Page 133 - Invited Paper Session (IPS) - Volume 2

IPS188 Bruno Tissot
assessing turning points. Indeed, an important objective of the Billion Prices
Project is not just to provide advance information on inflation, but also to
estimate higherfrequency (daily) inflation indicators for a large number of
countries, including advanced economies.
Third, the high granularity of big data sets can support to measure of
various dimensions of inflation, allowing for a better understanding of the
distribution of prices aggregates – eg across markets and/or locations. As
regards market differentiation, web-scrapped prices can be useful for
measuring inflation in very volatile sectors, such as fresh food prices, allowing
for a better measure of some specific components of the CPI. As regards
locations, large and granular data sets collected from commercial
advertisements can help to capture local patterns with sufficient precision, say
for instance to measure rents or property prices depending on zip codes or
even street names.
Fourth, various big data sources such as numbers on internet search
queries (eg Google Trends) and “soft” indicators computed from digitalised
textual information (eg displayed by social media posts like Twitter) can
provide interesting insights on economic agents’ sentiment and
expectations (Wibisono and Zulen (2019)). This can be particularly useful for
nowcasting exercises and short-term forecasts (say, for the next quarters), as
well as to assess the risks surrounding them. Traditional statistical surveys can
also provide this kind of information, but they typically focus on specific items,
eg firms’ production expectations and consumer sentiment. In contrast,
internet-based sources allow a wider range of indicators to be used. In
addition, they can be less intrusive than face-to-face statistical surveys, and
may therefore better reflect true behaviours and expectations (Rigobon
(2018)).Yet, almost by construction, big data-based forecasts can mainly be
used for short-term forecasts, since they depend on the flow of incoming
data. In practice, longer-term inflation forecasts (say beyond the current and
next year) have continued to remain model-based.
Fifth, new big data sources appear of increasing interest for measuring
the wide range of asset prices that are not easily covered by traditional
surveys because of the lack of statistics available and/or methodological
clarity. Cases in point relate to residential and commercial property markets,
which are often lacking reliable statistics, while alternative sources can be
easily found using big data (eg advertisements from property websites and
newspapers). In addition, these markets are characterised by a low and
infrequent number of transactions (compared to stocks) and by significant
heterogeneity across tangible assets, making the compilation of quality-
adjusted house price indices difficult. These challenges can be overcome by
capturing the various characteristics of the properties displayed in webbased
advertisements and the application of hedonic methods. Moreover, the

120 | I S I W S C 2 0 1 9