The 2008 Financial Crisis, Exploratory Analysis

The financial crisis beginning in late 2008 saw the sharpest declines beginning in early October. While the GCP formal analysis protocols don't lend themselves to long drawn out events, we can do probes and can look at trends in an exploratory manner. We also defined one formal event to probe the meltdown period, looking at the bailout vote on Sept 29.

William Treurniet produced some long-term graphical analyses which he describes in an email:

After the action started on Monday Oct 6, there is a continuous increase in the Stouffer-Z from the 6th to the 9th, and the p-value is .008 at the end of the period. Do you have an a priori rule to look at that kind of event? If not, it's post hoc but interesting nevertheless.

Are you aware of the linguistic prediction technique developed by the folks at "Half Past Human" (http://halfpasthuman.com/)? They analyze the language used on the internet to predict when significant events are likely to occur. These include social unrest and natural disasters. For example, the language indicated that a major event in the economic and military domains was likely to occur starting in October, 2008, and would carry through until about February, 2009, I believe. The premise is that the language used well before the event reflects psychic perceptions on a global scale. So the EGG network response to the events at the beginning of October are consistent with their predictions.

See http://projectcamelot.org/clif_high_half_past_human_26_sept_2008.html for a transcript of an interview by Project Camelot with Cliff High from Half Past Human. They discuss his technique in more depth. Or go to http://projectavalon.net/half_past_human_26_sept_2008.mp3 for the audio interview.

The following figure, suggested by William's analysis, shows October 6-10, the first week of the crisis. The trend is obvious, and compounds to a probability on the order of 0.016 that it is just chance fluctuation. The trends are relatively flat both before and after this week.

This figure is more impressive when we note that the strong deviation is a mirror reflection of the decline in the Dow Jones Industrial Average during that same week. The following figure shows mid September to early November. The deepening public awareness of the financial meltdown was driven by the precipitous drop in the DJI in the week of October 6 to 10. (Arguably, the meltdown was reciprocally driven by market perceptions.) The following figure shows the daily closings and a 5-day smoothed version (red) in which that week's decline is even more clear. The segment corresponding to the previous figure is just the central marked section starting with Oct 6.

Half Past Human Forecast

And we then had data that suggested that there might be a little tiny dip from 9/27 until October 7th in terms of the amount of that building tension, but it was basically still on that same plateau.

And then on October 7th... And I chose�... well, you know, the Universe moved me to choose it, I chose ten minutes after 7 in the morning, UTC time, which makes it ten after midnight my time here on the Pacific coast on October 7th would be the point at which we would slip into what I call Release Language.

Release Language is where everybody is letting out the emotion, as opposed to letting the emotion in, in effects on their body. They just can't take it any more, and they're expressing that emotion. Good, bad, or indifferent, they're expressing it. So release has to do with expression. Building has to do with input.

And we're going to get into a period that goes from October 7 until... Now the new data run is showing it moved into about mid-March of 2009, pretty straight-forward release language that entire period, with no little stair-steps, if you will, in the building-tension language.

This remarkably specific moment is shown in the following graph as the beginning of a 10-hour period that does show a significant trend. (It may look familiar at first glance -- the graph is almost fractally similar to that for 120 hours shown above.)

Please remember that these analyses are post hoc, not formal, so interpretation in an evidentiary mode is not appropriate. It is also important to keep in mind that we have only a tiny statistical effect, so that it is always hard to distinguish signal from noise. This means that every "success" might be largely driven by chance, and every "null" might include a real signal overwhelmed by noise. In the long run, a real effect can be identified only by patiently accumulating replications of similar analyses.