the contribution factor for each of the twelve COICOP

99

99.5

100

100.5

101

101.5

102

102.5

202004 202005 202006 202007 202008 202009 202010 202011 202012

Published CPI Experimental CPI

divisions for both index series. The contribution factors were calculated for the all-items

CPI annual rate of change for each month in 202011.

Figure 8 below present the difference between the contributing factor for the recalculated

experimental index series compared to the published index series, month-by-month, for

the annual rate of change in the all-items CPI. Looking at figure 8 we found some larger

deviations, in particular for COICOP division 01 Food and non-alcoholic beverages, and 07

Transport, with opposite signs. Both, as well known, severely affected by the pandemic.

The price development of COICOP division 01 Food and non-alcoholic beverages together

with 05 Furnishings, household equipment and routine maintenance showed the largest

deviation pulling the recalculated experimental index up compared to the published CPI.

We see that the recalculated experimental index for food and non-alcoholic beverages lies

above the published series throughout the period, not so surprisingly as the annual rate

of change for food and non-alcoholic beverages remained positive throughout 2020. We

find a larger impact in especially February and July, two months prone to price increases:

Increased prices in combination with larger weight shares in the recalculated

experimental series, the results are as expected. It should be noted that February was a

month not affected by the pandemic, the increased weight share in the experimental

series might therefore somewhat overstate actual consumption for food and non-

alcoholic beverages in February.

In addition to food and non-alcoholic beverages we also found a larger positive

contribution to the recalculated index series for COICOP division 05 Furnishings,

household equipment and routine maintenance, 04 Housing, water, electricity, gas and

other fuels, and 02 Alcoholic beverages and tobacco. COICOP division 05 received a larger

weight share in the experimental CPI, and in combination with an positive annual rate of

change we found that the contribution from division 05 on the experimental CPI were

increasing during the most part of 2020.

For COICOP division 04, the price decrease during 2020 for reasons explained above, in

combination with considerably less weight share on electricity for the recalculated 2020

weights also contributed to pulling the experimental CPI up. For division 02 Alcoholic

11 For elaboration of the calculation of contribution factor, see Nygaard (2017) and OECD (2022)

beverages and tobacco an increased weight share in combination with price increases in

particularly the beginning of the year, i.e. pre-pandemic months, was the main factor for

contribution to pulling the experimental CPI up compared to the published CPI.

In the opposite direction we found the largest contribution to pulling the experimental

CPI down compared to the published CPI in division 07 Transport. Expenditure on

transport showed a sharp decline during 2020 due to the government restrictions, pulling

the weight share in the experimental series down. For the price movements during 2020

it should be noted that division 07 was largely influence by having imputed prices by a

seasonal factor, the results must therefore be handled with some care.

Figure 8: Contributing factor, difference between recalculated experimental and published CPI

Source: Statistics Norway

5. Concluding remarks

The COVID-19 pandemic had large implications for the CPI compilation in 2020. A sudden

shift in consumption when the pandemic hit made it challenging to compile the CPI in a

-0.15

-0.1

-0.05

0

0.05

0.1

0.15

0.2 01 Food and non-alcoholic beverages

02 Alcoholic beverages and tobacco

03 Clothing and footwear

04 Housing, water, electricity, gas and other fuels

05 Furnishings, household equipment and routine maintenance

06 Health

07 Transport

08 Communications

09 Recreation and culture

10 Education

11 Restaurants and hotels

12 Miscellaneous goods and services

regular manner. The main challenges were large scale missing price observations, in

addition to having obsolete weights compared to the COVID-19 consumption pattern.

According to international recommendations, and the formula in general, the weight

shares were kept fixed during 2020. The challenges related to the CPI compilation were

resolved by imputing missing price observations, either by the all-items CPI of reliable

indices, or by imputing by a seasonal factor, depending on the presence of seasonality for

the product in question.

Having NA 2020 data on household final consumption expenditure enabled a study on the

differences between the weight shares used for CPI calculation in 2020 and the

recalculated weight shares based on actual consumption, according to NA 2020. As

expected, the recalculated weights showed an increase in consumption for activities

related to staying at the residence, and likewise a decrease in travel and leisure related

activities.

An experimental analysis on how the weight differences affected the CPI compilation

showed that the published CPI somewhat underestimated the inflation during 2020, when

compared to using weight shares according to NA 2020. These results should however be

treated with some care. 2020 was a year that contained months both heavily affected by

the pandemic and subsequent government restrictions, but also months not affected by

the pandemic. Neither a basket containing pre-pandemic weights nor a basket containing

the effects of the pandemic will fully be adequate for all months of the year 2020.

Nevertheless, the experimental study could give some insight on the size and sign of the

impact of the sudden shift in consumption that were experienced during the pandemic.

References

Eurostat (2020), «Guidance on the compilation of the HICP in the context of the COVID-

19 crisis”, Methodological note.

https://ec.europa.eu/eurostat/documents/10186/10693286/HICP_guidance.pdf

IMF et al. (2020), “Consumer Price Index Manual: Concepts and Methods”. https://www.imf.org/en/Data/Statistics/cpi- manual?msclkid=c5c3b0aca82211ec8ac1fb0e4654bab6

Inter-secretariat Working Group on Price Statistics, IWGPS, (2020). “Consumer Price

Index: Business Continuity Guidance”.

https://statswiki.unece.org/display/CCD2/Compilation+of+CPI+in+times+of+COVID-

19?preview=/278037166/279776928/IWGPS%20CPI%20Continuity%20Note.pdf

Lamboray, C., Evangelista, R., Konijn, P. (2020). Measuring inflation in the EU in times of

COVID-19. EURONA Issue 2020, Eurostat. https://cros-

legacy.ec.europa.eu/content/measuring-inflation-eu-times-covid-19-claude-lamboray-

rui-evangelista-and-paul-konijn_en

NVE (2020), «Kraftsituasjonen i Norge – 4 kvartal og året 2020» (Norwegian only).

https://www.nve.no/media/11490/kraftsituasjonenq4.pdf

Nygaard, R. (2017): “Hvor mye påvirker enkeltgrupper den totale prisendringen i KPI?»

