2021 Price setting in chile: Micro evidence froM consuMer on-line Prices during the social outbreak and covid-19
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Price setting in Chile: Micro
evidence from consumer on-line
2021
prices during the social
outbreak and Covid-19
Documentos de Trabajo
N.º 2112
Jennifer Peña and Elvira PradesPrice setting in Chile: Micro evidence from consumer on-line prices during the social outbreak and Covid-19
Price setting in Chile: Micro evidence from consumer on-line prices during the social outbreak and Covid-19 (*) Jennifer Peña (**) Central Bank of Chile Elvira Prades (***) Banco de España (*) We are very grateful for useful comments to internal seminar participants, both in BoS and the CBCh, and to participants of the Conference on “Non-traditional data and statistical learning with applications to macroeconomics” organized by the Bank of Italy and the Federal Reserve Board and to Isabel Sánchez García. The views expressed herein are those of the authors and do not necessarily reflect the views of Banco de España nor the Central Bank of Chile. (**) Central Bank of Chile (CBCh). E-mail: jpena@bcentral.cl. (***) Bank of Spain (BoS). This paper was written while visiting the Central Bank of Chile under the agreement signed with Banco de España. The hospitality of the Central Bank of Chile is gratefully acknowledged. E-mail: elvira.prades@bde.es. Documentos de Trabajo. N.º 2112 2021
The Working Paper Series seeks to disseminate original research in economics and finance. All papers have been anonymously refereed. By publishing these papers, the Banco de España aims to contribute to economic analysis and, in particular, to knowledge of the Spanish economy and its international environment. The opinions and analyses in the Working Paper Series are the responsibility of the authors and, therefore, do not necessarily coincide with those of the Banco de España or the Eurosystem. The Banco de España disseminates its main reports and most of its publications via the Internet at the following website: http://www.bde.es. Reproduction for educational and non-commercial purposes is permitted provided that the source is acknowledged. © BANCO DE ESPAÑA, Madrid, 2021 ISSN: 1579-8666 (on line)
Abstract In this paper we analyze the price setting behavior in Chile by using scraped data from public websites of the main retailers including supermarkets, a pharmacy retailer and car dealerships. The data collection started in July 2019 and the dataset covers two major recent events: (1) the social outbreak and (2) the state of emergency declaration due to Covid-19, both episodes led to disruptions in the economy. With information on product varieties that accounts for 22 % of the CPI basket, we document several empirical findings as regards price setting behaviour in terms of stickiness, that is, frequency, implied duration and the size of price adjustments. We find that in spite of facing large shocks, prices adjusted very little, at a lower frequency and at a smaller size than prior to these two events. We also find that there was a reduction on product variety availability on-line, a typical feature that also has been found during natural disasters such as earthquakes. The reduction in product availability poses additional difficulties to construct CPI indexes and to properly capture price rigidities, which are relevant for monetary policy. Keywords: on-line price data, CPI, prices stickiness, retail distribution. JEL classification: E01,E31,L81.
Resumen En este trabajo analizamos la fijación de precios en Chile utilizando técnicas de webscraping, donde recolectamos los precios de productos de consumo a través de las páginas web de los principales establecimientos de comercio minorista del país. La base de datos incluye precios de productos de supermercados, farmacias y comercializadoras de automóviles. La recolección de datos se inició en julio de 2019 y contiene información de precios mientras ocurrían dos hechos relevantes: 1) el estallido social, y 2) la declaración del estado de emergencia por el COVID-19. Ambos episodios conllevaron disrupciones en el funcionamiento normal de la actividad. Con información de productos que representan aproximadamente el 22 % de la canasta del índice de precios de consumo (IPC), documentamos el comportamiento de los precios en términos de frecuencia y duración y el tamaño de estos ajustes, indicadores que informan sobre la flexibilidad o la rigidez del ajuste ante perturbaciones de esta magnitud y comparamos con respecto a tiempos normales. Encontramos que, durante estos episodios, la rigidez en el ajuste aumentó, ya que los precios se ajustaron con menor frecuencia y las variaciones, en promedio, fueron de menor tamaño. También encontramos que se produjo una importante reducción en la variedad de productos disponibles durante la pandemia, un hecho que suele ocurrir durante eventos disruptivos, como los desastres naturales. Esta reducción en las variedades supone un reto adicional para la elaboración del IPC, así como una mayor complejidad para capturar las rigideces nominales, que son relevantes para la política monetaria. Palabras clave: precios on-line, índice de precios de consumo, rigideces de precios, distribución minorista. Códigos JEL: E01,E31,L81.
1 Introduction
The ability of monetary policy to influence in the short-run real economic activity it
much depends on nominal rigidities. This paper intends to contribute to understand
from a micro perspective the behavior of price setting in Chile during the 2019
social outbreak and the Covid-19 pandemic.1 The price setting behavior is even
more relevant in a context with heightened uncertainty as regards how inflation will
1 Introduction
behave over time.2
TheToability of monetary
this end policyuse
we will make to influence in the
of a dataset short-run
collected overreal economic
internet sitesactivity
of mainit
much depends on nominal rigidities. This paper intends to contribute to understand
retailers in Chile. The use of Big Data by policy makers and researchers became more
from a micro
popular perspective
and accepted the behavior
in recent years. Theof main
price source
settingisinInternet
Chile during the Since
data sets. 2019
1
social outbreak
online shopping and the Covid-19
substitutes, pandemic.
or at least The price
supplements, offlinesetting behavior
shopping, onlineisprices
even
more relevant in a context with heightened uncertainty as regards how inflation
can also be used as a substitute, or supplement, of offline prices. One concern may will
2
behave
lie overdifferences
on the time. between on-line and off-line prices, the discrepancy between
these two channels is expected to be small as in Chile third-party delivery services
are To this used
widely end weandwill make
have use ofpenetration
a higher a dataset collected
comparedover internet
to other sites of
countries. 3 main
retailers in Chile. The use of Big Data by policy makers and researchers became more
popular
Onlineanddata
accepted in recent
have many years. The
advantages in main source is
the context of Internet data sets.
