I do apologise for the late response.
I try to predict daily stock market movements of German DAX Naqd100 ETC, currencies, shares using the SVM. As input features in want to take the daily changes of several stock markets. The problem is that different stock markets have different holidays which results in missing data (on trading days of the DAX). I get the data from Yahoo Finance. For example the S&P500 has similar trading days as the DAX with just a few days missing (1-2 days in a row).
The trading days of SSE composite vary more significantly from the ones of the European markets, sometimes 5 days in a row are missing.
Studies use different approaches on that topic. Some take the linear interpolated data between two trading days. This seems problematic because in reality, investors do not have these informations of market trends when the SSE is closed. Other studies remove the missing data with the changes of the previous day. Here a sequence of days would have the same values which could cause difficulties.
What would you recommend to deal with this issue? Should I remove all data, where at least one stock market is missing, from the training and test set (~8% of data set) and use the changes to the previous trading day of every individual stock markets?