SlideShare a Scribd company logo
1 of 27
Download to read offline
Sampling schemes using scanner data for the
consumer price index
De Vitiis C., Guandalini A., Inglese F., Terribili M. D.
Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS
Rome, 19th November 2018
Outline
1. ISTAT and international context
2. The ISTAT sampling experiments of SD
3. Experiments of sampling from SD
4. Experiments of static and dynamic approaches
5. The sample design for the outlets
6. Conclusions and open issues
Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
 ISTAT started using the scanner data (SD) in 2018 to compute the
consumer price index (CPI) for retail modern distribution (food and grocery)
 SD are regularly provided since 2014 to ISTAT through the market research
company ACNielsen, for the main chains in Italy and for a sample of about
2,000 outlets, scattered throughout the country
 Nielsen provide to ISTAT the Dictionary which associates to items identifying attributes
and ECR classification codes, allowing to link to COICOP classification
 For 2018 the compilation of the CPI using scanner data is based on a fixed
basket perspective, but a gradual transition to a flexible basket is the goal
for ISTAT
 At the moment, for the traditional retail distribution of food and grocery, data
are collected by the current survey on field: traditional and scanner data are
combined through weights derived from the expenditure survey
 Aim of this presentation is showing and discussing the experiments
carried out by our research group using SD in a sampling perspective, to
compare sample schemes for different index calculation approaches
1. Context: the ISTAT project on Scanner Data
Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
 SD are used In several countries for the compilation of the CPI:
Switzerland, Norway, Netherlands and Sweden for some years, while
Belgium and Denmark only from 2016
 Scanner data are exploited in different ways:
 Replacing price collection in the stores, without changing the traditional
principles of computing the price indices (Swiss Federal Statistical
Office)
 As a universe from which samples of references can be selected
following different methods (Norway and Sweden)
 Extensively to compile price indices, without a strict sample selection,
but with consequences on the theoretical definition of the index
(Nederland )
1. Context: The international use of SD
Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
SCANNER DATA
 Electronic sales data, containing both prices and quantities sold for every
item purchased by consumers in modern distribution
Opportunities
 To introduce improvements in the CPI also with regards the sampling
perspective, considering a dynamic population approach and different index
formulas for the lowest aggregation level (elementary indices), allowing to take
into account the dynamicity of the market
 To offer a real possibility of computing more accurate indices, since SD allow,
potentially, to include in indices the expenditure share of each item
Issues to be addressed
 Attrition of products, temporary missing products, entry of new products,
relaunches and volatility of the prices and quantities due mainly to sales.
These aspects need to be addressed from both a theoretical and a practical
point of view (de Haan et al., 2016)
1. ISTAT and international context
Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
OBJECTIVE
 The general aim was evaluating the use of SD for the compilation of
elementary price indices (lowest aggregation level of price index calculus, on which
the subsequent aggregations are based, in the Italian case the consumption segments)
 from a sampling perspective
 considering both static and dynamic population
 following a simulative approach
The experiments were developed in two phases:
 In the first phase, following a fixed basket approach, probability and
nonprobability selection schemes of series (references individuated by
EAN and outlet codes) are compared
 In the second phase, an experiment of evaluating the differences
between a fixed and a flexible basket approach was carried out, trying
to measure the magnitude of sampling error and bias for different index
formulas in both approaches
2. The ISTAT sampling experiments of SD
Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
The DATA used for the experimental phase
 Data referred to the first Italian provinces for which ISTAT got data, food and
grocery sector (covering 11% of the total index, while modern distribution
covers the 55% of the sector)
 Weekly data on turnover and quantities per EAN-code (European Article
Number, or GTIN, Global Trade Item Number) and outlet allow obtaining
weekly unit values
 One series (=EAN by outlet) is considered present in a specific month if it has
associated a non zero value in at least one week out of the first full three
weeks
 Our experiments are based on a panel of PERMANENT SERIES
 The permanent series are defined as having non-zero turnover for at
least one relevant week (one of the first three full weeks) every month for
13 months
 Permanent series provide generally a good coverage of the total turnover
 Each experiment focused on one province, some consumption segments
(Italian COICOP6) or markets, permanent series
2. The ISTAT sampling experiments of SD
Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
2. The ISTAT sampling experiments of SD: the theoretical context
Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
FRAMEWORK of the experiments
 Two selection stages for each domain (the whole province, not only the chief
towns as the current survey on field)
o Outlets selected through a probability sampling from the archive
o Items (references) selected either by probability sampling or cut-off selection
 For the item the definition of the population and parameter of interest is crucial :
o Static population  fixed approach – fixed sample/ direct index (weighted or
unweigthed)
o The indices at the lowest aggregation level are obtained from the monthly
price ratios (micro-indices) with fixed base (December previous year)
o Dynamic population  dynamic approach – dynamic sample/ chained index
(weighted or unweigthed)
o The elementary indices are calculated on matched-items: only price
relatives of items that are sold in two consecutive months enter in the
index formulas (flexible basket)
The bilateral indices in the Static and Dynamic approach
 Static population  fixed basket/ direct index
 The loss of representativeness of the basket (shrinkage effect) is addressed through
the yearly base change of the index and the renewal of the basket
 The fixed basket approach do not allow to exploit all the information provided by
scanner data, do not take into account the population dynamics, new products are
ignored
 Dynamic population  matched model/ chained index
 Chain indices take into account the movements of prices within the considered time
interval and the shrinkage effect over time due to the attrition of a fixed basket of
products is solved
 Period-on-period chaining of weighted indices introduces chain drift - bias due to the
non-transitivity of the chained weighted price indices, nonsymmetric effect on quantities
sold and expenditures share before and after sale (de Haan and van der Grient, 2011;
Ivancic et al., 2011)
2. The ISTAT sampling experiments of SD: the theoretical context
Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
Aggregation formulas for the elementary indices
The price indices considered are:
 Fisher (superlative, weighted) index is preferred by the economic theory, it uses
quantities in different times and allows for substitution effects
 Törnqvist (superlative, weighted) index is a weighted geometric mean, weights are
the average of expenditure share calculated in two times
 Lowe index is a modified Laspeyres in which a base month for the price (December
of previous year) and a base year (previous year) for the quantities are taken into
account for the weight computation
 Jevons index unweighted geometric mean
 In the experiments unbiased estimators of these indices were defined
through the sampling weights
2. The ISTAT sampling experiments of SD: the theoretical context
Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
First experiments : Comparison of probability and nonprobability
sampling selection schemes
 Non-probability sample: Cut-off selection of series based on thresholds
of covered turnover; the index is compiled using all series covering 60%
or 80% of all turnover in each outlet for the consumption segment, in
previous year 2013; Reference method: most sold items in each outlet
for representative products (current fixed basket approach)
 Probability sampling: two-stage sampling design with stratification of the
first stage units
 Stratification of outlets by 6 chains (Conad, Coop, Esselunga,
Auchan, Carrefour, Selex) and outlet type (hypermarket and
supermarket)
 The size of the sample of outlet has been fixed at a number of 30
out of 121 outlets available for Turin, selected with equal
probabilities
 Sample size for the second stage (EANs) is given by a 5%
sampling rate and selection by Sampford pps
 The universe is the set of panel series
3. Experiments of sampling from SD
Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
 Three different sampling designs have been considered:
 Stratified one-stage sampling
 Cluster sampling (outlet)
 Two-stage sampling (outlet and EAN)
 The universe of EANs is stratified by market (ECR groups) in six consumption
segments (coffee, pasta, mineral water, olive oil, spumante and ice cream)
 Different criteria of sample allocation (proportional and Neyman) for both outlets
and elementary items (EANs) and different methods for the selection of sampling
units have been considered (SRS and PPS)
Further experiments: Comparison among probability sampling schemes
3. Experiments of sampling from SD
Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
The comparison between probability and nonprobability selection