the Covid-19 Since
Pandemic
online shopping
where there are substitutes, or at least
lockdowns, mobility supplements,
restrictions and offline shopping, online
social-distancing rules. prices
Data
can also be used as a substitute, or supplement, of offline prices. One concern
collection over the internet is called web scraping. This technique provides flexibility may
lie
andon the differences
extreme automation.between
Thereon-line and off-line
is abundant prices,
anecdotal the discrepancy
evidence that scraped between
prices
these two channels is expected to be small as in Chile third-party delivery
is a potentially useful tool for nowcasting and short-term forecasting of CPI inflation services
3
are widely
as well usedsales
as retail and variables.
have a higher
Onlinepenetration compared
CPI provides to other with
policy-makers countries.
a reasonably
good “pulse” as to the direction being taken by inflation in real time and at a higher
Online data
frequency. have many
The Billion Prices advantages
Project (BPP)in theatcontextMIT isofa the solidCovid-19
evidencePandemic
that has
where there are lockdowns,
been functioning for more than a decade. mobility restrictions
4 and social-distancing rules. Data
collection over the internet is called web scraping. This technique provides flexibility
andThe extreme automation.
objective Thereis istoabundant
of this paper anecdotal
provide evidence onevidence
the behavior that of
scraped prices
prices/price
is a potentially
setting in Chile useful
based tool fordataset
on the nowcasting and short-term
collected by the “Online forecasting
Prices of CPI inflation
Project” by the
as well as retail sales variables. Online CPI provides policy-makers
Central Bank of Chile. The dataset covers two episodes of particular interest that led with a reasonably
good “pulse” asintothe
to disruptions theeconomy,
directionparticularly
being takeninbyterms inflation
of thein degree
real time of and
priceatrigidities,
a higher
frequency.
by using online Theprices.
BillionWe Prices
wantProject
to answer(BPP) at MIT
questions suchis as:
a solid
How evidence that has
nominal rigidities
4
been functioning for more than a decade.
have changed during the social outbreak and the pandemia episode in Chile? Do we
observe product/sectoral differences? What is the role of seasonal/sales discounts in
observe
1 product/sectoral differences? What is the role of seasonal/sales discounts in
a context iswhere
Thereobjective
The a of
a renewed substantial
interest thisshare
this paper front of there
is to as products
provide are haveinitiatives
recent
evidence beenthedropped
to studyoutof from
pricing on-line
behaviour
afrom
context
a micro where a substantial
perspective such as share
the of products
Price-setting have on
Microdata been
Analysis
behavior
dropped out
Network
prices/price
from
(PRISMA) on-line
and
retailers?
setting in Chile based on the dataset collected by the “Online Prices Project” by the
retailers?
the Inflation Persistence Network (IPN) by the European Central Bank. PRISMA is a new ESCB
Central Bank ofestablished
research network Chile. The in dataset
2018 withcovers two episodes
a mandate from the of particular
Governing interest
Council, thisthat led
network
to disruptions
will In June
study in the economy,
2019 thebehaviour
price-setting particularly
“OnlineofPrices Project”
individual in terms
firms and of the
wasininitiated degree
the retail at of price
theusing
sector Centralrigidities,
microBank
price
In June 2019 the “Online Prices Project” was initiated at the Central Bank
by using
datasets.
of Chile. online
The prices.
purpose Wewaswanttotofollow-up
answer questions
price such as: Howfocused
developments nominalon rigidities
goods
of 2Chile.
See The (2020)
Blanchard purpose was to follow-up price developments focused on goods
https://voxeu.org/article/there-deflation-or-inflation-our-future.
have
typically
3
changed duringatthe
available social outbreak
supermarkets andweb
using the scraping
pandemiatechniques;
episode in Chile? Do we
this provides
typically availableonatthesupermarkets
For a discussion representativenessusing web scraping
of on-line techniques;
versus off-line this provides
data and implications see
flash
Cavallo
1 estimates
(2017). of CPI available onasathere
daily basis, that is, to with higher frequency
flash estimates
There of
is a renewed CPI available
interest this fronton a daily are basis, that is,
recent initiatives with higher
study pricingfrequency
behaviour
and 4 more timely that the one provided by the national statistical office.
fromFor more details
amore
micro see http://www.thebillionpricesproject.com/.
perspective This
and timely thatsuch
theasonethe provided
Price-settingbyMicrodata Analysis
the national Network (PRISMA)
statistical office. Thisand
article describes
the Inflation the Network
Persistence behavior(IPN)of online prices inCentral
by the European Chile, Bank.
taking a deeper
PRISMA lookESCB
is a new into
article
research
describes the behavior
network established
ofwith
online prices from in Chile,Governing
taking aCouncil,
deeper look into
the micro-data that we in 2018collected
have a mandate
on a dailythebasis
3on between Julythis2019 network
till
the
will micro-data
study that
price-setting we have
behaviour ofcollected
individual firms a daily
and in basis
the between
retail sector July
using 2019
micro till
price
November 2020, from the two of the largest retailers of fast moving consumer goods
November
datasets. 2020, from the two of the largest retailers of fast moving consumer goods
(groceries,
2 personal
See Blanchard
care
(2020)care
and household equipment), a health retailer and auto
https://voxeu.org/article/there-deflation-or-inflation-our-future.
(groceries, personal and household equipment), a health retailer and auto
dealers
3 in Chile.
For aindiscussion We provide
on provide evidence on the
the representativeness key features on price-setting behaviour.
dealers Chile. We evidence on of the on-line versus off-line
key features data and implications
on price-setting behaviour. see
Cavallo (2017).
4
For more details see http://www.thebillionpricesproject.com/.