(Norwegian only). https://www.ssb.no/priser-og-prisindekser/artikler-og-

publikasjoner/hvor-mye-pavirker-enkeltgrupper-den-totale-prisendringen-i-kpi

OECD (2022), «OECD calculation of contributions to overall annual inflation”.

https://www.oecd.org/sdd/prices-ppp/OECD-calculation-contributions-annual-

inflation.pdf

Reinsdorf, M. B. (2020): “COVID-19 and the CPI: Is Inflation Underestimated?”, IMF

Working Paper (WP/20/224), IMF.

https://www.imf.org/en/Publications/WP/Issues/2020/11/05/COVID-19-and-the-

CPI-Is-Inflation-Underestimated-49856

Reinsdorf, M. B., Tebrake, J., O’Hanlon, N. and Graf, B. (2020), “CPI Weights and COVID-

19 Household Expenditure Patterns”, Special Series on Statistical Issues in Response to

COVID-19, IMF. https://www.imf.org/en/Publications/SPROLLs/covid19-special-

notes#stats

UNECE (2021): “Guide on producing CPI under lockdown”.

https://unece.org/info/Statistics/pub/359134

United States Census Bureau, X-13ARIMA-SEATS Seasonal Adjustment Program. Last

updated: 11 July 2022. https://www.census.gov/data/software/x13as.html

2020

0 5 10 15 20 25

12 Miscellaneous goods and services

11 Restaurants and hotels

10 Education

09 Recreation and culture

08 Communications

07 Transport

06 Health

05 Furnishings, household equipment and routine maintenance

04 Housing, water, electricity, gas and other fuels

03 Clothing and footwear

02 Alcoholic beverages and tobacco

01 Food and non-alcoholic beverages

Published weights 2020 Recalculated weights 2020

2020

0 5 10 15 20 25

Mineral waters, soft drinks, fruit and vegetable juices

Coffee, tea and cocoa

Food products n.e.c.

Sugar, jam, honey, chocolate and confectionery

Vegetables

Fruit

Oils and fats

Milk, cheese and eggs

Fish and seafood

Meat

Bread and cereals

Recalculated weight 2020 Published weight 2020

2020

99.0

99.5

100.0

100.5

101.0

101.5

102.0

102.5

Published all-items CPI Recalculated all-items CPI

99

99.5

100

100.5

101

101.5

102

102.5

Published all-items CPI Recalculated all-items CPI

2020

-0.15

-0.10

-0.05

0.00

0.05

0.10

0.15

0.20

jan.20 feb.20 mar.20 apr.20 MAY20 jun.20 jul.20 aug.20 sep.20 OCT20 nov.20 DEC20

07 Transport

05 Furnishings, household equipment and routine maintenance

11 Restaurants and hotels

01 Food and non-alcoholic beverages

04 Housing, water, electricity, gas and other fuels

02 Alcoholic beverages and tobacco

09 Recreation and culture

* Difference in contribution

2020

Abstract Sharing economy or just utilization of new business models? Authors: Camilla Rochlenge, Randi Johannessen

The year 2019 was when the sharing economy and its collaborative consumption was starting to make a bigger impact on Norwegian society and way of life. With international hospitality and mobility services leading the way, also several digital platforms developed domestically saw noticeable growth in its users and revenue. New legislation was put in place to support an orderly transition to an economy that makes better use of idle resources. However, the COVID-19 pandemic caused a major temporary setback to this development.

The sharing economy offers a quick and cheap way of matching supply with demand for goods and services. The main innovation in the business model of the sharing economy lies in the technological platforms such as smartphone apps which

h bring demand and supply together. There are two main types of sharing platforms: peer-to-peer (P2P) and business-to-consumer (B2C). In P2P demand and supply are matched via a digital platform developed and operated by a third entity who usually charges a fee of a fixed percentage of each transactions’ payment. Typical examples are platforms such as Airbnb and Uber, two major players in the sharing economy. Due to the growing popularity of the P2P business models, more traditional commercial firms are also adapting their economic model to incorporate this concept of “sharing” into their companys portyfolio. This type of business (B2C) implies direct contact between the commercial provider and their customers via sharing platform apps or by adapiting the providers own app or platform.

The aim of the paper is to define and delineate sharing economy within the P2P and B2C plattforms. We find that although the underlying business model of the sharing economy keeps growing, the consumption within the P2P segment in Norway is still limited, while there is an increase in the B2C segment. Further, based on data from the Norwegian Tax Authority, the paper will demonstrate the limitations and the challenges of estimating a proper price index for accommodation within the sharing economy.

1 Introduction The sharing economy as a sizable phenomenon is relatively new and due to Norway being a small country with a small market, it may be subject to international companies operating with platforms based on mature technology after testing their set up in other countries first. New legislation was put in place in 2018 to support an orderly transition to an economy that makes better use of idle resources. And 2019 was the year when the sharing economy and collaborative consumption was starting to make a bigger impact on Norwegian society and way of life. With international hospitality and mobility services leading the way, also several digital platforms developed domestically saw noticeable growth in the numbers of users and income. However, the occurance of the COVID-19 pandemic in 2020 dealt a major temporary setback to the development.

The sharing economys business models utilization of technological platforms such as smartphone apps provides an enviroment where demand and supply can meet at “an instance” independent of time zones and geography. The business model is found in a wide range of sectors, although currently most noteworthy within tourist accommodation and personal transport, such as taxi services and sharing of vehicles. Since the term “sharing economy” appeared around 2008 the phenomenon has grown alongside the rise of peoples’ omnipresent connection to the web through smartphones, all while the activity within the sharing economy has evolved during the same period of time. In this paper we will describe multiple definitions existing in Norway of what is considered as sharing economy. We also aim to identify which economic activity is covered within the sharing economy platforms in Norway, and that the sharing economy business model is widespread both in the B2C and P2P segments, also showing that the P2P segment for the time being is rather small. We will furthermore demonstrate the limitations and challenges of estimating a proper price index for accommodation within the sharing economy based on data from the Norwegian Tax Authority. As of now it is still preferable to obtain data from Airbnb or other platforms directly. But in the future with sufficient adjustments the Norwegian Tax Authority data may prove useful as a source for price information.

2 Definition of sharing economy A random search online for the definition of the sharing economy results in “noun: an economic system in which assets or services are shared between private individuals, either free or for a fee, typically by means of the internet”. However in Norway other, more specific definitions exist, among them the definition by a Norwegian Official Report (Government.no, 2017:3) which states that the sharing economy is “economic activity enabled or facilitated via digital platforms that coordinate the provision of a service or the exchange of services, skills, assets, property, resources, or capital without transferring ownership and primarily between private individuals.”

Next, the sharing economy is defined by The Norwegian Tax Administration as “a business model where private individuals sell services or rent out assets directly or through intermediary companies” (The Norwegian Tax Authorities, 2022). Payment may be returned as services in kind, instead of money.

As a clear distinction between a hobby and a commercial activity is not defined, to identify what category the activity falls within the Tax Administration suggest the following assessment to be carried out in effort to identify whether the activity:

- is carried on at the business’s own expense and risk - has a certain scope - is likely to generate a surplus over time - is aimed at having a certain duration

Another definition that is based upon three key features that characterize the sharing economy is provided by Fafo (an independent social science research foundation associated with the largest Norwegian labour union) (Jesnes et al., 2016:7):

- An intermediary in the form of a digital platform. - Which helps to connect complementary players, which can be considered as providers and

customers. - Who exchange a set of benefits from the provider to the customer. There can be a wide variety

of benefits, from services and asset/property sharing to capital, expertise, and labour.

In all the above definition the peer-to-peer (P2P) transaction is a defining characteristic of the sharing economy, however in the last definition by Fafo it is the contact facilitation of the P2P transaction which defines the sharing economy, not the sharing element itself.

