The Euro Area Flash Estimation Procedure/media/Kontorer... · Pedro Martins Ferreira and Luca...

The Euro Area Flash Estimation ProcedurePedro Martins Ferreira and Luca Gramaglia

Eurostat, European [email protected]

Problem statement• Eurostat has been publishing the euro area Harmonised Index of Consumer Prices (HICP) flash estimate (all-items) every month since October 2001 but its relevance has been somewhat limited to the users

as only the HICP all-items was released and no flash indication on its main components were made available;

• Inflation main components, i.e., ’food’, ’non-energy industrial goods’ (igoodsxe), ’energy’ and ’services’ (serv) are more difficult to forecast. In addition, the outcome of the nowcasting procedure has to be internally consistent, i.e., the aggregation of the nowcasts for the main components has to equal the nowcast of the all-items;

• A completely new nowcasting system had to be developed. The new system combines early information from the euro area Member States with 1-step ahead forecasts and timely price data on specific energy products into a consistent set of nowcasts.

ConclusionFrom October 2012 onwards Eurostat produced timely and accurate flash

estimates. The maximum recorded error was 0.2 percentage points for energy, the most volatile component, and no error for the headline inflation

(all-items) was recorded so far.

Automatic model selection1. For each parameter model combination, perform a pseudo-

out-of-sample forecast over the last 24 months (n = 24). Compute and record ME and MSE;

2. Sort the results by MSE in ascending order;

3. Test if the ME of the model of the ordered list is significantly different from zero using the t-test. If the ME is significantly different from zero, discard that model and test the next model from the sorted list. Repeat while t-test is rejected;

4. Perform the pseudo-out-of-sample for the last 24 months using both the reference model and the candidate model and save the two vectors of forecasting errors;

5. Compare the forecast errors vectors using the Diebold-Mariano statistic. If the MSE of the forecast errors of the candidate model is significantly lower than that of the reference model, replace the reference model by the candidate model.

CalibrationThe calibration is made by a simple proportional allocation. Nevertheless, a new calibration procedure is being developed, which will be applied both to preliminary data and nowcasts.

� =

IAll items

t

aggregation

⇣IMain component

t

⌘

Figure 3: Calibration factor density plot. The calibration factor φ is the scalar that needs to be multiplied with the non-calibrated main components nowcasts to make them aggregate to the all-items nowcast. The closer φ is to 1, the less the main components nowcasts need to be changed to meet the consistency constraint.

Flash Estimate procedure

Results

Oct-12 Nov-12 Dec-12 Jan-13 Feb-13 Mar-13

All-itemsFE 2.51 2.25 2.23 2.02 1.84 1.75

All-items HICP 2.49 2.19 2.22 1.98 1.84 1.74All-items

Diff. 0.02 0.05 0.01 0.04 0.00 0.01

Food

FE 3.18 3.04 3.15 3.23 2.73 2.70

Food HICP 3.10 3.03 3.16 3.24 2.72 2.71Food

Diff. 0.08 0.01 -0.01 -0.01 0.01 -0.01

Non-energy industrial goods

FE 1.14 1.15 1.09 0.83 0.82 0.99Non-energy industrial goods

HICP 1.10 1.10 1.05 0.78 0.78 0.97

Non-energy industrial goods Diff. 0.04 0.05 0.05 0.05 0.04 0.02

Energy

FE 7.75 5.80 5.25 3.93 3.99 1.66

Energy HICP 7.96 5.72 5.23 3.90 3.92 1.67Energy

Diff. -0.21 0.08 0.01 0.04 0.07 -0.01

Services

FE 1.78 1.70 1.78 1.75 1.59 1.86

Services HICP 1.73 1.64 1.79 1.64 1.54 1.85Services

Diff. 0.04 0.06 -0.01 0.11 0.04 0.01

Table 2: Nowcast results. fe: flash estimate annual rate; hicp: final annual rate; diff.: difference in percentage points between the flash estimate and the final figure. igoodsxe: ’non-energy industrial goods’. Unrounded differences.

Acknowledgments: The authors would like to express their gratitude to Jarko Pasanen and Roberto Barcellan who gave valuable input through comments, ideas, suggestions, guidance or simply through their pertinent questions that made us think more deeply on the methodology being developed. The other part of the production process which involves, among other things, the publication material, would not have been possible without the participation of Svetoslava Pavlova and Sílvia Santos.

Disclaimer: The content of this poster does not reflect the official opinion of the Eurostat. Responsibility for the information and views expressed therein lies entirely with the authors.

