Score contribution per author:
α: calibrated so average coauthorship-adjusted count equals average raw count
In this paper, we propose a new method for combining forecasts based on complete subset least squares support vector regressions (LSSVRCS) that is applicable to both linear and nonlinear data generation processes. Our LSSVRCS is very flexible that it can incorporate other methods, like ridge regression or complete subset regression, as special cases. In a Monte Carlo simulation experiment, our LSSVRCS outperforms many other competing approaches. The out-of-sample performance of the LSSVRCS method is examined in an analysis for predicting Bitcoin realized volatility. The results favor our method relative to others.