3 B2C and P2P There are two main types of sharing platforms: peer to peer (P2P) and business to consumer (B2C). In P2P demand and supply are matched via a digital platform developed and operated by a third entity who usually charges a fee of a fixed percentage of each transactions payment. Typical examples are platforms such as Airbnb and Uber, two of the best-known examples of sharing economy models1.

Following the strong growth of P2P business models, two trends have occured. Some suppliers expand operations to investing in more rental units, thereby transcending from a P2P supplier to becoming an owner of several units and operating as a B2C supplier within the same platform. Simultaneously, the traditional commercial firms adapt their economic model to incorporate similar concepts of “sharing”. This type of B2C business implies direct contact between the commercial provider and their customers either via the providers own platform or through an established sharing economy platform. According to the definition from FAFO, these activities are not included in the definition of the sharing economy as these business models are relatively similar to those of traditional traders. Contact through well- established web sites such as booking.com between hotels and guests are easily defined as B2C,however when booking.com also include listings of lodging by private owners the distinction between the two segments becomes less clear as the web sites trancends into also providing stays P2P.

Given the connection between the National Accounts (NA) the Consumer Price Index (CPI)2 a collaboration between the price and the NA communities is preferable to ensure progress and consistency in both statistics. Digitalisation leads to a shift in the production boundary with more activities taking place within the household. The traditional assumption in NA is that firms create value added as producers, while households/individuals are consumers only. Due to the limited role of households as producers, their value added is recorded in the informal economy (IMF Committee on Balance of Payment, 2020). We now face an increasing number of individuals who participate directly as “producers” in activities related to the sharing economy. For instance, we see a growing trend of trading second-hand goods like clothing, furniture, electronics, books, etc. This trend is facilitated by the simplicity brought to the second-hand market by P2P sharing platforms. The practice of P2P in general not being measured in NA applies for every area of the economy with one exception;

1 Consumers are also using digital networks to lend office space, parking spots, boats, bicycles, cameras and more. 2 Throughout the paper CPI also refers to the Harmonised Index of Consumer Prices HICP)

for accommodation services where a correction is performed to the housing service by owner occupiers of houses which otherwise is registered as production in the NA.

4 Accommodation Arguably among the most well-known sharing economy models are Airbnb, which has been said to have disrupted the industry of accommodation when entering the market as a competitor operating under rules differing from the ones existing in the established market. Airbnbs P2P offered accommodation service may feel different from stays provided through traditional accommodation. Differences are present through the accomodations physical attributes and its less visible ones such as different requirement, such as for instance building risk assessment and other similar national regulations mandatory to the existing accomodation service while not required in the regular housing market. Hence, the two service options should be seen as different products in price statistics. As the market share of Airbnb and the likes differs between countries, the inclusion of these services in the CPI sample must be determined individually by each country. In Norway short-term rental services like Airbnb are still rather small. Based on figures from 2017, NA currently estimates the household expenditure share to be below 0.1 per cent of total household consumption, but there are indications that the share is steadily growing and is expected to grow in the years to come. The question about whether rentals through Airbnb are to be considered B2C or P2P remains to be answered.

Traditional accommodation services such as hotels, motels, inns, and their likes operate within a legal context supporting the supplier and consumers in the existing markets. However, the existing legislation did not fully cover the activity made possible by sharing platforms which enabled peers to easily offer lodging under the safeguarding of the platforms terms and condition all while connecting the host to the “whole world” in an instance. As platforms such as Airbnb offer user profiles at no fee, the barrier was lowered substantially for peers to put an offer out for lodging while at the same time increasing the awareness of these possibilities for potential hosts.

The economic efficiency from the sharing economy model, which make it easier to rent out underused assets, is in general welcomed by the Norwegian government who appointed a Sharing Economy Committee in March 2016. The committee was asked to evaluate opportunities and challenges presented by the emerging market phenomenon (Government.no, 2017). Among other things, the Committee was tasked with identifying and assessing regulatory provisions challenged by the sharing economy, identifying the consequences of the sharing economy on the labour market and finally, the Committee was requested to consider consumer protection rules and the objective of consumer safety.

In the wake of the committee’s findings the government took legislative action. In effect from 2 April 2019 a new short-term rental law was effectuated allowing apartments in housing cooperatives to be rented out for a total of 90 days per year, while previously these types of short rentals were not allowed at all. Furthermore, the law made it illegal to own more than one unit in each housing cooperative. The intention of the new law is to balance the interests of those who wish offer lodging in their home and their neighbouring residents. The new rules state that for rentals where the length of stay is less than 30 days for each individual letting, the revenue is taxable under the standard method i.e.: that revenue from rental up to NOK 10,000 (around 850 € April 2023) is tax-free, while 85 percent of the remaining surplus revenue is considered taxable income. Rental revenue equals the total fee paid by the renter to the host including all additional cost related to the individual letting (The Norwegian Tax authorities, 2021).

5 Transport services Among the most well-known and highest profile companies within the sharing economy are the ridesharing companies Uber and Lyft. However, Norway has chosen a legislation which acts counter- current to many other countries in the transportation field within the sharing economy.

5.1 Taxi services As of November 2020 a new taxi market reform took effect in Norway, postponed from July 2020 due to the pandemic (www.government.no, 2021). The main elements in this reform was linked to shifting rights and responcebilities from the taxi license holders to the taxi drivers, in addition to deregulationg the numerical restriction on number of licences.

The deregulation of the market due to the taxi reform has led to a huge increase in the number of taxi licenses by 45% on a national basis and as much as 69% only in Oslo. New companies, like Yango, Bolt and Uber, providing taxi services have entered the market since the reform took place. These new companies in the market are all foreign, which limits the Norwegian Statistics Act legal force to to oblige data delivery to Statistics Norway. Due to the taxi reform the taxi companies are not required to be connected to a dispatcher. This makes it difficult to obtain data for taxi rides, probably even for the regular taxi rides in the future, as comprehensive data ideally should be obtained from each taxi driver.

Uber provided their services prior to the reform, but had to abandon their operation in Norway after a damaging court case in 2017. There are reasons to believe that the increase in the number of taxi licenses is connected to the establishment of these new platform companies. However, since the taximeter requirement is not yet removed, none of the new companies can operate entirely within the P2P business model, as one is required to holding a professional taxi licence3 as well as registering the vehicle as a taxi. The latter requirement includes a yearly EU periodic roadworthiness check, as opposed to a biannual check of roadworthiness required for a car purposed for private usage. Regular taxi drivers have also started driving for the new companies in addition to dispatching central. Yango and Bolt have registered as transport companies and not as taxi operation as given in NACE, probably to avoid some of the rules that a taxi driver/taxi company is subject to, among others the requirement of a taximeter.

According to the Tax Authorities, if you decide to make driving your main source of income, you must follow the general tax and reporting rules that apply to businesses. The general rule will then be that the income from the driving is taxable from the first NOK, and the expenses associated with the driving are deductible.

