<<

Towards an empirically founded behavioural macro

Exploiting the Internet as a Data Source for Social Science

Nikos Askitas, IZA – Institute of Labor Economics November 24, 2017

Workshop Nowcasting & Big Data, Banco Central de la Rep´ublicaArgentina Table of contents

1. Introduction

2. The Internet

3. Unemployment and Health

4. Shares: referenda, Case/Shiller Index

5. Hourly Data: Traffic Jams, [Askitas, 2016a]

6. Toll Index

7. A couple of thoughts

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics1 Introduction • Emerging data contains so much more than just economics • Public debate is shifting from being about economics to being about (identity) politics. • All of the non-economics parts in the data feed back into the economy so they are essential even for economists.

Behavioural Macro

“Macro” is often perceived as short for macroeconomics but I mean it in a rather more general sense. Although I am a Mathematician working as an economist it would be a waste to not think of macroscience more generally. Many reasons to do so:

• Economics is a derivative discipline founded on Math, Psychology, Sociology etc

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics2 • Public debate is shifting from being about economics to being about (identity) politics. • All of the non-economics parts in the data feed back into the economy so they are essential even for economists.

Behavioural Macro

“Macro” is often perceived as short for macroeconomics but I mean it in a rather more general sense. Although I am a Mathematician working as an economist it would be a waste to not think of macroscience more generally. Many reasons to do so:

• Economics is a derivative discipline founded on Math, Psychology, Sociology etc • Emerging data contains so much more than just economics

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics2 • All of the non-economics parts in the data feed back into the economy so they are essential even for economists.

Behavioural Macro

“Macro” is often perceived as short for macroeconomics but I mean it in a rather more general sense. Although I am a Mathematician working as an economist it would be a waste to not think of macroscience more generally. Many reasons to do so:

• Economics is a derivative discipline founded on Math, Psychology, Sociology etc • Emerging data contains so much more than just economics • Public debate is shifting from being about economics to being about (identity) politics.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics2 Behavioural Macro

“Macro” is often perceived as short for macroeconomics but I mean it in a rather more general sense. Although I am a Mathematician working as an economist it would be a waste to not think of macroscience more generally. Many reasons to do so:

• Economics is a derivative discipline founded on Math, Psychology, Sociology etc • Emerging data contains so much more than just economics • Public debate is shifting from being about economics to being about (identity) politics. • All of the non-economics parts in the data feed back into the economy so they are essential even for economists.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics2 Why Macro? – I

The first answer is that this is the nature of the data I am working with. But a second deeper answer which vindicates us working with this type of data might go as follows. The standard reductionist model says I can understand your shopping habits from your molecules if I can deconstruct how they are put together into creating you and how they respond to environmental impulses (i.e. “once a micro level is fixed, macro levels are fixed too; supervenience”). The reason then why this approach is not followed is that our (e.g. computing) means cannot capture it all or that it is too complex. So we need “theory” or “models”, so reductionism has it, as a means of speculatively filling the gaps and we hence confine ourselves to macro data as a kind of compromise.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics3 • ”the most important part of learning is actually forgetting”, [Tishby et al., 2000] (bottleneck theory in deep learning / artificial neural networks). • “macro beats micro in causal inference as measured by Effective Information”, [Hoel et al., 2013] (e.g. stochasticity of hidden layers of a neural network vs the more sentient nature of the network itself: causality is determined at the micro but first at the macro level.

Why Macro? – II

While most of causal economics research is of the “micro” kind we keep coming up short in putting it together into a macro picture and in fact this might be the case not just because of complexity but for more profound reasons.

• “Forgetting” information is fundamental for “remembering” it ([Askitas, 2018]) (e.g. the tangent space of a smooth manifold, the Taylor expansion of a function, digital vs. analogue technology, or impressionist paintings).

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics4 • “macro beats micro in causal inference as measured by Effective Information”, [Hoel et al., 2013] (e.g. stochasticity of hidden layers of a neural network vs the more sentient nature of the network itself: causality is determined at the micro but first at the macro level.

Why Macro? – II

While most of causal economics research is of the “micro” kind we keep coming up short in putting it together into a macro picture and in fact this might be the case not just because of complexity but for more profound reasons.

• “Forgetting” information is fundamental for “remembering” it ([Askitas, 2018]) (e.g. the tangent space of a smooth manifold, the Taylor expansion of a function, digital vs. analogue technology, or impressionist paintings). • ”the most important part of learning is actually forgetting”, [Tishby et al., 2000] (bottleneck theory in deep learning / artificial neural networks).

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics4 Why Macro? – II

While most of causal economics research is of the “micro” kind we keep coming up short in putting it together into a macro picture and in fact this might be the case not just because of complexity but for more profound reasons.

• “Forgetting” information is fundamental for “remembering” it ([Askitas, 2018]) (e.g. the tangent space of a smooth manifold, the Taylor expansion of a function, digital vs. analogue technology, or impressionist paintings). • ”the most important part of learning is actually forgetting”, [Tishby et al., 2000] (bottleneck theory in deep learning / artificial neural networks). • “macro beats micro in causal inference as measured by Effective Information”, [Hoel et al., 2013] (e.g. stochasticity of hidden layers of a neural network vs the more sentient nature of the network itself: causality is determined at the micro but first at the macro level.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics4 Why Behavioural?

In trying to forecast a time series it is (most of the time) very hard to beat an ARIMA model as seasoned forecasting practitioners know. On the other hand of course nothing will fail as miserably as an ARIMA model in predicting a break from past patterns. In those cases we need models which use ARIMA predictions as a baseline and are informed by behaviourally determined variables as captured by various measuring devices.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics5 Empirically Founded

Of course in some sense all of our models are behavioural and empirically founded at least in the sense that we postulate behavioural determinants which we concoct and then we seek to validate with our data. We know however (see ”rationality”) that our assumptions have been too naive. I am thinking about behavioural data captured in real time in the internet hence empirically founded behavioural macro.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics6 Empirically founded behavior: porn vs math

3 months worth of daily searches in the US 80

0 6 6 0 6 6 6 6 6 1 0 6 0 6 6 0 0 0 0 0 0 6 4 1 0 6 5 6 0 2 3 4 2 0 5 3 5 1 1 60 1 4 2 3 5 4 5 2 1 1 5 1 1 2 3 4 5 3 4 5 3 5 4 5 1 2 2 4 3 4 2 3 2 4 1 4 3 3 5 3 5 2 2 4 1 2 3 4 5 1

40 Google Searches for Porn Google Searches for Math

5 5 5 5 5 5 4 4 5 5 4 4 2 4 3 3 4 4 2 3 5 4 3 2 2 3 2 3 3 3 2 4 5 4 2 1 3 1 1 2 1 1 1

20 2 3 1 5 1 4 2 3 1 1 2 5 4 3 1 2 1 4 5 6 0 6 0 6 0 6 6 0 6 0 6 0 6 0 6 1 6 0 0 6 0 6 0 0 6 0 0 01aug2017 01sep2017 01oct2017 01nov2017 date Data from Google Trends and own calculations

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics7 Empirically founded behavior: traffic jams

7 days worth of hourly Stau Searches in 7 100

80 6

7 7 8 7 15 6 60 16 6 6 14 17 5 8 8 16 8 6 17 7 16 40 13 17 15 1617 17 16 4 18 17 5 15 5 8 'Stau' (traffic jam) 12 9 15 15 12 131415 1718 3 18 5 9 11 1415 9 9 14 10 19 910 12 16 13 18 9 5 78 16 12 19 18 14 13 10 20 6 14 4 1012 19 17 11 10 13 101213 4 11 20 18 910 111213 4 1012 19 11 4 21 78 20 18 11 19 1 3 4 56 3 3 20 1921 24 2223 192122 4 21 20 2 2122 3 202123 2 2223 24123 20 2324123 222301 2122230 23012 2224 3 2 1 1 2 0 06nov2017 12:34:1106nov2017 03nov2017 12:59:42 03nov2017 12:51:12 04nov2017 12:42:41 05nov2017 12:25:40 07nov2017 12:17:10 08nov2017 12:08:40 09nov2017 12:00:09 10nov2017 date Data from Google Trends and own calculations

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics8 Empirically founded behavior: VW defeat device, dieselgate

Efficient Markets and Hidden Data 100 250 VW stock in € VW on Google Search 200 90 150 € 80 VW stock in Dieselgate Starts Dieselgate VW on Google Search VW Google on 100 70 50 60 0 01jan2004 01jan2006 01jan2008 01jan2010 01jan2012 01jan2014 01jan2016 01jan2018 date © askitas.com | Data from Yahoo Finance, Google Trends and own calculations

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics9 Empirically founded behavior: S&P 500

Google 'short' searches in category Financial Markets

100 Google 'short' searches 1.500e+10 2500 90 day moving avg S&P Trading Volume 90 day moving avg S&P Index 80 2000 1.000e+10 60 Lehmann Brothers Lehmann 1500 S&P Index 40 5.000e+09 1000 20 0 0 500 01jul2003 01jan2007 01jul2010 01jan2014 01jul2017 date Data from Google Trends, Yahoo Finance and own calculations

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 10 The Data Scientist as a Historian

One of the ways humanity has tried to make sense of the world is through measurement. One of the ways we have used measurement is to do it in regular intervals and to record the results. Recording time series whether temperature, precipitation and pollution or population sizes, prices, GDP or market sentiment is very much like a quantitative form of history. The reason we do so is so that we may protect the environment we live in, stir the economy that sustains us or evolve the social structures that underlie the economy. Not unlike Herodotus, a data scientist records data points in time so that the causes of our troubles are not forgotten and when possible they can be avoided.

“...neither the deeds of men may be forgotten by lapse of time, nor the works great and marvellous, which have been produced some by Hellenes and some by Barbarians, may lose their renown; and especially that the causes may be remembered for which these waged war with one another.” Herodotus

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 11 The diary of Merer (4,500 years old) describes the daily life of workers who took part in the building of the Great Pyramid of Giza

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 12 • Accuracy: Are we measuring precisely enough? • Timeliness: Is our measurement complete on time? • Heisenberg 1: Are we changing what we are measuring? • Heisenberg 2: In case we are forecasting are we changing the past if we announce our forecast of the future? • Counterfactuals: How bad is it if our measurements are bad? (Can’t tell because no counterfactuals).

Problems with measuring things

• Quality: Are we measuring the right thing?

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 13 • Timeliness: Is our measurement complete on time? • Heisenberg 1: Are we changing what we are measuring? • Heisenberg 2: In case we are forecasting are we changing the past if we announce our forecast of the future? • Counterfactuals: How bad is it if our measurements are bad? (Can’t tell because no counterfactuals).

Problems with measuring things

• Quality: Are we measuring the right thing? • Accuracy: Are we measuring precisely enough?

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 13 • Heisenberg 1: Are we changing what we are measuring? • Heisenberg 2: In case we are forecasting are we changing the past if we announce our forecast of the future? • Counterfactuals: How bad is it if our measurements are bad? (Can’t tell because no counterfactuals).

Problems with measuring things

• Quality: Are we measuring the right thing? • Accuracy: Are we measuring precisely enough? • Timeliness: Is our measurement complete on time?

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 13 • Heisenberg 2: In case we are forecasting are we changing the past if we announce our forecast of the future? • Counterfactuals: How bad is it if our measurements are bad? (Can’t tell because no counterfactuals).

Problems with measuring things

• Quality: Are we measuring the right thing? • Accuracy: Are we measuring precisely enough? • Timeliness: Is our measurement complete on time? • Heisenberg 1: Are we changing what we are measuring?

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 13 • Counterfactuals: How bad is it if our measurements are bad? (Can’t tell because no counterfactuals).

Problems with measuring things

• Quality: Are we measuring the right thing? • Accuracy: Are we measuring precisely enough? • Timeliness: Is our measurement complete on time? • Heisenberg 1: Are we changing what we are measuring? • Heisenberg 2: In case we are forecasting are we changing the past if we announce our forecast of the future?

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 13 Problems with measuring things

• Quality: Are we measuring the right thing? • Accuracy: Are we measuring precisely enough? • Timeliness: Is our measurement complete on time? • Heisenberg 1: Are we changing what we are measuring? • Heisenberg 2: In case we are forecasting are we changing the past if we announce our forecast of the future? • Counterfactuals: How bad is it if our measurements are bad? (Can’t tell because no counterfactuals).

