Wednesday, January 6, 2016
Emory Anderson Data Incubator proposal
In 2011 when I became responsible for Citigroup's small business cards portfolio, I examined the losses witnessed by this portfolio during the 2008 financial crisis. My initial response was to direct my team towards building a far more robust set of analytic acquisitions policies that incorporated previously unused commercial variables.
Another question that I began to examine was whether it was possible to predict the onset of an entirely new loss regime in a portfolio. In the case of the small business cards portfolio, it was observed that, during 2008/2009, the usual controls leveraged to reign in mounting losses were insufficient and, even, irrelevant to controlling losses that were following a previously unwitnessed trajectory. The work I will do at Data Incubator will attempt to identify leading indicators of a pending financial crisis, which at the highest level I define as now longer obeying historic statistical patterns.
In this first image I've superimposed actual small business loss data over that same data windowed via a Hanning (to suppress sidelobe artifacts from the FFT):
In this image I've taken the FFT and then linearized the plot. As can be seen, there is evidence for the existence of nonlinear 1/f noise, along with a "dragon king" event, corresponding of course to the 2008 financial crisis.
For my DI project I plan on pulling in publically available loss data and allowing a user to move a window to dynamically see the 1/f noise and the possible onslaught of a financial crisis impacting the portfolio (which may or may not correspond to a larger global financial crisis).
Another question that I began to examine was whether it was possible to predict the onset of an entirely new loss regime in a portfolio. In the case of the small business cards portfolio, it was observed that, during 2008/2009, the usual controls leveraged to reign in mounting losses were insufficient and, even, irrelevant to controlling losses that were following a previously unwitnessed trajectory. The work I will do at Data Incubator will attempt to identify leading indicators of a pending financial crisis, which at the highest level I define as now longer obeying historic statistical patterns.
In this first image I've superimposed actual small business loss data over that same data windowed via a Hanning (to suppress sidelobe artifacts from the FFT):
In this image I've taken the FFT and then linearized the plot. As can be seen, there is evidence for the existence of nonlinear 1/f noise, along with a "dragon king" event, corresponding of course to the 2008 financial crisis.
Subscribe to:
Posts (Atom)