A Big Data approach to understand Central Banks / 1

A Big Data approach

to understand
Central Banks
Big Data Spain 2018

November 2018
 A Big Data approach to understand Central Banks /2

Summary

01 Why is the use of NLP important in economics

and Monetary policy?
The data and methodology

02 Understanding Central Banks: “What”, “How”

and “Who” is talking (or writing) about?
 A Big Data approach to understand Central Banks /3

01
Why is the use of NLP important in
economics and Monetary policy?

The data and methodology

 A Big Data approach to understand Central Banks / 4

Why is the use of NLP important in economics and Monetary policy?

Text as a key source of information to enrich economic analysis

Text mining
80
of the total amount of web
pages on the internet is given to extract meaning from
% by textual or unstructured data strings of letters

The potential use of textual information and It helps us to understand

text sources improves the understanding of what drives monetary
economic and financial systems policy decisions
 A Big Data approach to understand Central Banks / 5

20 %
of used data

80 %
of available data
 A Big Data approach to understand Central Banks / 6

The data and methodology

From Extraction to Sentiment Analysis

Information Pre-Processing Text mining Sentiment

Transformation
extraction and text parsing and NPL analysis
Documents Extract words Text filtering Analysis and Apply sentiment
Machine learning dictionaries
Web pages Identify parts of Indexing to quantify
speech text in lists of Topics extraction Semantic analysis
term counts (LDA) and classification
Tokenization and
multi-word tokens Create the Clustering Clustering
Document-term
Stopword Removal matrix Modelling
(STM and DTM)
Stemming Weighting matrix
Case-folding Factorization
(SVD)
 A Big Data approach to understand Central Banks / 7

The data and methodology

Analyzing central banks’ communication:
Examined documents Information extraction

Statements / Press Releases

Immediately after the meeting on monetary policy, a short report about the decision on
interest rates is released. If there’s a press conference, the president of the CB explains the
decision and answer questions from journalists

Minutes
A more detailed document explaining the monetary policy decision containing an overview of
financial market, economic and monetary developments

Speeches
Collection of speeches and articles by senior central bank officials published in the central
bank websites
 A Big Data approach to understand Central Banks / 8

The data and methodology

Analyzing central banks’ communication:
Cleaning and transforming the text Pre-Processing and text parsing

Transformation

Working with text in numbers Extracting and organizing the data

Create the Document-term matrix Extract words
Weighting matrix Identify parts of speech
Factorization Stopword Removal
Case-folding

Preparing it for the analysis Converting it into numbers

Text filtering Stopword Removal
Indexing to quantify text in lists Stemming
of term counts Tokenization and multi-word tokens
 A Big Data approach to understand Central Banks / 9

The data and methodology

Analyzing central banks’ communication:
Dynamic topic models Text mining and NPL

Latent Dirichlet Allocation (LDA) and Dynamic Topic Model (DTM)

 A Big Data approach to understand Central Banks / 10

The data and methodology

Analyzing central banks’ communication:
Sentiment analysis Text mining
Sentiment analysis
and NPL

Loughran and McDonald FED Financial Stability dictionary

(2011) (2017)

Positive words Negative words Positive words Negative words

Positive words Negative words Positive words Negative words
achieve progress bankruptcy fallout benefit improve adverse escalate
benefit stabilize bottleneck imbalance enhance upgraded challenge stagnation
efficiency strength corrupt monopolize stabilise smooth deteriorate vulnerability
outperform versatility downgrade stagnant favorable strengthened downgrade worsen

𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒 𝑤𝑜𝑟𝑑𝑠 − 𝑁𝑒𝑔𝑎𝑡𝑖𝑣𝑒 𝑤𝑜𝑟𝑑𝑠

𝐀𝐯𝐞𝐫𝐚𝐠𝐞 𝐭𝐨𝐧𝐞 =
𝑇𝑜𝑡𝑎𝑙 𝑤𝑜𝑟𝑑𝑠
 A Big Data approach to understand Central Banks / 11

Main outputs
Analyzing the Central Bank of Turkey, European Central Bank
and Federal Reserve
 A Big Data approach to understand Central Banks / 12

Main outputs
More than words: Getting the relation between words…
In the case of CBRT:
 A Big Data approach to understand Central Banks / 13

Main outputs
…their evolution over time…

Most frequent words by year in the analyzed documents (the case of CBRT)
2014 2015 2016 2017 2018
 A Big Data approach to understand Central Banks / 14

Main outputs
…as well as the topical content covered in the text
 A Big Data approach to understand Central Banks / 15

02
Understanding Central Banks:
“What”, “How” and “Who”
is talking (or writing) about
 A Big Data approach to understand Central Banks / 16

We go Inside of the CB Reports to identify the topics using Machine

Learning and Dynamic Topic Models. They can be different…

A Big Central Bank (ECB) A Central Bank of a EM Country as Turkey (CBRT)

Economy EMU Integration Monetary Policy Global Flows Activity

Financial Crisis Banking Union Quantitative Easing Monetary Policy Inflation

Each word cloud represents the probability distribution of words within a given topic. The size
of the word and the color indicates its probability of occurring within that topic
Source: BBVA Research
 A Big Data approach to understand Central Banks / 17

Topics are dynamic and can change over time…and the picture
can change…particularly if important events hit the economy

European Central Banks: Evolution of Topics Central Bank Of Turkey: Evolution of Topics

100% 100%
90% 90%
80% 80%
70% 70%
60% 60%
50% 50%
40% 40%
30% 30%
20% 20%
10% 10%
0% 0%
2015 mar

2016 sep
2006 jan

2007 feb

2008 feb

2009 feb

2010 feb

2011 feb

2012 feb

2013 feb

2014 feb

2006

2007

2008

2009

2010

2011

2012

2013

2014

2015

2016

2017

2018
Global Flows Economic Activity
Economy EMU integration Labor Market Fiscal & Structural Policies
Banking Union Financial crisis Inflation Core Monetary Policy
Standard MP Non-standard MP Other

Source: BBVA Research Source: BBVA Research

 A Big Data approach to understand Central Banks / 18

Networks are a useful tool to show the interconnectedness &

complexity…helping us to understand “How” the Central Banks talk..

