Previous studies have found that time series of the financial terms have non-linearity and could be predicted. Among the previous forecasting models, we pay our attention to the nearest neighbor forecasting model. In this study, we first present the limitation of the Euclidean distance in the nearest neighbor forecasting model and propose the uncentered correlation as an alternative. This study considers not only the daily price change but also the daily volume traded which contains the important information for forecasting return.
Using KOSPI, we predict daily stock return with (i) the nearest neighbor forecasting model with Euclidean distance, (ii) the nearest neighbor forecasting model with uncentered correlation, and (iii) the multivariate (including volume) nearest neighbor forecasting model with uncentered correlation. The results of the prediction show that the uncentered correlation is more effective than the Euclidean distance and that the multivariate model including the volume traded is superior to the univariate model.