As the current regulation does not allow a taxi service purely through the P2P segment, the platforms are not able to fully make use of the P2P business models. In the future, if the taximeter requirement are replaced by digital platforms, P2P offered transportation services may reach significant market shares.

The question of whether or not taxi fares from the regular taxi companies work as a proxy for prices in the P2P segment still remains unanswered. Most likely, to gain market share in the Norwegian market the price level for taxi rides by the new companies will not surpass the fares in the existing taxi market. However, we do not have any information whether the price development differ from the regular taxi rides.

3 The fee for getting a taxi license issued is at present NOK 3400 (around 300 € April 2023)

Following the changes in the market brought to us by the new reform, some political parties are now advocating a reversion of the reform due to complaints about too many licenses in the market, leaving drivers without enough clients to reach decent wages. According to economic theory prices should drop in the face of supply overbidding demand, and the digitalisation within the taxi market has opened for more differeansiated prices (Aftenposten, 2023).

5.2 Vehicle sharing Several companies in Norway are offering vehicle sharing within the B2C segment. The companies are a mix of Norwegian and foreign.

As an alternative to private car ownership, organised carsharing is a system that offers people to rent cars locally available at any time and for any duration. Carsharing has existed in Norway for over two decades, however the number of users is still limited. The first carsharing providers in Norway were member-owned cooperatives in Norways three largest cities: Oslo, Bergen, and Trondheim. The carsharing stations were almost always located in central areas with a high enough residential or business density to sustain a viable customer base. Currently, seven carsharing providers within the B2C segment operate in Norway. It is estimated that around half of the members are passive members. One platform offers carsharing within the P2P segment, with about 10 000 cars in their registry (figures from 2021).

Carsharing users generally tend to be more urban, wealthy, educated and younger than the general population (link.springer.com, 2023). A typical user is between 30-40 years old, has higher education and fewer cars in the household. The biggest motivations for memberships are related to convenience, the financial aspect and the environment. Carsharing is primarily used for holiday and leisure trips as well as for shopping heavy goods, and rarely used for everyday travel such as commute. As more of the following generations grow up in families who do not own a car the phenomenon of carsharing may increase.

Allthough the station based cooperative model is the most established model, newer types of platforms, both in terms of organizational model and operational model, have entered the market since 2015. What remains to be seen is who will be the dominant players and what the dominant platforms will be in the future.

According to the Tax Authorities you do not have to pay tax on renting out your car if your rental income is up to NOK 10,000 (around 850 € April 2023) per year. It makes no difference whether you rent out the vehicle yourself privately or through an agent.

Smaller vehicles such as e-scooters was legalized in Norway in 2018, and since then several e-scooter sharing companies have established themselves in Norwegian cities offering around 20 000 vehicles. In Oslo, and elsewhere, unregulated e-scooter markets create challenges with respect to traffic safety and littering of excess vehicles resulting in an introduction of a new regulation in 2021 which limmit the number of companies in Oslo to three and the number of vehicles reduced to 8000, down from previously 23 000. All companies that operate in Norway are within the B2C segment, as the vehicles are owned by commercial companies. The same is the case for bicycles, both regular and electric,which also operate commercially.

Table 1. Types of sharing economy and share of total private consumption

Type of service Expenditure share, CPI (%)

P2P’s share of expenditure

Comment

Accommodation 0.8 Not significant* Both B2P and P2P at play, legislation adapted to both business models

Taxi services 0.3 Not significant* P2P business models not fully utilized yet as the requirement are like regular taxies

Carsharing (rental car) 0.1** Not significant* One companies offering P2P services. Several B2C companies are established in the market

*Less than 0.1 % of total private consumption according to NA **The expenditure share is for rental cars

6 The Covid-19 pandemic and the sharing economy The rapid development of the sharing economy in Norway was dealt a major setback due to the COVID- 19 pandemic of 2020 (Halvorsen, 2021). In 2023 it is still not fully clear what of the pandemics impacts remain permanent, and whether the rapid changes experienced pre-pandemic may soon return when the COVID-19 pandemic converges towards an endemic stage.

Figures from Statistics Norway’s accommodation statistics show a sharp decline in guest nights at commercial accommodation establishments in 2020. Norwegian guest nights declined by 17 per cent, while foreign guest nights declined by 69 per cent. Increased guest nights by Norwegians in the summer, especially at camping sites and holiday dwellings and youth hostels, did not compensate for the absence of foreigners. As restrictions were loosened and the willingness to travel domestically rose, the number of guest nights increased by 14 per cent from 2020 to 2021. In 2022, when most restrictions were liftet worldwide, the total number of guest nights rose to almost 3 per cent above the pre pandemic level in the year of 2019.

Similar data for Airbnb lodging in Norway is not public, but figures for Airbnb nights & bookings worldwide (FourWeekMBA, 2023) describe a substantial rise from 2017 to 2019, while figures dropped to reach the 2017 level in 2020, before once again climbing steeply through 2021 to reach an all-time number of bookings in 2022. Although the rules and regulations differed between countries throughout the pandemic, some similarities were present; rules and regulations which serve to limit the contact between people and reduce the likelihood of spreading the virus. It is likely that the decline observed in commercial accommodation in Norway corresponds to a decrease in the activity facilitated by Airbnb worldwide.

Also the transportation services were hit hard by the Covid-19 pandemic. The level of restrictions induced a reduction in demand, with activity increasing during the summer months of 2020, although variations between different segments were observed; the street segment was hit hardest, while the contract market segment4 seems to have performed better.

4 The taxi service industry can be divided into two segments, the single trip segment, and the contract segment, e.g. contract driving for public authorities or companies who negotiate fares for multiple trips. In the single trip segment customers either order a taxi through a dispatching service companies or hail a taxi from a taxi rank or from the street.

For taxi owners and employed drivers, the reduced demand in the early phase of the pandemic led to many temporary lay-offs and parked cars. Many taxi owners applied for compensation, with those who own multiple cars having a much higher chance of getting their claim for compensation accepted. Some of the temporarily laid-off drivers likely received an equal or larger sum in unemployment benefits than they would have been paid in wages if they were to continue to work in a market with severely reduced demand. Combined with the deregulation of the taxi market in November 2020, the pandemic made many taxi owners and employed drivers leave the industry.

While sales of new cars in Norway faced new records in 2021 and the government instructed people to avoid the use of crowded public transport as an attempt to stop the spread of the Covid-19 culminating in historic low passenger demand for public transport use especially in the capital of Oslo, there are reasons to believe that private driving increased during the pandemic, and therefore also the use of carsharing in 2020 and 2021.

No data are found for city-bicycles during the pandemic period in Norway, however dealerships of new electrical bicycles reportet new sales records during this period.

7 Taxable income data, - a possible data source? Legislative action was introduced in 2018 to address short-term rentals (defined as rental periods of 30 days or less) resulting in a softening of the regulation of the housing market. The deregulation opened up for subletting apartments in housing cooperatives for 90 days per year, as opposed to earlier restriction which forbid renting out these types of self-owned apartments.