schemes was made for each price index taking the corresponding
universe index value (panel series SD) as reference true value
Figure 1 – First experiments: Jevons, Lowe and Fisher indices computed on samples of series selected with
different selection schemes - Coffee and pasta segments, Turin province, year 2014
COFFEE PASTA
Jevons
Lowe
Fisher
Lowe
Fisher
Results of the experiments in the first phase
Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
Absolute Relative Bias and Relative Sampling Error distributions of monthly Lowe, Fisher and Jevons
indices for consumption segment (Sampford sampling)
Results of the experiments in the first phase
Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
Index
Coffee consumption segment
Min Q1 Me Q3 Max
Lowe
ARB -0.0005 -0,0014 -0,0013 -0,0008 -0,0017
RSE 1,32 1,38 1,41 1,43 1,52
Fisher
ARB 0,0005 0,0015 0,0021 0,0030 0,0039
RSE 1,14 1,36 1,53 1,73 1,98
Jevons
ARB -0,0199 -0,0485 -0,0400 -0,0353 -0,0609
RSE 1,05 1,09 1,14 1,20 1,24
Index
Pasta consumption segment
Min Q1 Me Q3 Max
Lowe
ARB 0.0008 -0,0003 -0,0001 0,0005 -0,0007
RSE 1,13 1,20 1,28 1,32 1,38
Fisher
ARB -0,0009 -0,0043 -0,0034 -0,0025 -0,0051
RSE 1,21 1,61 1,67 1,83 2,06
Jevons
ARB -0,0017 0,0031 0,0105 0,0209 0,0314
RSE 0,85 0,92 1,04 1,10 1,22
The results highlight the following evidences about the performances of different
series selection schemes
 Probability pps sample produces approximately unbiased estimates for indices using
weights (Lowe and Fisher)
 Increasing the sampling rate produce a remarkable improvement of the bias in all
indices, in addition to an obvious reduction of sampling error
 Stratified One stage sampling produces estimators more efficient, but for practical
problems we need to pass through outlets.
 Cluster sampling produces estimators having good performances, but the high
number of products sold in each outlets rapidly makes increase the size of SSU.
 Two stage sampling produces estimators with quite good performances, probably
because of the most of variability is among outlets.
 The performance depends in general on type of turnover distribution of items for the
product segments
I simulation study: summary of the evidences
Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
 The second experimental phase is an attempt to make a comparison
between static and dynamic approach, evaluating the total error of the
estimates of indices
 Simulation of alternative population scenarios
 artificial population: starting from the panel, new products have been
introduced considering the monthly birth rates and old products have been
removed in accordance to a survival function (both monthly birth and survival
rates have been estimated on the real open population)
 From a sampling perspective the two approaches refer to different
sampling designs:
 STATIC two-stage sampling (outlets, EANs)
 DYNAMIC cluster sampling (outlets)
4. Experiments of Static and Dynamic approaches (second phase)
Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
(A) Panel (Static approach)
All the products enter in the
computation of indices
(B) Static approach without
disappearing products
The products presented in December
of the base month are followed during
all the year. The new products are not
considered by definition. However,
disappearing products are assumed to
be present
(C) “Pure” Static approach
Only the fixed basket approach is
applied. The new products are not
considered
Scenarios implemented under Static approach
Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
(A) Panel (dynamic approach)
All the products enter in the
computation of indices.
(D) Dynamic approach
Only the matched products in two
months in a row enter in the
computation of price indices.
Scenarios implemented under Dynamic approach
Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
STATIC vs DYNAMIC – Jevons with and without Treshold
Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
Comments on the experiments on
static and dynamic approaches
 The simulation allowed to experiment the two approaches and to
highlight the practical and theoretical implications
 Non-sampling error seems to affect the estimators of price indices
under the dynamic approach more than the under the static
approach
 The sampling error of the estimator under the dynamic approach
is much lower than in the static approach due to the difference in
sample size
Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
5. The sample design for the selection of outlets from the Nielsen universe
 For 2017 a random sample of outlets has been selected from the Nielsen
universe of hyper and supermarkets covering all Italian provinces and 16
chains, to start experimenting the index production
 To allocate efficiently the sample of 2100 outlets, the variability in the
universe of the prices indices referred to the initial set of 37 provinces and
6 chains has been studied
 The values of monthly price indices with Jevons, Laspeyres and Lowe
aggregation formula have been computed for each available outlet of the
37 prov.
o Models for analysing the variance of the outlet monthly indices
o Study of distributions of the standard deviation of outlet monthly indices
Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
Jevons
Laspeyres
Lowe
Standard deviation of monthly indices by provinces, 2015
Variability of interest parameter in the population of outlets
 The three aggregation formulas provide the same results.
 There is a variability among provinces, but there is not a clear tendency we can use for improving the
allocation.
Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
The allocation of the sample of outlets from Nielsen universe
The Nielsen universe for 2017:
16 chains (out of 25) gave to Nielsen the authorization to provide data to
ISTAT, covering nearly 80% of total turnover (but not all outlets are
available)
 Stratification of stores in around 1000 strata, cross-classification of
 Province (107)
 chain (16)
 type (hyper /supermarket)
 No clear evidences could direct the definition of the allocation of the sample
 we chose for 2017 a compromise allocation:
 2100 sample outlets distributed among strata with a compromise
allocation: uniform, proportional to turnover, proportional to the
number of outlets
 In 2018 a slightly reduced sample of outlet was used for the index compilation
following the current fixed basket approach
5. The sample design for the selection of outlets from the Nielsen universe
Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
Allocation of outlets sample
is a compromise between:
 a uniform allocation
 proportional allocation to turnover
 proportional allocation to the
number of outlets
ROME
CHAIN 1 16
HYP 2
SUP 14
CHAIN 2 8
HYP 2
SUP 6
CHAIN 3 2
SUP 2
CHAIN 4 9
HYP 2
SUP 7
CHAIN 5 12
HYP 2
SUP 10
CHAIN 6 4
HYP 2
SUP 2
CHAIN 7 6
SUP 6
CHAIN 8 4
SUP 4
The aim of the first sample of outlets,
drawn for 2017 and 2018, is to gather
information on the variability of the
indices for allocating more efficiently
the sample of outlets for 2019.
Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
Conclusion and open issues
 The experiments on sampling designs showed the feasibility of the
procedures for the selection of items and the index elaboration
 Computational and process issues related to the implementation of static
and dynamic approach should be taken into account
 We are planning of using resampling methods to evaluate the variance of
the estimates
 One limit of the current implementation of dynamic approach is the use of
unweighted indices, whereas SD would allow using prices and quantities
 Further studies are needed to evaluate the properties of multilateral
indices, which could allow to overcome the chain drift and other issues of
weighted indices in the dynamic approach
 Since May 2018 we did not work on this project but we are planning to
follow some of the suggestions of the Committee to study an optimal
design for the selection of outlets for 2019 for the implementation of the
dynamic approach which is under experimentation at the survey sector
Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
Chessa A. G., Verburg J. and Willenborg L. A (2017). Comparison of Price Index Methods for Scanner Data.
de Haan, J., Opperdoes, E., Schut, C.M. (1999). Item selection in the Consumer Price Index: Cut-off versus probability
sampling. Survey Methodology, 25(1), 31-41.
de Haan, J. and van der Grient, H. A. (2011). Eliminating chain drift in price indexes based on scanner data. Journal of
Econometrics, 161, 36-46.
de Haan, J., Willemborg, L. and Chessa, A. G. (2016). An overview of price index methods for scanner data.
Feldmann B. (2015). Scanner-data-current-practice, http://www.istat.it/en/files/2015/09/5-WS-Scanner-data-Rome-1-2-
Oct_Feldmann-Scanner-data-current-pratice.pdf.
Gábor, E. and Vermeulen, P. New evidence in elementary index bias. (2014) ILO, IMF, OECD, Eurostat, United Nations,
World Bank Consumer Price Index Manual: Theory and Practice, Geneva: ILO Publications (2004)
Ivancic, L., Diewert, W.E., Fox, K.J. Scanner Data (2011). Time Aggregation and the Construction of Price Indexes.
Journal of Econometrics, 161(1), 24-35.
Nygaard, R. Chain drift in a monthly chained superlative price index. Workshop on scanner data, Geneva, 10 may
(2010)
Norberg A. (2014). Sampling of scanner data products offers in the Swedish CPI. Draft version 8 – Statistics Sweden.
van der Grient, H. A. and de Haan J. (2010). The Use of Supermarket Scanner Data in the Dutch CPI. Paper presented
at the Joint ECE/ILO Workshop on Scanner Data, Eurostat.
van der Grient, H. and J. de Haan (2011). Scanner Data Price Indexes: The “Dutch” Method versus Rolling Year GEKS”.
Paper presented at the twelfth meeting of the Ottawa Group, 4-6 May 2011, Wellington, New Zealand.
Vermeulen, B. C. and Herren, H. M. (2006). Rents in Switzerland: sampling and quality adjustment. 11th Meeting -
Ottawa Group - Neuchâtel 27-29 May (2006)
References
Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
Thank you for your attention !

