This document discusses applying daily inflation forecasts to predict Brazil's Broad Consumer Price Index (IPCA). It motivates the importance of forecasting CPI for monetary policy. Currently an "Inflation Monitor" provides daily estimates but has limitations in predicting the first 6 days of collection periods. The authors propose using SARIMA models on daily data to improve initial forecasts. Their results found SARIMA models produced better predictions than the Inflation Monitor for each of the first 6 days. Future work could involve forecasting index components individually and aggregating them or estimating SARIMA models directly on daily data.

1. May, 2015
Applying the daily inflation to
forecast the Broad Consumer
Price Index (IPCA)
Pedro Costa Ferreira
Juliana Carneiro
André Braz

2. 2
Agenda
o Motivation
o Inflation Monitor
Ardeo, Quadros and Picchetti (2013) – A daily frequency inflation measure and its
information content on forecasts
o Monitor “limitation”
o Proposed Idea
o Results
o Conclusions and Future studies
Center for Statistical and Computational Methods

3. 3
Motivation
o The importance of Consumer Price Index
o It is used by the Brazilian Central Bank as the guideline for achieving its inflation
targets policy;
o In Brazil, owing to the hyperinflation episode in the 1980s, there have been many
studies done on this issue.
o Inflation Monitor and first six days forecasts issue
o The importance of CPI predictability and the SARIMA models
Center for Statistical and Computational Methods

4. 4
Inflation Monitor
o What is the Inflation Monitor?
o It’s a daily inflation estimate of the Broad Consumer Price Index (IPCA),
the official inflation index in Brazil.
o How does it work?
o The inflation rates estimations are calculated for moving periods of 30
days based on the preceding 30 days.
Center for Statistical and Computational Methods
Apr, 2015 May, 2015 June, 2015
IPCA
release
070203 0304
o What is the obvious problem?

5. 5
Proposed Idea
o We have used the one step ahead forecast by employing SARIMA models;
o But, we didn’t make daily forecasts. We have built 6 month TS with the daily
information
Center for Statistical and Computational Methods
3 4 5 6 7 8 9 ... 31 1 2 ... 30 1 2 ... 31 1 2 ... 30 1 2 ... 31 1 2 ... 30 ...
april may jun jul aug sep
 Daily
Inflation
0
apr/05
may05
jun/05
nov/13
dec/13
apr/05
may/05
jun/05
nov/13
dec/13
Second days of
collection period
First days of
collection period

6. 6
Results
o SARIMA model had best results for every six days.
Center for Statistical and Computational Methods
Days of
collection
period
SARIMA Model
1 (0,0,2)(0,1,1)12
2 (0,0,1)(0,1,2)12
3 (0,0,2)(0,1,1)12
4 (0,0,2)(0,1,1)12
5 (0,0,2)(0,1,1)12
6 (0,0,2)(0,1,1)12
Days of
collection
period
Monitor SARIMA
1 0,048 0,043
2 0,044 0,031
3 0,041 0,037
4 0,044 0,038
5 0,042 0,039
6 0,039 0,038
Adjusted models Mean Absolute Error (MAE)

7. 7
Conclusion and future studies
o We achieved the core purpose of this work, that is: we reduced the forecast
error of daily IPCA done by the Inflation Monitor
o We keep working on this research problem;
o Better results can be achieved by:
o forecasting the components of the index themselves and, afterwards, by
aggregating them in an index as a whole
o Estimating the SARIMA using daily data
Center for Statistical and Computational Methods

8. www.fgv.br/ibre
ありがとう
Thank you!!
Obrigado
Pedro Costa Ferreira
pedro.guilherme@fgv.br