Cambridge, Massachusetts-based parallel processing systems builder Thinking Machines Corp reckons its technology is as good at financial data analysis as it is at scientific and industrial applications, and it has already won the endorsement of one US finance company. Company scientists Dr Xiru Zhang and Jim Hutchinson, using Connection Machine supercomputers, produced the best results […]
Cambridge, Massachusetts-based parallel processing systems builder Thinking Machines Corp reckons its technology is as good at financial data analysis as it is at scientific and industrial applications, and it has already won the endorsement of one US finance company. Company scientists Dr Xiru Zhang and Jim Hutchinson, using Connection Machine supercomputers, produced the best results in the prediction categories of chaos modelling and currency exchange rates in the international Santa Fe Institute in competition with 50 other top researchers in time series forecasting. The physics problems they tackled involved 100,000 observations of an unknown system. Their task was to determine the hidden correlations within the data in order to make assumptions about the underlying model. They attributed their success to the speed with which Connection Machines run programs. As well as saving time, the ability to train a large artificial neural network in half an hour instead of a few days could even change the approach to problem-solving, they said. Portfolio management company Grantham, Mayo, Van Otterloo & Co, was sufficiently impressed with the performance to consider a possible joint research venture with Thinking Machines. Xiru and Jim’s results are the first concrete evidence I’ve seen that sophisticated mathematical models coupled with supercomputers could give us a competitive edge, the company said: The next step is to resolve real problems in the financial world with this approach.