More Related Content

Similar to Session I - Big Data De Vitiis, Sampling schemes using scanner data for the consumer price index

IJSCAI PAPER.pdf
IJSCAI PAPER.pdfIJSCAI PAPER.pdf
IJSCAI PAPER.pdfijscai
 
MODEL OF MULTIPLE ARTIFICIAL NEURAL NETWORKS ORIENTED ON SALES PREDICTION AND...
MODEL OF MULTIPLE ARTIFICIAL NEURAL NETWORKS ORIENTED ON SALES PREDICTION AND...MODEL OF MULTIPLE ARTIFICIAL NEURAL NETWORKS ORIENTED ON SALES PREDICTION AND...
MODEL OF MULTIPLE ARTIFICIAL NEURAL NETWORKS ORIENTED ON SALES PREDICTION AND...ijscai
 
Presentation on the state of play for a recommendation on processing scanner ...
Presentation on the state of play for a recommendation on processing scanner ...Presentation on the state of play for a recommendation on processing scanner ...
Presentation on the state of play for a recommendation on processing scanner ...Istituto nazionale di statistica
 
Detecting outliers at the end of the series using forecast intervals
Detecting outliers at the end of the series using forecast intervals Detecting outliers at the end of the series using forecast intervals
Detecting outliers at the end of the series using forecast intervals Dario Buono
 
SESSION IV Quality issues and generalized business processes - F.Rocci - R.V...
SESSION IV Quality issues and generalized business processes - F.Rocci  - R.V...SESSION IV Quality issues and generalized business processes - F.Rocci  - R.V...
SESSION IV Quality issues and generalized business processes - F.Rocci - R.V...Istituto nazionale di statistica
 
Real sale prices vs. displayed prices : an issue to consider before expanding...
Real sale prices vs. displayed prices : an issue to consider before expanding...Real sale prices vs. displayed prices : an issue to consider before expanding...
Real sale prices vs. displayed prices : an issue to consider before expanding...Istituto nazionale di statistica
 
HLEG thematic workshop on measuring economic, social and environmental resili...
HLEG thematic workshop on measuring economic, social and environmental resili...HLEG thematic workshop on measuring economic, social and environmental resili...
HLEG thematic workshop on measuring economic, social and environmental resili...StatsCommunications
 
Polish scanner data project - first steps - Anna Bobel, Tomasz Pietras
Polish scanner data project - first steps - Anna Bobel, Tomasz PietrasPolish scanner data project - first steps - Anna Bobel, Tomasz Pietras
Polish scanner data project - first steps - Anna Bobel, Tomasz PietrasIstituto nazionale di statistica
 
O1_Analysis KETs_DEF
O1_Analysis KETs_DEFO1_Analysis KETs_DEF
O1_Analysis KETs_DEFa b
 
Spain Diagnostic Imaging Market Outlook to 2018 - Ultrasound Systems, MRI Sys...
Spain Diagnostic Imaging Market Outlook to 2018 - Ultrasound Systems, MRI Sys...Spain Diagnostic Imaging Market Outlook to 2018 - Ultrasound Systems, MRI Sys...
Spain Diagnostic Imaging Market Outlook to 2018 - Ultrasound Systems, MRI Sys...ReportsnReports
 
Brochure format research english
Brochure format research englishBrochure format research english
Brochure format research englishformatresearch
 
GI2016 ppt charvat workshop geoss & conference inspire2016
GI2016 ppt charvat workshop geoss & conference inspire2016GI2016 ppt charvat workshop geoss & conference inspire2016
GI2016 ppt charvat workshop geoss & conference inspire2016IGN Vorstand
 
Advanced statistical methods for the analysis.pdf
Advanced statistical methods for the analysis.pdfAdvanced statistical methods for the analysis.pdf
Advanced statistical methods for the analysis.pdfcomsats786
 
The Swedish experience with scanner data from sampling to index calculation -...
The Swedish experience with scanner data from sampling to index calculation -...The Swedish experience with scanner data from sampling to index calculation -...
The Swedish experience with scanner data from sampling to index calculation -...Istituto nazionale di statistica
 
For a new model of distribution in italian ho reca market
For a new model of distribution in italian ho reca marketFor a new model of distribution in italian ho reca market
For a new model of distribution in italian ho reca marketfebo leondini
 
APPLICATION OF FACEBOOK'S PROPHET ALGORITHM FOR SUCCESSFUL SALES FORECASTING ...
APPLICATION OF FACEBOOK'S PROPHET ALGORITHM FOR SUCCESSFUL SALES FORECASTING ...APPLICATION OF FACEBOOK'S PROPHET ALGORITHM FOR SUCCESSFUL SALES FORECASTING ...
APPLICATION OF FACEBOOK'S PROPHET ALGORITHM FOR SUCCESSFUL SALES FORECASTING ...ijnlc
 