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 13 • CBOE quickly became one of the most important financial derivates exchanges in the world. • [Black and Scholes, 1973] and [Merton, 1973] published their options pricing theory for which Scholes and Merton (Black died in 1995) won the Nobel prize in 1997. • Their formula expressed the price of an option as a function of observable parameters and of the unobservable volatility of the price of the underlying asset. • At the beginning there were deviations of real and predicted prices of up to 30-40%. • 1976-1978: deviations reduced to within 2%! - ”the most successful theory in all of economics”...

Heisenberg: self-fulfilment ([Ferraro et al., 2005])

• CBOE (Chicago Board of Options Exchange) opened in 1973.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 14 • [Black and Scholes, 1973] and [Merton, 1973] published their options pricing theory for which Scholes and Merton (Black died in 1995) won the Nobel prize in 1997. • Their formula expressed the price of an option as a function of observable parameters and of the unobservable volatility of the price of the underlying asset. • At the beginning there were deviations of real and predicted prices of up to 30-40%. • 1976-1978: deviations reduced to within 2%! - ”the most successful theory in all of economics”...

Heisenberg: self-fulfilment ([Ferraro et al., 2005])

• CBOE (Chicago Board of Options Exchange) opened in 1973. • CBOE quickly became one of the most important financial derivates exchanges in the world.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 14 • Their formula expressed the price of an option as a function of observable parameters and of the unobservable volatility of the price of the underlying asset. • At the beginning there were deviations of real and predicted prices of up to 30-40%. • 1976-1978: deviations reduced to within 2%! - ”the most successful theory in all of economics”...

Heisenberg: self-fulfilment ([Ferraro et al., 2005])

• CBOE (Chicago Board of Options Exchange) opened in 1973. • CBOE quickly became one of the most important financial derivates exchanges in the world. • [Black and Scholes, 1973] and [Merton, 1973] published their options pricing theory for which Scholes and Merton (Black died in 1995) won the Nobel prize in 1997.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 14 • At the beginning there were deviations of real and predicted prices of up to 30-40%. • 1976-1978: deviations reduced to within 2%! - ”the most successful theory in all of economics”...

Heisenberg: self-fulfilment ([Ferraro et al., 2005])

• CBOE (Chicago Board of Options Exchange) opened in 1973. • CBOE quickly became one of the most important financial derivates exchanges in the world. • [Black and Scholes, 1973] and [Merton, 1973] published their options pricing theory for which Scholes and Merton (Black died in 1995) won the Nobel prize in 1997. • Their formula expressed the price of an option as a function of observable parameters and of the unobservable volatility of the price of the underlying asset.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 14 • 1976-1978: deviations reduced to within 2%! - ”the most successful theory in all of economics”...

Heisenberg: self-fulfilment ([Ferraro et al., 2005])

• CBOE (Chicago Board of Options Exchange) opened in 1973. • CBOE quickly became one of the most important financial derivates exchanges in the world. • [Black and Scholes, 1973] and [Merton, 1973] published their options pricing theory for which Scholes and Merton (Black died in 1995) won the Nobel prize in 1997. • Their formula expressed the price of an option as a function of observable parameters and of the unobservable volatility of the price of the underlying asset. • At the beginning there were deviations of real and predicted prices of up to 30-40%.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 14 Heisenberg: self-fulfilment ([Ferraro et al., 2005])

• CBOE (Chicago Board of Options Exchange) opened in 1973. • CBOE quickly became one of the most important financial derivates exchanges in the world. • [Black and Scholes, 1973] and [Merton, 1973] published their options pricing theory for which Scholes and Merton (Black died in 1995) won the Nobel prize in 1997. • Their formula expressed the price of an option as a function of observable parameters and of the unobservable volatility of the price of the underlying asset. • At the beginning there were deviations of real and predicted prices of up to 30-40%. • 1976-1978: deviations reduced to within 2%! - ”the most successful theory in all of economics”...

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 14 Revisit prior slide: Is the market going to crash any time soon?

Google 'short' searches in category Financial Markets

100 Google 'short' searches 1.500e+10 2500 90 day moving avg S&P Trading Volume 90 day moving avg S&P Index 80 2000 1.000e+10 60 Lehmann Brothers Lehmann 1500 S&P Index 40 5.000e+09 1000 20 0 0 500 01jul2003 01jan2007 01jul2010 01jan2014 01jul2017 date Data from Google Trends, Yahoo Finance and own calculations

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 15 Heisenberg: self-fulfilment ([MacKenzie and Millo, 2003])

[MacKenzie and Millo, 2003] show that this increasing accuracy resulted because people and organisations acted as if the theory were true, which made its predictions come true. Interviews with market participants revealed, for example, that traders started to use the theoretical value sheets obtained from the Black- Scholes equation to determine their bids.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 16 Countless measurement problems in Economics

In social science measurement is a Sisyphian task. A survey (or the announcement of its results) may change the behaviour it is meant to understand, an answer may be strategic so that it does not reveal the truth, you might just have a proxy to a measurement, you might have been keeping tabs on the wrong variables, you may introduce breaks in your historical series on technical or other grounds, the meaning of a variable might change. These and many more problems make economic forecasting both frustrating and beautiful.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 17 The Internet • It places networked computing into an increasing number of objects that are neatly integrated into our daily lives thus creating a data driven society and a data driven economy. • Perspective of a complete recording of all aspects of our life. Since demand and supply, all activities of companies and individuals, as well as the matching process are documented in the internet, not only the complete market economy but also all social aspects are reflected in a huge data cloud (big data). • When made accessible to social scientists, it provides a universe of research potential. Not only do we remember the past but we can rewind and replay recordings of the socioeconomic process. [Askitas and Zimmermann, 2015b]

The Internet

• Starting in the 1990s, digitisation and the internet in particular is impacting our world in a completely new and different way and will continue to be doing so in the foreseeable future.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 18 • Perspective of a complete recording of all aspects of our life. Since demand and supply, all activities of companies and individuals, as well as the matching process are documented in the internet, not only the complete market economy but also all social aspects are reflected in a huge data cloud (big data). • When made accessible to social scientists, it provides a universe of research potential. Not only do we remember the past but we can rewind and replay recordings of the socioeconomic process. [Askitas and Zimmermann, 2015b]

The Internet

• Starting in the 1990s, digitisation and the internet in particular is impacting our world in a completely new and different way and will continue to be doing so in the foreseeable future. • It places networked computing into an increasing number of objects that are neatly integrated into our daily lives thus creating a data driven society and a data driven economy.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 18 • When made accessible to social scientists, it provides a universe of research potential. Not only do we remember the past but we can rewind and replay recordings of the socioeconomic process. [Askitas and Zimmermann, 2015b]

The Internet

• Starting in the 1990s, digitisation and the internet in particular is impacting our world in a completely new and different way and will continue to be doing so in the foreseeable future. • It places networked computing into an increasing number of objects that are neatly integrated into our daily lives thus creating a data driven society and a data driven economy. • Perspective of a complete recording of all aspects of our life. Since demand and supply, all activities of companies and individuals, as well as the matching process are documented in the internet, not only the complete market economy but also all social aspects are reflected in a huge data cloud (big data).

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 18 The Internet

• Starting in the 1990s, digitisation and the internet in particular is impacting our world in a completely new and different way and will continue to be doing so in the foreseeable future. • It places networked computing into an increasing number of objects that are neatly integrated into our daily lives thus creating a data driven society and a data driven economy. • Perspective of a complete recording of all aspects of our life. Since demand and supply, all activities of companies and individuals, as well as the matching process are documented in the internet, not only the complete market economy but also all social aspects are reflected in a huge data cloud (big data). • When made accessible to social scientists, it provides a universe of research potential. Not only do we remember the past but we can rewind and replay recordings of the socioeconomic process. [Askitas and Zimmermann, 2015b]

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 18 Markets go online

Internet data can be applied to a very wide range of research areas including forecasting (e.g. of unemployment, consumption goods, tourism, festival winners and the like), nowcasting (obtaining relevant information much earlier than through traditional data collection techniques), detecting health issues and well-being (e.g. flu, malaise and ill-being during economic crises), documenting the matching process in various parts of individual life (e.g. jobs, partnership, shopping), and measuring complex processes where traditional data have known deficits (e.g. international migration, collective bargaining agreements in developing countries).

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 19 Refugees (https://www.weforum.org/agenda/2017/06/how-google-searches-could-help-governments-

map-immigration-flows)

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 20 Matching

The reason offline markets rapidly migrate online and new markets are directly born online is that the internet and digitisation are much more efficient with matching of supply and demand than analogue implementations. Moreover they provide market participants with non invasive access to the market at hand and they have the ability to record transactions seamlessly. This latter feature makes online market invaluable for socioeconomic research.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 21 Matching in the Information Market

The best known example of course is Google itself which matches the demand for with the supply of documents. In the information market attention is the scarce resource as noticed first by Herbert A. Simon: ”...in an information-rich world, the wealth of information means a dearth of something else: a scarcity of whatever it is that information consumes. What information consumes is rather obvious: it consumes the attention of its recipients. Hence a wealth of information creates a poverty of attention and a need to allocate that attention efficiently among the overabundance of information sources that might consume it” This is the reason for the early success of Google’s business model.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 22 Google Search and the state of the Google user

The demand for information on a certain topic by an individual reveals information about the state of that individual. If you search for “symptoms of depression” you might be self diagnosing or acting in diagnosing a person close to you. If you search for “side effect of prosac” you or a loved one is probably taking the antidepressant. In aggregate form we might then be able to develop indicators on the proliferation of certain phenomena. In the remainder of this talk I will show you some examples. I summarised the properties of the Google Trends data in [Askitas, 2015b] (https://wol.iza.org/articles/ google-search-activity-data-and-breaking-trends).

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 23 Google searches are utterances

Google searches may be viewed as continuous, irregular, unsolicited surveys with a Global cross section and a longitudinal resolution which is timestamped to the second. In traditional surveys we devise a question to which we seek an answer and get a (possibly) strategic one whereas in Google search we have unsolicited and hence more revealing answers to which we seek to reverse engineer the fitting question. Example of strategic answering: we know that people search for “child porn” but nobody would admit doing so in a traditional survey.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 24 Unemployment and Health Kurzarbeit (short-work)

Kurzarbeit Official resgistrations vs. Google Search 100

800000 Num. registered persons Searches for kurzarbeit 80 600000 60 400000 40 Searches for kurzarbeit Searches Num. registered persons Num. registered 200000 20 0 0 01jan2008 01jan2009 01jan2010 01jan2011 date Data from Bundesagentur für Arbeit, Google Trends and Own calculations

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 25 Gallup [Askitas and Zimmermann, 2015a]

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 26 Symptoms and Side Effects

Use Google searches for “symptoms” as a proxy to self diagnosis and searches for “side effects” as a proxy for medication. Test what happens around the TARP date (Troubled Asset Relief Program – October 3, 2008), [Askitas and Zimmermann, 2015a].

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 27 “high blood pressure symptoms” [Askitas and Zimmermann, 2015a]

Figure 1: TARP – Troubled Asset Relief Program – (October 3, 2008)

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 28 “symptoms” searches in the Health Category, US and DE.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 29 “symptoms” breakdown

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 30 • It can be used as we will see in an other example later to know whether your identification strategy is good enough. • In this case we see that after taking things out which might be less prone to respond to the crisis shock the spike remains as pronounced.

Remarks on breakdown

• The break down is useful to disentangle various causes for deviations from the mean (H1N1).

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 31 • In this case we see that after taking things out which might be less prone to respond to the crisis shock the spike remains as pronounced.

Remarks on breakdown

• The break down is useful to disentangle various causes for deviations from the mean (H1N1). • It can be used as we will see in an other example later to know whether your identification strategy is good enough.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 31 Remarks on breakdown

• The break down is useful to disentangle various causes for deviations from the mean (H1N1). • It can be used as we will see in an other example later to know whether your identification strategy is good enough. • In this case we see that after taking things out which might be less prone to respond to the crisis shock the spike remains as pronounced.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 31 Symptoms searches vs Labor Market, US.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 32 Side effects searches , US vs Germany: phase difference

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 33 Symptoms searches in the G8

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 34 Side Effects in the G8

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 35 Recovering “Theorems” of Health Economics

This work is relevant for Business cycle forecasting in the tradition of Joseph Schumpeter who in his 1939 volume “Konjunkturzyklen” includes church going in his list of 41 variables to monitor the business cycle: ”number of church goers is in inverse proportion to the degree of economic development ” Schumpeter would have a ball with internet data!