Monetary Policy in the North (ECB) and in the EM (Turkey): Complexity and interconnectedness
(Networks)
Pre Lehman (1999-2007) Financial Crisis (2007 -2012) QE & Post Crisis (2013-2018)

Source: BBVA Research

 A Big Data approach to understand Central Banks / 19

Sentiment analysis reinforces the analysis by describing

“How” the Central Bank talks (“tone”)

Turkey (CBRT) : Turkey (CBRT):

Economic Activity & Inflation Tone Economic Activity & Employment Tone
(Tone economic activity and Inflation jn the MP Minutes) (Tone economic activity and employment jn the MP Minutes)

3 3
Positive

Positive
2 2

1 1

0 0

-1 -1
Negative

Negative
-2 -2

-3 -3
2006

2007

2008

2009

2010

2011

2012

2013

2014

2015

2016

2017

2018

2006

2007

2008

2009

2010

2011

2012

2013

2014

2015

2016

2017

2018
Economic Activity Inflation Economic Activity Employment

Source: BBVA Research

 A Big Data approach to understand Central Banks / 20

Through Sentiment Analysis we can check the monetary policy

stance… how “Tight” or “Ease” is the Wording of the reports

Central Bank of Turkey: Monetary Policy Sentiment

(Standardized, estimated through Big Data LDA and STM Techniques from Minutes & Statements)

Monetary Policy “Statements” Monetary Policy “Minutes”

3 3

2 Tightening 2

1 1

0 0

-1 -1

-2 -2

-3 Easing -3

-4 -4
2006

2007

2008

2009

2010

2011

2012

2013

2014

2015

2016

2017

2018

2006

2007

2008

2009

2010

2011

2012

2013

2014

2015

2016

2017

2018
A more formal Statement… More extensive and analytical…

Source: BBVA Research

 A Big Data approach to understand Central Banks / 21

And how the market rates react to the Central Bank changes in
monetary policy sentiment …

Response to Short term and Long term interest rates to positive/Negative changes in Sentiment CB Turkey
(Response of interbank deposits rates and 2Y BondSwaps to mild and strong chnages in sentiment. Changes relative to t-1. T=event)

Bond Swaps Response to a Positive Change in Sentiment Bond Swaps Response to Negative Change in Sentiment

Source: BBVA Research

 A Big Data approach to understand Central Banks / 22

Remember that in the case of Sentiment Analysis, we are using

unsupervised methods so…always cross-check for Robustness

Monetary Policy in Turkey: Experts vs Algorithms in Turkey:

Experts vs Algorithms Size of Surprises & Sentiments
(Sentiments fron LDA Algorithm and MP Surprises by (Sentiments fron LDA Algorithm and MP Surprises by
Demiralp et Al. 1=Hawkish, 0= Neutral, -1=Dovish) Demiralp et Al)

Source: BBVA Research

 A Big Data approach to understand Central Banks / 23

Last… but not least … we are working on the Federal Reserve Board
(FED) Topics and Stance…

Federal Reserve Board (FED) Topics definition FED Hawkish/Dovish index and Fed Funds rate
(Moving average)
0.6 7

Hawkish
0.4 6

0.2 5

0 4

-0.2 3

-0.4 2

Dovish
-0.6 1
Inflación
-0.8 0
Tasa de paro

1996-June

1998-October

2003-June

2005-October

2010-June

2012-October

2017-June
1999-December

2006-December

2013-December
2002-April

2009-April

2016-April
1997-August

2001-February

2004-August

2008-February

2011-August

2015-February

2018-August
BBVA Fed Sentiment Index (12-month moving average, left)
Fed. Funds Rate (right)

Source: BBVA Research

 A Big Data approach to understand Central Banks / 24

…complementing our “What” and “How” the Central Banks talk

with “who” is talking…

General and Governor FED Hawkish/Dovish index by speaker over time

(Moving average 12 months)
Tightening
Easing

1998
2000
2002
2004
2006
2008
2010
2012
2014
2016
2018

1998
2000
2002
2004
2006
2008
2010
2012
2014
2016
2018

1998
2000
2002
2004
2006
2008
2010
2012
2014
2016
2018
1998
2000
2002
2004
2006
2008
2010
2012
2014
2016
2018
General Index Greenspan General Index Bernanke General Index Yellen General Index Powell

From a EM Crisis To a Governor To a Lady managing To a Normalization

Reactive and tigtening… Managing the crisis… the Exit Strategy… Policy
(Mr Greenspan) (Mr Bernanke) (Mrs Yellen) (Mr Powell)
1987-2003 2016-2014 2014-2018 2018-

Source: BBVA Research

 A Big Data approach to understand Central Banks / 25

You can find us at:

Thank you!
www.bbvaresearch.com

Alvaro Ortiz
Tomasa Rodrigo

@alvaroortiz1968
@TomasaRodrigo
 A Big Data approach to understand Central Banks / 26

A Big Data approach

to understand
Central Banks
Big Data Spain 2018

November 2018