Application of Facebook's Prophet Algorithm for Successful Sales Forecasting ...
Application of Facebook's Prophet Algorithm for Successful Sales Forecasting ...Application of Facebook's Prophet Algorithm for Successful Sales Forecasting ...
Application of Facebook's Prophet Algorithm for Successful Sales Forecasting ...AIRCC Publishing Corporation
 

Similar to Session I - Big Data De Vitiis, Sampling schemes using scanner data for the consumer price index (20)

IJSCAI PAPER.pdf
IJSCAI PAPER.pdfIJSCAI PAPER.pdf
IJSCAI PAPER.pdf
 
MODEL OF MULTIPLE ARTIFICIAL NEURAL NETWORKS ORIENTED ON SALES PREDICTION AND...
MODEL OF MULTIPLE ARTIFICIAL NEURAL NETWORKS ORIENTED ON SALES PREDICTION AND...MODEL OF MULTIPLE ARTIFICIAL NEURAL NETWORKS ORIENTED ON SALES PREDICTION AND...
MODEL OF MULTIPLE ARTIFICIAL NEURAL NETWORKS ORIENTED ON SALES PREDICTION AND...
 
Presentation on the state of play for a recommendation on processing scanner ...
Presentation on the state of play for a recommendation on processing scanner ...Presentation on the state of play for a recommendation on processing scanner ...
Presentation on the state of play for a recommendation on processing scanner ...
 
Detecting outliers at the end of the series using forecast intervals
Detecting outliers at the end of the series using forecast intervals Detecting outliers at the end of the series using forecast intervals
Detecting outliers at the end of the series using forecast intervals
 
Market Studies and Competition - Amelia Fletcher - Professor of Competition P...
Market Studies and Competition - Amelia Fletcher - Professor of Competition P...Market Studies and Competition - Amelia Fletcher - Professor of Competition P...
Market Studies and Competition - Amelia Fletcher - Professor of Competition P...
 
SESSION IV Quality issues and generalized business processes - F.Rocci - R.V...
SESSION IV Quality issues and generalized business processes - F.Rocci  - R.V...SESSION IV Quality issues and generalized business processes - F.Rocci  - R.V...
SESSION IV Quality issues and generalized business processes - F.Rocci - R.V...
 
Real sale prices vs. displayed prices : an issue to consider before expanding...
Real sale prices vs. displayed prices : an issue to consider before expanding...Real sale prices vs. displayed prices : an issue to consider before expanding...
Real sale prices vs. displayed prices : an issue to consider before expanding...
 
HLEG thematic workshop on measuring economic, social and environmental resili...
HLEG thematic workshop on measuring economic, social and environmental resili...HLEG thematic workshop on measuring economic, social and environmental resili...
HLEG thematic workshop on measuring economic, social and environmental resili...
 
Polish scanner data project - first steps - Anna Bobel, Tomasz Pietras
Polish scanner data project - first steps - Anna Bobel, Tomasz PietrasPolish scanner data project - first steps - Anna Bobel, Tomasz Pietras
Polish scanner data project - first steps - Anna Bobel, Tomasz Pietras
 
O1_Analysis KETs_DEF
O1_Analysis KETs_DEFO1_Analysis KETs_DEF
O1_Analysis KETs_DEF
 
Spain Diagnostic Imaging Market Outlook to 2018 - Ultrasound Systems, MRI Sys...
Spain Diagnostic Imaging Market Outlook to 2018 - Ultrasound Systems, MRI Sys...Spain Diagnostic Imaging Market Outlook to 2018 - Ultrasound Systems, MRI Sys...
Spain Diagnostic Imaging Market Outlook to 2018 - Ultrasound Systems, MRI Sys...
 
Brochure format research english
Brochure format research englishBrochure format research english
Brochure format research english
 
GI2016 ppt charvat workshop geoss & conference inspire2016
GI2016 ppt charvat workshop geoss & conference inspire2016GI2016 ppt charvat workshop geoss & conference inspire2016
GI2016 ppt charvat workshop geoss & conference inspire2016
 
Advanced statistical methods for the analysis.pdf
Advanced statistical methods for the analysis.pdfAdvanced statistical methods for the analysis.pdf
Advanced statistical methods for the analysis.pdf
 
The Swedish experience with scanner data from sampling to index calculation -...
The Swedish experience with scanner data from sampling to index calculation -...The Swedish experience with scanner data from sampling to index calculation -...
The Swedish experience with scanner data from sampling to index calculation -...
 
For a new model of distribution in italian ho reca market
For a new model of distribution in italian ho reca marketFor a new model of distribution in italian ho reca market
For a new model of distribution in italian ho reca market
 
Methodologies to measure market competition – Romania – Feb 2021 OECD Workshop
Methodologies to measure market competition – Romania – Feb 2021 OECD WorkshopMethodologies to measure market competition – Romania – Feb 2021 OECD Workshop
Methodologies to measure market competition – Romania – Feb 2021 OECD Workshop
 
Final assignment
Final assignmentFinal assignment
Final assignment
 
APPLICATION OF FACEBOOK'S PROPHET ALGORITHM FOR SUCCESSFUL SALES FORECASTING ...
APPLICATION OF FACEBOOK'S PROPHET ALGORITHM FOR SUCCESSFUL SALES FORECASTING ...APPLICATION OF FACEBOOK'S PROPHET ALGORITHM FOR SUCCESSFUL SALES FORECASTING ...
APPLICATION OF FACEBOOK'S PROPHET ALGORITHM FOR SUCCESSFUL SALES FORECASTING ...
 
Application of Facebook's Prophet Algorithm for Successful Sales Forecasting ...
Application of Facebook's Prophet Algorithm for Successful Sales Forecasting ...Application of Facebook's Prophet Algorithm for Successful Sales Forecasting ...
Application of Facebook's Prophet Algorithm for Successful Sales Forecasting ...
 

More from Istituto nazionale di statistica

More from Istituto nazionale di statistica (20)

Censimenti Permanenti Istituzioni non profit
Censimenti Permanenti Istituzioni non profitCensimenti Permanenti Istituzioni non profit
Censimenti Permanenti Istituzioni non profit
 
Censimenti Permanenti Istituzioni non profit
Censimenti Permanenti Istituzioni non profitCensimenti Permanenti Istituzioni non profit
Censimenti Permanenti Istituzioni non profit
 
Censimenti Permanenti Istituzioni non profit
Censimenti Permanenti Istituzioni non profitCensimenti Permanenti Istituzioni non profit
Censimenti Permanenti Istituzioni non profit
 
Censimenti Permanenti Istituzioni non profit
Censimenti Permanenti Istituzioni non profitCensimenti Permanenti Istituzioni non profit
Censimenti Permanenti Istituzioni non profit
 
Censimenti Permanenti Istituzioni non profit
Censimenti Permanenti Istituzioni non profitCensimenti Permanenti Istituzioni non profit
Censimenti Permanenti Istituzioni non profit
 
Censimenti Permanenti Istituzioni non profit
Censimenti Permanenti Istituzioni non profitCensimenti Permanenti Istituzioni non profit
Censimenti Permanenti Istituzioni non profit
 
Censimento Permanente Istituzioni Pubbliche
Censimento Permanente Istituzioni PubblicheCensimento Permanente Istituzioni Pubbliche
Censimento Permanente Istituzioni Pubbliche
 