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 36 Hardship searches

In the course of this work we noticed, as we had expected, and is known in the literature that searches for ”chocolate cake”, ”inspirational quotes” or ”obituaries” also respond to the crisis shock. We also noticed that searches for hardship also spike.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 37 Hardship Letter

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 38 What is Hardship Letter?

A hardship letter is a loan modification request you write to your mortgage bank to explain that you are in temporary financial distress in order to avoid foreclosure.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 39 Google results for ”hardship letter”

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 40 Angelo Mozilo

In May of 2008, Angelo Mozilo, then Chairman of the board and CEO of a mortgage company called Countrywide Financial, made the news by accidentally hitting “reply” instead of “forward” in response to an e-mail from a distressed homeowner in North Carolina. The homeowner, who apparently was facing or anticipating an inability to make mortgage payments on time, had written a letter to request a loan modification and emailed it directly to the Office of the President of Countrywide. Angelo Mozilo who apparently thought he was forwarding the email to one of his people commented it as follows: “This is unbelievable. Most of these letters now have the same wording. Obviously they are being counseled by some other person or by the internet. Disgusting.”

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 41 Angelo Mozilo

Apparently, the homeowner who received the inadvertent reply with the comment made the email exchange public. Countrywide Financial, which in 2006 was the “largest mortgage lender” in the US, was by then failing and was purchased by Bank of America. Mozilo who at that moment must have been confronted with a prolonged rise in loan delinquencies saw it right. As more and more homeowners were entering financial hardship they sought help by writing loan modification requests to the banks holding their mortgages and the internet was their prime resource: a number of sites offered advice on the subject matter, loansafe.org being among the most prominent. What if we were able to count Mozilo’s letters?

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 42 Counting hardship letters

We can do one better than counting the letters. We can count the number of people who are in the process of fearing they might eventually need to write one!

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 43 Nowcasting Mortgage Delinquencies

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 44 Shares: referenda, Case/Shiller Index • So if your target variables are shares e.g. yes or no, sell or buy, put or call and you have a good identification strategy for representing them in terms of searches you might hope to get exact numbers. • E.g. in a yes no referendum you need to know Y /N where Y is the yes vote and N is the no vote. • So if you can proxy the yes voting intent and the no voting intent with g(Y˜) and g(N˜) for some keywords Y˜ and N˜ respectively then g(Y˜ ) Y˜ /T Y˜ Y t = t T = t ≈ g(N˜t ) N˜t /TT N˜t N for t close to voting date.

General Strategy

• Google trends data are relative data i.e. for a keyword K in time

unit t, Google Trends delivers g(K) = Kt /Tt where Kt are the number of searches for K at time t and Tt are the total number of searches at time t.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 45 • E.g. in a yes no referendum you need to know Y /N where Y is the yes vote and N is the no vote. • So if you can proxy the yes voting intent and the no voting intent with g(Y˜) and g(N˜) for some keywords Y˜ and N˜ respectively then g(Y˜ ) Y˜ /T Y˜ Y t = t T = t ≈ g(N˜t ) N˜t /TT N˜t N for t close to voting date.

General Strategy

• Google trends data are relative data i.e. for a keyword K in time

unit t, Google Trends delivers g(K) = Kt /Tt where Kt are the number of searches for K at time t and Tt are the total number of searches at time t. • So if your target variables are shares e.g. yes or no, sell or buy, put or call and you have a good identification strategy for representing them in terms of searches you might hope to get exact numbers.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 45 • So if you can proxy the yes voting intent and the no voting intent with g(Y˜) and g(N˜) for some keywords Y˜ and N˜ respectively then g(Y˜ ) Y˜ /T Y˜ Y t = t T = t ≈ g(N˜t ) N˜t /TT N˜t N for t close to voting date.

General Strategy

• Google trends data are relative data i.e. for a keyword K in time

unit t, Google Trends delivers g(K) = Kt /Tt where Kt are the number of searches for K at time t and Tt are the total number of searches at time t. • So if your target variables are shares e.g. yes or no, sell or buy, put or call and you have a good identification strategy for representing them in terms of searches you might hope to get exact numbers. • E.g. in a yes no referendum you need to know Y /N where Y is the yes vote and N is the no vote.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 45 General Strategy

• Google trends data are relative data i.e. for a keyword K in time

unit t, Google Trends delivers g(K) = Kt /Tt where Kt are the number of searches for K at time t and Tt are the total number of searches at time t. • So if your target variables are shares e.g. yes or no, sell or buy, put or call and you have a good identification strategy for representing them in terms of searches you might hope to get exact numbers. • E.g. in a yes no referendum you need to know Y /N where Y is the yes vote and N is the no vote. • So if you can proxy the yes voting intent and the no voting intent with g(Y˜) and g(N˜) for some keywords Y˜ and N˜ respectively then g(Y˜ ) Y˜ /T Y˜ Y t = t T = t ≈ g(N˜t ) N˜t /TT N˜t N for t close to voting date.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 45 Greek Referendum 2015: yes no searches

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 46 Greek Referendum 2015: identification

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 47 Greek Referendum 2015: yes no timeline

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 48 Greek Referendum 2015: ratio timeline

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 49 • I tried the method to a number of national elections and it did NOT work although it captured surprising outcomes some times. • Reason: not alway possible to proxy reliably. E.g. in Greek, German French elections searches for the Populist Right were topping the results not because it meant support for them but because everybody searched for the dystopian option. In the Greek and Irish referendum there was a high degree of polarisation, it was simpler due to dichotomy and could identify the two camps very well. • I did find another dichotomy to apply the method to and returned interesting results

Irish Gay Marriage Referendum 2015: ratio timeline

• I applied the method of [Askitas, 2015a] to the Irish Gay Marriage referendum in [Askitas, 2015c] using an undocumented access to the data shared with me by the Google Trends team and it was successful as well.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 50 • Reason: not alway possible to proxy reliably. E.g. in Greek, German French elections searches for the Populist Right were topping the results not because it meant support for them but because everybody searched for the dystopian option. In the Greek and Irish referendum there was a high degree of polarisation, it was simpler due to dichotomy and could identify the two camps very well. • I did find another dichotomy to apply the method to and returned interesting results

Irish Gay Marriage Referendum 2015: ratio timeline

• I applied the method of [Askitas, 2015a] to the Irish Gay Marriage referendum in [Askitas, 2015c] using an undocumented access to the data shared with me by the Google Trends team and it was successful as well. • I tried the method to a number of national elections and it did NOT work although it captured surprising outcomes some times.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 50 • I did find another dichotomy to apply the method to and returned interesting results

Irish Gay Marriage Referendum 2015: ratio timeline

• I applied the method of [Askitas, 2015a] to the Irish Gay Marriage referendum in [Askitas, 2015c] using an undocumented access to the data shared with me by the Google Trends team and it was successful as well. • I tried the method to a number of national elections and it did NOT work although it captured surprising outcomes some times. • Reason: not alway possible to proxy reliably. E.g. in Greek, German French elections searches for the Populist Right were topping the results not because it meant support for them but because everybody searched for the dystopian option. In the Greek and Irish referendum there was a high degree of polarisation, it was simpler due to dichotomy and could identify the two camps very well.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 50 Irish Gay Marriage Referendum 2015: ratio timeline

• I applied the method of [Askitas, 2015a] to the Irish Gay Marriage referendum in [Askitas, 2015c] using an undocumented access to the data shared with me by the Google Trends team and it was successful as well. • I tried the method to a number of national elections and it did NOT work although it captured surprising outcomes some times. • Reason: not alway possible to proxy reliably. E.g. in Greek, German French elections searches for the Populist Right were topping the results not because it meant support for them but because everybody searched for the dystopian option. In the Greek and Irish referendum there was a high degree of polarisation, it was simpler due to dichotomy and could identify the two camps very well. • I did find another dichotomy to apply the method to and returned interesting results

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 50 Trend-Spotting in the Housing Market

In [Askitas, 2016b] I formed what I called the BUSE index (i.e. BUy SEll index) by taking the point wise ratio of buy to sell Google searches in the Real Estate category.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 51 Nowcasting the Case/Shiller Housing Index

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 52 Trade Balance Argentina

Quick and dirty calculations, very crude, so please check on your own...

Argentinian Trade Balance with Google Searches

Google: exports/imports 12 weeks moving avg 1.4 exports/imports 1.2 1 .8 .6 .4

01jan2004 01jan2006 01jan2008 01jan2010 01jan2012 01jan2014 01jan2016 01jan2018 date © askitas.com : Data from Google Trends, Worldbank and own calculations

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 53 Hourly Data: Traffic Jams, [Askitas, 2016a] The German Highway Network

Autobahnen und Bundesstraßen 2017 Stand: Mai 2017

DÄNE- Westerland Sylt Süderlügum

5 Flensburger Kolding Klixbüll 200 Förde Niebüll R OSTSEE Leck FLENSBURG MARK 199 199 Föhr Steinberg Gelting

Wyk Norder- Wanderup Amrum aue Tarp Kappeln 5 Süderbrarup Bredstedt 200 F Süderaue e 201 hm Kieler a Hooge rn 203 b e Hidden- Hattstedt l Sagard Norder- t see Sassnitz Pell- Treia Schleswig 96 oogsand worm 201 76 Nord- Bucht 96b Husum Eckernförder Puttgarden Dubnitz strand 7 76 Eckernförde Bucht Kleiner Windebyer Surendorf Fehmarn Jasmunder Süder- m Noor stro L285 L45 Bodden oogsand Hever 5 Burg Bergen 76 96 77 Karow 203 Fehmarn- 207 Friedrichstadt Gettorf 196 Wittensee 503 Rügen sund

TR 76 S Göhren Heiligenhafen Samtens TR Schönberg L165 t Garding Saaler r e 96 202 502 Passader See Hohwachter Großenbrode Lunden Bodden la 202 Rendsburg Bucht - NORDSEE St.-Peter-Ording Tönning Selenter 215 See Löbnitz Rügischer Bodden E 210 105 id Erfde Großer e 76 KIEL Binnensee 202 r Oldenburg Mecklenburger 202 202 Borgwall- sund 5 Lütjenburg Westensee r see 215 e Helgoland Hamdorf id -Damgarten Tellingstedt E 203 501 Ribnitz- Heide 404 Preetz 430 Greifswalder 76 105 Rein- berg al 194 Kan Kellersee Bodden ee- 77 105 203 sts Nortorf Plön 1 Rövershagen Pommersche -O Bucht rd Ascheberg 96 Büsum No 103 5 Eutin Grömitz Grimmen Großer 76 Bad Doberan Zinnowitz NEU- 430 Meldorfer 431 Hohenwestedt Plöner See Neustadt Tribsees 109 Deutsche Bucht L328 MÜNSTER 19 Bucht Meldorf TR TR Lübecker Sanitz Wolgast Bucht 430 TR 103 110 20 Kröpelin 430 430 109 Trischen 23 Bornhöved Pönitz 111 105 ROSTOCK Schenefeld Bucht Scharbeutz Hochdonn TR Neubukow 111 205 Ahrensbök 194 Stör Poel z Schar- it 5 77 432 n Loitz Karlsburg 76 k Gnoien hörn c Marne 7 e 111 Wismar- R Ahlbeck Kellinghusen TRM T bucht 20 Jarmen Wilster 206 Bad Schwartau TR 110 Gützkow 109 Murchin 75 Laage Neuwerk Itzehoe 110 5 Bad 206 20 Dassow Demmin St. Bad 226 WISMAR TR 110 110 Brunsbüttel Bramstedt Segeberg CUXHAVEN Margarethen Dargun Liepen Kritzkow 110 Usedom T 105 106 o 108 l 21 104 le Anklam Stettiner Kaltenkirchen 75 104 n TR TR 73 431 Zurow 103 se Wangerooge 75 Schönberg Grevesmühlen TR Reinfeld Warnow Teterower 199 Otterndorf Bad 75 LÜBECK TR Neuklostersee Haff Spiekeroog TR See Peene 432 Oldesloe 207 20 192 194 Langeoog R Warin Güstrow Neuhaus 73 Barmstedt TR Kummerower Ducherow Glückstadt Teterow Baltrum Wischhafen Niendorf See Norderney Mellum Oste Rehna Elmshorn 431 1 Großer 208 Norderney 4 75 Großer 104 104 104 Ueckermünde Elmen- Ratzeburger See 104 106 Malchin 431 Sternberger See Prüzen Inselsee Lalendorf Juist 495 23 horst 208 Brüel Carolinensiel Quickborn TR TR 207 197 Elbe E 109