The dataset collected includes products corresponding to expenditure categories
The dataset collected includes products corresponding to expenditure categories
that cover around 22% of the official Chilean CPI. In spite the short-time span, due
that cover around 22% of the official Chilean
3 CPI. In spite the short-time span, due
to7the recent start of this project, our sample
BANCO DE ESPAÑA DOCUMENTO DE TRABAJO N.º 2112
covers two major events that took
to the recent start of this project, our sample covers two major events that took
place in the Chilean economy. Along this period two major shocks occurred which
place in the Chilean economy. Along this period two major shocks occurred which
are unambiguous examples of exogenous and unanticipated aggregate shocks, which
are unambiguous examples of exogenous and unanticipated aggregate shocks, which
makes them very interesting to analyze how prices are being set and to anticipateIn
In June
June 2019
2019 the
the “Online
“Online Prices
Prices Project”
Project” waswas initiated
initiated atat the
the Central
Central Bank
Bank
of Chile. The purpose was to follow-up price developments
of Chile. The purpose was to follow-up price developments focused on goods focused on goods
typically
typically available
available at
at supermarkets
supermarkets using
using web
web scraping
scraping techniques;
techniques; this
this provides
provides
flash estimates of CPI available on a daily basis, that is, with higher
flash estimates of CPI available on a daily basis, that is, with higher frequency frequency
and
and more
more timely
timely that
that the
the one
one provided
provided byby the
the national
national statistical
statistical office.
office. This
This
article
article describes the behavior of online prices in Chile, taking a deeper look
describes the behavior of online prices in Chile, taking a deeper look into
into
the
the micro-data that we have collected on a daily basis between July 2019 till
micro-data that we have collected on a daily basis between July 2019 till
November
November 2020,
2020, from
from the
the two
two of
of the
the largest
largest retailers
retailers of
of fast
fast moving
moving consumer
consumer goods
goods
(groceries,
(groceries, personal
personal care
care and
and household
household equipment),
equipment), aa health
health retailer
retailer and
and auto
auto
dealers in Chile. We provide evidence on the key features on price-setting
dealers in Chile. We provide evidence on the key features on price-setting behaviour. behaviour.
The
The dataset
dataset collected
collected includes
includes products
products corresponding
corresponding to to expenditure
expenditure categories
categories
that
that cover around 22% of the official Chilean CPI. In spite the short-time
cover around 22% of the official Chilean CPI. In spite the short-time span,
span, due
due
to the recent start of this project, our sample covers two major
to the recent start of this project, our sample covers two major events that tookevents that took
place
place in
in the
the Chilean
Chilean economy.
economy. Along
Along this
this period
period two
two major
major shocks
shocks occurred
occurred which
which
are
are unambiguous examples of exogenous and unanticipated aggregate shocks,
unambiguous examples of exogenous and unanticipated aggregate shocks, which
which
makes them very interesting to analyze how prices are being set
makes them very interesting to analyze how prices are being set and to anticipateand to anticipate
what
what could
could bebe the
the implications
implications for for CPI
CPI developments.
developments. Firstly,
Firstly, the
the 18
18 of
of October
October
(18-O)
(18-O) the
the social
social outbreak
outbreak started,
started, during
during aa week
week aa severe
severe curfew
curfew was
was set
set and
and from
from
this date on regular demonstrations took place with different intensities.
this date on regular demonstrations took place with different intensities. All these All these
actions
actions led
led to
to some
some disruptions
disruptions in in economic
economic activity,
activity, mainly
mainly in in the
the retail
retail sector.
sector.
Strikes
Strikes and demonstrations dramatically reallocate store traffic and opening hours
and demonstrations dramatically reallocate store traffic and opening hours
which led to an increase of on-line sales and delivery
which led to an increase of on-line sales and delivery services. services.
Then,
Then, aa second
second shock
shock took
took place,
place, this
this time
time ofof aa global
global nature,
nature, the
the Covid-19
Covid-19
pandemia.
pandemia. The first Covid-19 case was registered in Chile in early March
The first Covid-19 case was registered in Chile in early March andand
by mid-March, in order curb the spread of the virus and to avoid
by mid-March, in order curb the spread of the virus and to avoid the collapse the collapse
of
of the
the health
health system,
system, thethe government
government introduced
introduced some some measures
measures in in order
order to
to
maintain
maintain social
social distance.
distance. To To this
this end,
end, the
the government
government announced
announced dynamic
dynamic andand
selective
selective quarantines,
quarantines, curfews,
curfews, the
the closure
closure of
of restaurants,
restaurants, malls,
malls, schools,
schools, universities
universities
and,
and, when possible, stay at home work was encouraged. Few activities
when possible, stay at home work was encouraged. Few activities such
such as
as
supermarkets
supermarkets and pharmacies where allowed to function normally. And, again, it
and pharmacies where allowed to function normally. And, again, it
was registered a heightened demand for online shopping. Both episodes
was registered a heightened demand for online shopping. Both episodes implied the implied the
closure
closure of
of some
some retail
retail stores
stores and
and an
an increased
increased demand
demand forfor delivery
delivery services.
services. The
The
pandemia
pandemia also put some pressures from the supply side. Under this episode we can
also put some pressures from the supply side. Under this episode we can
also
also expect
expect some
some supply
supply disruptions
disruptions asas some
some manufacturing
manufacturing activities
activities require
require high
high
social
social interaction.
interaction.
We
We end
end up
up with
with aa dataset
dataset following
following prices
prices onon aa daily
daily basis
basis for
for approximately
approximately
47
47 product categories out of the 305 included by the National Statistical
product categories out of the 305 included by the National Statistical Office,
Office,
and 4
and start following around 4, 755 product-varieties that account for 22% of the CPI
start following around 4, 755 product-varieties
4 that account for 22% of the CPI
Chilean basket.
Chilean basket. The
The main
main stylized
stylized facts
facts on
on the
the price
price setting
setting was
was calculated
calculated out
out
for
for four differentiated periods: (1) “pre-shocks”, that goes from June 2019 to 18
four differentiated periods: (1) “pre-shocks”, that goes from June 2019 to 18
of October, (2) “social outbreak” from the end of October till mid-March
of October, (2) “social outbreak” from the end of October till mid-March and (3) and (3)
“pandemia”
“pandemia” fromfrom 18
18 March
March and
and (4)
(4) from
from the
the partial
partial uplift
uplift of
of mobility
mobility restrictions
restrictions
from July till November 2020 (502 days at this point). We start
from July till November 2020 (502 days at this point). We start by constructing by constructing
common statistics
common statistics of
of price
price setting
setting based
based onon averages
averages across
across products
products and
and time.
time.