Censimento Permanente Istituzioni Pubbliche
Censimento Permanente Istituzioni PubblicheCensimento Permanente Istituzioni Pubbliche
Censimento Permanente Istituzioni Pubbliche
 
Censimento Permanente Istituzioni Pubbliche
Censimento Permanente Istituzioni PubblicheCensimento Permanente Istituzioni Pubbliche
Censimento Permanente Istituzioni Pubbliche
 
Censimento Permanente Istituzioni Pubbliche
Censimento Permanente Istituzioni PubblicheCensimento Permanente Istituzioni Pubbliche
Censimento Permanente Istituzioni Pubbliche
 
14a Conferenza Nazionale di Statisticacnstatistica14
14a Conferenza Nazionale di Statisticacnstatistica1414a Conferenza Nazionale di Statisticacnstatistica14
14a Conferenza Nazionale di Statisticacnstatistica14
 
14a Conferenza Nazionale di Statistica
14a Conferenza Nazionale di Statistica14a Conferenza Nazionale di Statistica
14a Conferenza Nazionale di Statistica
 
14a Conferenza Nazionale di Statistica
14a Conferenza Nazionale di Statistica14a Conferenza Nazionale di Statistica
14a Conferenza Nazionale di Statistica
 
14a Conferenza Nazionale di Statistica
14a Conferenza Nazionale di Statistica14a Conferenza Nazionale di Statistica
14a Conferenza Nazionale di Statistica
 
14a Conferenza Nazionale di Statistica
14a Conferenza Nazionale di Statistica14a Conferenza Nazionale di Statistica
14a Conferenza Nazionale di Statistica
 
14a Conferenza Nazionale di Statistica
14a Conferenza Nazionale di Statistica14a Conferenza Nazionale di Statistica
14a Conferenza Nazionale di Statistica
 
14a Conferenza Nazionale di Statistica
14a Conferenza Nazionale di Statistica14a Conferenza Nazionale di Statistica
14a Conferenza Nazionale di Statistica
 
14a Conferenza Nazionale di Statistica
14a Conferenza Nazionale di Statistica14a Conferenza Nazionale di Statistica
14a Conferenza Nazionale di Statistica
 
14a Conferenza Nazionale di Statistica
14a Conferenza Nazionale di Statistica14a Conferenza Nazionale di Statistica
14a Conferenza Nazionale di Statistica
 
14a Conferenza Nazionale di Statistica
14a Conferenza Nazionale di Statistica14a Conferenza Nazionale di Statistica
14a Conferenza Nazionale di Statistica
 

Recently uploaded

Holdier Curriculum Vitae (April 2024).pdf
Holdier Curriculum Vitae (April 2024).pdfHoldier Curriculum Vitae (April 2024).pdf
Holdier Curriculum Vitae (April 2024).pdfagholdier
 
Grant Readiness 101 TechSoup and Remy Consulting
Grant Readiness 101 TechSoup and Remy ConsultingGrant Readiness 101 TechSoup and Remy Consulting
Grant Readiness 101 TechSoup and Remy ConsultingTechSoup
 
Class 11th Physics NEET formula sheet pdf
Class 11th Physics NEET formula sheet pdfClass 11th Physics NEET formula sheet pdf
Class 11th Physics NEET formula sheet pdfAyushMahapatra5
 
social pharmacy d-pharm 1st year by Pragati K. Mahajan
social pharmacy d-pharm 1st year by Pragati K. Mahajansocial pharmacy d-pharm 1st year by Pragati K. Mahajan
social pharmacy d-pharm 1st year by Pragati K. Mahajanpragatimahajan3
 
Nutritional Needs Presentation - HLTH 104
Nutritional Needs Presentation - HLTH 104Nutritional Needs Presentation - HLTH 104
Nutritional Needs Presentation - HLTH 104misteraugie
 
IGNOU MSCCFT and PGDCFT Exam Question Pattern: MCFT003 Counselling and Family...
IGNOU MSCCFT and PGDCFT Exam Question Pattern: MCFT003 Counselling and Family...IGNOU MSCCFT and PGDCFT Exam Question Pattern: MCFT003 Counselling and Family...
IGNOU MSCCFT and PGDCFT Exam Question Pattern: MCFT003 Counselling and Family...PsychoTech Services
 
APM Welcome, APM North West Network Conference, Synergies Across Sectors
APM Welcome, APM North West Network Conference, Synergies Across SectorsAPM Welcome, APM North West Network Conference, Synergies Across Sectors
APM Welcome, APM North West Network Conference, Synergies Across SectorsAssociation for Project Management
 
Sanyam Choudhary Chemistry practical.pdf
Sanyam Choudhary Chemistry practical.pdfSanyam Choudhary Chemistry practical.pdf
Sanyam Choudhary Chemistry practical.pdfsanyamsingh5019
 
Paris 2024 Olympic Geographies - an activity
Paris 2024 Olympic Geographies - an activityParis 2024 Olympic Geographies - an activity
Paris 2024 Olympic Geographies - an activityGeoBlogs
 
Presentation by Andreas Schleicher Tackling the School Absenteeism Crisis 30 ...
Presentation by Andreas Schleicher Tackling the School Absenteeism Crisis 30 ...Presentation by Andreas Schleicher Tackling the School Absenteeism Crisis 30 ...
Presentation by Andreas Schleicher Tackling the School Absenteeism Crisis 30 ...EduSkills OECD
 
Interactive Powerpoint_How to Master effective communication
Interactive Powerpoint_How to Master effective communicationInteractive Powerpoint_How to Master effective communication
Interactive Powerpoint_How to Master effective communicationnomboosow
 
Measures of Central Tendency: Mean, Median and Mode
Measures of Central Tendency: Mean, Median and ModeMeasures of Central Tendency: Mean, Median and Mode
Measures of Central Tendency: Mean, Median and ModeThiyagu K
 
Unit-IV- Pharma. Marketing Channels.pptx
Unit-IV- Pharma. Marketing Channels.pptxUnit-IV- Pharma. Marketing Channels.pptx
Unit-IV- Pharma. Marketing Channels.pptxVishalSingh1417
 
Student login on Anyboli platform.helpin
Student login on Anyboli platform.helpinStudent login on Anyboli platform.helpin
Student login on Anyboli platform.helpinRaunakKeshri1
 
Introduction to Nonprofit Accounting: The Basics
Introduction to Nonprofit Accounting: The BasicsIntroduction to Nonprofit Accounting: The Basics
Introduction to Nonprofit Accounting: The BasicsTechSoup
 
1029-Danh muc Sach Giao Khoa khoi 6.pdf
1029-Danh muc Sach Giao Khoa khoi  6.pdf1029-Danh muc Sach Giao Khoa khoi  6.pdf
1029-Danh muc Sach Giao Khoa khoi 6.pdfQucHHunhnh
 
Kisan Call Centre - To harness potential of ICT in Agriculture by answer farm...
Kisan Call Centre - To harness potential of ICT in Agriculture by answer farm...Kisan Call Centre - To harness potential of ICT in Agriculture by answer farm...
Kisan Call Centre - To harness potential of ICT in Agriculture by answer farm...Krashi Coaching
 

