The classification models aim to predict whether the monthly stock market return is positive or negative. I wonder how you modify a classification problem when some errors are more costly than others (other than just treating it as a regression problem). Predicting the stock market goes up is more costly when it falls 10% than when it falls 1%.
The classification models aim to predict whether the monthly stock market return is positive or negative. I wonder how you modify a classification problem when some errors are more costly than others (other than just treating it as a regression problem). Predicting the stock market goes up is more costly when it falls 10% than when it falls 1%.
The working paper "Predicting Stock Market Returns with Machine Learning" at https://sites.google.com/site/albertorossifrb/research attacks the same problem using machine learning regression models.