Income from the sharing economy is liable to taxes, and as of February 2021 all platforms providing connection and facilitating payment between parties involved in renting lodging services in Norway, both Norwegian and foreign, are obliged to report information about each unique rental, regardless of the duration of the stay. Statistics Norway was granted full access to these data from the year 2020 through an agreement with the Norwegian Tax Authorities.

In its most severe form travel bans due to the Covid19 pandemic, effectuated in the spring of 2020, restricted inhabitants to stay within their registered municipality. Hence the figures derived from the Norwegian Tax Authority data for 2020 must be viewed as highly affected by the rules and regulations imposed on international travel and national movement in the period the data covers. In comparison several hotels shut down during the early stages of the pandemic. Most likely the following year is also affected by the pandemic as restrictions were imposed with variable strength and strictness in Norway and the rest of world throughout 2021 . This hypothesis is supported by figures for number of nights conveyed through Airbnb throughout the last six years, where the number of nights in 2021 accumulated to less than the most recent pre-pandemic year of 2019. Analysis on the year of 2021 will be released by Statistics Norway later in 2023, henceforth this paper will only describe figures from the first pandemic year of 2020.

In total slightly more than 400 000 unique rentals5 were registered in the Tax Authorities data for the year 2020 with about 90 per cent of them related to short-term rentals for up to 90 consecutive days. The numbers do not include information about rentals where the platform only arrange for the connection between the provider and the buyer of lodging services as verifiable transaction prices related to them registered in their system, as these platforms are relieved from the duty to report.

5 Unique rentals are equivalent to each transaction between the one that rents out and the ones renting. Every transaction is considered unique. The data does not identify the renter leaving no option for Statistics Norway to identify renters that repeats their rental on several occasions.

Since then COVID-19 hit, Airbnb launched its “Live Anywhere theme” in 2020, and said: As a result of the pandemic, millions of people can now live anywhere. They’re using Airbnb to travel to thousands of towns and cities, staying for weeks, months, or even entire seasons at a time. We want to design for this new world by making it even easier for guests to live on Airbnb. We believe that living somewhere enables deeper connections to local communities and the people who live there. In Q4 2022 stays of at least 28 nights accounted for 22% of gross nights booked, while 47 per cent of gross nights booked were from stays of at leas 7 nights (rentalscaleup, 2022).

8 Analysing the data from the Norwegian Tax Autorities 8.1 Number of stays and revenue In economic figures revenues from lodging reported to the Norwegian Tax Authorities was 1.7 billion NOK for 2020, while the corresponding total revenue for hotel accommodation from official statistics was 9.4 billion NOK. On average the price for lodging was 1100 NOK per night stay, while the average price per night in a hotel room in 2020 was 979 NOK (Statistics Norway, 2021). Be aware that these figures do not say anything about the size and location of the rental, not the number of people staying in the unique rental object; all factors that may influence the observed prices.

Further drilling in to the length of stay dimension in the Tax Authority data show that most rentals are substantially shorter than 90 days6. This corresponds to the general right of vacation days granted within the EU being 4 weeks, some member states and EFTA members, like Norway, operate with more vacation days, while the US and Canada have considerable weaker standard rights to paid vacation days (EurDev, 2021). The observed data showed a boost in the length of stay at exactly one-week rentals, while the numbers consistently decreased for each more added night of stay. By selecting only stays consisting of one-week rentals or less we are left with about 84 per cent of the original data material.

Table 2. Share of stays by lenght

All platforms providing rental agreement in Norway are represented in the Tax Authority data. In this analysis we aim to identify the ones represented by platforms as defined by the sharing economy phenomenon. As defined above an important aspect of these exchanges is the distinction of transaction made “primarily between private individuals”.

6 The Norwegian Tax Authority’s definition of short-term rental (less than 90 days per stay) is adapted to the purpose of tax liability.

Night(s) stay Percent Cumulative Percent 1 27,3 27,4 2 23,3 50,7 3 14,3 64,9 4 7,6 72,6 5 3,9 76,5 6 1,9 78,4 7 5,8 84,3 8 0,7 85,0 9 0,6 85,6

The better part of the data were rentals arranged via Airbnb. Booking.com were well represented in the data too. However, booking.com operate also in the segment were rentals made through them are targeted at established brands and entrepreneurs of all sizes (Booking.com, 2022).

8.2 Rental object number The data included information about rental object number. To secure the aspect of transactions “primarily between private individuals” we assume that private individuals most likely do not operate with several rental objects and decided to include only rental object numbers of one or two. The higher rental object numbers from 3 and up covered about 10 per cent of the original data material, leaving almost 90 per cent for further analysis.

By selecting only rentals with a length of stay of a week or less, only Airbnb and securing the number of objects rented out by each host to be no more than 2 we believe we are left with a subset of data that is well within the definition of sharing economy where transactions are made “primarily between private individuals” as well as it represents lodging acquired by private households in Norway through this channel. The subset of data after this selection is done accounts for more than 180 000 unique rentals, equivalent to about 45 per cent of the original dataset. The number of nights of stay in the subset of data were about 500 000, about 1/3 of the original data material, totalling up to 500 million, about 30 per cent of the total revenue in the full data set.

In compiling a price index the first step was to derive a unit price per night per unique rental. As the same rental object may have been rented out several times within one period (month) the unit price was aggregated to a monthly unit price before a timeline per unique rental object was constructed.

Rental numbers vary throughout 2020. About 60 000 unique rental subjects have a monthly unit price registered which are unequally distributed throughout the months of 2020.

Table 3. Monthly overview of rentals, per cent.

Action taken by the government during the spring of 2020 to restrict the spread of COVID-19 is visible through the low activity seen in the spring months of March, April and May. Followed by a summer where the mobility within Norway was unrestricted and numbers again rose, the lodging numbers declined as COVID-19 numbers rose through the fall and the government once again enforced strict regulationsto reduce the spread of the virus to a maintainable scale as the year moved towards Christmas celebration. Most likely also non-pandemic figures would vary throughout the year, with high numbers associated / coinciding with national holidays and summer vacation in Norway mid-June to mid-August, and the following summer holiday season for southern Europe lasting through August.

When measuring hotel prices, the services followed are consistent over time with regards to location, interior and amnesties included. It is to be assumed that the same rooms are either rented out or

offered for rent. This is opposite to the sharing economy lodging which offers non-commercial accommodation by private households at the time when the rental object is available for the hosts to offer the public. Whereas it is possible to measure the same service over time offered by traditional accommodation services at hotels or other established facilities, the very idea of sharing economy imposes challenges through its diversity in object offered or actually rented out which may differ greatly between periods.