Recently uploaded (20)

Holdier Curriculum Vitae (April 2024).pdf
Holdier Curriculum Vitae (April 2024).pdfHoldier Curriculum Vitae (April 2024).pdf
Holdier Curriculum Vitae (April 2024).pdf
 
Grant Readiness 101 TechSoup and Remy Consulting
Grant Readiness 101 TechSoup and Remy ConsultingGrant Readiness 101 TechSoup and Remy Consulting
Grant Readiness 101 TechSoup and Remy Consulting
 
Class 11th Physics NEET formula sheet pdf
Class 11th Physics NEET formula sheet pdfClass 11th Physics NEET formula sheet pdf
Class 11th Physics NEET formula sheet pdf
 
social pharmacy d-pharm 1st year by Pragati K. Mahajan
social pharmacy d-pharm 1st year by Pragati K. Mahajansocial pharmacy d-pharm 1st year by Pragati K. Mahajan
social pharmacy d-pharm 1st year by Pragati K. Mahajan
 
Mattingly "AI & Prompt Design: Structured Data, Assistants, & RAG"
Mattingly "AI & Prompt Design: Structured Data, Assistants, & RAG"Mattingly "AI & Prompt Design: Structured Data, Assistants, & RAG"
Mattingly "AI & Prompt Design: Structured Data, Assistants, & RAG"
 
Nutritional Needs Presentation - HLTH 104
Nutritional Needs Presentation - HLTH 104Nutritional Needs Presentation - HLTH 104
Nutritional Needs Presentation - HLTH 104
 
IGNOU MSCCFT and PGDCFT Exam Question Pattern: MCFT003 Counselling and Family...
IGNOU MSCCFT and PGDCFT Exam Question Pattern: MCFT003 Counselling and Family...IGNOU MSCCFT and PGDCFT Exam Question Pattern: MCFT003 Counselling and Family...
IGNOU MSCCFT and PGDCFT Exam Question Pattern: MCFT003 Counselling and Family...
 
APM Welcome, APM North West Network Conference, Synergies Across Sectors
APM Welcome, APM North West Network Conference, Synergies Across SectorsAPM Welcome, APM North West Network Conference, Synergies Across Sectors
APM Welcome, APM North West Network Conference, Synergies Across Sectors
 
Sanyam Choudhary Chemistry practical.pdf
Sanyam Choudhary Chemistry practical.pdfSanyam Choudhary Chemistry practical.pdf
Sanyam Choudhary Chemistry practical.pdf
 
Mattingly "AI & Prompt Design: The Basics of Prompt Design"
Mattingly "AI & Prompt Design: The Basics of Prompt Design"Mattingly "AI & Prompt Design: The Basics of Prompt Design"
Mattingly "AI & Prompt Design: The Basics of Prompt Design"
 
Paris 2024 Olympic Geographies - an activity
Paris 2024 Olympic Geographies - an activityParis 2024 Olympic Geographies - an activity
Paris 2024 Olympic Geographies - an activity
 
Advance Mobile Application Development class 07
Advance Mobile Application Development class 07Advance Mobile Application Development class 07
Advance Mobile Application Development class 07
 
Presentation by Andreas Schleicher Tackling the School Absenteeism Crisis 30 ...
Presentation by Andreas Schleicher Tackling the School Absenteeism Crisis 30 ...Presentation by Andreas Schleicher Tackling the School Absenteeism Crisis 30 ...
Presentation by Andreas Schleicher Tackling the School Absenteeism Crisis 30 ...
 
Interactive Powerpoint_How to Master effective communication
Interactive Powerpoint_How to Master effective communicationInteractive Powerpoint_How to Master effective communication
Interactive Powerpoint_How to Master effective communication
 
Measures of Central Tendency: Mean, Median and Mode
Measures of Central Tendency: Mean, Median and ModeMeasures of Central Tendency: Mean, Median and Mode
Measures of Central Tendency: Mean, Median and Mode
 
Unit-IV- Pharma. Marketing Channels.pptx
Unit-IV- Pharma. Marketing Channels.pptxUnit-IV- Pharma. Marketing Channels.pptx
Unit-IV- Pharma. Marketing Channels.pptx
 
Student login on Anyboli platform.helpin
Student login on Anyboli platform.helpinStudent login on Anyboli platform.helpin
Student login on Anyboli platform.helpin
 
Introduction to Nonprofit Accounting: The Basics
Introduction to Nonprofit Accounting: The BasicsIntroduction to Nonprofit Accounting: The Basics
Introduction to Nonprofit Accounting: The Basics
 
1029-Danh muc Sach Giao Khoa khoi 6.pdf
1029-Danh muc Sach Giao Khoa khoi  6.pdf1029-Danh muc Sach Giao Khoa khoi  6.pdf
1029-Danh muc Sach Giao Khoa khoi 6.pdf
 
Kisan Call Centre - To harness potential of ICT in Agriculture by answer farm...
Kisan Call Centre - To harness potential of ICT in Agriculture by answer farm...Kisan Call Centre - To harness potential of ICT in Agriculture by answer farm...
Kisan Call Centre - To harness potential of ICT in Agriculture by answer farm...
 

Session I - Big Data De Vitiis, Sampling schemes using scanner data for the consumer price index