l 104

NORDER- b Stavenhagen e

Hemmoor - 208 Schweriner Malchiner STEDT L Gadebusch Sternberg See Borkum ü Ahrens- TK See 432 TR b Friedland Memmert 461 Uetersen burg e Ratzeburg Krakower Ferdinandshof c 192 431 k 104 See 27 - L203 104 Borkum 404 K Moltzow Rosenow Galenbecker Hechthausen Pinneberg a 433 n 14 See Norden Langen 73 a Barniner 104 HAMBURG l Mölln 106 See Wittmund WILHELMS- Stade Dobbertin Goldberger 108 194 Jatznick Ley- 495 K78 SCHWERIN HAVEN Wedel 207 321 See bucht 72 Upgant- Jever 210 BREMER- 1 Schaalsee Goldberg 19 Torgelower 210 Dümmer 321 Schott K291 431 5 See 104 20 HAVEN 24 See 392 Drewitzer 212 26 Crivitz 192 See 210 4 192 Pasewalk Schortens 74 W Kölpin- Glinde Zarrentin 321 a NEU- Strasburg Bremer- 255 5 207 TRM rn see 192 192 404 195 ow Karow Fleesen- vörde ( ) 192 see BRANDENBURG 104 104 Aurich Nordenham Horneburg Buxte- 252 Waren Penzlin Linken Ems-Jade-Kanal Jadebusen ( ) Schwarzenbek R M 24 Woldegk 104 73 hude 321 103 Tollensesee 210 Großes Friedeburg 212 L235 TR M 5 207 TR 436 7 TRM Malchow Rederangsee Löcknitz Meer Gallin 96 71 Beverstedt 3 73 73 253 25 404 M Lübz üritz-El 191 193 437 Neu- de- 192 72 Basdahl 75 Geesthacht 209 195 W Plau 29 Wulmstorf asserstra Käbelicksee Wiesmoor Varel TR ße 113 EMDEN 1 321 Parchim Plauer Müritz Bredenfelde 198 109 436 437 3 437 Weser Hagenow 191 See Großer Roden- 261 TK Maschen lbe 5 Woterfitz- See kirchen 71 E 198 210 31 Lauenburg Stuer see Neustrelitz Krackow 5 TR Röbel RM 404 5 Boizenburg Ems Carpin ( ) Pritzier Neustadt- 321 103 198 75 TR Prenzlau Brake R M 3 R Handorf Glewe 11 Dollart Hesel ( ) 74 Buchholz 39 209 5 Ludwigslust 14 Marnitz Woblitz- Carwitzer 2 Oste TR M See 70 436 212 198 see 113 72 R Bardowick 198 28 Meyenburg 198 Zeven TRM Neuhaus Mirow 211 Unter- Grabow Wesenberg 96 uckersee Westerstede Elsfleth 4 24 Rätz- 109 Gartz 5 see Bunde 31 75 Welle 195 Haßleben Leer Osterholz- 1 LÜNEBURG Gramzow 72 K Platlinsee 280 436 70 Scharmbeck 122 Folmhusen OLDEN- 74 191 Egestorf Elbe Fürsten- Zwischenahner 293 BURG 14 berg Ober- 2 Groningen Stolpsee 166 Weener Meer 270 Kaarßen uckersee Passow 438 69 Scheeßel 3 103 Hunte Wittstock Strücklingen 71 Dahlenburg 216 Karstädt 70 281 Großer BREMEN 212 Pritzwalk Stechlinsee Templin 198 Dömitz Havel Papenburg 27 75 189 Greiffenberg 401 Rotenburg 5 Groß R Rheinsberger 72 6 TR TRM Amelinghausen 189 Lübbe- 2 Schwedt Wardenburg DELMEN- TRM 209 Pankow See Rheinsberg 4 R see TK HORST TK 191 Aschendorf 28 281 166 TR Zernien 195 Lenzen Perleberg 71 Dannenberg 107 Wolletzsee 401 Friesoythe TK 75 7 TR L121 Großer 109 Küstenkanal Ganderkesee 215 Neuenkirchen 191 189 Garz Wentowsee TR Schnackenburg 195 5 103 Angermünde 401 29 6 248 72 TR Gransee Grimnitzsee Achim Soltau 493 122 213 440 Visselhövede Munster Wittenberge Gudelack- Zehdenick Dörpen TR 493 Obersee see 51 198 439 Wese R r 71 Uelzen 5 TR 96 158 Werbellin- TR Lüchow 70 1 Neuruppin Alt Ruppin see 189 Lathen Untersee TR Parsteiner Kyritz See TR Verden 5 167 213 27 71 Arendsee 167 167 TR Rütenbrock 51 248 Ruppiner Oderberg Wildeshausen Bad Falling- 4 Seehausen Elbe See Löwenberg TR 408 s 107 Wusterhausen m Walsrode bostel 167 Finowfurt E Bergen 167 215 209 Liebenwalde Cloppen- 190 Arendsee 24 167 EBERSWALDE 158a 72 E anal Haren l Havel-K burg 6 b Oder- e Bützsee 158 408 Breitenhees - Salzwedel Havelberg 11 167 Hoya S 96 POLEN Bergen e Hunte

i t Lastrup e Havel n Osterburg l er Kan 213 209 k na ppin al Bad 70 Rethem 3 a Sandau 5 ka Ru A n hin 168 Freienwalde Oder lle Sprakensehl a R Wandlitz r l 102 TR 273 402 Vechta 51 61 158 Wriezen Löningen Friesack Oranienburg 402 Eschede 191 4 Winterfeld 31 68 Wittingen 189 Rhinow TR 96 NIEDER- Meppen 69 . Quakenbrück Rohrberg K Gülper - Haselünne Bernau s TR 244 See Velten 10 Tiefensee 167 Eschebrügge Marklohe 215 TR 248 Groß Hohennauener m Winsen 71 Schwechten T Prötzel E See - 214 273 158 d Kakerbeck T Emlichheim n 214 Schwarm-

70 u 188 Hennigsdorf TR Werneuchen stedt 214 Klietz

m 214 Nienburg l V t 213 402 96 e r 5 e 111 Strausberg Neuhardenberg v c o 214 Kirchdorf 107 168 Küstrin- h a

