Is it ok to estimate with a model that has insignificant variables?

[ruixue-li]

In PS4 part (f), we were asked to predict a value, but some variables in that model were insignificant. I'm wondering is it ok to make predictions with such a model?
Thanks!

[khan1792]

It is not a problem if insignificant variables can improve the model fit or predictive accuracy. Insignificance is often relatively tricky since it only means "we cannot reject null hypothesis" rather than "accepting null hypothesis" (especially in datasets with small sample). Parameter's p-value is a widely used but actually problematic indicator, especially in predictive model. Normally, sample size, distribution, variance of parameter, bias, theory and many other factors should be comprehensively considered before we decide which variable should be used.