  • 1. Sampling schemes using scanner data for the consumer price index De Vitiis C., Guandalini A., Inglese F., Terribili M. D. Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS Rome, 19th November 2018
  • 2. Outline 1. ISTAT and international context 2. The ISTAT sampling experiments of SD 3. Experiments of sampling from SD 4. Experiments of static and dynamic approaches 5. The sample design for the outlets 6. Conclusions and open issues Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
  • 3.  ISTAT started using the scanner data (SD) in 2018 to compute the consumer price index (CPI) for retail modern distribution (food and grocery)  SD are regularly provided since 2014 to ISTAT through the market research company ACNielsen, for the main chains in Italy and for a sample of about 2,000 outlets, scattered throughout the country  Nielsen provide to ISTAT the Dictionary which associates to items identifying attributes and ECR classification codes, allowing to link to COICOP classification  For 2018 the compilation of the CPI using scanner data is based on a fixed basket perspective, but a gradual transition to a flexible basket is the goal for ISTAT  At the moment, for the traditional retail distribution of food and grocery, data are collected by the current survey on field: traditional and scanner data are combined through weights derived from the expenditure survey  Aim of this presentation is showing and discussing the experiments carried out by our research group using SD in a sampling perspective, to compare sample schemes for different index calculation approaches 1. Context: the ISTAT project on Scanner Data Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
  • 4.  SD are used In several countries for the compilation of the CPI: Switzerland, Norway, Netherlands and Sweden for some years, while Belgium and Denmark only from 2016  Scanner data are exploited in different ways:  Replacing price collection in the stores, without changing the traditional principles of computing the price indices (Swiss Federal Statistical Office)  As a universe from which samples of references can be selected following different methods (Norway and Sweden)  Extensively to compile price indices, without a strict sample selection, but with consequences on the theoretical definition of the index (Nederland ) 1. Context: The international use of SD Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
  • 5. SCANNER DATA  Electronic sales data, containing both prices and quantities sold for every item purchased by consumers in modern distribution Opportunities  To introduce improvements in the CPI also with regards the sampling perspective, considering a dynamic population approach and different index formulas for the lowest aggregation level (elementary indices), allowing to take into account the dynamicity of the market  To offer a real possibility of computing more accurate indices, since SD allow, potentially, to include in indices the expenditure share of each item Issues to be addressed  Attrition of products, temporary missing products, entry of new products, relaunches and volatility of the prices and quantities due mainly to sales. These aspects need to be addressed from both a theoretical and a practical point of view (de Haan et al., 2016) 1. ISTAT and international context Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
  • 6. OBJECTIVE  The general aim was evaluating the use of SD for the compilation of elementary price indices (lowest aggregation level of price index calculus, on which the subsequent aggregations are based, in the Italian case the consumption segments)  from a sampling perspective  considering both static and dynamic population  following a simulative approach The experiments were developed in two phases:  In the first phase, following a fixed basket approach, probability and nonprobability selection schemes of series (references individuated by EAN and outlet codes) are compared  In the second phase, an experiment of evaluating the differences between a fixed and a flexible basket approach was carried out, trying to measure the magnitude of sampling error and bias for different index formulas in both approaches 2. The ISTAT sampling experiments of SD Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
  • 7. The DATA used for the experimental phase  Data referred to the first Italian provinces for which ISTAT got data, food and grocery sector (covering 11% of the total index, while modern distribution covers the 55% of the sector)  Weekly data on turnover and quantities per EAN-code (European Article Number, or GTIN, Global Trade Item Number) and outlet allow obtaining weekly unit values  One series (=EAN by outlet) is considered present in a specific month if it has associated a non zero value in at least one week out of the first full three weeks  Our experiments are based on a panel of PERMANENT SERIES  The permanent series are defined as having non-zero turnover for at least one relevant week (one of the first three full weeks) every month for 13 months  Permanent series provide generally a good coverage of the total turnover  Each experiment focused on one province, some consumption segments (Italian COICOP6) or markets, permanent series 2. The ISTAT sampling experiments of SD Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
  • 8. 2. The ISTAT sampling experiments of SD: the theoretical context Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018 FRAMEWORK of the experiments  Two selection stages for each domain (the whole province, not only the chief towns as the current survey on field) o Outlets selected through a probability sampling from the archive o Items (references) selected either by probability sampling or cut-off selection  For the item the definition of the population and parameter of interest is crucial : o Static population  fixed approach – fixed sample/ direct index (weighted or unweigthed) o The indices at the lowest aggregation level are obtained from the monthly price ratios (micro-indices) with fixed base (December previous year) o Dynamic population  dynamic approach – dynamic sample/ chained index (weighted or unweigthed) o The elementary indices are calculated on matched-items: only price relatives of items that are sold in two consecutive months enter in the index formulas (flexible basket)
  • 9. The bilateral indices in the Static and Dynamic approach  Static population  fixed basket/ direct index  The loss of representativeness of the basket (shrinkage effect) is addressed through the yearly base change of the index and the renewal of the basket  The fixed basket approach do not allow to exploit all the information provided by scanner data, do not take into account the population dynamics, new products are ignored  Dynamic population  matched model/ chained index  Chain indices take into account the movements of prices within the considered time interval and the shrinkage effect over time due to the attrition of a fixed basket of products is solved  Period-on-period chaining of weighted indices introduces chain drift - bias due to the non-transitivity of the chained weighted price indices, nonsymmetric effect on quantities sold and expenditures share before and after sale (de Haan and van der Grient, 2011; Ivancic et al., 2011) 2. The ISTAT sampling experiments of SD: the theoretical context Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
  • 10. Aggregation formulas for the elementary indices The price indices considered are:  Fisher (superlative, weighted) index is preferred by the economic theory, it uses quantities in different times and allows for substitution effects  Törnqvist (superlative, weighted) index is a weighted geometric mean, weights are the average of expenditure share calculated in two times  Lowe index is a modified Laspeyres in which a base month for the price (December of previous year) and a base year (previous year) for the quantities are taken into account for the weight computation  Jevons index unweighted geometric mean  In the experiments unbiased estimators of these indices were defined through the sampling weights 2. The ISTAT sampling experiments of SD: the theoretical context Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
  • 11. First experiments : Comparison of probability and nonprobability sampling selection schemes  Non-probability sample: Cut-off selection of series based on thresholds of covered turnover; the index is compiled using all series covering 60% or 80% of all turnover in each outlet for the consumption segment, in previous year 2013; Reference method: most sold items in each outlet for representative products (current fixed basket approach)  Probability sampling: two-stage sampling design with stratification of the first stage units  Stratification of outlets by 6 chains (Conad, Coop, Esselunga, Auchan, Carrefour, Selex) and outlet type (hypermarket and supermarket)  The size of the sample of outlet has been fixed at a number of 30 out of 121 outlets available for Turin, selected with equal probabilities  Sample size for the second stage (EANs) is given by a 5% sampling rate and selection by Sampford pps  The universe is the set of panel series 3. Experiments of sampling from SD Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
  • 12.  Three different sampling designs have been considered:  Stratified one-stage sampling  Cluster sampling (outlet)  Two-stage sampling (outlet and EAN)  The universe of EANs is stratified by market (ECR groups) in six consumption segments (coffee, pasta, mineral water, olive oil, spumante and ice cream)  Different criteria of sample allocation (proportional and Neyman) for both outlets and elementary items (EANs) and different methods for the selection of sampling units have been considered (SRS and PPS) Further experiments: Comparison among probability sampling schemes 3. Experiments of sampling from SD Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018 The comparison between probability and nonprobability selection schemes was made for each price index taking the corresponding universe index value (panel series SD) as reference true value