Preliminary dataPreliminary data sent by Member States has a 7 to 10 times lower RMSE than 1-step ahead forecasts described in the literature, which corroborates our assumption that preliminary data is a better forecast than any model-based forecast.

Aggregate Period Coverage MAPE max RMSE Rel RMSE

All-itemsP1 95% 0.051 0.117 0.034 0.073

All-itemsP2 98% 0.053 0.089 0.028 0.060

food P2 85% 0.053 0.189 0.066 0.216

igoodsxe P2 85% 0.110 0.305 0.106 0.285

energy P2 85% 0.136 0.358 0.148 0.086

serv P2 85% 0.100 0.207 0.070 0.149

Table 1: Preliminary data accuracy. P1: from Jan-2011 to Feb-2012; P2: from Mar-2012 to Jan-2013; MAPE: weighted mean absolute percentage error; max: maximum absolute difference in percentage error; RMSE: root mean square error; Rel. RMSE: root mean square error relative to a benchmark model.

Time constraintTheoretical framework

Nt = ��1 (B)w (B)Xt +��1n (B) ✓n (B)

� (B)s � (Bs)�Ds �d

at

: aggregate of countries that didn’t provide preliminary data : set of regressors (energy prices and / or aggregate of countries which did provide preliminary data

Nt

Xt

Forecasting EnergyEnergy is the most volatile component...

−0.15

−0.10

−0.05

0.00

0.05

0.10

0.15

2007 2008 2009 2010 2011 2012 2013

all

food

igoodsxe

energy

serv

Figure 1: ∆12 log for the euro area HICP for the all-items and main components. Energy is clearly the most volatile main component.

Figure 2: HICP versus energy prices. Since COICOP 0722 is essentially a mix of ’diesel’ and ’petrol’, a linear combination of these two energy products was calculated as Ct = α · diesel + (1 − α) · petrol, where α was chosen in order to maximize the correlation between ∆logHICP and ∆logCt.

...but energy prices published weekly by DG ENER, European Commission are good regressors of the HICP ‘energy’ component.

Consistency constraint

!1.5%

!0.5%

0.5%

1.5%

2.5%

3.5%

4.5%

Jan!06%

Apr!06%

Jul!06%

Oct!06%

Jan!07%

Apr!07%

Jul!07%

Oct!07%

Jan!08%

Apr!08%

Jul!08%

Oct!08%

Jan!09%

Apr!09%

Jul!09%

Oct!09%

Jan!10%

Apr!10%

Jul!10%

Oct!10%

Jan!11%

Apr!11%

Jul!11%

Oct!11%

Jan!12%

Apr!12%

Jul!12%

Oct!12%

Jan!13%

food% igoodsxe% energy% serv% all%

Deeper insight into inflation...

...in a more timely manner

More information available

• Eurostat webpage• News release• Statistics explained

http://ec.europa.eu/eurostat

51/2013 - 3 April 2013

Flash estimate - March 2013

Euro area annual inflation down to 1.7%

Euro area1 annual inflation2 is expected to be 1.7% in March 2013, down from 1.8% in February3, according to a

flash estimate4 from Eurostat, the statistical office of the European Union.

Looking at the main components of euro area inflation, food, alcohol & tobacco is expected to have the highest

annual rate in March (2.7%, stable compared with February), followed by services (1.9% compared with 1.5% in

February), energy (1.7% compared with 3.9% in February) and non-energy industrial goods (1.0% compared with

0.8% in February).

Euro area annual inflation and its components, %

Weight (‰)

2013 Mar 2012 Oct 2012 Nov 2012 Dec 2012 Jan 2013 Feb 2013 Mar 2013

All-items HICP 1000.0 2.7 2.5 2.2 2.2 2.0 1.8p 1.7e

Food, alcohol & tobacco 193.7 3.3 3.1 3.0 3.2 3.2 2.7p 2.7e

Energy 109.6 8.5 8.0 5.7 5.2 3.9 3.9p 1.7e

Non-energy industrial goods 273.6 1.4 1.1 1.1 1.0 0.8 0.8p 1.0e

Services 423.0 1.8 1.7 1.6 1.8 1.6 1.5p 1.9e

Source: Eurostat e = estimate p = provisional

1. The euro area consists of Belgium, Germany, Estonia, Ireland, Greece, Spain, France, Italy, Cyprus, Luxembourg, Malta,

the Netherlands, Austria, Portugal, Slovenia, Slovakia and Finland.

