Big Data, Innovations and New Business Models Predicting Human Behaviour with

Tobias Preis

@t_preis [email protected] Our team

Tobias Preis Steven Bishop

Philip Treleaven Suzy Moat Nick Chater A map of the world built only from GPS locations of Flickr photos Preis & Moat (2013)

Internet data opens up opportunities for new dynamic, adaptive economic models Data Revolution OutperformanceGoogle Trading Strategy for Keyword “debt”

Buy and Hold Strategy Trends Strategy 300 μ+σ, μ, μ−σ of Random Strategy 326 200 ‘debt’ 100 0 16 2004 2005 2006 2007 2008 2009 2010 2011 Pro f t and Loss [%] Time, t [Years]

Preis, Moat & Stanley, Scientifc Reports 3, 1684 (2013) http://www.nature.com/srep/2013/130425/srep01684/full/srep01684.html

Covered by QuantifyingWikipedia: Patterns Financial topics 1.00

Wikipedia Edits 0.75 Financial Topics Views Financial Topics 0.50 Density Random 0.25 Strategy Views strategies proftable 0.00 −2 0 2 Return [Std. Dev. of Random Strategies]

Moat, Curme, Avakian, Kenett, Stanley & Preis, Scientifc Reports 3, 1801 (2013) http://www.nature.com/srep/2013/130508/srep01801/full/srep01801.html

Covered by QuantifyingHourly Flickr Patternsdata and Hurricane Sandy 0.10 A 0.08 0.06 Landfall of Hurricane Sandy 0.04 Flickr Photos with Hurricane Related Text 0.02 0.00 20 Oct 25 Oct 30 Oct 05 Nov 10 Nov 15 Nov 20 Nov Normalised Number of Photos Time

B Landfall of Hurricane Sandy 1020 1000 Averaged Pressure in 980 US State New Jersey 960

20 Oct 25 Oct 30 Oct 05 Nov 10 Nov 15 Nov 20 Nov

Atmospheric Pressure [mbar] Time Preis, Moat, Bishop, Treleaven & Stanley, Scientifc Reports 3, 3141 (2013) http://www.nature.com/srep/2013/131105/srep03141/full/srep03141.html

Covered by What Google Knows About Houses House Price Forecast based on Data Outperformance Training Period Out-of-Sample Forecast 190 ● ●● ●● ● ● ● ●● ● ●● ● 180 ● ●● ● ● ●●● ● ● ● ● ●● ● ● ●● ● ● ●● ● ●● ● ● ● ●● x ● ● ● ●● ● ● ●● ● ● ●● e ● 170 ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ●● ●

ice Ind ● ●

r ● 160 ● ● ● ●● ● ● ●●● ●●● ● ● ● ●● ●● Predicted Value 150 ●● ●● ● ● Observed Value House P ● ● 80% Prediction Interval 140 ● 95% Prediction Interval ● ●●

2004 2005 2006 2007 2008 2009 2010 2011 2012 2013

Preis, Bishop, Stanley & Moat, Under Review House Price Forecast based on Google Trends Data

Outperformance Baseline Advanced IN-SAMPLE FORECAST MAE 1.48 1.37 4 RMSE 1.89 1.75

2

0

−2 orecast Error [%] F 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013

Baseline Advanced OUT-OF-SAMPLE FORECAST MAE 1.74 1.66 4 RMSE 2.09 1.99

2

0

−2 orecast Error [%] F 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 Time [Months] Preis, Bishop, Stanley & Moat, Under Review Summary Home Summary 1. Large technological systems have become a central part of our society.

2. These systems form a world wide sensor network, measuring human behaviour on a massive scale.

3. We can use this data to: • measure what is happening in our world right now • anticipate what will happen in our world later