0

One of my friends got this question during an interview. He asked about my opinion. Honestly I dont have any clue. The question is as follows:

Assume we fit a machine learning model to predict the stock prices of the stock XXX. The model was trained and evaluated before COVID. It was then used to predict the future prices(Before COVID). The data during COVID is pretty much new to this model. (The model hasn't seen that type of incident before). What should we do to get the accurate predictions during the COVID?

student_R123
  • 751
  • 4
  • 13
  • @StephanKolassa This talks about how to "build" models with the effect of COVID. In my case, the model was already built (before COVID). It has been used to predict future values. Does the model should use to predict the data during COVID? if so, how? – student_R123 Dec 14 '21 at 05:03
  • What lets you assume that ths problem has a solution? – cdalitz Dec 14 '21 at 06:59
  • 1
    I don't think that makes all that much of a difference. In the linked threads, there were already models up and running (and predicting), and suddenly COVID came along. So the models had to be modified, possibly retrained. Something similar would need to happen in your case. – Stephan Kolassa Dec 14 '21 at 07:50
  • One possible answer is "be humble and don't assume your models will continue to work--at all." You could follow that up with suggestions about how to capitalize on what was learned with previous models (*e.g.,* start by using features that have been useful for prediction), how to evaluate the accuracy of existing models (propose and discuss suitable loss functions, perhaps), and consider displaying some of your general knowledge by citing previous disasters, such as [Google's flu prediction model.](https://en.wikipedia.org/wiki/Google_Flu_Trends) – whuber Dec 14 '21 at 18:29

0 Answers0