These include measures focused on three dimensions: frequency
These include measures focused on three dimensions: frequency of price changes, of price changes,
implied
implied duration
duration and
and the
the magnitude
magnitude of of price
price changes.
changes.
The
The main
main conclusions
conclusions are
are that
that both
both product
product availability
availability and
and the
the price
price setting
setting
behavior changed during Covid-19 and the social outbreak episodes. In
behavior changed during Covid-19 and the social outbreak episodes. In particular, particular,
the frequency
the frequency of
of price
price changes
changes fell
fell during
during both
both episodes,
episodes, suggesting
suggesting that
that firms
firms
were delaying their price adjustments. In sum, after the two shocks prices
were delaying their price adjustments. In sum, after the two shocks prices became became
moderately
BANCO DE ESPAÑA 8
moderately more sticky.
more sticky. InIn terms
DOCUMENTO DE TRABAJO N.º 2112 terms ofof availability
availability we
we found
found that
that there
there was
was aa
significant reduction
significant reduction in
in online
online availability
availability of
of products
products during
during pandemia
pandemia andand that
that
they are recovering at a slow pace. This reduction occurred after March
they are recovering at a slow pace. This reduction occurred after March 18, the 18, the
date in
date in which
which the
the state
state of
of alarm
alarm was
was declared
declared and
and confinement
confinement measures
measures where
whereThe main conclusions are that both product availability and the price setting
behavior changed during Covid-19 and the social outbreak episodes. In particular,
the frequency of price changes fell during both episodes, suggesting that firms
were delaying their price adjustments. In sum, after the two shocks prices became
moderately more sticky. In terms of availability we found that there was a
significant reduction in online availability of products during pandemia and that
they are recovering at a slow pace. This reduction occurred after March 18, the
date in which the state of alarm was declared and confinement measures where
put in place in Chile due to Covid-19. The goods that disappear quickly from the
stores are mostly perishable and some nonperishable items that people are likely to
Weinend
hoard fearupof with
futurea supply
datasetdisruptions,
following prices
that isontoasay,
daily basis on
a focus foressential
approximately
goods.
47 productduring
However, categories out ofoutbreak
the social the 305 theincluded by the
reduction in National Statisticalwas
online availability Office,
not
and start following
significant. around 4, 755 product-varieties that account for 22% of the CPI
Chilean basket. The main stylized facts on the price setting was calculated out
for Along
four differentiated
these episodes periods:
the CLP (1)also
“pre-shocks”,
registered athat goes fromagainst
depreciation June 2019 to 18
the USD.
of October, (2) “social outbreak” from the end of October
Therefore, if passed through, we may expect imported goods to become more till mid-March and (3)
“pandemia”
expensive. To from 18 March
explore this,and
we (4)
havefrom the partial
identified the uplift
brand offormobility restrictions
each product and
from July till November 2020 (502 days at this point). We start
therefore we can infer the country of origin. This allows us to explore additional by constructing
common statistics
features such as theofrole
price setting based
of exchange on averages or
rate pass-through across
whetherproducts
there and
have time.
been
These includedisruptions.
supply-chain measures focused on three dimensions: frequency of price changes,
5
implied duration and the magnitude of price changes.
This paper is organized as follows. After this introduction, in section 1, we
reviewThe themain conclusions
literature are that both
on microdata product
on prices andavailability
webscraping andinthe price setting
section 2. In
behavior
section 3changed we describeduringthe Covid-19 and the social
data collection method outbreak
and theepisodes. In particular,
characteristics of the
the
scrapedfrequency
data. of In price
section changes
4 we fell reportduring
the both
main episodes,
results onsuggesting that firms
product availability,
we conclude
were
frequencydelayingand with
sizethe
their of preliminary
price adjustments.
adjustments results Indrawn
comparing so4far.
sum,allafter the two Finally,
periods. shocks prices became
in section 5
moderately more sticky. In terms
we conclude with the preliminary results drawn so far. of availability we found that there was a
significant
5
We are awarereduction
that this in assumption
online availability of products
may be simplistic in a contextduringwhere pandemia and firms
some domestic that
they are recovering at a slow pace. This reduction occurred after March 18, the
might be using imported goods and at the same time we are identifying the country of origin of
final goods, that at the same time if involved in global value
date in which the state of alarm was declared and confinement measures where chains the true origin of the product
2 be Literature
can from other or from multiple Review origins.
put in place in Chile due to Covid-19. The goods that disappear quickly from the
2 Literature
stores are mostly perishable Review and some nonperishable items that people are likely to
Price setting analysis based on official micro data.– The existence of price
hoard in fear of future supply disruptions,5 that is to say, a focus on essential goods.
rigidities
Price setting is a determinant of the oneffectiveness of monetary policy. In general,
However, duringanalysis
the social based
outbreak official micro
the reduction data.–
in online The existence
availability of price
was not
the literature
rigidities is a refers
determinantto two of ways
the toeffectiveness
measure theof degree monetary of price
policy.flexibility:
In one
general,
significant.
constitutes
the literature a macro
refers to approach
two ways through modelling
to measure the asdegreein Gali and Gertler
of price flexibility:(2000),
one
and another
constitutes is
a macrothe micro
approachapproach.
through The latter is based
modellingaasdepreciation on
in Gali and four types
Gertler of data
Along these episodes the CLP also registered against the(2000),
USD.
sources:
and on surveys
another is the directed
micro at companies,
approach. The onlatter
microdata
is basedcompiled
on fourin each
types reporting
of data
Therefore, if passed through, we may expect imported goods to become more
establishment
sources: on surveys and for each
directed variety of article by the agency in charge of generating
expensive. To explore this,atwe companies,
have identifiedon microdata
the brand compiled
for each in each reporting
product and
the consumer price
establishment and indicator,
for each on scanner
variety of databyand
article themoreagency recently
in web scraped
charge of data.
generating
therefore we can infer the country of origin. This allows us to explore additional
This paper focuses
the consumer price on the micro on approach using scraped online data.
features such as theindicator,
role of exchange scanner ratedata and
pass-throughmoreor recently
whether web scraped
there have data.
been
This paper focuses
supply-chain disruptions. on the5 micro approach using scraped online data.
