Page 136 - Invited Paper Session (IPS) - Volume 2

IPS188 Bruno Tissot
Central banks’ experiences have underlined the benefits of having an
encompassing framework for making the most of all the data available
to enhance their regular monitoring of economic activity in general and of
price developments in particular. These approaches draw on big data analytics
to effectively summarise the information contained in large data sets through
a small number of common factors and to capture time-varying effects so as
to incorporate continuously the new information that becomes available (Bok
and al (2017)). But, interestingly, they primarily rely on “traditional” survey
statistics, while the use of web-based indicators has remained limited,
suggesting that they may work less well in nowcasting/forecasting exercises
compared to “traditional” statistics and confidence surveys. In the United
States, for instance, the price nowcasting exercises conducted at the Federal
Reserve Bank of Cleveland (Knotek and Zaman (2014)) mainly rely on past
observations of inflation (eg monthly developments in CPI, core CPI, CPI for
food, etc) complemented by higher-frequency information on oil prices (eg
retail gasoline prices and daily crude prices); this reflects the fact that core
inflation tends to be relatively stable in the short-run and that most of the
volatility observed in headline inflation is driven by changes in food and
energy prices. The parallel approach set up at the ECB relies on similar
mechanisms when trying to predict short-term development in inflation
(Modugno (2011): the high-frequency data retained for nowcasting inflation
are mainly related to the daily prices of raw materials including energy. Such
mixed approaches – eg using big data analytics to digest the large amount of
incoming information while still having as input relatively “traditional”
statistics – are reported to perform relatively well compared to, for instance,
consensus forecasts. From a different perspective, the GDP nowcasting
exercises conducted by the Federal Reserve Bank of New York have
highlighted the information content of past inflation when estimating
economic activity in advance.
Whether such approaches will allow for the integration of more “big
data-type” information when measuring and predicting inflation remains
to be seen. Certainly, there are important challenges to be considered, related
in particular to: the methodological choices to be retained; the need to clean
the vast of amount of the new information available, not least to deal with
duplicates, outliers and other quality issues; the difficulty to measure real
transaction prices and avoid capturing obsolete information (since internet-
based prices can remain on the web for a long time); and the issues posed by
matching different datasets at a quite granular level. Looking forward, key is
to clarify/document the type of information available and allow researchers
and policy analysers to test it in a transparent way.

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