D 96a t 214 Kietz

H e Stendal 214 Celle Nauen 114 Steinfeld 7 Brome 1 ( ) 215 Wienhausen Ehra- 2 61 4 Rathenow Falkensee Seelow LINGEN Bersen- TR M TRM Leine Lessien 188 5 109 TR Fürstenau 214 Dümmer 6 188 TR Rathstock 403 Neuenhaus TR brück Landes- 111 158 bergen 188 Tangermünde Gardelegen 96 TR Münche- 213 214 1 248 Uchtspringe berg 1 Stolzenau Premnitz Wustermark 5 10 112 214 Aller K 100 Freren 68 Leese 1 5 167 Uchte 441 213 239 Neustadt Groß- 3 Lüderitz Jerichow 102 10 5 51 Steinhuder 244 Havelsee BERLIN 1 Heinersdorf NORD- 218 Meer burgwedel 96a Herzfelde 442 188 Beetzsee TRM 100 Großer 215 r HORN 352 188 Mieste Müggelsee 61 e 6 LANGEN- 188 2 100 103 s 107 213 70 Hunte K332 Meinersen Gifhorn 115 Spree Lebus e HAGEN 188 113 Erkner Ems Letzlingen 96 Bramsche ittellan Rahden Uetze 189 Havel 273 TRM 403 M d W 441 K331 1 LANDE ka TRM 522 37 96a 168 5 112 n Burgdorf BRANDENBURG al ( ) 443 188 Dolle Wunstorf 441 Oebisfelde 1 nal Jeserig 218 TR M TR K92 71 l-Ka 444 ave 1 1 Teltow T 117 482 Mittellandkanal GARBSEN 6 214 4 Aller -H POTSDAM 101 FRANKFURT 112 5 30 68 Velpke e 102 T 442 441 WOLFS- lb Genthin Rietzer See L40 96a 87 Espelkamp Petershagen 3 R E Breitling- 1 2 TK M 12 Bad Lehrte BURG see L40 Fürstenwalde HANNOVER i 403 65 t Werder Mahlow t Templiner Hengelo Hörstel 51 Nenndorf 96 113 TR 61 443 e Elbe Schwielow- TRM l l See TR Bad Preußisch Stadthagen 65 Mittellandkanal T Poznan 65 65 Lehre 39 244 a see Bentheim Belm 65 Oldendorf Lübbecke TR Sehnde n Bad Essen MINDEN 6 TR 2 d Scharmützel- Ibbenbüren 68 442 Ronnenberg 443 k 1 TR Peine TR a 14 TR see 65 n 179 Wolziger Neuen- 65 3 37 a 10 RHEINE TR OSNABRÜCK l Hohenseeden See Oder kirchen 30 Haldensleben TR Königs 168 Müllrose Obernkirchen 443 65 392 107 Golzow 101 Wusterhausen Storkow Gronau 481 239 Ziesar 2 Rangsdorfer 96 87 112 TR ( ) 65 R 391 Burg See Ochtrup 70 219 7 39 Wolmir- TR 246 Wettringen TRM TR M Bückeburg Pattensen 494 Ilsede 61 2 1 stedt 246a 54 Georgs- Melle Bad 83 TR Königslutter 245 71 1 2 102 Beelitz R 246 33 482 442 444 1 BRAUN- TR 246 Saer- marien- TK Kirchlengern Oeynhausen Springe 217 6 1 TR Erxleben 54 beck Leine 1 Motzener Emsdetten 475 hütte 30 TR 3 SCHWEIGT TR 189 246 Zossen R Großer Beeskow Helmstedt Brück Trebbin See Selchower See Hessisch- Bad T TR 246 Mellen- Pätzer 51 TK 39 T 2 see See Grunow Vechte Löhne Burg- Borghorst 475 Oldendorf Münder Nord- 1 TR Wünsdorf 179 481 stemmen K39 WOLFENBÜTTEL 1 Görzke 246 70 steinfurt 61 514 238 Rinteln 244 TR 246a 2 EISEN- 219 239 83 Hachmühlen 1 1 Heek Vlotho 444 245a Biederitz 107 Märkisch Dissen HERFORD 217 K30 245 Bad Belzig Spree Ranzig HÜTTENSTADT 442 6 TK Schöningen Möckern Buchholz 475 Coppen- 82 TR 96 Teupitzer Ahaus Borg- 79 Eilsleben 1 246 246 Greven holzhausen brügge Elze HILDES- K12 248 Schöppen- TR See 70 54 Glandorf 246a Loburg 102 9 TR stedt Barneberg 71 246a 1 HEIM SALZ- 184 246 Wiesenburg 13 Neu Lübbenau Vreden 238 HAMELN 1 3 Seehausen Treuenbrietzen 101 87 168 112 Legden Halle BIELEFELD BAD 243 395 Altenberge TR GITTER 102 Luckenwalde 51 SALZUFLEN 245 179 e Ostbevern 6 L472 Wanzleben MAGDE- Gommern Baruth ß 475 Aerzen Gronau Leitzkau Nedlitz i Stadtlohn 1 Versmold 61 Emmern Hamersleben BURG Reuden Niemegk e 474 481 476 68 239 81 Schwieloch- N 2 82 246a 246a 107 see 1 TR Lemgo 83 W 240 246 184 102 320 66 79 244 31 Sassenberg 66 e Oschersleben Lieberose 320 Guben 66 66 s Bockenem Schladen 180 Petkus 54 51 Ems e Schönebeck 87 Lage r 246a Arnhem Coesfeld Telgte 513 2 115 525 64 475 Harsewinkel Barntrup 240 Alfeld 243 248 245 81 Alten- 525 Nottuln TR TR 61 66 238 14 Kropstädt Jüterbog 115 Lübben 112 525 L712 6 82 Schwanebeck weddingen Golßen 1 Bodenwerder Zerbst 97 Elten R 51 Warendorf GÜTERSLOH Vienenburg Dardesheim MÜNSTER 239 3 Langels- 70 DETMOLD Blomberg 240 82 Egeln TK 8 474 43 64 R Eschershausen heim 241 Straupitz R -Bad TR 79 81 96 -Clarholz 252 TR Gröningen 184 168 Isselburg 54 Leine 248 82 Peitz Emmerich Rhede Borken Senden 61 Schloß Holte- Meinberg 6 187a 107 67 Herzebrock- 239 243 Goslar 244 101 87 9 220 67 L600 TR Stukenbrock 83 Bad Witten- Dahme TK 9 Horn- Steinheim 6 187 Coswig 102 67 67 64 Gandersheim Seesen 498 Stapelburg Halberstadt e -Roßlau berg Ennigerloh 64 l Lübbenau Brunsen a Kranen- Reken 241 Elbe 187 97 Bad a Kleve 3 235 Rheda- TR s 64 burg BOCHOLT 70 Dülmen 242 Staßfurt S 67 Em 64 Harzburg 81 180 Luckau 475 Oker- 184 112 57 473 474 Ascheberg 239 79 Elster Rees Raesfeld 58 Drensteinfurt Rietberg 1 Holzminden stausee 4 Concordia- DESSAU- 2 504 -Wiedenbrück see Aken TR 54 Einbeck 445 248 6 Eutzsch Vetschau 9 TR 58 AHLEN 2 33 Bad 242 Wernigerode Jessen Brandis 169 Kalkar 70 R R R 55 Clausthal- 6 187a 187 Niers 58 Lippspringe Oranienbaum 224 Haltern 243 Zellerfeld 6 Calau 168 474 Lüdinghausen 64 Delbrück 252 3 244 185 184 S Schlabendorfer Forst 67 Rh 1 Langenberg 242 Gremminer chwarze Elster COTTBUS Goch e 58 63 Höxter Bad Grund 27 Bernburg See 6 Mulde i 58 185 107 See 87 n L652 Olfen Beckum Bad Blankenburg Quedlinburg 100 473 58 PADER- 497 7 241 Elbingerode 27 Kemberg 101 Gräbendorfer 15 TK L612 Selm Driburg 242 180 Boxmeer 58 DORSTEN BORN Aschersleben Köthen See 57 WESEL 70 498 81 64 Brakel 27 58 Lippe 235 248 Rappbode- 185 Gräfen- Schlieben Groß Oßnig TK Lippe 54 Lippe Braunlage talsperre heinichen Elbe Sonnewalde Xanten 225 43 Datteln 236 64 Northeim Osterode Pretzsch Herzberg ( ) Werne HAMM 64 241 Ballenstedt MARL Altdöberner Talsperre Wroclaw TRM 224 LIPPSTADT 83 241 183 2 Drebkau 97 115 9 TR M 225 241 180 182 96 See Spremberg 58 8 52 54 TRM 241 243 Hasselfelde 184 100 169 Kevelaer RECKLING- LÜNEN Borchen Uslar 3 Katlenburg Döbern 233 475 Beverungen 241 Hardegsen 242 Mulde- 107 Finsterwalde GLAD- HSN. 1 Herzberg Hettstedt TR stausee 57 55 27 Oder- 4 Könnern -Wolfen BECK HERTEN 2 TRM Erwitte Geseke 446 DINSLAKEN 247 stausee 242 Harzgerode TR Dommitzsch Löhsten 101 Tschernitz T 226 TR TR 83 81 Bitterfeld- Schwemsal SPREMBERG Issum 63 Willebadessen Hardenberg 27 183 Großräschen 8 CASTROP- 241 Bad 183 87 156 K. KAMEN 44 Lichtenau 68 252 Bad Lauterberg Zörbig Herne- Karlshafen TR Bad Düben 96 Sedlitzer 58 Kamp- Rhein- 54 Werl Soest TR 446 242 TR Großer Bad Muskau TR -RAUXEL 236 Borgent- 3 Gieboldehausen Mansfeld Goitzschesee Seelhausener TR TR See Geldern Lintfort berg 59 516 Rhein- 42 UNNA reich 100 See BOTTROP 224 1 TK TK Anröchte 27 183 Senften- 226 DORT- L663 Ilfeld Bergheider HERNE L821 See berg 44 27 27 184 Weißwasser 58 528 GELSENK. 235 MUND 247 183a 183 156 L474 42 T 1 TK TK 55 Scherfede Diemel 80 86 14 Torgau 169 9 510 OBERHSN. T Bremen Bad 83 183a 13 227 40 T Wickede 229 Körbecke 252 446 243 Mockrehna S MOERS 1 Wünnenberg 7 GÖTTINGEN Mackenrode Eisleben 87 Bad p Neukirchen- 231 236 516 Weser 3 Straelen ESSEN 226 233 480 Dransfeld 4 100 Lands- 107 Neuwieser Lohsaer re DUIS- 3 TR 54 TR 63 445 TR 180 Delitzsch Liebenwerda Lauchhammer e Vluyn 448 TR Ruhr 516 Warburg ( ) berg 9 See 97 See Niers 448 HALLE 58 7 252 3 Nordhausen Riestedt 2 Elsterwerda Senften- 96 156 223 40 7 Möhne- Möhne 516 Marsberg 7 TR TRM TR M Duderstadt 6 169 berger See 221 WITTEN 45 515 ARNS- Warstein Hofgeismar Krostitz Belgern L237 SCHWERTE see L 27 Sangerhausen 80 40 1 227 BOCHUM BERG Rhoden e 184 Lauta Scheibesee 226 MENDEN i R n 80 Eilenburg 101

BURG 54 229 e Ruhland Rietschen

Venlo MÜLHEIM Herdecke Werbeliner K Kempen 59 224 233 247 HOYERS- 44 43 236 7 55 480 7 Bleicherode Berga 91 See Bärwalder 252 44 38 TR 143 6 182 WERDA 9 1 HATTINGEN Wetter 7 Hannoversch Friedland Gröditz See 115 Nette- 52 r Talsperre 509 288 Ruh 46 Hemer Münden 80 Kelbra Groß Särchen Venlo tal 57 524 234 46 Brilon Diemel Calden TR 87 107 515 Bad Arolsen TR 180 Weiße Bernsdorf ter KREFELD VELBERT 229 T 3 496 Elste 169 Els TR 7 ISERLOHN Bestwig 7 r Schkeuditz e 8 227 7 38 85 Merse- 2 14 Wurzen 97 rz TRM 27 80 Bad Bad a TK 1 L Olsberg Hede- TR 4 6 101 9 Sondershausen Querfurt Taucha 6 w 57 e Balve Diemelsee TK Frankenhausen Lauchstädt burg Niesky RATINGEN n Zeithain h TK HAGEN münden c 61 236 n 80 Artern Ruhr KASSEL Heiligenstadt Leinefelde Talsperre WUPPER- Meschede 181 S 221 326 e Sorpesee Luppa Commerau VIERSEN 44 535 Gevelsberg K 450 251 181 96 Quitzdorf 46 TK Altena 252 Witzen- 91 186 87 Schönfeld 229 55 hausen 71 6 Elbe 98 TK 8 7 K Schwelm 54 Hennesee Willingen Mark- 107 98 Kamenz 156 Mettmann K 45 U e 57 TAL K Wolfhagen 251 n S Oschatz Königsbrück 38 a ß 7 Werdohl 3 Heldrungen s Geiseltalsee ranstädt Riesa 98 Talsperre i t 250 a LEIPZIG Trebsen Großenhain 52 7 r 2 169 e 1 249 u l Mark- TR TK 480 251 Dingelstädt t e Bautzen L70 N DÜSSEL- 7 483 Eslohe 7 27 180 87 kleeberg 52 230 57 NEUSS 3 251 247 Ebeleben TR