Three studies have analyzed the patterns of price setting of Chile, evidence from
1999-2005 has been
Three studies havereported
analyzedbythe Medina
patterns et of
al.price
(2007) usingof micro
setting Chile, data
evidenceprovided
from
by This
the paper
national is organized
statistical as follows.
office, and hasAfter
beenthis introduction,
recently updated in bysection
Canales 1,and
we
1999-2005 has been reported by Medina et al. (2007) using micro data provided
review
López the literature on microdata on prices and webscraping in section 2. In
by the (2021)
national with information
statistical office,upand to has
March 2020.
been recentlyBothupdated
studies by findCanales
substantial
and
section
heterogeneity 3 we describe theofdata collection method and divisions.
the characteristics ofetthe
López (2021) in withfrequency
information adjustment
up to March across product
2020. Both studies Chaumontfind substantial al.
scraped
(2011) data. In
provideinevidence section with4 we report
scanner data the main
2007-2009 results on
in divisions. product
the supermarket availability,
industry,
heterogeneity frequency of adjustment across product Chaumont et al.
frequency
coverage ofand
7.23%sizeofofthe adjustments
CPI basket. comparing all 4 periods. Finally, in section 5
(2011) provide evidence with scanner data 2007-2009 in the supermarket industry,
coverage
5 of 7.23% of the CPI basket.
We are aware that this assumption may be simplistic in a context where some domestic firms
Price
might be using setting
imported analysis
goods andbased
at the sameon webscraping techniques.–
time we are identifying the country This paper
of origin of
contributes
finalPrice to up-date
setting
goods, that the
at the analysis facts
same time ifbasedfor Chile
involvedon but with
webscraping
in global online or web
value chainstechniques.– scraped
the true origin ofThisdata. One
paper
the product
of
can the firstother
be from
contributes one to from
or
to up-date use multiple
online
the factsdatafor to
origins. construct
Chile but with a price
onlineindex or web in scraped
Chile was Cavallo
data. One
(2013)
of the firstand onedue to to use
its good
onlineperformance
data to construct and it ahas priceproven
indextoinbeChile a representative
was Cavallo
source
BANCO DE ESPAÑA 9
(2013) of information.
and due to its good
DOCUMENTO DE TRABAJO N.º 2112 Online data has5and
performance proved
it hasto proven
providetomany be a advantages
representative in
particular during the social
source of information. distancing
Online data has episode.
proved The practicemany
to provide of social distancing
advantages in
means staying at home and away from others as much
particular during the social distancing episode. The practice of social distancing as possible to help prevent of
Covid-19.
means staying Online data and
at home also away
has advantages
from othersinasthe muchcontext of natural
as possible to help disasters,
preventforof(2011) provide evidence with scanner data 2007-2009 in the supermarket industry,
coverage of 7.23% of the CPI basket.
Price setting analysis based on webscraping techniques.– This paper
contributes to up-date the facts for Chile but with online or web scraped data. One
of the first one to use online data to construct a price index in Chile was Cavallo
(2013) and due to its good performance and it has proven to be a representative
source of information. Online data has proved to provide many advantages in
particular during the social distancing episode. The practice of social distancing
means staying at home and away from others as much as possible to help prevent of
Covid-19. Online data also has advantages in the context of natural disasters, for
more details see Cavallo et al. (2014), where they analyze the price setting behavior
after the earthquake in Chile in 2010 and in Japan 2011. Notwithstanding, online
data have also limitations. Even when retail sales are carried online, their services
could stop operating during crisis, so there is no way to scrape data.
The applications in empirical literature of online or web scraped data are in
general three-folded and include: (1) inflation measurement, (2) inflation forecasting
(including nowcasting) and (3) the analysis of micro price setting mechanism. In
2008, the Billion Prices Project was created at MIT. It has remained the largest
project focused on web scraping and online prices analysis, for more details see
Cavallo
Cavallo and
and Rigobon
Rigobon (2016).
(2016). The The usage
usage ofof big
big data
data inin Central
Central Banks
Banks also
also is
is
becoming a common practice. In 6
becoming a common practice. In addition to the above mentioned PRISMA led
addition to the above mentioned PRISMA led
by
by ECB.
ECB. Some
Some central
central banks
banks have
have used
used big
big data
data tools
tools to
to track
track prices
prices online.
online.
Macias
Macias and Stelmasiak (2019), they evaluate the ability of web scraped data to
and Stelmasiak (2019), they evaluate the ability of web scraped data to
improve
improve nowcasts
nowcasts of of Polish
Polish food
food inflation.
inflation. They
They also
also explore
explore product
product selection
selection
and
and classification
classification problems,
problems, their
their importance
importance in in constructing
constructing web
web price
price indices
indices and
and
other
other limitations of online datasets. The Sveriges Riksbank, Hull et al.
limitations of online datasets. The Sveriges Riksbank, Hull et al. (2017),
(2017),
they
they developed
developed anan automatic
automatic internet
internet data
data collection
collection process
process to
to collect
collect sales
sales prices
prices
daily for selected fruits and vegetables from a number of Swedish online
daily for selected fruits and vegetables from a number of Swedish online retailers. retailers.
Their
Their results
results indicate
indicate that
that the
the information
information from
from the
the daily
daily data
data could
could increase
increase the
the
precision in short-term inflation forecasts in
precision in short-term inflation forecasts in Sweden.Sweden.
