The aim of this research is to discover a reliable and scientific precursor that is theoretically able to predict all earthquakes, within specified parameters, within a few days, in this case within two days, of the earthquakes - this is more efficient than having a prediction window of months or years. This method has accurately predicted 14 out of 15 earthquakes within specified parameters of location, magnitude and depth, with no false predictions-this is a success rate of 93%. Deviations in the times for a simple pendulum to complete 30 oscillations were analysed and these deviations were used to make earthquake predictions. This is related to plate motion and changes in ‘g’. The parameters of earthquakes predicted include those that occurred in north-eastern Colombia, of magnitude M 4.0 and higher, and depth 100 km or more. Also included in the predictions are earthquakes of similar magnitude, originating at depths of 25 km and more in Antigua and Barbuda, Montserrat and Guadeloupe. Another parameter is that for an earthquake to be accurately predicted it must not occur within 24 hours of the previous one. These earthquakes were predicted within two days of their occurrences. In addition to predicting earthquakes, the data supports the rebound theory, since deviations in times for 30 oscillations return to zero within a day or two after an earthquake-this allows for calculations of rebound velocities. The data also shows that there is a net transfer of lithospheric matter away from the observation point at St. Augustine, Trinidad. This method can be applied anywhere on earth-it is simple to set up, inexpensive, scientific and reliable.