Back to overview

Households' Risk through Deep Data Analysis: Mobility, Credit, Consumption, Mortality, and Voting

English title Households' Risk through Deep Data Analysis: Mobility, Credit, Consumption, Mortality, and Voting
Applicant De Giorgi Giacomo
Number 182243
Funding scheme Project funding (Div. I-III)
Research institution Département d'économie Université de Genève
Institution of higher education University of Geneva - GE
Main discipline Economics
Start/End 01.12.2018 - 30.11.2022
Approved amount 504'661.00
Show all

Keywords (5)

Credit; Gentrification and De-Grentrification; Delinquency; Consumer behaviour; Big and Deep Data

Lay Summary (Italian)

Il mercato immobiliare e del credito sono di grande rilevanza nel determinare il comportamento di individui e famiglie. Attraverso l’uso di dati amministrativi sul comportamento nel mercato del credito, ci proponiamo di capire in che modo le fluttuazioni del mercato immobiliare e del credito influiscano su varie dimensioni del comportamento delle famiglie e individui.
Lay summary
Il progetto si focalizza sull’analisi del comportamento delle famiglie a fronte dell’incertezza nel mercato del lavoro, del credito, e sullo stato di salute.
Ci chiediamo come le famiglie rispondano a shocks dei prezzi immobiliari, e quali siano le traiettorie di mobilità a seguito di questi shocks. Allo stesso tempo studiamo se, e in che misura, l’incertezza nell’accesso al credito, così come, il rischio di bancarotta e insolvenza possano determinare comportamenti di voto e facilitare l’emergere di partiti e candidati populisti. Proporremo dei contributi metodologici per la predizione dello stato di salute e comportamento sul mercato del credito attraverso l’uso di big or deep data.
Direct link to Lay Summary Last update: 22.11.2018

Responsible applicant and co-applicants


Project partner


The current project aims at investigating three main topics of research utilizing the wealth of information available through the novel big (or deep) data on consumers’ credit behaviour provided by Experian. In a nutshell, the Experian data cover all credit transactions through credit reports at the monthly frequency from 2004 to this day, so that individuals, and households are followed overtime. The cross-sectional coverage is virtually the entire adult US population. The intent is that of establishing a unique research agenda, and research group able to tackle several issues concerning consumers’ behaviour for the years to come. The acquisition of the complete data, ultimately every credit transaction of the entire US population is recorded in such data, will allow to establish a unique research group and research agenda which would make us the leaders in Europe in these areas of research. To my knowledge very few researchers around the world have access to such wealth of information, I had access to similar data in the past as a Senior Economist at the NYFed and the Federal Reserve System, one of the very few institutions worldwide that has access to similar data, so that I am fully aware of the research potential they can unlock. Acquiring these data will give us the ability to lead the way in Europe and will make the research group a focal point for the Worldwide research community. The possibility of research given by the data are seemingly endless, while in this proposal we focus on three closely related topics, those are just a tiny subset of what is possible with the brainpower and resources we can put together overtime. To assure the success of such an ambitious project the current proposal includes the request for funding for computing equipment as well as personnel that might look a bit out of the standard requests. Let me reiterate that the data provided by Experian cover virtually the entire population of the US, potentially at the monthly frequency with thousands of transaction attributes, this means that the complete database is quite large, in the order of tens of terabytes. Handling such a vast wealth of data requires appropriate computing power as well as personnel. To be clear, while the Experian data focus on recording credit transactions, delinquencies, and so on, they also record major life events such as mobility, family formation, death. As the data have a very detailed geographic identifier (precise address) and individual social security number, it is obvious that we can merge them with other data sources and potentially with other administrative data if we receive the due authorizations to do so.In the current proposal, we will develop three closely related areas of research:1.Credit and Default behaviour in the wake of the Great Recession;2.The effects of “Gentrification” and “De-Gentrification” on consumers;3.Big (Deep) Data and their contribution to the analysis of credit behaviour.In the first line of research we will investigate the relationship between borrowing behaviour, house price dynamics, and delinquencies in the wake of the great recession. The current narrative, mostly based on work by Mian and Sufi (2009, 2015), points towards the negative effects of too lax borrowing criteria which have generated a large amount of bad loans opened by sub-prime or lower credit score borrowers. Those loans have then determined the large financial crisis and the great recession due to the changes in the housing price dynamics. However, a closer look at the actual borrowing behaviour through individual data will show that the old narrative is, at least, incomplete. That older narrative misses the fundamental life-cycle dynamics of borrowing behaviour. Our initial results, based on data that were accessible to this researcher as an employee of the Federal Reserve, show that in fact it is younger and better borrowers we should look into to explain the large growth in credit between 2003 and 2007 and not the subprime borrowers. These facts are crucial in understanding the nature of the financial crisis and subsequent great recession and therefore essential for the formulation of policy responses. Clearly the house price dynamics, housing supply, credit, and labour market conditions should be jointly investigated to fully understand the drivers of consumers’ behaviour and delinquencies. In this view, in the second research line, we then analyse the effects of large increase or fall in house prices within a local area (commuting zone or zip code) and their crucial consequences on consumers. On the one hand the effect of Gentrification (or a steady and large increase in house prices) should have a positive wealth effect on home-owners, therefore potentially improving their consumption profiles and lowering their delinquencies rates. At the same time renters might suffer from the increase in house prices as rents will respond to that. To our knowledge very little is known about the consequences of gentrification and de-gentrification in terms of geographic mobility, consumption, and delinquency as resulting from steady and large increase or fall in housing prices. We therefore aim at filling the gap in the literature by combining data on credit behaviour with data on prices and labour market in the US. Credit data will give us a unique opportunity to consider delinquency directly, and as the amount of consumption out of credit cards is now around 50% of total expenditure (much higher in large metropolitan areas) in the US, we can use credit card balances to infer a measure of total consumption through a dynamic budget constraint formulation. Moreover, the dynamics of geographic mobility across owners and renters are crucial forces for equalization of differentials across markets and deserve a central attention in policy making.A natural extension of the two areas of research above is our third line of research in this proposal. We aim at the development of predictive models of delinquency, consumption, and mobility outcomes exploiting the wealth of data acquired for this project. The aim is to build forecasting models of behaviour based on promising new approaches developed within the big (deep) data revolution. The current frontier in econometrics combines methods from statistics and machine learning and has shown great promises. Of particular interest are Deep Learning models which involve neural networks and have been shown to be extremely powerful tool for prediction and outlier detection (LeCun et. al. 2015, Schmidhuber, 2015) Furthermore, they can be used for unsupervised learning, which can provide novel insights into how high-dimensional models are to be constructed, the number of consumer types in a population or developing new approaches to delinquency (and the other aforementioned outcomes) in credit data.