Once
Once wewe have
have mentioned
mentioned the the penetration
penetration that that web
web scraping
scraping hashas had
had in
in central
central
banking as a great ally of big data, we want to use its benefits to
banking as a great ally of big data, we want to use its benefits to provide evidence provide evidence
on price
on price setting
setting and
and inflation
inflation during
during Covid-19
Covid-19 pandemic
pandemic lockdowns.
lockdowns. A A lockdown
lockdown
implies that
implies that many
many of of typical
typical products
products thatthat people
people buybuy areare no
no longer
longer available.
available.
During this
During this period
period there
there are
are changes
changes bothboth onon the
the supply
supply andand on on the
the demand
demand
side
side that could be reflected in prices, generating either inflationary pressures or
that could be reflected in prices, generating either inflationary pressures or
deflation. Cavallo (2020) has found that consumers spend relatively
deflation. Cavallo (2020) has found that consumers spend relatively more on food more on food
and other
and other categories
categories with
with rising
rising inflation,
inflation, and
and relatively
relatively less
less on
on transportation
transportation and and
other categories
other categories experiencing
experiencing significant
significant deflation.
deflation. TheThe coronavirus
coronavirus pandemic
pandemic has has
radically changed
radically changed what
what households
households are are buying.
buying. People
People are
are spending
spending aa relatively
relatively
more
more money on food at grocery stores and less on transportation, healthcare and
money on food at grocery stores and less on transportation, healthcare and
clothes. Also, since overall spending is down, housing costs now
clothes. Also, since overall spending is down, housing costs now represent a larger represent a larger
share of
share of most
most people’s
people’s expenditures;
expenditures; see see Diewert
Diewert andand Fox
Fox (2020),
(2020), Carvalho
Carvalho et et al.
al.
(2020), Dunn
(2020), Dunn etet al.
al. (2020)
(2020) and
and Cavallo
Cavallo (2020)
(2020) for
for evidence
evidence ofof dramatically
dramatically changing
changing
consumer expenditure
consumer expenditure patterns
patterns arising
arising from
from the
the pandemic.
pandemic.
Similarly, by
Similarly, by using
using scanner
scanner data
data from
from UK
UK supermarkets
supermarkets covering
covering the
the Great
Great
Lockdown episode
Lockdown episode Jaravel
Jaravel and
and O’Connell
O’Connell (2020)
(2020) for
for fast-moving
fast-moving consumer goods
consumer goods
also report a sharp decline in product variety and a reduction in promotions.
also report a sharp decline in product variety and a reduction in promotions.
BANCO DE ESPAÑA 10 DOCUMENTO DE TRABAJO N.º 2112
3
3 Data
Data description
description3 Data description
The “Online Prices Project” was initiated by in June 2019 by the Central Bank of
Chile, with the objective of collecting public prices posted online by main retailers
in Chile. At the initial stages of the project the aim was to construct an index
based on fast-arriving online prices of food and alcoholic beverages and other fast
moving consumer goods usually sold in supermarkets. By December 2019, the
such asofclothing,
scope productsfootwear, medicines
was expanded andasstarted
well astovehicles.
scrap theIn prices
this paper we will
of other focus
product
on
suchgoods typically
as clothing, availablemedicines
footwear, in two supermarkets, a pharmacy
as well as vehicles. In this and
papertenwecar
willdealers
focus
where
on goodswe typically
have collected for more
available than
in two a 7year of daily
supermarkets, data.
a pharmacy and ten car dealers
where we have collected for more than a year of daily data.
Although online transactions are still a small fraction of all retail sales in most
countries,
Although we online
are confident that the
transactions aredata
still collected online also
a small fraction of allprovides information
retail sales in most
about offline
countries, we prices and product
are confident availability.
that the In Chile
data collected delivery
online servicesinformation
also provides are widely
used
aboutand moreover,
offline prices anddueproduct
to the availability.
health crisesInderived from Covid-19
Chile delivery services the switch
are widely
towards
used andon-line shopping
moreover, due towasthe
more widespread
health amongfrom
crises derived consumers
Covid-19 thantheusual in
switch
6
order
towardsto avoid
on-linesocial contact.
shopping was more widespread among consumers than usual in
order to avoid social contact.6
3.0.1 Products and retailers selection
3.0.1 Products and retailers selection
To select which products to track, we comply the criteria suggested by ?: (1) the
product
To selectshould
which be from atoretailer
products track, with a highthe
we comply market share
criteria and (2)bythis
suggested ?: retailer
(1) the
should exploit different channels to sell their products (multi-channel).
product should be from a retailer with a high market share and (2) this retailer We only
track official pages of retailers and not of intermediaries.
should exploit different channels to sell their products (multi-channel). We only
track official pages of retailers and not of intermediaries.
With this in mind, for grocery products we track prices from the main largest
retailers
With in thisChile.
in mind, These are theproducts
for grocery two main weplayers in thefrom
track prices retail
thefor
main food and
largest
supermarket suppliesThese
retailers in Chile. with operations
are the twoalong mainthe country.
players in theBoth account
retail for more
for food and
than 70% market share. For health products we obtain
supermarket supplies with operations along the country. Both account for more prices from the main
pharmaceutical
than 70% market retailer in Chile
share. for 25%products
For health of the market share in
we obtain the pharma
prices from the market.
main
For car dealers we have collected ten models of passenger cars
pharmaceutical retailer in Chile for 25% of the market share in the pharma market. and sport utility
vehicle
For car(suv).
dealers we have collected ten models of passenger cars and sport utility
vehicle (suv).
