Page 120 - Invited Paper Session (IPS) - Volume 2

IPS 188 G. P. Samanta
resources to track inflation behaviour. Accordingly, this paper examines
empirically if the data on internet search queries using Google engine can be
gainfully employed to predict inflation for India. The plan of the paper is as
follows. Section 2 provides a literature review on predicting inflation based on
Google search index and the consumer-theory based approach adopted by
Guzmán (2011). Section 3 presents data and empirical results, and Section 4
concludes.

2. Methodology
As this paper focuses on assessing the information content of Google
search index for inflation rate, the dataset consists of two main components:
First, the official statistics on price index which forms the basis of estimating
inflation. Second, the Google search index for suitable keywords. Monthly data
on these series are collected for a period of seven years from April 2012 to
March 2019.
2.1 Data on Price Index
In India, inflation is now measured by annual percentage changes in
Consumer Price Index (CPI) compiled by National Statistics Office (NSO),
Ministry of Statistics and Programme Implementation (MoSPI), Government of
India. Monthly data are released under three broad heads, viz., CPI-Urban (CPI-
U), CPIRural (CPI-R) and CPI-Combined (CPI-C). While CPI-R and CPI-U
represent price index for rural and urban India, respectively, the CPI-C is overall
price index arrived at by combining CPI-R and CPI-U.
2.2 Google Search Indicators/Indices
Each Google trend series is characterised by two important features
(Guzmán, 2011; Seabold and Coppola 2015): First, the numbers at various time
points over the data period do not provide the absolute search volumes on
the given keywords. Instead, they represent relative estimates in a sense that
the time point with maximum search interest over the entire enquiry period is
assigned a value 100 and the actual search volumes in other time points are
rescaled accordingly. Second, the time series replica of search index on given
key words for a specified period depends on the date when the search enquiry
was made. Thus, time series data may change with search date even when the
reference period for search index remain unchanged. These typical issues with
Google search index have been handled by a two-step process. We first gather
replicas of time series data on Google search Index on the keyword ‘inflation'
(with location: India) for the period April 2012 to March 2019 on three different
dates of May 2019. As expected, the replica of relative measures on three
different dates appears numerically different. We constructed an overall
replica of time series by taking geometric-mean of the replicas obtained in
three different dates and denote these geometric-mean based overall indices
for the keywords ‘inflation' and ‘price' as GMInfl and GMPrice, respectively.

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