About a correlation between currency exchange rate and astronomical parameters of the Solar system celestial bodies
November 1st, 2006©Alexander Trunev (Toronto, Canada)
©Victor Okhonin (Toronto, Canada)
A currency exchange rate simulation is one of the popular problems of financial astrology. There are several approaches to such a problem. The most substantial results we can reach by using intellectual systems based on neural network computer applications. We have analysed an exchange rate dependence among 20 countries (see: Table1) according to astronomical parameters of Solar system’s celestial bodies: The Sun, Moon, Mercury, Venus, Mars, Jupiter, Chiron, Saturn, Uranus, Neptune and Pluto in period from January 1, 2000 up to June 15, 2006. For the simulation we have constructed a neural network which was able to compare a relative contribution of input parameters. Among the input parameters there were following:
- Astronomical parameters of celestial bodies;
- Newton’s linear time (in seconds);
- Calendar time: year, month, day of a month and day of a week.
Among astronomical parameters were used longitude sine and longitude cosine, latitude and a distance between the Earth and every celestial body. The database was prepared as a matrix where there were exchange-rate rows as well as astronomical parameters, linear time and calendar time. It has been used back-propagation neural network, with 60 outputs, 189 input, and 14940 non-linear neurones. As algorithm of optimization it was used conjugate gradients method, modified for elimination of effect of conversion training. On an output neural network reproduced daily trends on 20 hard currencies, for three next days, on an input there were daily trends on the same currencies for seven last days, and the block of 49 time parameters, including time under the physical standard, calendar day, day of week, month and year, and actually astronomical 44 parameters. Other variants of neural networks were approved also. All entrance and target parameters were preliminary shifted on zero average on sample and normalized on an individual root-mean-square deviation on sample that excluded dependence of neural network parameters from scale of a signal.