Another element to consider when selecting the goods to track is the definition
of variety
Anotherand product.
element The National
to consider when selectingStatistical Officeto(Instituto
the goods track is theNacional de
definition
Estadı́stica
of variety and- INE) defines The
product. a variety as “aStatistical
National good or service
Office that forms Nacional
(Instituto the basic de or
elemental
Estadı́sticaunit
- INE)of the IPC abasket
defines varietyand is defined
as “a good oraccording
service thatto aforms
set oftheattributes
basic or
or pre-established specifications, such as the brand, description,
elemental unit of the IPC basket and is defined according to a set of attributes size, content,
packaging and provenance,
or pre-established among other
specifications, such as specific characteristics”,
the brand, description,whilesize,
a product
content,is
the “generic term to refer to a good or service that it has a defined
packaging and provenance, among other specific characteristics”, while a product is purpose and for
which it has term
the “generic an expense
to referweight. In terms
to a good or serviceof the CPI,
that it isa the
it has representation
defined purpose and of for
an
elementary
which it hasaggregate”.
an expense Aweight.
product
In isterms
madeofup theofCPI,
varieties, as an
it is the example, we of
representation have
an
the Pisco product that has Pisco 35 grades or 40 grades; from 700
elementary aggregate”. A product is made up of varieties, as an example, we have to 1000 cc varieties.
the Pisco product that has Pisco 35 grades or 40 grades; from 700 to 1000 cc varieties.
6
For a discussion on the representativeness of on-line versus off-line data and implications see
Cavallo
6 (2017).
For a discussion on the representativeness of on-line versus off-line data and implications see
Cavallo (2017).
BANCO DE ESPAÑA 11 DOCUMENTO DE TRABAJO N.º 2112 8
83.0.2 Web scraping
We have collected information through web scraping techniques conducted in Python
using Selenium, Beautiful Soup as well as auxiliary libraries to fetch and process
data. We collect online prices every day and we perform data acquisition while
minimizing the burden on web stores owner. We build an automated procedure that
on a daily basis scans the retailers web and searches for the web code associated
with each product variety and collects the price, and other characteristics and stores
this information. As illustrated in figure 1 we proceed in three steps:
• First step. Each day, between 10:00 p.m and midnight, we download from
each selected url the information on the varieties and prices.
• Second step. The code finds the relevant information. An example of a
product-variety in our data is “Brand Name, Pisco, 1 liter”. The scraping
software automatically collects data on every single product on display on the
retailer’s website each day. We only extract goods are listed on the website
if they are in stock, are available for on-line sale and that are not labelled as
“agotado”, that is out of stock. Some out of stock items are labelled as such
or they immediately disappear from the website, and only reappear on the day
that they are offered for sale again (if ever). We can therefore build statistics
to study how the set of goods available for purchase changes over time.
• Third step. The code stores the following information for each product:
its price, the unit of measure (e.g. kilograms, millimeters,..), the product
description, whether it is under a promotion, the Stock Keeping Unit (SKU)
and the date. The price includes taxes and sale discounts, to account for the
final price paid by the consumer. We do not include shipping/delivery costs,
which may vary according to the location of the purchaser and are an exclusive
feature of the online purchasing experience. In addition, missing prices in the
data were often caused by scraping errors on days when the software fails to
automatically start.
In sum, the collection of data follows the best practices for statistical collection
delaying the access to time frames where little traffic is expected and we use Python
given the large amount of information we handle and its flexibility. In table 1 we
summarize the coverage by division compared with the official methodology.
3.0.3 Data cleaning and filters
When analyzing prices at high frequencies, it is observed that the actual prices
they often temporarily move away from a slow-moving trend line called the regular
price, but after a temporary price change, the nominal price often returns exactly
to its pre-existing level. These distinctive features imply that even though an
individual price series has a great deal of 9high-frequency price flexibility (the actual
price changes frequently), the series also has a great deal of low-frequency price
stickiness (the regular price changes infrequently).
Before we can address how to treat sales in constructing a measure of price
stickiness, we must first be able to accurately identify sales. Sales or promotions are
short-lived price changes that are reverted. Constructing a dependable algorithm
for12identifying sales has proven challenging for researchers. Researchers often turn
BANCO DE ESPAÑA DOCUMENTO DE TRABAJO N.º 2112
to other methods of detecting temporary price changes, such as asserting that sale
prices have a particular shape. A common identifier of sales is V-shaped temporary
discounts, similar to those employed by Nakamura and Steinsson (2008). We alsoBefore we can address how to treat sales in constructing a measure of price
stickiness, we must first be able to accurately identify sales. Sales or promotions are
short-lived price changes that are reverted. Constructing a dependable algorithm
for identifying sales has proven challenging for researchers. Researchers often turn
to other methods of detecting temporary price changes, such as asserting that sale
prices have a particular shape. A common identifier of sales is V-shaped temporary
discounts, similar to those employed by Nakamura and Steinsson (2008). We also
to its pre-existing level. These distinctive features imply that even though an
apply the running mode filter proposed by Kehoe and Midrigan (2015)’s algorithm
individual price series has a great deal of high-frequency price flexibility (the actual
categorizes each price shift as either temporary or regular, based on each price’s
price changes frequently), the series also has a great deal of low-frequency price
relative position to the mode price over a given window of time within a particular
stickiness (the regular price changes infrequently).
time series; notably, the algorithm differentiates between price increases and price
decreases.
Before we can address how to treat sales in constructing a measure of price
stickiness, we must first be able to accurately identify sales. Sales or promotions are
the To account
number
short-lived price
for missing information
of observations,
changes that forare
more we take
details
reverted. on the
theprice
Constructing
information
calculations see the
a dependable
from previous
appendix.
algorithm
periods. For the price index calculations, if a good disappears and later reappears
for identifying sales has proven challenging for researchers. Researchers often turn
in our sample,availability.–
Product we fill missing prices with the previously available values
along until a new
to other methods of detectingOne of the most
temporary
7 priceremarkable
changes, such features
as asserting thethat
sample
sale
price
is theisdecline
observed within 150
in on-line days.availability across division and groups, see figure
product
prices have a particular shape. A common identifier of sales is V-shaped temporary
2. In the sample
discounts, similaroftoour thosetwoemployed
supermarkets we observeand
by Nakamura a decline
Steinsson of 55.7%
(2008). on Weaverage
also
8
with respect to the average available before March
apply the running mode filter proposed by Kehoe and Midrigan (2015)’s algorithm 18. In spite of the different
nature of the
categorizes eachshock,price shiftwe can compare
as either this drop
temporary or with
regular, the based
evidence on eachon natural
price’s
4 Empirical
disasters provided by findings
Cavallo et al. (2014) for Chile and
relative position to the mode price over a given window of time within a particular Japan. They show that the
earthquake
time series; that
notably,tookthe place in Chiledifferentiates
algorithm in 2010 had between an immediate impact onand
price increases product
price
The data
availability.