2. Annual inflation is the change of the price level between the current month and the same month of the previous year. For

further information on the euro area inflation flash estimate, see the Statistics Explained article on the Eurostat website:

http://epp.eurostat.ec.europa.eu/statistics_explained/index.php/Inflation_in_the_euro_area

3. See News Release 41/2013 of 15 March 2013.

4. The euro area inflation flash estimate is issued at the end of each reference month. The complete set of harmonised indices

of consumer prices (HICP) for the euro area, EU and Member States is released around the middle of the month following

the reference month. The next release with full data for March 2013 is scheduled for 16 April 2013.

Issued by Eurostat Press Office

Tim ALLEN

Tel: +352-4301-33 444

[email protected]

For further information on data:

Svetoslava PAVLOVA

Tel: +352-4301-34 406

[email protected]

Eurostat News Releases on the internet: http://ec.europa.eu/eurostat

Selected Principal European Economic Indicators: http://ec.europa.eu/eurostat/euroindicators

Register | Links | Contact | Important legal noticeEnglish (en)

European Commission > Eurostat > Home

Home

Statistics

Publications

About Eurostat

User support

Basic figures on the EU -

Summer 2013 editionScience, technology and

innovation in Europe - 2013

edition

Handbook on Residential

Property Prices Indices

(RPPIs)

Key figures on Europe 2012

GDP per capita in PPS

Real GDP growth rate

Total population

Unemployment rate

Employment rate, by sex

Inflation (monthly)

Inflation rate (annual)Macroeconomic Imbalance

Procedure (MIP)

Europe 2020 indicators

Euroindicators/PEEIs

Sustainable development

indicatorsGovernment finance

Prices (HICP)Eurostat yearbook

Regional yearbook

22.04.2013

Euro area and EU27 government deficit at 3.7% and 4.0% of GDP

respectively

19.04.2013

Proportion of underemployed part-time workers up to 21.4% in the EU27

in 2012

19.04.2013

EU27 current account surplus 31.4 bn euro

17.04.2013

Euro area production in construction down by 0.8%

16.04.2013

Euro area annual inflation down to 1.7%

15.04.2013

Spain, Italy and France: top destinations for holiday trips abroad of EU27

residents in 2011

15.04.2013

Euro area international trade in goods surplus 10.4 bn euro

Full list

24.04.2013

New: Road freight transport statistics - cabotage

23.04.2013

Updated: Tourism statistics - occupancy rates in hotels and similar

establishments

18.04.2013

Updated: Unemployment statistics

16.04.2013

Updated: Packaging waste statistics

16.04.2013

Updated: Housing price statistics - house price index

22.04.2013

In the fourth quarter of 2012, euro area and EU-27 seasonally adjusted

government deficits remain stable - Issue number 11/2013

22.04.2013

Support for financial institutions increases government deficits in 2012 -

Issue number 10/2013

04.04.2013

General government expenditure in 2011 – Focus on the functions ‘social

protection’ and ‘health’ - Issue number 9/2013

22.03.2013

Living standards falling in most Member States - Issue number 8/2013

20.03.2013

Continued recovery in volume of goods handled in EU ports - Issue

number 7/2013

27.02.2013

Economic ebb and flow in maritime sectors - Issue number 5/2013

27.02.2013

Enterprises making slow progress in adopting ICT for e business

integration - Issue number 6/2013

Full list

Search

Give us your opinion on

the quality of Eurostat's

data!User satisfaction survey 2012

(results)

Financial Crisis

GDP and beyond

Calls for tenders

Subscribe to the "Alert" systemSelection procedure for Eurostat in the

domain of national and accounts and

excessive deficit procedure.

closing date: 17.05.2013 (12.00 a.m)

Your guide to European statistics

Now available for

iPhone, iPad and Android!

Your guide to inflation statistics

Statistics Database

Release Calendars

Most popular database tables

Selected Statistics

Selected Publications

Latest news releases

Press centre | RSS

Latest updates of Statistics Explained

Statistics in focus

RSS

Log in | Register | Log off

User survey

In the spotlight

Job opportunities

Statistics Explained

Country Profiles

Inflation Dashboard

Test your knowledge

Table 1: Euro area annual in

flation and its main

components (%), 2

013, March

2012 and October 2

012

- March

2013

Source: E

urostat (p

rc_hicp

_manr)

(http://ec.e

uropa.eu/eurostat/product?

code=prc_hicp

_manr&language=en&mode=view)

Figure 1: Euro area annual in


components (%), 2

002-2013-03

Source: E

urostat (p

rc_hicp

_manr)

(http://ec.e


code=prc_hicp


Figure 2: Euro area annual in


components (%), S

eptember 2010 - M

arch 2013

Source: E

urostat (p

rc_hicp

_manr)

(http://ec.e


code=prc_hicp


Figure 3: Weights o

f the main co

mponents of th

e euro

area HICP (2013)

Source: E

urostat (p

rc_hicp

_inw)

(http://ec.e


code=prc_hicp

_inw&language=en&mode=view)

Inflatio

n in th

e euro area

From Statistic

s Explained

Data from M

arch 2013. M

ost rece

nt data: Furth

er Eurosta

t inform

ation, Main tables

and Database.