8.3 Matching unique rental objects Aggregating the about 60 000 unit price observation per night per unique rental to an annual time series leaves us with shortly less than 20 000 unique hosted lodgings during the year. Among these slightly 30 per cent of the unique lodgings were present during only one of the months in the year of 2020. To measure prices over time the lodging object must at least be present in two consecutive periods (months) or more. The data shows that only very few object (less than 1 per cent) were rented out throughout every month of 2020, with an increasing percentage of lodging objects appearing when moving from occurring in twelve months during the year towards only twice.

Table 4. Unique rentals per month, per cent.

As a large proportion of the data are only present in one month of 2020 only a small share are available for a match with a previous period. The figures illustrate the increase in matches when shifting from matches towards a fixed base period to matches between two consecutive months.

Numbers of matches during a full year can at maximum reach 11 for one unique lodging object. More than 70 percent of the lodging object does not match with a fixed base period of January. While the range differs between a good 6 per cent for a match between January and two other months during the year of 2020 and below 1 per cent for a match between a unique lodging object in January and the following eleven months.

Table 5. Overview of unique rental objects rented per month and in January

The numbers of matches increase for unique lodging object when matches are made for two consecutive months. Almost 36 per cent are unique lodging object rented out only during one of the months in 2020, while about 23 per cent are found to have been rented out for two consecutive months and almost 17 per cent were rented out for two consecutive months twice7. The numbers evenly decrease for each added possible match with only 0,2 per cent of the unique lodging object being rented out for 11 of the all years twelve months, and a slight rise to 0,7 per cent of all unique lodgings accounted for were rented out at least once in every month of 2020.

Table 6. Overview of unique rental objects rented in a particular month and the month before, per cent.

Having price observations for the same service is one step along the way to compiling a price index. We also need to make sure the services we measure prices for in the whole universe of services offered and consumed are representative services consumed by private households, both with regards to location, length of stay, size of lodging and the standard of the service provided. When we have access to data which accounts for all activity which fall under the Norwegian Tax Authority terms of sharing economy for lodging, covering the geographical boundaries of Norway and channelled both through nationally and abroad owned platforms, the challenge moves from traditional sample issues towards more limitations in the data source. Other P2P rentals are probably prevalent throughout the year, but

7 This can either be 3 consecutive months or two consecutive months twice.

rentals which are channelled via a sharing app or website are most likely registered in these data unless the platform operates under illicit terms. Hence, the Tax Authority data provides a complete overview over the accommodation activity in Norway under the terms of sharing economy apps and websites.

Lacking in the data is information on the purpose of the stay, if it is to be considered business or recreational purpose. The lack of such information is well known when measuring prices with the intention to include in a price index. As long as there is no discrimination between the two consumer purposes, leaving one of the two with a different price development then the prices measured are accurate enough. However, if the rental objects are strongly related to the purpose of business, which does not fall the scope of the CPI, then a bias towards including non-relevant rental objects may be introduced when the corresponding prices are not properly identified and excluded from the price material that enters the index. For instance, when booking at Airbnb.com there is an option to mark an Airbnb reservation as a business trip resulting in Airbnb providing an receipt for expenses while also providing Airbnb with valuable information about the purpose of the stay. This information is not conveyed to the Norwegian Tax Authority. In the fall of 2018 Airbnb launched a work program aiming for an increase from 15 per cent of total bookings stemming from business travel to a 30 per cent share in 2020 (curbed.com, 2018). For the period of the Tax Authority data which we analyse in this paper, it is safe to assume that if the business segment it present, then it is limited as the rules and restriction for work forced typical travel activity to become digital. However, in a situation with no pandemic restriction these are issues that should be addressed.

The most basic form of an unweighted monthly chained price index for the year 2020 shows a more volatility and a higher index level throughout the year of 2020 compared with the published series for 11201 Hotels, motels, inns and similar accommodation services.

Chart 1. Price index lodging by Airbnb 2020. January 2020=100.

The experimental index is a monthly chained index for the year 2020. The published index is a weighted Laspeyre, with December the previous year as the price reference month. For comparability the published index is re-referenced to January 2020=100 from the official 2015=100.

As described above the experimental index is tainted by several challenges, while the published index series for some of the months are affected by the pandemic directly through how we treated consumption which fell close to zero in the period.

First and foremost, the number of prices entering the experimental index varies strongly between the periods. In general, missing observations in the published index series are imputed in line with rules according to the principle of nearest neighbour imputation; starting with the most detailed level within the region the missing price is observed, then drilling upward in the hierarchy . In the experimental index no such imputation is performed, and prices enter where they exist. Additionally, the published index is affected by imputation of the overall index of the CPI consisting of the remaining consumption based on real price observations for the (pandemic) period (Statistics Norway, 2020). With regards to homogeneity, in the published series homogeneity are ensured as the respondents are asked to price a representative service of a specified standard, equally stated to all respondents who provide these services. In the Airbnb data the aspects of the rental object beyond regionality is not registered. The variation of unique rental object whose prices enter the index may vary substantially both within a month and over time time. Not performing imputation of the missing basic data in the experimental index forfeit the possibility to follow the unique rental object over time as missing price observation in one period introduces a breach to the timeline.

9 Further work The sharing economy within the P2P platforms is for the time being rather small in Norway. Most of what is described as sharing economy is within the B2C plattforms, just indicating that traditional business are utilizing the new business models. Worldwide, we find accommodation and transport services as major services within the P2P plattforms. Through NA we are able to identify an expenditure share for Airbnb, which is still less than 0.1 per cent of total private consumption. No data is available to identify a significant expenditure share for taxi services from the platforms operating within the P2P segment. However, restrictions in the taxi market, making it not so easy to use your own car, indicates an almost non-existing market share of the total taxi market. The experimental work on the Norwegian Tax Authority data shows the new possibilities that occur when access to a new data source appears. Although several challenges remain unanswered, the data available from the Norwegian Tax Authorities are detailed enough to compile a simple version of a price index retrospectively.

Data for the following year are yet to be analysed, however we are already aware of more granulated details introduced in the data for 2021. Utilizing the added level of detail in the data source are expected to enable further improvements in the processing and delineation of the data, maybe even increasing the subset of data potentially entering a price index as the added granularity of detail may prove usefull to subtract the P2P arranged stays from the B2P segment for platforms such as booking.com which currently are catgorized as fully operating withing the B2P segment in lack of information to categorize a stay differently.

The primary challenge with the Norwegian Tax Authority data stems from timeliness, as these data are a one-time extraction for the whole year of 2020, this does not satisfy the timeliness needed in a price index which should register the prices in the period the service commences.

If or when this data source may be of the right timeliness and quality to be used as a source of price information to produce the CPI is too early to conclude on. However, these data will be a much- needed new source of information for NA in their calculation of the production level for Airbnb- related activities in Norway. And the data in its current set up does shed light on aspects regarding traditional sampling issues such as specifying the population and selecting a sample with regards to regionality.

Even though the aggregated expenditure shares for the variety of services provided though the P2P measured by the NA are yet less than of 0.1 per cent of total consumption, we anticipate a future need to measure these prices as the sharing economy activity in Norway most likely will become more prevalent.

