Humans have a tendency to tunnel which is also known as "herding".
If given the task to predict a certain event and give a number or date on when a certain event would happen, then most people would guess close to each other than close to the actual number. Even complicated financial models are used but the results are always herded close together. If economists are asked to predict the GDP growth rate next year, then everyone would give very similar numbers which would in general be very far from the actual number.
Herding or tunneling is specifically dangerous in extremistan where there are large deviations.
http://www.math.nyu.edu/fellows_fin_math/gatheral/extremistan.pdf
http://www.soa.org/library/newsletters/risk-management-newsletter/2010/march/jrm-2010-iss18-mills.pdf
Expert Tunneling
A second extremely dangerous idea is "expert tunnelling". Experts tend to tunnel more precisely with an even smaller standard deviation.
If a taxi driver is asked to predict tomorrow's weather than he'll predict it with a wide standard deviation. On the other hand, a weather expert would have a very small standard deviation and more confidence in his predictions. The weather expert would suffer more from unexpected and large deviations compared to a taxi driver who would not rely on his prediction and would consider worst case scenarios.
Tunneling is dangerous when you try to predict complicated phenomenon that are inherently unpredictable, or which are only predictable for some time and become unpredictable later on.
If given the task to predict a certain event and give a number or date on when a certain event would happen, then most people would guess close to each other than close to the actual number. Even complicated financial models are used but the results are always herded close together. If economists are asked to predict the GDP growth rate next year, then everyone would give very similar numbers which would in general be very far from the actual number.
Herding or tunneling is specifically dangerous in extremistan where there are large deviations.
http://www.math.nyu.edu/fellows_fin_math/gatheral/extremistan.pdf
http://www.soa.org/library/newsletters/risk-management-newsletter/2010/march/jrm-2010-iss18-mills.pdf
Expert Tunneling
A second extremely dangerous idea is "expert tunnelling". Experts tend to tunnel more precisely with an even smaller standard deviation.
If a taxi driver is asked to predict tomorrow's weather than he'll predict it with a wide standard deviation. On the other hand, a weather expert would have a very small standard deviation and more confidence in his predictions. The weather expert would suffer more from unexpected and large deviations compared to a taxi driver who would not rely on his prediction and would consider worst case scenarios.
Tunneling is dangerous when you try to predict complicated phenomenon that are inherently unpredictable, or which are only predictable for some time and become unpredictable later on.
No comments:
Post a Comment