Roermond L418 Korbach Nieder- DORF 229 Helsa Nebra Lützen Laußnitz TR MÖNCHEN- 1 TR 451 107a krüchten Radevormwald 236 251 TRM Bad Sooden- Kindelbrück 91 38 101 Radeburg 6 L74 REM- W 186 97 GLADBACH 229 TR LÜDEN- Pletten- Allendorf Schlotheim 176 Grimma 4 228 229 e 4 6 224 TR Wroclaw 230 46 SCHEID 511 r Störmthaler Ottendorf- ( ) 229 SCHEID berg 7 r 2 14 HILDEN 229 Waldeck a 249 Halver 450 T Greußen 86 Zwenkau See M 169 Okrilla Bautzen 46 TR TR L141 TR M 51 Bever- Verse- Finnentrop 55 Winterberg T 44 Eigenrieden 85 176 Freyburg u Zehren 57 483 249 249 Weißenfels 95 K8353 ld 59 TR talsperre 54 talsperre Eschwege Mühlhausen e 178 Rur TK LANGENFD. 237 Schmallen- 485 Hessisch 84 Kölleda Bad Bibra Rötha 221 TR SOLINGEN TRM 249 176 2 6 6 44 477 TK Hückeswagen berg Eder Lichtenau 27 87 Hainer See 107 Meißen 98 Wupper Kierspe 49 83 Rothenberga 180 Pegau Bad GÖRLITZ Heinsberg Erkelenz 1 Wermels- 237 Lennestadt Lenne 236 254 452 Uns Sömmerda Bischofs- DOR- 237 Ederstausee 7 Wanfried tr Straußfurt 250 Naumburg Lausick 96 9 59 Leich- 51 237 252 Treffurt u 176 Borna 101 werda Neukirch Berzdorfer 542 kirchen Biggesee 236 Fritzlar Melsungen 487 247 t 176 176 175 46 540 MAGEN lingen Wipper- Meinerz- Eckarts- 97 e See 61 Hallenberg Bad Wildungen 253 Waldkappel 87 Speicher 176 ( ) e Löbau 54 253 180 TR R 6 DRESDEN r 57 hagen berga 170 p fürth Kirchhundem Wabern TR 91 Borna Döbeln 6 98 F GREVEN- 59 TR M S u Gebesee Rommers- 55 236 Spangenberg 250 Bad Langensalza 107 Oppach 99 56 Große 253 l Hartha Gangelt BROICH LEVER- 480 d 2 169 6 kirchen Dhünntalsperre Frankenberg a Sontra 175 4 Berg- Bad Berleburg 7 Buttelstedt 175 173 221 57 TR KUSEN 506 GUMMERS- 55 517 253 485 176 180 Nossen TRM neustadt 3 254 9 Geithain 7 Ostritz 477 1 BACH 400 Camburg Froh- Rochlitz Wilsdruff 170 Geilen- 44 TR 59 8 3 Löhlbach Homberg 4 87 Meuselwitz 93 kirchen Bedburg 256 54 253 83 27 84 85 Zeitz burg 178 9 Engelskirchen Drols- Olpe der TR Creuzburg S177 BERGISCH 55a E TR Apolda ter 180 56 hagen Els 7 175 BERGHEIM KÖLN Hilchenbach Battenberg L3149 323 TR e 59 GLADBACH 55 252 247 WEIMAR eiß Pirna 96 K K 4 Jesberg W 173 lbe TR Behringen Wiederau E 55 TR 55a 54 62 Erndte- 236 Hainichen 178 56 55 51 508 TR 19 84 71 K35 88 Jülich brück Rotenburg 7 2 Altenburg 101 172a TK 253 Münchhausen Bebra 7 Umpferstedt 72 169 Bad Schandau 477 L124 Kreuztal 254 7 Eisenach 4 Alsdorf TR 4 265 Overath 107 17 56 R 56 256Wiehl- a TK 87 7 Eisenberg 172 Frechen 62 62 rr GOTHA Aldenhoven 559 talsperre 54 e Wutha 7 180 170 L496 Hürth T Sieg W 7 Penig Zittau 57 Kerpen 1 T 484 Much Bad Laasphe 3 ERFURT TR TR 4 R Biedenkopf 252 TR 7 175 Freiberg Dippoldiswalde TR TR 51 8 45 TRM Schwalmstadt 85 JENA Schmölln Heerlen TR n 3 Obergeis 27 4 Marksuhl 4 93 180 173 264 61 ei Waldbröl 4 TR Flöha 265 h TR Göttingen Bad Berka 88 R Brand-Erbisdorf 4 R TROIS- ( ) 169 264 507 478 SIEGEN Neukirchen 84 Waldenburg TR 264 59 56 Bad 324 Friedewald 173 F Eschweiler Wesseling DORF 62 62 19 247 TK M 4 TR Meerane Oederan r DÜREN Brühl 54 88 e 544 454 i R GERA b 1a 8 Siegburg( ) Lahn Hersfeld Friedrichroda Kranichfeld TRM 180 95 1 Erftstadt Wilns- 454 Ronneburg e 553 256 e CHEMNITZ Eibelshausen 253 62 Arnstadt l TR r TR M Kirchhain Neustadt Kieselbach g 477 555 dorf 453 62 a e Frauenstein Ohrdruf a AACHEN Stolberg TRM 478 62 3 r 254 S R M 264 57 Weiler- 560 Windeck Stadtallendorf 62 Barchfeld 92 Glauchau u 170 44 399 56 62 85 Kahla 175 174 180 lde Altenberg Praha 265 565 Sieg Betzdorf RM Crimmitschau 173 Rur swist Vacha 62 56 Wissen TRM 87 175 BONN Hennef Gladenbach Niederaula Bad 175 93 95 Clausnitz MARBURG Ohm Hürtgenwald Roth Kirtorf 101 R TR 84 Salzungen 175 180 72 Liège 565 Dillenburg 255 Alsfeld 62 Stadtilm Zschopau 256 88 88 2 Weida 56 562 277 62 285 19 258 Zülpich 8 Homberg 27 Neustadt 56 56 Haiger Ohra- Triptis 173 Lichtenstein Stoll- Burkhardts- 171 42 54 TR 180 174 565 277 Schmalkalden Talsperre talsperre berg dorf 56 Königs- 414 255 F TR Schmalwasser 92 Pockau 265 477 TR 5 u Geisa L1026 Bad ZWICKAU l Alten- TR d 399 winter TR 84 Dermbach TR Rudolstadt 281 93 a Blankenburg 266 9 kirchen 414 3 254 71 Olbernhau 266 Bad Flammersfeld 8 Herborn Rurtalsperre EUSKIRCHEN TR 414 L1118 Pößneck Talsperre Zeulen- 173 Thum Honnef 413 255 7 88 3 256 Hachenburg Lollar 49 Wasungen Ilmenau 85 Zeulenroda roda Greiz 95 171 Hünfeld 278 62 1 Höchstenbach Lauterbach Königsee 169 101 Marien- TRM Tann Talsperre 94 266 477 L3474 480 Kaltennordheim Zella-Mehlis Saalfeld Reichen- 93 berg 266 K 255 L3053 ( ) 85 Hohenwarte 94 L83 Linz TRM 45 27 bach Monschau 265 Kall Bad Rennerod TR Aue 174 573 61 277 TR M SUHL 9 92 72 Zw Annaberg- Schleiden Münstereifel 257 413 Flensungen 19 ic 399 266 8 429 49 49 Talsperre 281 Schneeberg k Buchholz Sinzig TR 480 Schleiz a 267 485 Leibis-Lichte u 266 49 Grünberg 73 Talsperre er 258 Bad 571 254 Meiningen Leutenberg M 9 49 457 Hilders Pöhl 94 u Nettersheim Altenahr Neuenahr- 42 R lde 101 Lahn GIESSEN 276 Reichmannsdorf Olef- Ahrweiler 256 282 169 Schwarzen- talsperre K 54 WETZLAR Syrau 283 berg 258 R 27 Probstzella Talsperre 413 255 275 Fladungen 169 265 412 NEU- 49 Lich FULDA 458 278 Talsperre Eibenstock Weilburg 5 Schotten Themar PLAUEN Blanken- 257 457 Bleiloch Auerbach BELGIEN TR WIED 8 Neuhaus TR Eibenstock heim TR 9 TR Schleusingen Kempenich Bendorf 3 Hungen 285 92 169 48 Ludwigs- Saale- 51 Andernach Rhe 42 455 276 Grebenhain 95 412 in 256 TRM 284 stadt 90 stausee Laacher K Montabaur 457 89 281 Falkenstein Oberwiesen- Müsch 489 279 Wurzbach See TR 279 Oelsnitz thal 258 Adenau 456 278 Bad 256 9 49 Butzbach Nidda Gersfeld TR TR 421 42 Mellrichstadt Eisfeld Lobenstein TRM Stadtkyll Limburg 455 85 W 256 Gedern TR 2 Scheid 262 KOBLENZ TK 279 Hildburghausen Werra TR e TR TR iß 258 416 Bad e 92 Saale 457 275 E Klingenthal Mayen 417 Diez Bischofsheim l 49 261 Nauheim 279 s 258 R Bad Ems Schlüchtern t 421 257 410 R 275 Sonneberg e TR TR 455 45 89 r Hillesheim Kelberg 260 260 Nassau Ortenberg 276 72 51 Boos 49 54 3 8 ( ) Usingen 2 265 262 Lahn Friedberg 275 73 Adorf 283 TR M Naila 173 410 Polch 327 9 Lahnstein 275 457 411 TRM Bad 275 455 Bad Bad Neustadt Stockheim HOF TR TR 3 4 TR Rhens Kirberg Camberg Büdingen Soden 66 Bad Königshofen 15 St. Vith Gerolstein Kaisersesch 456 Bad ale 93 Dock- 257 TK 260 417 521 Sa 410 274 Brückenau he 71 15 weiler Esch sc Wallenfels Prüm 421 1 48 521 TR i 416 45 k 279 85 173 Boppard BAD HOMBURG TR n 4 92 TK TR ä Rehau 275 286 r Kronach 3 F Münner- Karden 49 Brodenbach 42 Nidderau Coburg 303 Pronsfeld 51 Daun Nastätten Oberursel T Geln- 7 stadt 54 Idstein 8 Bad 457 Sonnefeld 289 259 49 hausen 303 Münch- 274 455 287 303 Bad Mosel T Vilbel 45 Kassel Seßlach 303 berg Brambach 410 257 9 260 Taunusstein Königstein Bad 303 Bad 521 Küps Selb St. Goarshausen FRANK- 5 Kissingen Marolds- Schwalbach 275 661 HANAU 289 173 Stadt- 289 Cochem St. Goar FURTTR weisach Lichten- steinach 421 455 66 85 Meisburg 417 L3005 3 286 fels 275 54 287 49 8 T 8 L3209 Flörsbach Hofheim 173 Burgkunstadt Eger 455 519 T 9 60 TK 43 289 289 327 WIES- TR Hofheim 66 648 44 43a Hammelburg 303 Dasburg Kyllburg Alf T T 45 Ebern W TK TR OFFEN- 276 e Kastellaun Bacharach BADEN 287 Poppen- Bad Main iß 42 262 54 40 3 e 303 53 Zell Blankenrath Lorch 260 T BACH 448 hausen Staffelstein r M 49 61 455 T 40 ain 303 421 66 43 TRM Frammersbach 19 286 4 Kulmbach 51 1 45 Gemünden SCHWEIN- 279 85 Bad 257 T T Drei- ( ) 49 Kinder- 3 303 303 FURT Berneck Wunsiedel Arzberg Wittlich beuern eich 3 9 42 TK 643 40 TR M 27 Bitburg Simmern Rüdes- 42a Rhein 671 519 Seligen- 26 73 Main 43 45 stadt Landscheid 421 heim 661 459 Röder- Goldbach 303 53 50 TR 60 MAINZ 43 Mör- R 50 486 70 o Waldsassen 40 43 felden mark t Marktredwitz 299 Zeltingen Ingelheim 486 Lohr 26a er TR 60 Langen 486 Arnstein 70 Breiten- M Bingen 8 26 a 51 Traben- Kirchberg 469 güßbach i Bernkastel-Kues Stromberg Gau-Algesheim RÜSSELS- 44 26 Werneck 85 n Mitterteich Trarbach Sohren TR TR 26 257 50 TRM 286 Eltmann Scheßlitz l HEIM 26 Weibersbrunn Karlstadt ose 50 5 3 ASCHAFFEN- M 9 67 45 26 48 ( ) Main 22 TSCHECHISCHE Groß-Gerau BURG TRM 26 Hollfeld 422 53 42 DARM- TR M F Bad Kreuznach 44 Dieburg Gerolzhofen BAYREUTH i Bergtheim c Tirschenreuth Schweich T M Neumagen 269 41 STADT Groß- h Gries- 672 a t 418 53 421 63 26 469 BAMBERG el i Kemnath n 327 heim 45 Umstadt TRM n 27 a dnaab 51 Wörrstadt 26 7 19 22 ab al 26 Obern- Stegaurach R W 41 428 420 Oppen- 3 burg e Erbendorf LUXEM- 38 gn 22 Roßdorf 3 Ebrach i e 420 heim 449 Main Marktheidenfeld t 299 TR TR 53 Morbach ah 426 z Creußen 15 602 N TRM 426 WÜRZ- Thalfang 327 Kirn Stock- TR Prichsenstadt 49 41 48 426 426 64 269 stadt Reinheim BURG 22 Hirschaid 2 422 Wörth Schwarzach 22 286 Burgebrach 299 TRIER 420 44 426 Pfungstadt Höchst 8 Ebermann- 418 19 8 TR Wiesentheid 505 470 Luxembourg 49 1 stadt Pressath 22 93 51 Hochstätten 45 468 27 73 Kirchen- Obermoschel 9 TR TR Schlüsselfeld( ) ( ) thumbach TR 41 Alzey Gernsheim 38 TR 3 Gößweinstein TRM TR 3 Bad König Wertheim TR M 470 Konz 469 TR M Eschenbach Neustadt Wellen Idar-Oberstein 420 TR Kitzingen Pegnitz 268 270 Rhein BURG 51 52 327 269 41 47 8 Höch- 470 TRM REPUBLIK Kirchheimbolanden 271 61 44 67 TR 47 Iphofen stadt 85 Auerbach Grafenwöhr Hermeskeil Nahe Michel- 286 470 Forchheim 419 Saarburg Lauterecken L425 stadt Miltenberg Zerf Birkenfeld 48 Bür- Bensheim 41 Mons- WORMS Lindenfels 9 Weiden 407 stadt T Hai Rockenhausen 47 heim TR Erbach a de Waidhaus 407 41a TR 47 u Giebelstadt n 47 TR 460 b Gräfenberg 2 a 22 Trassem 420 TR e a 47 460 470 b Vohenstrauß Plzen 47 45 Tauber- r Ochsenfurt 299 5 3 Amorbach 13 Wolfstein 63 47 Lampertheim bischofsheim TK 8 62 38 47 TK 19 407 51 271 27 Grünstadt 6 406 41 Kusel Winnweiler 47 9 44 85 Beerfelden Walldürn 290 Wern- 419 Losheim 420 270 TR TR TR 6 T Weinheim ERLANGEN berg 423 Lauda- -Königs- Neustadt 268 48 hofen Uffenheim 14 Perl Enkenbach- MANNHEIM 659 Buchen St. Wendel 6 271 Frankenthal 45 470 Hirschau Alsenborn 9 44 T Lauf 14 Luxembourg Merzig 269 3 7 Bad 8 Langen- 73 Hersbruck 420 13 Windsheim zenn FÜRTH LUDWIGSHAFEN 38 290 Bad Dürkheim 38a 292 14 Sulzbach 650 Boxberg TR 4 3 8 Saar 269 41 37 Eberbach Creglingen 8 2 14 Oberviechtach 423 36 27 Bad 299 Lebach 37 37 Weikersheim Tauber TR G108 4R 10 Mergentheim 470 85 1 Ottweiler Hoch- Weidenthal 44 6 HEIDELBERG Amberg Nabburg 269 G109 4R 268 6 Landstuhl speyer 656 N Osterburken Dillingen NEUN- TRM 271 TR eckar 81 TR KAISERS- TR 9 36 37 Reichels- NÜRNBERG 85 KIRCHEN 65 37 hofen Oberdachstetten N4 4 22 LAUTERN 3 292 Stein 8 Naab 8 39 535 TR 6 41 Schwetzingen Dörzbach 6 535 Neckargemünd Elztal Adelsheim 73 Saar- 268 TR Homburg 38 13 Altdorf Kastl 61 TR Leimen Feucht 405 louis 270 36 291 TR Rothenburg ( ) T 423 2 299 85 623 290 14 TR M TRM T St. Ingbert Speyer Mosbach Heilsbronn Rötz Völk- NEUSTADT Meckesheim Schwabach M 51 62 48 Hocken- TR TK Furth Walldorf TK 19 a lingen 39 i 8 41 39 heim TR n Schwandorf 620 45 A - TRM 6 gst D