References (

Aasestad K, K. J. (2021, November 30). Accommodation offered via online collaborative economy platforms. Norway 2020. Retrieved from https://www.ssb.no/en/transport-og- reiseliv/reiseliv/artikler/accommodation-offered-via-online-collaborative-economy- platforms.norway-2020

Aftenposten. (2023, 05 12). Retrieved from https://www.aftenposten.no/meninger/kommentar/i/9zMBaq/smarte-drosjekunder-har- faatt-det-bedre

Andreotti, A. A. (2017). bo.edu. Retrieved from European Perspectives on Participation in the Sharing Economy: https://www.bi.edu/globalassets/forskning/h2020/participation-working-paper- final-version-for-web.pdf

Booking.com. (2022). Booking.com. Retrieved from About Booking.com: https://www.booking.com/content/about.html?aid=318615&label=Norwegian-NO- 131246328204- NGQNXKWfg44q8FRM%2A4MHhwS562363086939%3Apl%3Ata%3Ap1%3Ap2%3Aac%3Aap% 3Aneg%3Afi2657853280%3Atidsa- 1227182654382%3Alp1010826%3Ali%3Adec%3Adm&sid=3ef3ce6a69ac1cce78b18a7b5f

curbed.com. (2018, OCT 4). Retrieved from Airbnb expands services to corner profitable business travel market: https://archive.curbed.com/2018/10/4/17938076/hotel-airbnb-meeting- business-travel

EurDev. (2021, January 22). EurDev. Retrieved from Paid Vacation Days Europe 2021: https://blog.eurodev.com/paid-vacation-days-europe-2021

Eurostat. (2018, November). ec.europa.eu. Retrieved from Harmonised Index of Consumer Prices (HICP) Methodological Manual: https://ec.europa.eu/eurostat/documents/3859598/9479325/KS-GQ-17-015-EN- N.pdf/d5e63427-c588-479f-9b19-f4b4d698f2a2

Eurostat. (2018). Harmonised Index of Consumer Prices (HICP). Methodological manual. Retrieved from https://ec.europa.eu/eurostat/documents/3859598/9479325/KS-GQ-17-015-EN- N.pdf/d5e63427-c588-479f-9b19-f4b4d698f2a2

FourWeekMBA. (2023, February 19). Retrieved from https://fourweekmba.com/airbnb-bookings/

Government.no. (2017). Retrieved from NOU 2017: 4 Sharing Economy - Opportunities and challenges: https://www.regjeringen.no/en/dokumenter/nou-2017-4/id2537495/

Government.no. (2017:3). Retrieved from NOU Norges offentlige Utredninger: Delingsøkonomien - muligheter og utfordringer : https://www.regjeringen.no/contentassets/1b21cafea73c4b45b63850bd83ba4fb4/no/pdfs/ nou201720170004000dddpdfs.pdf

Government.no. (2021, 10 14). Retrieved from Spørsmål og svar om nytt drosjeregelverk: https://www.regjeringen.no/no/tema/transport-og-kommunikasjon/ytransport/sporsmal- og-svar-om-nytt-drosjeregelverk/id2641640/

Government.no. (2021). Retrieved from The coronavirus situation: https://www.regjeringen.no/en/topics/koronavirus-covid-19/id2692388/

Halvorsen, T. C. (2021, OCTOBER). sharingandcaring.eu. Retrieved from The Sharing Economy in Norway: Emerging Trends and: https://sharingandcaring.eu/sites/default/files/files/ebook/Chapter_18_The_Sharing_Econo my_in_Norway_Emerging_Trends_and_Debates.pdf

IMF Committee on Balance of Payment. (2020). Statistical Framework for the Informal Economy. Retrieved from https://www.unescwa.org/sites/default/files/event/materials/Informal%20Economy%20Tas k%20Team-concept-note.pdf

Jesnes et al. (2016:7). Retrieved from Aktører og arbeid i delingsøkonomien: https://www.fafo.no/images/pub/2016/10247.pdf

link.springer.com. (2023, 03 25). Retrieved from https://link.springer.com/article/10.1007/s11116- 023-10386-0

Newlands, G. L. (2019). The conditioning function of rating mechanisms fro consumers in the sharing economy. Retrieved from biopen.bi.no: https://biopen.bi.no/bi- xmlui/handle/11250/2602833

Ranzini, G. E. (2017 - II). bi.edu. Retrieved from Privacy in the Sharing Economy: European perspective: https://www.bi.edu/globalassets/forskning/h2020/privacy-survey-working- paper-for-web.pdf

Ranzini, G. N. (2017 - I). bi.edu. Retrieved from Millennials and the sharing economy: European perspectives.: https://www.bi.edu/globalassets/forskning/h2020/focus-group-working- paper.pdf

rentalscaleup. (2022, 02 17). Retrieved from https://www.rentalscaleup.com/2022-airbnb-strategy/

Statistics Norway. (2020). Retrieved from Corona consequences for CPI: https://www.ssb.no/en/priser-og-prisindekser/artikler-og-publikasjoner/corona- consequences-for-cpi

Statistics Norway. (2021). ssb.no. Retrieved from 12897: Revenue and utilisation of rooms at hotels, by region, contents and month: https://www.ssb.no/en/statbank/table/12897/tableViewLayout1/

Statistics Norway. (2021). Travel Survey . Retrieved from https://www.ssb.no/en/transport-og- reiseliv/reiseliv/statistikk/reiseundersokelsen

Thackway, W. T. (2021). Airbnb during COVID-19 and what this tells us about Airbnb’s Impact on Rental Prices. Retrieved from Findings: https://findingspress.org/article/23720-airbnb- during-covid-19-and-what-this-tells-us-about-airbnb-s-impact-on-rental-prices

The Norwegian Tax authorities. (2021). Tax rules for short-term letting of homes and holiday homes. Retrieved from https://www.skatteetaten.no/en/person/taxes/get-the-taxes-right/property- and-belongings/houses-property-and-plots-of-land/letting-of-houses-and-property/short- term-letting-of-dwellings-and-holiday-homes/tax-rules-for-short-term-letting-of-homes-and- holida

The Norwegian Tax Authorities. (2022). Sharing economy. Retrieved from www.skatteetaten.no: https://www.skatteetaten.no/en/person/taxes/get-the-taxes-right/employment-benefits- and-pensions/hobby-odd-jobs-and-extra-income/sharing-economy/

Utleiemegleren.no. (2022). Retrieved from Om Utleiemegleren: https://www.utleiemegleren.no/om- oss

www.government.no. (2021, 10 14). Retrieved from https://www.regjeringen.no/no/tema/transport- og-kommunikasjon/ytransport/sporsmal-og-svar-om-nytt-drosjeregelverk/id2641640/

www.ssb.no. (2021). Retrieved from https://www.ssb.no/en/omssb/lover-og- prinsipper/statistikkloven

Ydersbond. (2023). Erfaringer med lov om utleie av små elektriske kjøretøy på offentlig grunn. Retrieved from https://www.toi.no/publikasjoner/erfaringer-med-lov-om-utleie-av-sma- elektriske-kjoretoy-pa-offentlig-grunn-article38033-8.html

https://one.oecd.org/document/STD/CSSP/WPNA(2017)9/En/pdf

Appendix Examples of sharing economy in Norway

Following is a list of some of the sharing apps in the Norwegian market which ranges from singular focused sharing apps to the all-consumer area apps:

Lodging services: www.airbnb.com

Child care services: www.sitly.no

Transportation by car: www.uber.com

Cleaning services: www.weclean.no

FINN online market (almost everything): https://finn.no

Book market (used and new): https://bookis.com (skal brukt være med?)