decreases. sampleA was
large broken
share ofinto
goodsfour periods
went out labelled
of stock as follows:
within (1)
days. “pre-shocks”,
The fall was
that
gradualgoesbutfromlarger June 2019 (than
in Chile up toin18Japan),
of October,
where the (2) number
“social outbreak”
of productsfrom mid
available
October
fell To till
by account mid-March,
32% in for themissing
first two (3) “pandemia”
months after
information from
the the
we take 18 March
earthquake, till end of
recoveringfrom
price information June
slowly and (4)
after
previous
from
that. the
periods. “gradual
Supply
For shocks
the uplift”
price of mobility
originate
index in the restrictions
calculations, destruction
if a good inofdisappears
July till November
production and later2020.
capacity and The
reappears the
dataset
disruption
in includes
our sample, weprice
of supply information
fill chains.
missing pricesofwith
Demand food,thedrinks,
shocks are toiletries,
linked
previously to thecleaning
available panic that
values products
consumers
until and
a new
household
may equipment.
experience after
price is observed within 150 days.natural disasters,
7 with people rushing to the stores to purchase
basic necessities and hoarding goods that they fear could become unavailable in the
daysWe to end
come. up9 with three price series for each product-variety, that is, (1) the raw
dataset, (2) the price filtered following Nakamura and Steinsson (2008) (NS) and
(3) The
the price
lowestfiltered
availabilityseries of using Kehoe
on-line and Midrigan
products (2015) (KM).
was registered during Our Aprilcollected
and by
4
data Empirical
allows us to observefindingsthat one of the retailers
the end of the month it began to recover, but without reaching the levels shown under study follows a more
dynamic pricing strategy.
at the beginning of March. We identify retailersinasavailability
The recovery Ri , i = 1, 2,occurred
3 and 4. in To food
calculate
and
The data sample was broken into four periods labelled as follows: (1) “pre-shocks”,
the different
beverages, measures
while technology of frequency
and white of line
priceproducts
changescontinued
each product at loweris weighted
levels, with by
that goes from June 2019 up to 18 of October, (2) “social outbreak” from mid
respect
the7 During
numberto those registered atforthe
ofpandemia
observations, morebeginning of
on data
detailsgoods the collection. This
calculations seeoftheperiod of less
appendix.
October till
themid-March, (3) “pandemia”
a substantial amount offrom 18wereMarch till end
not being offered June
or 10 and
taking (4)
longer
availability coincides with the months of greatest imputation by the INE.
from the “gradual uplift” of mobility restrictions in July till November 2020. The
to reappear, this parameter has been revised at each update.
Product
dataset includes availability.–
price information One ofofthe most
food, remarkable
drinks, toiletries,features
cleaning along the sample
products and
is We
the also
decline have
household equipment. in information
on-line product as regards
availability
10the brands
across supplied
division and in each
groups, supermarket
see figure
as
2. well
In the assample
the number of ouroftwo varieties suppliedwebyobserve
supermarkets each brand.a decline Weofobserve
55.7% on that there
average
is aWe
with reduction
end uptoin
respect thethree
the
with number
average price ofseries
varieties
available before
for offered
each March by 18.each In
8
product-variety, supplier
spite
thatof but
is,the
(1)wedifferent
do raw
the not
observe such
nature of(2)the
dataset, a decline
theshock, in the number
we can following
price filtered of suppliers.
compare Nakamura
this drop and This points
withSteinsson to
the evidence a decline
(2008)on(NS) in the
natural
and
varieties
disasters
(3) possibly
provided
the price with
filtered the aim
by series
Cavallo etofal.Kehoe
using focusing
(2014)and their
for production
Chile
Midrigan and(2015)
Japan.on (KM).
their
TheybasicOur products,
show that the
collected
see figure
earthquake 3.
data allows us to observe that one of the retailers under study follows product
that took place in Chile in 2010 had an immediate impact on a more
availability. A large share of goods went out of stock
dynamic pricing strategy. We identify retailers as Ri , i = 1, 2, 3 and 4. To calculate within days. The fall was
Analytical
gradual but larger classification
in Chile (than of
in goods.–
Japan), In
where
the different measures of frequency of price changes each product is weighted by order
the to
number capture
of differentiated
products available
behaviour
fell by 32%we in classify
the first product-varieties
two months afteraccording the earthquake, to theirrecovering
nature: slowly emergency, after
7
whether
During it
the is perishable,
pandemia a nonperishable
substantial amount of or
goodsdurable.
were
that. Supply shocks originate in the destruction of production capacity and the not A
being first
offered group
or taking include
longer
to reappear, of
disruption thissupply
parameter has been
chains. revisedshocks
Demand at eachareupdate.
linked to the panic that consumers
8
This reduction is computed by comparing the average number of items available before March
may experience after natural disasters, with people rushing to the stores to purchase
18, the date in which the state of alarm was declared and confinement measures where introduced
basic necessities and hoarding goods that10they fear could become unavailable in the
in Chile.
9 to come.9
days During the first days after the state of alarm and given the increased demand in fear of shortages
or13theDOCUMENTO
BANCO DE ESPAÑA
derived uncertainty some retailers focused on essential products, in a way of simplifying the
DE TRABAJO N.º 2112
delivery process.
The lowest availability of on-line products was registered during April and by
10
During the pandemia the National Statistical Office faced several problems to collect the
the end of the month it began to recover, but without reaching the levels shown
information of prices and used imputation methods.
at the beginning of March. The recovery in availability occurred in food andYou can also read