a Rednitz 269 51 268 Zweibrücken 3 292 J Blaufelden lt 299 SAAR- Gundelsheim cher m o Blieskastel 9 Wiesloch Ko n TR TR ü Ansbach TR a BRÜCKEN Künzelsau h u 20 l 2 - 13 K TR F a Neumarkt TK 8 Sinsheim 27 ränki 93 85 272 39 sc n TR ( 6) Germersheim TR TR he a Bad l Cham 423 424 10 h 6 Re 466 Roth Bruck T K 10 10 ueic 39 Friedrichshall Jagst za Weiß Metz 406 Hornbach Q 292 TR t er Regen Annweiler 6 TR TR Pirmasens 35 Östringen 85 Landau Bad Neckar- Merkendorf Regen 51 48 5 ( ) sulm 3 Burg- Nittenau Graben- Schönborn Öhringen TR lengenfeld Roding Dahn Neudorf TRM 3 39 16 TR M 14 6 Feuchtwangen 13 299 38 35a HEILBRONN Großer 427 293 39 19 Brombachsee 65 9 35 Bruchsal Crailsheim Bayerisch Eppingen 39a Altmühlsee Regenstauf Bad Bergzabern 427 Rhein 36 Schwäbisch Hall Gunzenhausen 2 Eisenstein Löwenstein Berching 85 38 Wörth 35 293 Lauffen 39 Pleinfeld 15 20 Kandel 3 14 Dinkelsbühl Viechtach S Schweigen- 13 Ellingen chw 11 Rechtenbach 9 arz 27 er Zwiesel REGENSBURG R KARLS- Bretten 19 TK e 10 N TR Weißenburg TR g 9 e e TR Donau RUHE Ascha n 293 c Beilngries Besigheim k 25 a 81 290 r Gaildorf 3 35 L605 294 14 3 Regen TR Pfatter Mühlacker Bietigheim- TR 466 36 10 Ellwangen Altm TR TR Bissingen ü hl 11 Durmers- Backnang 8 Ettlingen 10 Oettingen 85 heim 3 Vaihingen Kelheim TR TR LUDWIGSBURG Gschwend Bad 19 299 Straubing Philippsreut 5 Eichstätt Altmannstein Abbach 20 Grafenau PFORZ- 27 Kocher Altm 10 298 2 ühl HEIM 533 12 Rastatt Korn- Winnenden 29 16 Deggendorf 533 8 westheim 13 Schierling 27a 14 AALEN 29 TRM 462 TR Schönberg 500 Neuenbürg 10 Bopfingen TK Abensberg 8 Gaggenau 294 Fellbach 7 Nördlingen 15 Freyung 463 29 Monheim 3 Leonberg 295 29 Plattling 14 WAIB- Rems Lorch INGOLSTADT BADEN- Neustadt Gernsbach STUTTGART LINGEN Schorndorf 29 25 20 BADEN L1180 16 299 Ilz SCHWÄBISCH 466 16a 15 TR Murg GMÜND Vohburg Bad Wildbad 295 ESSLINGEN 19 301 Lichtenau 500 296 14 N 297 Neuburg Donau 85 464 TR 831 eck Neresheim Neufahrn Pilsting 27 10 ar 16 93 Calw Sindelfingen Plochingen GÖPPINGEN Isar 12 Forbach Ostfildern Uhingen Donauwörth Osterhofen Bühl 16 300 3 Böblingen TR 10 Eislingen Böhmenkirch 2 13 8 Donau 3 313 92 Landau Wendlingen 10 Pfeffenhausen Rhein Leinfelden- 27 Filder- 466 Rain Geisenfeld Vilshofen Achern 296 312 stadt 297 Reichertshofen Echterdingen 313 Mainburg 15 Wegscheid 462 Simmersfeld 463 HEIDENHEIM Kirchheim 19 300 299 Wörth Dingolfing PASSAU 294 TR 297 Geislingen TR TR 5 14 TR 297 388 Renchen Schönmünzach Altensteig 464 Nürtingen 8 16 Donau TR 27 Herbrechtingen Simbach 12 TR 466 Lech 28 28 TR TR 492 Dillingen 15 Vils Appenweier Herren- Donau Kehl berg 297 313 Schrobenhausen FRANK- Oberkirch 312 TR Au 20 500 81 28 465 LANDSHUT TÜBINGEN Lauingen Meitingen TRM 3 28 Nagold 27 Metzingen 10 300 Pfaffenhofen 33 TR Gundelfingen 301 28 Baiersbronn 463 464 28 OFFEN- 28a 2 33a 28a 28 REUT- Bad Urach Langenau 16 11 299 512 BURG 14 28a L370 28 Laichingen Moosburg Bad Birnbach 28 LINGEN 9 Pfarrkirchen 28a Vilsbiburg 388 Neu- -Griesbach Freuden- Horb Neckar Günzburg Aichach 388 haus L370 28 r stadt Pfullingen mpe Bad Peterstal- 32 13 A Gangkofen K Blau- Burgau 15 Eggenfelden in 312 92 Pocking z beuren ULM i TR 33 g 27 Münsingen 28 10 Isar 3 TRM TR Hohenkammer Linz 12 Engstingen 10 8 TRM Freising 299 Sulz Schelklingen 465 NEU- Zusmars- 17 388 Neumarkt- 36 Biberach 294 Hechingen 2 300 Taufkirchen Velden Inn hausen St. Veit 463 ULM Malching Lahr 415 492 28 TR Alpirsbach 300 588 TR TR Erbach Friedberg 20 94 33 Wolfach 27 313 16 AUGS- Erding 12 294 311 BURG Balingen Senden 301 Dorfen Simbach Schiltach Oberndorf 32 300 Haslach Hausach 17 9 388 Ampfing REICH Iller 94 92 13 12 312 Ehingen Thannhausen 8 15 Mühldorf Schramberg Dachau 12 30 h Mering 471 388 Altötting Hornberg Zwiefalten c TR 81 Gammertingen D u a 3 ona Illertissen rt 471 12 462 Ebingen 311 Krumbach e 2 Ismaning 20 Laupheim W 33 TR 27 Schömberg 471 465 471 99 Speichersee Elzach 463 13 99 Burghausen 32 Fürsten- 304 St2331 Rottweil 7 Lech Emmendingen Kloster- feldbruck Hohenlinden Inn 5 294 Triberg Veringenstadt lechfeld Haag St. Georgen K MÜNCHEN 2R 299 2 94 Forstinning 12 33 Riedlingen Babenhausen 2 471 99 n 27 312 i TR TR 16 Waldkirch 17 er TR TR 15 e p 304 h 300 Am 2R

R -SCHWENNINGEN 14 Sigmaringen Ertingen 96 Zorneding 294 523 Spaichingen 471 Breisach L114 TK Furtwangen Biberach Ochsenhausen 304 31 311 VILLINGEN- 32 TR Inning 13 Tittmoning 33 27 312 99 Ebersberg Salzach 500 Breg 95 523 313 Mindelheim TR Wörthsee 8 Wasser- 3 Buchloe Bad Dürrheim Mengen burg Trostberg 31a Brigach au Bad Saulgau 30 Landsberg FREIBURG n 995 304 20 31 Do 311 96 864 Starnberg 11 TR 31 523 Ammer- 952 R Tuttlingen 311 27 Meßkirch Oberessendorf see Traunreut 14 Donaueschingen Memmingen Isar 15 Bad Krozingen -Neustadt 311 Aitrach 304 Waginger Laufen Hüfingen 31 TR TR ÖSTER- 31 313 Bad Waldsee 465 See Hinterzarten Titisee- 491 14 32 Wolfratshausen 3 Geisingen Bad Wurzach 12 17 2 31 81 16 r TR TR 317 e 11 Chiemsee Traunstein Löffingen m Holzkirchen Teisen- 27 Kaufbeuren 8 ROSEN- TRM 30 m 304 dorf TRM A HEIM Simssee Freilassing Todtnau Stockach TR Starnberger TR 306 378 Müllheim 315 Blumberg Engen Iller K 304 Bonndorf TR See Geretsried 318 15 15 8 Leutkirch Weilheim Schluchsee 98 31 Weingarten TRM TK Bernau TR 20 5 314 Schongau Peißenberg 11 472 Siegsdorf 27 Peiting 13 Miesbach 306 31 472 95 305 Grassau 313 TK Mulhouse 317 81 33 12 472 Inzell 314 Überlingen 30 Ravensburg 7 472 T 500 96 TR Penzberg 472 318 Marquartstein Bad 21 472 Marktoberdorf Ruhpolding Salzburg Stühlingen Gmund Singen Radolf- 12 h Reichenhall 5 315 34 31 c 472 Bad Tölz zell 33 e 307 Inn 307 Zell Markdorf L 472 Bad Brannenburg 305 Mecken- 32 Heilbrunn Tegernsee Tegernsee 3 33 Meers- beuren KEMPTEN 93 305 Schneizlreuth 21 467 Staffelsee 2 Bad Wiessee burg 19 21 305 314 h 318 307 Reit Schopfheim KONSTANZ FRIEDRICHS- Wangen Isny c Steingaden 20 30 980 a 16 Rottach- HAFEN Tettnang t 23 33 12 r 17 Kochel- Lenggries 317 -Tiengen 31 e Egern RO52 98 Wehr 98 31 32 Murnau Bayrischzell Waldshut- 33 W see 13 999 Jestetten 12 Saulgrub Berchtesgaden 319 Lörrach 518 34 27 Kochel 532 Weil 31 Forggen- Kreuth Bodensee 19 Kiefersfelden ( ) 305 REICH 34 500 Oy see Urfeld 20 34 32 Ober- 307 TRM TR M TR TR 316 Bad 12 Lindenberg 7 Königssee 98 Rh Kress- ammergau Walchensee Basel 861 Säckingen ein 11 Kufstein Rhein- bronn 31 308 Sylvenstein- 34 felden Alpsee 310 310 Walchensee Königssee mer see Pratteln 518 Lindau Ober- Am 2 12 staufen 307 Rheinfelden 308 Füssen Oberau Immenstadt Reutte Vorderriß Bregenz Isar Sonthofen Garmisch- Krün 308 Partenkirchen Bad 2 Hindelang 23

19 Mittenwald SCHWEIZ 2 © 2007-2017 Patrick Scholl r Oberstdorf le l I Maßstab (im Druck) 1 : 600 000 0 10 20 30 40 50 60 km

© 2007-2017 Patrick Scholl

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 54 • ”stau Radio Station Name” or • ”stau TV Station Name” • This means they are drivers preparing to inject themselves into the traffic • Combined with IP geolocation we have their complete itinerary. • Google searches are an open ledger of upcoming itineraries.

Traffic Jam (stau) searches

• These searches take the form ”stau Highway Number” or

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 55 • ”stau TV Station Name” • This means they are drivers preparing to inject themselves into the traffic • Combined with IP geolocation we have their complete itinerary. • Google searches are an open ledger of upcoming itineraries.

Traffic Jam (stau) searches

• These searches take the form ”stau Highway Number” or • ”stau Radio Station Name” or

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 55 • This means they are drivers preparing to inject themselves into the traffic • Combined with IP geolocation we have their complete itinerary. • Google searches are an open ledger of upcoming itineraries.

Traffic Jam (stau) searches

• These searches take the form ”stau Highway Number” or • ”stau Radio Station Name” or • ”stau TV Station Name”

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 55 • Combined with IP geolocation we have their complete itinerary. • Google searches are an open ledger of upcoming itineraries.

Traffic Jam (stau) searches

• These searches take the form ”stau Highway Number” or • ”stau Radio Station Name” or • ”stau TV Station Name” • This means they are drivers preparing to inject themselves into the traffic

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 55 • Google searches are an open ledger of upcoming itineraries.