Carsharing: https://nabobil.no/en

Services provided by neighbours: www.obos.no/Nabohjelp

Clothes, decoration and furniture (used and redesign): https://tiseit.com

Sharing goods: https://www.hygglo.no/

Finding family relations: About quality issues regarding family immigration

statistics BY CHRISTIAN SOERLIEN MOLSTAD

DIVISION OF POPULATION STATISTICS, STATISTICS NORWAY (SN)

Agenda

1. Statistics on reason for immigration

• Data sources

• Family immigration and the person of reference (PoF)

2. Directive 2004/38/EC and its consequences

• The registration scheme and loss of data

3. Some considerations on potential solutions

1. Statistics on reason for immigration

Data sources for reason for immigration

• The Central Population Register (CPR)

Managed by the Norwegian Tax Administration (Skatteetaten)

• The Aliens Register (UDB)

Managed by the Norwegian Directorate of Immigration (UDI)

CPR BEREG + AR= The annual statistics on reasons for immigration

5

The Central Population Register (CPR)

Daily transactions of a copy of the

CPR

BeReg

The Aliens Register in DI

Annual data on detailed reason for immigration

Statistical files

Main reasons for immigration Five main categories of reasons for immigration (inngrunn1) on the finished file:

• Labour

• Family

• Refuge

• Education

• Other

The file only includes non-Nordic immigrants who arrived in 1990 or later

The person of reference (PoF) and background variables about her/him • The person of reference (PoF): the person to which family migrant is

immigrating

• Background information regarding the PoF is added from data derived the

CPR:

◦ Country of birth (rpfodeland)

◦ First citizenship (rpforststatsborg)

◦ Immigrant background (rpinvkat)

◦ Country background (rplandbak3gen)

◦ Registry status (rpregstatus)

etc.

Family immigration is too a large extent an extension of other types of immigration

0

5 000

10 000

15 000

20 000

25 000

30 000

1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2020

Labour

Family

Refuge

Education

Other

Missing

Immigrations to Norway, by reason, 1990-2020

Source: Division for Population Statistics

… and integration of family immigrants seem to vary with the background of the PoF

0

10

20

30

40

50

60

70

80

90

100

2 3 4 5 6 7 8 9 10

All

Labor migrant

Refugee

Family immigrant

Person without immigrant background

Family immigrants (arrived in 2006-2008) by PoF’s reason for immigration/immigrant background. Percentage employed 2-10 years after immigration

Source: Molstad et al, 2022, p. 65

2. Directive 2004/38/EC and its consequences

Directive 2004/38/EC

• The directive grants citizens of the European Union “the right to move and

reside freely within the territory of the Member States”. This right is “also

granted to their family members, irrespective of nationality” (EUR-Lex, 2022,

pp. 78-79)

• A simplified registration scheme was introduced by UDI in 2009, through

which EU/EEA-citizens and family members of these have been required to

register within three months of arrival

Increase in missing PoF among family migrants from both inside and outside the EU/EEA

0

10

20

30

40

50

60

70

80

90

100

1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2020

Total

Non-EU/EEA

EU/EEA

Family immigrations to Norway, percentage missing person of reference, 1990-2020

Source: Division for Population Statistics

Total: 55 700 (17 percent) family migrants with missing PoF

3. Some considerations on potential solutions

The project and its scope • A project, financed by the Ministry of Labour and Social Inclusion (AID), has been

initiated to attempt to cover the information loss

• Project deadline: March 2023

• Methods for imputation of missing data, applying a near-neighbour methodology, have

been used for linking households and dwellings (Zhang & Hendriks, 2012) and

imputing missing data among immigrants in the Register of the Population’s Level of

Education (BU) (Jentoft, 2014)

• In the current project we hope to avoid wholesale imputation and to be able to

successfully identify the actual PoF for each family immigrant using data derived from

administrative records

Starting point for development of method(s)

Preliminary steps to be performed:

1. Identify the possible constellations of family relations that qualify for family immigration

2. Identify data available, either in UDB or CPR or in other administrative data sources, that

may be utilized to distinguish family migrants according to their probable relation to their

person of reference as identified in step 1

3. Identify data that can act to utilize this relation to link the family migrant to a person

registered in CPR

The result is most probably a range of methods, each different according to

each “class” of family relation

Example: Existing method for family establishments

• Family migration due to spouse is possible to identify through the variable

inngrunn, derived from data from UDB

• Marital data concerning the identity of married individuals’ spouses is present

in CPR and available to Statistics Norway through the variable ekt_fnr

• This data is used to identify the spouse (i.e. PoF) of the family immigrant and

add variables regarding the background of this person to the file

The result: Few family establishment with missing PoF

0

10

20

30

40

50

60

Establishment Reunion Establishment Reunion

EU/EEA Non-EU/EEA

Percentage family immigrants with missing person of reference, by type of family immigration

Source: Division for Population Statistics

Thank you for your attention!

References EUR-Lex (2022). Directive 2004/38/EC of the European Parliament and of the Council of 29 April 2004 on the right

of citizens of the Union and their family members to move and reside freely within the territory of the Member

States, 47 77 (2022). https://eur-lex.europa.eu/legal-content/EN/TXT/PDF/?uri=CELEX:32004L0038&from=EN

Jentoft, S. (2014). Imputation of missing data among immigrants in the Register of the Population's Level of Education

(BU) (Notater/Documents 2014/27). Statistisk Sentralbyrå. https://www.ssb.no/utdanning/artikler-og-

publikasjoner/_attachment/183364?_ts=146a98197c8

Molstad, C. S., Gulbrandsen, F. B., & Steinkellner, A. (2022). Familieinnvandring og ekteskapsmønster 1990-2020.

(2022/03). Statistisk Sentralbyrå. https://www.ssb.no/befolkning/innvandrere/artikler/familieinnvandring-og-

ekteskapsmonster-1990-2020

Zhang, L.-C., & Hendriks, C. (2012). Micro integration of register-based census data for dwelling and household.

United Nations Economic Commission for Europe (UNECE): Conference of European Statisticians. Work Session on

Statistical Data Editing https://unece.org/fileadmin/DAM/stats/documents/ece/ces/ge.44/2012/16_Norway.pdf