The researchers developed a new method to predict stock market volatility by adjusting their model based on how accurate their previous predictions were. They tested this method on data from 40 stocks in the S&P 500 and found that using log variances gave the most accurate forecasts compared to other methods.