Traffic Jam (stau) searches

• These searches take the form ”stau Highway Number” or • ”stau Radio Station Name” or • ”stau TV Station Name” • This means they are drivers preparing to inject themselves into the traffic • Combined with IP geolocation we have their complete itinerary.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 55 Traffic Jam (stau) searches

• These searches take the form ”stau Highway Number” or • ”stau Radio Station Name” or • ”stau TV Station Name” • This means they are drivers preparing to inject themselves into the traffic • Combined with IP geolocation we have their complete itinerary. • Google searches are an open ledger of upcoming itineraries.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 55 Traffic Jams

Adverse road conditions such as traffic congestion are known to cause a host of undesired side effects. These vary from increased pollution (e.g. gas emissions), energy waste, additional transportation and production costs to waste of labor and delays in product deliveries. Traffic congestion is also known to impact public health negatively, to cause stress and road rage and to even affect the unborn. If we thought of a city, a region, a country or any other social unit as a large living organism road traffic would be one of its circadian rhythms and traffic jams would be an obstruction to its entrainment. It is hence not surprising that obstructing the smooth flow of traffic sends ripples of negative effects deep into many aspects of socioeconomic life. Understanding, forecasting and preventing adverse road conditions is therefore important for the benefit of the drivers but also obviously for economic reasons.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 56 Forecasting Road Conditions two hour in advance

Since I did not have a target variable to predict, in order to demonstrate the informational value of these data, I created one by programmatically harvesting the number of traffic jams every 5 minutes as reported by the German automobile club ADAC on their website. I did this for several months and I also collected the hourly stau searches for the same time period. I then wrote down a model which included day of the week and hour of day fixed effects plus the Google data. The main result of the paper is that after controlling for time-of-day and day-of-week effects we can still explain a significant additional portion of the variation of the number of traffic jam reports with Google Trends and we can thus explain well over 80% of the variation of road conditions using Google search activity. A one percent increase in Google stau searches implies a .4 percent increase of traffic jams.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 57 Regression Results

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 58 Toll Index The MAUT system

Starting in 2007 long vehicles on German highways pay Toll fees in an automated fashion using a system called MAUT. Each truck has an On Board Unit which keeps track of kilometres travelled by means of GPS and dumps the data on a central server over the air (G3 network) every three days or shortly before exiting the country. The data is deleted as mandated by the corresponding law a couple of months after the bill is settled. This is very unfortunate but some data is rescued in reports published monthly by the BAG (Federal Bureau of Goods Transport).

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 59 • RESULT ([Askitas and Zimmermann, 2013]): Forecast GIP a month in advance of the first preliminary estimate by the German Federal Ministry for Economic Affairs and Energy. • NOTE: In reality forecast may be effectively more than 3 months in advance since GIP numbers are revised while the MAUT numbers need no revision. • We call the resulting index Toll Index (TI).

The MAUT data

• IDEA: Leverage transportation data to forecast German Industrial Production (GIP) exploiting that due to the prevalence of just-in-time-delivery transportation and production activities tend to be coincident.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 60 • NOTE: In reality forecast may be effectively more than 3 months in advance since GIP numbers are revised while the MAUT numbers need no revision. • We call the resulting index Toll Index (TI).

The MAUT data

• IDEA: Leverage transportation data to forecast German Industrial Production (GIP) exploiting that due to the prevalence of just-in-time-delivery transportation and production activities tend to be coincident. • RESULT ([Askitas and Zimmermann, 2013]): Forecast GIP a month in advance of the first preliminary estimate by the German Federal Ministry for Economic Affairs and Energy.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 60 • We call the resulting index Toll Index (TI).

The MAUT data

• IDEA: Leverage transportation data to forecast German Industrial Production (GIP) exploiting that due to the prevalence of just-in-time-delivery transportation and production activities tend to be coincident. • RESULT ([Askitas and Zimmermann, 2013]): Forecast GIP a month in advance of the first preliminary estimate by the German Federal Ministry for Economic Affairs and Energy. • NOTE: In reality forecast may be effectively more than 3 months in advance since GIP numbers are revised while the MAUT numbers need no revision.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 60 The MAUT data

• IDEA: Leverage transportation data to forecast German Industrial Production (GIP) exploiting that due to the prevalence of just-in-time-delivery transportation and production activities tend to be coincident. • RESULT ([Askitas and Zimmermann, 2013]): Forecast GIP a month in advance of the first preliminary estimate by the German Federal Ministry for Economic Affairs and Energy. • NOTE: In reality forecast may be effectively more than 3 months in advance since GIP numbers are revised while the MAUT numbers need no revision. • We call the resulting index Toll Index (TI).

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 60 Models

AR model: . reg S1.ln(GIP) i.MOY L1.S1.ln(GIP) MAUT model: . reg S1.ln(GIP) i.MOY S1.ln(TI)

The MAUT model raises the Adj. R2 from .83 to .93 and the prediction reduces the MSE from .022 to .009 (reduction of avg pct error from 2.37 percent to 1.57) pointing in the right direction 87 percent of the time.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 61 Plots 6 5.5 5 4.5 4 MAUT forecast GIP

date .1 .05 0 -.05

Benchmark model MAUT model -.1 date

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 62 Operationalisation: Toll Index published on www.askitas.com

Figure 2: On November 14, the German Federal Statistical Office announced GDP growth of .8% in 3Q2017, on track for well over 2% growth for the year.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 63 A couple of thoughts • Official vs Internet Data: Typically we seek to validate the latter using the former. This might some times be not useful. Think of official data on hospital visits vs symptoms Google searches in an economic downturn. • Data Tax: Companies whose business models involve recording personal data of masses of people (Social Media, Credit Cards, online market operators, etc) could be taxed in the form of to make aggregate forms of their data available, in a form suitable for policy making, very much in the spirit of what Google does with Google Trends. • Data Lobbying: Researchers who work in data based policy advice ought to successfully communicate to policy makers that successful forecasting does not start with the forecaster but with the policy maker: every project financed by taxpayer money needs to have a data retention and reuse plan built in.

It’s all about data!

• Regulatory arbitrage: In 2014 the CEO of the German Telekom (built the MAUT system in 2007) pointed out that by not retaining MAUT data the government wasted a chance: many applications e.g. road conditions. While German data protection law imposes severe restrictions on data retention (history...) in an effort to protect individual privacy, Germans offer their personal data to Google, Facebook etc. voluntarily which then export the data to US based data refineries, selling the resulting data based services back to German companies (e.g. Google Traffic).

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 64 • Data Tax: Companies whose business models involve recording personal data of masses of people (Social Media, Credit Cards, online market operators, etc) could be taxed in the form of having to make aggregate forms of their data available, in a form suitable for policy making, very much in the spirit of what Google does with Google Trends. • Data Lobbying: Researchers who work in data based policy advice ought to successfully communicate to policy makers that successful forecasting does not start with the forecaster but with the policy maker: every project financed by taxpayer money needs to have a data retention and reuse plan built in.

It’s all about data!

• Regulatory arbitrage: In 2014 the CEO of the German Telekom (built the MAUT system in 2007) pointed out that by not retaining MAUT data the government wasted a chance: many applications e.g. road conditions. While German data protection law imposes severe restrictions on data retention (history...) in an effort to protect individual privacy, Germans offer their personal data to Google, Facebook etc. voluntarily which then export the data to US based data refineries, selling the resulting data based services back to German companies (e.g. Google Traffic). • Official vs Internet Data: Typically we seek to validate the latter using the former. This might some times be not useful. Think of official data on hospital visits vs symptoms Google searches in an economic downturn.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 64 • Data Lobbying: Researchers who work in data based policy advice ought to successfully communicate to policy makers that successful forecasting does not start with the forecaster but with the policy maker: every project financed by taxpayer money needs to have a data retention and reuse plan built in.

It’s all about data!

• Regulatory arbitrage: In 2014 the CEO of the German Telekom (built the MAUT system in 2007) pointed out that by not retaining MAUT data the government wasted a chance: many applications e.g. road conditions. While German data protection law imposes severe restrictions on data retention (history...) in an effort to protect individual privacy, Germans offer their personal data to Google, Facebook etc. voluntarily which then export the data to US based data refineries, selling the resulting data based services back to German companies (e.g. Google Traffic). • Official vs Internet Data: Typically we seek to validate the latter using the former. This might some times be not useful. Think of official data on hospital visits vs symptoms Google searches in an economic downturn. • Data Tax: Companies whose business models involve recording personal data of masses of people (Social Media, Credit Cards, online market operators, etc) could be taxed in the form of having to make aggregate forms of their data available, in a form suitable for policy making, very much in the spirit of what Google does with Google Trends.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 64 It’s all about data!

• Regulatory arbitrage: In 2014 the CEO of the German Telekom (built the MAUT system in 2007) pointed out that by not retaining MAUT data the government wasted a chance: many applications e.g. road conditions. While German data protection law imposes severe restrictions on data retention (history...) in an effort to protect individual privacy, Germans offer their personal data to Google, Facebook etc. voluntarily which then export the data to US based data refineries, selling the resulting data based services back to German companies (e.g. Google Traffic). • Official vs Internet Data: Typically we seek to validate the latter using the former. This might some times be not useful. Think of official data on hospital visits vs symptoms Google searches in an economic downturn. • Data Tax: Companies whose business models involve recording personal data of masses of people (Social Media, Credit Cards, online market operators, etc) could be taxed in the form of having to make aggregate forms of their data available, in a form suitable for policy making, very much in the spirit of what Google does with Google Trends. • Data Lobbying: Researchers who work in data based policy advice ought to successfully communicate to policy makers that successful forecasting does not start with the forecaster but with the policy maker: every project financed by taxpayer money needs to have a data retention and reuse plan built in. Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 64 Thank your for your attention Questions?

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 64 ReferencesI

Askitas, N. (2015a). Calling the greek referendum on the nose with google trends. IZA Discussion Paper Series No 9569. Askitas, N. (2015b). Google search activity data and breaking trends. IZA World of Labor. Askitas, N. (2015c). Predicting the irish ”gay marriage” referendum. IZA Discussion Paper Series No 9570. Askitas, N. (2016a). Predicting road conditions with internet search. PLoS ONE11(8): e0162080. https://doi.org/10.1371/journal.pone.0162080.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 65 ReferencesII

Askitas, N. (2016b). Predicting the irish ”gay marriage” referendum. Cityscape: A Journal of Policy Development and Research, 18(2):165–178. Askitas, N. (2018). On the interplay of forgetting and remembering. Thouvenin, Florent, Peter Hettich, Herbert Burkert and Urs Gasser (Eds.): Remembering and Forgetting in the Digital Age: An interdisciplinary approach to a complex phenomenon, Springer-Verlag, Heidelberg,. Askitas, N. and Zimmermann, K. F. (2013). Nowcasting business cycles using toll data. Journal of Forecasting, 32(4):299–306.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 66 ReferencesIII

Askitas, N. and Zimmermann, K. F. (2015a). Health and well-being in the great recession. International Journal of Manpower, 36(1):26–47. Askitas, N. and Zimmermann, K. F. (2015b). The internet as a data source for advancement in social sciences. International Journal of Manpower, 36(1):2–12. Black, F. and Scholes, M. (1973). The pricing of options and corporate liabilities. Journal of political economy, 81(3):637–654.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 67 ReferencesIV

Ferraro, F., Pfeffer, J., and Sutton, R. I. (2005). Economics language and assumptions: How theories can become self-fulfilling. Academy of Management Review, 30(1):8–24. Hoel, E. P., Albantakis, L., and Tononi, G. (2013). Quantifying causal emergence shows that macro can beat micro. Proceedings of the National Academy of Sciences, 110(49):19790–19795. MacKenzie, D. and Millo, Y. (2003). Constructing a market, performing theory: The historical sociology of a financial derivatives exchange. American journal of sociology, 109(1):107–145.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 68 ReferencesV

Merton, R. C. (1973). Theory of rational option pricing. The Bell Journal of economics and management science, pages 141–183. Tishby, N., Pereira, F. C., and Bialek, W. (2000). The information bottleneck method. arXiv preprint physics/0004057.

Nikos Askitas | IDSC – Research Data Center of IZA – Institute of Labor Economics 69