A new method was developed to detect changes in how well we can predict stock market returns using certain ratios. By looking at the errors in our predictions, we found that these changes happened over time. When we applied this method to US stock data from 1927 to 2013, we saw clear patterns of instability in the predictability of stock returns, especially before the mid-50s. This means that the ability to forecast stock market movements using these ratios was not consistent over the entire period.