  • 13. Figure 1 – First experiments: Jevons, Lowe and Fisher indices computed on samples of series selected with different selection schemes - Coffee and pasta segments, Turin province, year 2014 COFFEE PASTA Jevons Lowe Fisher Lowe Fisher Results of the experiments in the first phase Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
  • 14. Absolute Relative Bias and Relative Sampling Error distributions of monthly Lowe, Fisher and Jevons indices for consumption segment (Sampford sampling) Results of the experiments in the first phase Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018 Index Coffee consumption segment Min Q1 Me Q3 Max Lowe ARB -0.0005 -0,0014 -0,0013 -0,0008 -0,0017 RSE 1,32 1,38 1,41 1,43 1,52 Fisher ARB 0,0005 0,0015 0,0021 0,0030 0,0039 RSE 1,14 1,36 1,53 1,73 1,98 Jevons ARB -0,0199 -0,0485 -0,0400 -0,0353 -0,0609 RSE 1,05 1,09 1,14 1,20 1,24 Index Pasta consumption segment Min Q1 Me Q3 Max Lowe ARB 0.0008 -0,0003 -0,0001 0,0005 -0,0007 RSE 1,13 1,20 1,28 1,32 1,38 Fisher ARB -0,0009 -0,0043 -0,0034 -0,0025 -0,0051 RSE 1,21 1,61 1,67 1,83 2,06 Jevons ARB -0,0017 0,0031 0,0105 0,0209 0,0314 RSE 0,85 0,92 1,04 1,10 1,22
  • 15. The results highlight the following evidences about the performances of different series selection schemes  Probability pps sample produces approximately unbiased estimates for indices using weights (Lowe and Fisher)  Increasing the sampling rate produce a remarkable improvement of the bias in all indices, in addition to an obvious reduction of sampling error  Stratified One stage sampling produces estimators more efficient, but for practical problems we need to pass through outlets.  Cluster sampling produces estimators having good performances, but the high number of products sold in each outlets rapidly makes increase the size of SSU.  Two stage sampling produces estimators with quite good performances, probably because of the most of variability is among outlets.  The performance depends in general on type of turnover distribution of items for the product segments I simulation study: summary of the evidences Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
  • 16.  The second experimental phase is an attempt to make a comparison between static and dynamic approach, evaluating the total error of the estimates of indices  Simulation of alternative population scenarios  artificial population: starting from the panel, new products have been introduced considering the monthly birth rates and old products have been removed in accordance to a survival function (both monthly birth and survival rates have been estimated on the real open population)  From a sampling perspective the two approaches refer to different sampling designs:  STATIC two-stage sampling (outlets, EANs)  DYNAMIC cluster sampling (outlets) 4. Experiments of Static and Dynamic approaches (second phase) Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
  • 17. (A) Panel (Static approach) All the products enter in the computation of indices (B) Static approach without disappearing products The products presented in December of the base month are followed during all the year. The new products are not considered by definition. However, disappearing products are assumed to be present (C) “Pure” Static approach Only the fixed basket approach is applied. The new products are not considered Scenarios implemented under Static approach Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
  • 18. (A) Panel (dynamic approach) All the products enter in the computation of indices. (D) Dynamic approach Only the matched products in two months in a row enter in the computation of price indices. Scenarios implemented under Dynamic approach Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
  • 19. STATIC vs DYNAMIC – Jevons with and without Treshold Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
  • 20. Comments on the experiments on static and dynamic approaches  The simulation allowed to experiment the two approaches and to highlight the practical and theoretical implications  Non-sampling error seems to affect the estimators of price indices under the dynamic approach more than the under the static approach  The sampling error of the estimator under the dynamic approach is much lower than in the static approach due to the difference in sample size Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
  • 21. 5. The sample design for the selection of outlets from the Nielsen universe  For 2017 a random sample of outlets has been selected from the Nielsen universe of hyper and supermarkets covering all Italian provinces and 16 chains, to start experimenting the index production  To allocate efficiently the sample of 2100 outlets, the variability in the universe of the prices indices referred to the initial set of 37 provinces and 6 chains has been studied  The values of monthly price indices with Jevons, Laspeyres and Lowe aggregation formula have been computed for each available outlet of the 37 prov. o Models for analysing the variance of the outlet monthly indices o Study of distributions of the standard deviation of outlet monthly indices Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
  • 22. Jevons Laspeyres Lowe Standard deviation of monthly indices by provinces, 2015 Variability of interest parameter in the population of outlets  The three aggregation formulas provide the same results.  There is a variability among provinces, but there is not a clear tendency we can use for improving the allocation. Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
  • 23. The allocation of the sample of outlets from Nielsen universe The Nielsen universe for 2017: 16 chains (out of 25) gave to Nielsen the authorization to provide data to ISTAT, covering nearly 80% of total turnover (but not all outlets are available)  Stratification of stores in around 1000 strata, cross-classification of  Province (107)  chain (16)  type (hyper /supermarket)  No clear evidences could direct the definition of the allocation of the sample  we chose for 2017 a compromise allocation:  2100 sample outlets distributed among strata with a compromise allocation: uniform, proportional to turnover, proportional to the number of outlets  In 2018 a slightly reduced sample of outlet was used for the index compilation following the current fixed basket approach 5. The sample design for the selection of outlets from the Nielsen universe Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
  • 24. Allocation of outlets sample is a compromise between:  a uniform allocation  proportional allocation to turnover  proportional allocation to the number of outlets ROME CHAIN 1 16 HYP 2 SUP 14 CHAIN 2 8 HYP 2 SUP 6 CHAIN 3 2 SUP 2 CHAIN 4 9 HYP 2 SUP 7 CHAIN 5 12 HYP 2 SUP 10 CHAIN 6 4 HYP 2 SUP 2 CHAIN 7 6 SUP 6 CHAIN 8 4 SUP 4 The aim of the first sample of outlets, drawn for 2017 and 2018, is to gather information on the variability of the indices for allocating more efficiently the sample of outlets for 2019. Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
  • 25. Conclusion and open issues  The experiments on sampling designs showed the feasibility of the procedures for the selection of items and the index elaboration  Computational and process issues related to the implementation of static and dynamic approach should be taken into account  We are planning of using resampling methods to evaluate the variance of the estimates  One limit of the current implementation of dynamic approach is the use of unweighted indices, whereas SD would allow using prices and quantities  Further studies are needed to evaluate the properties of multilateral indices, which could allow to overcome the chain drift and other issues of weighted indices in the dynamic approach  Since May 2018 we did not work on this project but we are planning to follow some of the suggestions of the Committee to study an optimal design for the selection of outlets for 2019 for the implementation of the dynamic approach which is under experimentation at the survey sector Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
  • 26. Chessa A. G., Verburg J. and Willenborg L. A (2017). Comparison of Price Index Methods for Scanner Data. de Haan, J., Opperdoes, E., Schut, C.M. (1999). Item selection in the Consumer Price Index: Cut-off versus probability sampling. Survey Methodology, 25(1), 31-41. de Haan, J. and van der Grient, H. A. (2011). Eliminating chain drift in price indexes based on scanner data. Journal of Econometrics, 161, 36-46. de Haan, J., Willemborg, L. and Chessa, A. G. (2016). An overview of price index methods for scanner data. Feldmann B. (2015). Scanner-data-current-practice, http://www.istat.it/en/files/2015/09/5-WS-Scanner-data-Rome-1-2- Oct_Feldmann-Scanner-data-current-pratice.pdf. Gábor, E. and Vermeulen, P. New evidence in elementary index bias. (2014) ILO, IMF, OECD, Eurostat, United Nations, World Bank Consumer Price Index Manual: Theory and Practice, Geneva: ILO Publications (2004) Ivancic, L., Diewert, W.E., Fox, K.J. Scanner Data (2011). Time Aggregation and the Construction of Price Indexes. Journal of Econometrics, 161(1), 24-35. Nygaard, R. Chain drift in a monthly chained superlative price index. Workshop on scanner data, Geneva, 10 may (2010) Norberg A. (2014). Sampling of scanner data products offers in the Swedish CPI. Draft version 8 – Statistics Sweden. van der Grient, H. A. and de Haan J. (2010). The Use of Supermarket Scanner Data in the Dutch CPI. Paper presented at the Joint ECE/ILO Workshop on Scanner Data, Eurostat. van der Grient, H. and J. de Haan (2011). Scanner Data Price Indexes: The “Dutch” Method versus Rolling Year GEKS”. Paper presented at the twelfth meeting of the Ottawa Group, 4-6 May 2011, Wellington, New Zealand. Vermeulen, B. C. and Herren, H. M. (2006). Rents in Switzerland: sampling and quality adjustment. 11th Meeting - Ottawa Group - Neuchâtel 27-29 May (2006) References Workshop of the ADVISORY COMMITTEE ON STATISTICAL METHODS, Rome, 19th November 2018
  • 27. Thank you for your attention !