The data

in this artic

le show the m

ost rece

nt annual r

ates o

f change fo

r the e

uro area h

eadline in

flation an

d its four m

ain co

mponents issu

ed by Eurostat. The fi

gures

presented

are ac

tual HICP fig

ures.

Contents

1 Main

statist

ical fi

ndings

1.1 Flash est

imate an

d full HICP data

1.2 Main

components o

f inflat

ion

2 Further Eurostat

information

2.1 Database

2.2 Dedicated

section

2.3 Meth

odology / Meta

data

3 See also

Main statistic

al findings

The euro are

a annual i

nflation rat

e was 1

.7% in Marc

h 2013, down from 1.8% in February

. A year

earlie

r the

rate was 2

.7%. The flash

estimate

for Marc

h, published on 3 April

2013, was c

onfirmed. Further i

nformation on

the accu

racy of th

e flash

estimate

s can be fo

und in the artic

le Inflat

ion – methodology of th

e euro are

a flash

estimate

.

Looking at the m

ain co

mponents of th

e euro are

a inflat

ion, 'Food, al

cohol & tobacc

o' (2.7%, st

able compare

d to

February) had the h

ighest annual r

ate in M

arch, fo

llowed by 'Service

s' (1.8%, up fro

m 1.5% in February),

'Energy' (1

.7%, down from 3.9% in February

) and 'N

on-energ

y industrial g

oods' (1.0%, up fro

m 0.8% in

February).

Flash estimate and fu

ll HICP data

The euro are

a inflat

ion flash est

imate is

issued at

the end of ea

ch refere

nce month or sh

ortly aft

er. When the

complete set

of harmonised

indices o

f consumer p

rices (H

ICP) is rel

eased aro

und the middle o

f each month

following the refere

nce month, es

timate

d data are

replac

ed by actual d

ata deriv

ed from the M

ember S

tates

figures. Further i

nformation on the a

ccurac

y of the fl

ash est

imates ca

n be found in the a

rticle I

nflation –

methodology of th

e euro are

a flash

estimate

.

As from Septem

ber 2012, th

e flash

estimate

releas

e inclu

des, in ad

dition to the h

eadline in

flation, es

timate

s of

its four m

ain co

mponents.

The main

components a

nd their rela

tive weig

hts for 2013 are

presented

in Figure 3 an

d Table 2, re

spectively

.

Main components of infla

tion

Each of th

e main

components c

ontributes

in varying degree

to the head

line inflat

ion in the euro are

a. In ter

ms

of weig

ht, set a

t 1000 for the a

ll-item

s HICP, se

rvices is

the largest

component, acco

unting for around 42.3% of

individual consumption ex

penditure in the e

uro area. I

t is followed by non-en

ergy industri

al goods w

ith around

27.4%.

Food, alcohol &

tobacco an

d energ

y account fo

r 19.4% and 11.0%, re

spectively

. Together,

they comprise

less

than one third of eu

ro area e

xpenditure, but th

ey can have si

gnificant im

pacts o

n the head

line inflat

ion as their

levels

tend to flu

ctuate

signific

antly more than the o

ther components.

The brea

kdown of the H

ICP into four main

components d

oes not fo

llow the standard

COICOP classif

ication,

but groups item

s from diffe

rent ex

penditure class

es into four broad product c

ategories

. For further d

etails

on the

composition/su

b-indices, p

lease s

ee Eurostat

's COICOP/HICP cla

ssifica

tion

(http://ec.e

uropa.eu/eu

rostat/ram

on/nomenclatures

/index.cfm?

TargetU

rl=LST_NOM_DTL&StrN

om=HICP_2000&StrLanguageCode=EN&IntPcK

ey=&StrLayoutCode=EN) .

Further E

urostat inform

ation

Database

mailto:[email protected]

mailto:[email protected]



The Euro Area Flash Estimation Procedure/media/Kontorer... · Pedro Martins Ferreira and Luca...

Documents

Transcript of The Euro Area Flash Estimation Procedure/media/Kontorer... · Pedro Martins Ferreira and Luca...